20731 40 C1 6May 2000 Volume two 0- FD esigning 0< Household Survey OQ Questionnaires for n 0 Developing Countries _ Lessons from 15 years of the Living Standards Measurement Study m ca 0. Edited by Margaret Grosh and Paul Glewwe PX c The World Bank Oxford -The World Bank Volume two T es ign ing Household Survey Questionnaires for Developing Countries Lessons from 1 5 years of the Living Standards Measurement Study Edited by Margaret Grosh and Paul Glewwe U7 The World Bank "Household surveys are essential for the analysis of most policy issues. This book has carefully assessed recent experience and developed today's best-practice technique for household surveys. Indeed, much of this technique was developed and pioneered by the authors.This book is clear, systematic, and well structured. It is also wise and scholarly. It will be indispensable to anyone involved in carrying out or analyzing household surveys, and thus it is required reading for all those who wish to take evidence seriously when they think about policy." -Nicholas Stern, senior vice president, Development Economics and chief economist, the World Bank "This book is an ambitious undertaking, but it quickly exceeded my expectations. It has many strengths: ... com- prehensiveness, ...emphasis on practical application, ...and a sense of balance. For both my domestic and interna- tional survey research, this volume will serve as a valued reference tool that I will consult regularly." -David R.Williams, professor of sociology and senior research scientist, Survey Research Center, University of Michigan "This is a comprehensive guide to planning household surveys on a range of socioeconomic topics in develop- ing countries. It is authoritative, clear, and balanced. The work is a valuable addition to the library of any survey statistician or data analyst concerned with socioeconomic surveys in the developing world." -William Seltzer, former head, United Nations Statistical Office Household survey data are essential for assessing the impact of development policy on the lives of the poor.Yet for many countries household survey data are incomplete, unreliable, or out of date. This handbook is a compre- hensive treatise on the design of multitopic household surveys in developing countries. It draws on 15 years of experience from the World Bank's Living Standards Measurement Study surveys and other household surveys conducted in developing countries. The handbook covers key topics in the design of household surveys, with many suggestions for customizing sur- veys to local circumstances and improving data quality. Detailed draft questionnaires are provided in written and electronic format to help users customize surveys. This handbook serves several audiences: * Survey planners from national statistical and planning agencies, universities, think tanks, consulting firms and international organizations. * Those working on either multitopic or topic-specific surveys. * Data users, who will benefit from understanding the challenges, choices, and tradeoffs involved in data collection. Volume two tD esigning Household Survey Questionnaires for Developing Countries Lessons from 15 years of the Living Standards Measurement Study Edited by Margaret Grosh and Paul Glewwe Copyright (© 2000 The International Bank for Reconstruction and Development/THE WORLD BANK 1818 H Street, N.W Washington, D.C. 20433, U.S.A. All rights reserved Manufactured in the United States of America First printing May 2000 The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they repre- sent. The World Bank does not guarantee the accuracy of the data included in this publication and accepts no responsibility for any conse- quence of their use.The boundaries, colors, denominations, and other information shown on any map in this volume do not imply on the part of the World Bank Group any judgment on the legal status of any territory or the endorsement or acceptance of such boundaries. The material in this pubhcation is copyrighted. The World Bank encourages dissemination of its xvork and will normally grant per- mission promptly. Permission to photocopy items for internal or personal use, for the internal or personal use of specific clients, or for educational class- room use, is granted by the World Bank, provided that the appropriate fee is paid direcdy to Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, U.S.A., telephone 978-750-8400, fax 978-750-4470. Please contact the Copyright Clearance Center before photocopying items. For permission to reprint individual articles or chapters, please fax your request with complete information to the Republication Department, Copyright Clearance Center, fax 978-750-4470. All other queries on rights and licenses should be addressed to the World Bank at the address above or faxed to 202-522-2422. ISBN:0-19-521595-8 Library of Congress Cataloging-in-Publication Data has been applied for. Contents Foreword ix Acknowledgments xi Contributors xiii Volume I Part I Survey Design 1. Introduction 5 Margaret Grosh and Paul Glewwe 2. Making Decisions on the Overall Design of the Survey 21 Margaret Grosh and Paul Glewwe 3. Designing Modules and AssemblingThem into Survey Questionnaires 43 Margaret Grosh, Paul Glewwe, andJuan Muiioz Part 2 Core Modules 4. Metadata-Information about Each Interview and Questionnaire 77 Margaret Grosh andJuan Munoz 5. Consumption 91 Angus Deaton and Margaret Grosh 6. Household Roster 135 Paul Glewwe 7. Education 143 Paul Glewwe 8. Health 177 Paul J. Gertler, Elaina Rose, and Paul Glewwe 9. Employment 217 Julie Anderson Schaffner 10. Anthropometry 251 Harold Alderman v CONTENTS 11. Transfers and Other Nonlabor Income 273 Andrew McKay 12. Housing 293 Stephen Malpezzi 13. Community and Price Data 315 Elizabeth Frankenberg Volume 2 Part 3 Additional Modules 14. Environmental Issues 5 Dale Whittington IS. Fertility 31 lndu Bhushan and Raylynn Oliver 16. Migration 49 Robert E. B. Lucas 17. Should the Survey Measure Total Household Income? 83 Andrew McKay 18. Household Enterprises 105 Wim P M. Vijverberg and Donald C. Mead 19. Agriculture 139 Thomas Reardon and Paul Glewwe 20. Savings 183 Anjini Kochor 21. Credit 211 Kinnon Scott 22. Time Use 249 Andrew S. Harvey and Maria Elena Taylor Part 4 Special Topics 23. Recommendations for Collecting Panel Data 275 Paul Glewwe and Hanan Jacoby 24. Intrahousehold Analysis 3 15 Nobuhiko Fuwa, Shahidur R. Khandker, Andrew D. Mason, and Tara Vishwanath 25. Qualitative Data Collection Techniques 337 Kimberly Chung 26. Basic Economic Models and Econometric Tools 365 Jere R. Behrman and Raylynn Oliver Volume 3 Draft Questionnaire Modules Introduction I Module for Chapter 4: Metadata 5 Margaret Grosh andJuan Munoz Module for Chapter 5: Consumption 15 Angus Deaton and Margaret Grosh vi CONTENTS Module for Chapter 6: Household Roster 31 Paul Glewwe Module for Chapter 7: Education 37 Paul Glewwe Module for Chapter 8: Health 73 Paul J. Gertler, Elaina Rose, and Paul Glewwe Module for Chapter 9: Employment 147 Julie Anderson Schaffner Module for Chapter I0:Anthropometry 219 Harold Alderman Module for Chapter I I:Transfers and Other Nonlabor Income 221 Andrew McKay Module for Chapter 12: Housing 229 Stephen Malpezzi Module for Chapter 13: Community Data 247 Elizabeth Frankenberg Module for Chapter l4: Environment 285 Dale Whittington Module for Chapter 15: Fertility 325 lndu Bhushan and Raylynn Oliver Module for Chapter 16: Migration 333 Robert E B. Lucas Module for Chapter 18: Household Enterprise 349 Wim P. M. Vijverberg and Donald C. Mead Module for Chapter 19:Agriculture 407 Thomas Reardon and Paul Glewwe Module for Chapter 20: Savings 453 Anjini Kochar Module for Chapter 21: Credit 461 Kinnon Scott Module for Chapter 22:Time Use 483 Andrew S. Harvey and Maria Elena Taylor Module for Chapter 23: Panel Data 495 Paul Glewwe and Hanan Jacoby vii Foreword Multitopic household surveys have become an indis- The household surveys treated in this book truly pensable instrument for understanding development. are multitopic surveys, covering such topics as house- They are fundamental to serious microeconomic analy- hold size and composition, education, health, anthro- sis of the incentive and distributional aspects of policy, pometry, fertility, income and consumption, employ- and therefore to the analysis of most policy issues. ment, agricultural production, household enterprises, Researchers draw on them to test behavioral theories. transfers and nonlabor income, savings and credit, Policymakers need them to assess public interventions. housing, the environment, migration, and time use. The development community uses them to locate the The editors have greatly increased the value of the poor. Developing counties, without adequate house- basic approach by incorporating chapters on commu- hold survey data, are forced to make policy decisions in nity data, panel data, and the allocation of resources an environment with many blind spots, where crucial within the household. information can be seen only dimly or not at all. As the World Bank and other development Household surveys are also expensive, both in organizations increase their efforts to reduce poverty terms of money and institutional capacity. Ultimately and raise living standards in developing countries in their value depends on their design and execution. the 21st century, the need for comprehensive, reliable Errors in their design or execution are wasteful, and and up-to-date information on economic and social can lead to policies that are harmful to the poor. It is conditions in these countries will be greater than therefore important to design and implement surveys ever. The vast store of knowledge in this book will correctly from the outset. contribute significantly to meeting this need. Failure Margaret Grosh and Paul Glewwe have put to use this knowledge will consign policymakers to together one of the most comprehensive and inform- making their decisions without adequate information ative documents ever written on the design, imple- for many years to come, while systematic use of this mentation, and use of household surveys in develop- knowledge will do much more for the poor than the ing countries. If you are engaged in any of these tasks, innumerable speeches made and summits convened this book is essential reading. on their behalf. Lyn Squire Director, Global Development Network World Bank ix Acknowledgments A project of this size and scope depends on many peo- the surveys, the many agencies that provided techni- ple playing many roles. Space limitations preclude us cal assistance and funding, and the academic partici- from naming all of the hundreds of people who made pants who provided advice and criticism over the contributions along the way, but we would like to years. acknowledge some of the most important. The project as a whole was strongly supported The authors of the individual chapters deserve from original vision to final printing by our immedi- thanks for their gracious willingness to go through ate manager for most of that time, EmmanuelJimenez, many rounds of revisions, spread over a longer time who provided us with useful technical input and a than anyone originally envisioned. Producing a book great deal of enthusiasm, patience, and bureaucratic on the design of multitopic questionnaires requires support. We also greatly appreciate the support of his much more cooperation among authors and several directors, Lyn Squire and Paul Collier.The project was more iterations than does the standard edited volume. primarily financed by a grant from the World Bank We are extremely grateful for the forbearance of the Research Committee (679-61), managed by Greg authors in this difficult process.The authors themselves Ingram and administered by Clara Else. were helped by a large number of peer reviewers.They Many people reviewed the book and project as a are recognized in the individual chapters, but we whole. We gready appreciate these contributions by would like to extend our thanks to them here as well. Pat Anderson, Jere Behrman, Elisa Lustosa Caillaux, Much of the work in these volumes was based on Courtney Harold, John Hoddinott, Anna Ivanova, past practice in LSMS and other household surveys Alberto Martini, Raylynn Oliver, Prem Sangraula, including, but not limited to, the World Fertility/ Salman Zaidi, and three anonymous reviewers.To have Demographic and Health Surveys, the RAND Family input on the project as a whole from these outsiders Life surveys, the Social Dimensions of Adjustment was very helpful. In addition, participants at three surveys, and several special topic surveys such as workshops held at the World Bank, plus various train- household budget surveys, water and sanitation sur- ing events sponsored jointly by the World Bank and veys, housing surveys, and time use surveys. While the the Inter-American Development Bank, critiqued the authors pulled together the lessons from past experi- project while it was in progress. ence, it is also important to acknowledge the irre- In the course of creating the book, Diane Steele placeable contributions made by the thousands and answered questions from all authors on the details of thousands of household members who served as LSMS data sets. Fiona Mackintosh edited early drafts and respondents, the dozens of agencies that implemented helped to transform the disparate chapters into a single xi ACKNOWLEDGMENTS whole. Lyn Tsoflias provided us with valuable research Communications Development Inc. Communication assistance.Word processing and conference logistics were with the World Bank's Publications Comuittee and ably handied by Thomas Hastings, Patricia Sader, Jim with the publishers and printers was efficiently handled Schafer, and Daniel O'Connell. Questionnaire layout by Paola Scalabrin and Randi Park. was mastered by Thomas Hastings, Andrea Ramirez, and Finally, effusive and endless thanks to our families Heidi Van Schooten. Contracting support from Liliana and friends who put up with the excessively long hours Longo, Selina Khan, and Patricia Sader was timely and that we spent on this project, who cheered and calmed organized. The final editing, layout, and design were us through the frustrating times, and who helped us to handled dextrously by Meta de Coquereaumont,Wendy bring this long project to a successful conclusion with- Guyette, Kate Hull, Daphne Levitas, Heidi Manley, out completely losing track of other important aspects Laurel Morais, and Derek Thurber, all with of our personal and professional lives. xii Contributors Harold Alderman World Bank Jere R. Behrman University of Pennsylvania Indu Bhushan Asian Development Bank Kimberly Chung Michigan State University Angus Deaton Princeton University Elizabeth Frankenberg RAND Nobuhiko Fuwa World Bank Paul J. Gertler University of California, Berkeley Paul Glewwe World Bank and University of Minnesota Margaret Grosh World Bank Andrew S. Harvey St. Mary's University, Halifax, N.S. Canada Hanan Jacoby World Bank Shahidur R. Khandker World Bank Anjini Kochar Stanford University Robert E.B. Lucas Boston University Stephen Malpezzi University ofWisconsin Andrew D. Mason World Bank Andrew McKay University of Nottingham, United Kingdom Donald C. Mead Michigan State University Juan Mufioz Sistemas Integrales Raylynn Oliver Consultant,World Bank Thomas Reardon Michigan State University Elaina Rose University ofWashington Julie Anderson Schaffner Fletcher School of Law and Diplomacy,Tufts University Kinnon Scott World Bank Maria Elena Taylor St. Mary's University, Halifax, N.S. Canada Wim P.M.Vijverberg University of Texas at Dallas TaraVishwanath World Bank Dale Whittington University of North Carolina at Chapel Hill xiii Volume 2 Part 3 Additional Modules 4< j Environmental Issues 1t 4 Dale Whittington The purpose of this chapter is to examine what information is needed and can realistically be collected in a household survey (such as an LSMS survey) to support sound environmental policy analysis. Previous LSMS surveys have contained few questions specifically designed to gather household information on environmental issues and concerns, so this area is something of a clean slate. The challenge is to establish what data are most needed to allow analysts to conduct envi- ronmental policy analysis, then to establish what are the best questions to include in LSMS-type surveys to gather these data. The job of designing environmental questions to be physical and biological systems function and how included in LSMS and similar surveys is different in human interventions can alter natural and manmade two important ways from the task of designing (and processes.This understanding can be gained both from redesigning) questions on other topics. First, the num- creative theorizing and the testing of theories with ber of important policy issues that fall under the rigorous experiments and from personal experience. umbrella of "environment and natural resources poli- Survey designers should carefully consider what infor- cy" is very large. They include issues as diverse as air mation about environmental issues households might pollution, contamination of drinking water, soil degra- know that would be of policy relevance. Without the dation, rainforest loss, national parks, watershed man- resources or training to carry out scientific experi- agement, fisheries, wildlife conservation, urban sanita- ments, ordinary people may not be well informed tion, acid rain, ozone depletion, global warming, about the causes or consequences of some environ- hazardous waste disposal, misuse of pesticides, mineral mental problems, which means that some of their depletion, and malaria control-and this is by no actions may have unintended and unanticipated means an exhaustive list. No single household survey results. Thus household surveys are not the most can query respondents in depth about their knowl- appropriate way to collect the information required to edge of, attitudes toward, and practices regarding all of develop a rigorous understanding of the physical and these environmental issues or even a substantial num- biological aspects of environmental systems. However, ber of these issues. Thus the designers of LSMS-type they are generally considered a good and appropriate surveys must set priorities and ask questions about means of collecting data for the investigation of only a limited number of environmental issues. human behavior. Second, analyzing environmental policies typical- Because the level of households' scientific under- ly requires a thorough scientific understanding of how standing of environmental issues may vary greatly 5 DALE WHITTINGTON from location to location, it may be appropriate to ask capita well-being requires that environmental some questions in one location but not in another. resources-air, water, and land, and the life that Thus the approach taken by this chapter is to intro- depends on them for existence-be managed and duce a set of environment modules that illustrate the conserved in wise ways. This means that governments types and range of questions that can be included in an and individuals must think carefully about the most LSMS-type survey. These modules, presented in appropriate use of the capital stock of both nonre- Volume 3, cover important environmental issues in newable and renewable resources over time, and that urban and rural areas.The modules also collect data on policies, projects, and regulations must be chosen and households' attitudes toward the environment and per- implemented in ways that produce the desired envi- ceptions of urban air quality and on their use of water ronmental, economic, and social results (Pearce and services, sanitation services, and fuel. In addition, the others 1994, chapter 1). modules cover the willingness of households to pay for The sound selection and implementation of envi- improved water services in both urban and rural areas, ronmental policies requires that analysts design various for improved sanitation in urban areas, and for the policy alternatives and then work out how these alter- improvements in urban air quality that would result in natives will affect various groups of people (or other better health outcomes.And the environment modules affected parties) in terms of several important criteria. are designed to collect the data needed to estimate Table 14.1 presents a simple classification of issues household's rates of time preference-in other words, in terms of the spatial range of the problem (which the relative value that a household places on costs and can be anything from very local to global) and the benefits at different times. Designers of future LSMS- extent to which the resource in question is renewable type surveys can choose the submodules most relevant (like fish or forests) or nonrenewable (like oil or cop- to the circumstances in the country of the survey, then per). Renewability here refers only to the physical modify and customize them as needed to match the characteristics of the resource; all renewable resources policy goals of the survey. are ultimately exhaustible in the sense that they can be The first section of this chapter summarizes some overused and thereby extinguished. Determining the of the most important environmental policy issues and extent of the impact of the problem (whether local, the tasks involved in environmental policy analysis. regional, or global) depends not only on the location The second section discusses what data are needed to of the resource but also on its significance. For exam- analyze these policy issues and what kinds of data ple, a tropical forest can be considered a global LSMS and similar surveys are best suited to collect. resource if its value is global. The third section introduces the draft environment Households naturally know more about some modules (provided in Volume 3), and the fourth sec- kinds of environmental problems than about others. tion both explains why these modules were crafted the LSMS-type surveys should attempt to collect informa- way they were and points out important issues that are tion on the important environmental problems about likely to arise when these modules are administered. which respondents are best informed. In general, respondents are likely to be most knowledgeable about Environmental Policy Issues the damage that they suffer from the degradation of air, water, and land resources, as well as about the local, The basic message of the 1992 United Nations renewable resources that they use and depend upon for Conference on Environment and Development was their livelihood and sustenance-such as local forests, that the environment is an important factor in suc- fisheries, and groundwater (see the upper left cell in cessful economic and social development. Raising per Table 14.1). Local communities often have extensive Table 14.1 The Classification of Environmental Policy Issues Local Reg onal Global Renewable resources Local water sources and fisheries: Shared water basin systems; Ozone layer; major tropical forests; local forests; groundwater systems; airsheds genetic information fuelwood Nonrenewable resources Mineral deposits; oil, coal Shared natural gas reservoirs Source: Author's examples. 6 CHAPrER 14 ENVIRONMENTAL ISSUES experience in dealing with local pollution problems ownership of natural resources do not accrue to the and managing renewable resources. Such communities owner, because owners cannot transfer property rights may be equally experienced in dealing with local, non- to another owner in a voluntary exchange, or because renewable resources, but it is more likely either that the there are no penalties to prevent others from encroach- local population does not use these kinds of resources ing on or taking over an owner's property rights.1 or that these resources are owned or controlled by only In many places the rights to land or to a natural a small number of individuals. In either case local peo- resource (for example, the atmosphere and much of ple would have httle knowledge of these resources to the world's oceans) are not allocated or defined at all. share with the survey interviewers. Even where property rights exist, they may not be As the spatial scale of the environmental manage- enforceable so that in reality an open access situation ment problem increases, ordinary people are less like- exists-creating a risk that the resource may be over- ly to understand how the full ecological system oper- exploited. Households may be the best source of ates. This does not mean that households fail to information on the prevailing property rights to local understand the costs they bear from the degradation of renewable resources. such environmental resources as air and water. For example, upstream users of a river may not fully appre- Environmental Assets as Economic Inputs ciate the costs that they impose on downstream users Another insight afforded by environmental economics due to the flooding of the river basin that causes is that environmental assets are capital just as much as deforestation and soil erosion. However, the down- are machines, roads, or factories. Information from stream users are likely to fully understand the damages household surveys can be quite useful for establishing imposed upon them. It is even more likely that people the economic value of environmental assets such as will not fully understand how their actions may exac- drinking water supplies and forestry products. Because erbate global environment problems such as global capital stocks are needed to sustain development, the warming or the depletion of the ozone layer. Thus the overall maintenance of such stocks is an integral part of answers that respondents give to questions about such a sound environmental and economic policy. A tropical global issues are hkely to be less informed and less use- forest has a great many ecological functions, a large ful to analysts than their responses to questions about number of which have economic value. For example, environmental problems that are closer to home. the forest protects the watershed system; if the forest is removed, the watershed system will be damaged (to Property Rights and the Causes of Environmental varying degrees depending on what land-use system Degradation replaces the forest). At the global level the ozone layer Environmental economists have developed a powerful protects humans from unhealthy amounts of ultraviolet conceptual tool for identifying the causes of environ- radiation, so damage to the ozone layer is dangerous to mental degradation, with important implications for human health and productivity. Just as sound develop- the design of the environment modules of household ment planning requires an understanding of the value surveys. This tool is the analysis of property rights. of traditional capital assets, such planning also requires One potential role for the environment modules is to information on the value of environmental assets. collect information that will enable analysts to deter- mine the property rights regimes that govern impor- Taking Account of the Environment When Appraising tant natural resources and find out whether conditions Development Projects exist for the efficient allocation of property rights. The environment provides many inputs to economic A property right is an entitlement on the part of activities, and the residuals from production processes an owner to a resource or good that can be socially can affect the environment. In order to carry out enforced. In most countries an entitlement to use or sound project appraisals, analysts need to value both consume a natural resource is qualified by various legal these inputs and residuals properly and price them and customary restrictions embodied in law or in the accordingly. For example, if energy prices fail to reflect prevailing moral code. In many situations environmen- the pollution damage associated with energy produc- tal problems arise because property rights are not clear- tion and use, the goods produced using that energy ly assigned, because the benefits and costs arising from will be underpriced. 7 DALE WHITTINGTON When appraising projects, analysts must bear in state of affairs about which the public or technical mind that the prices of environmental inputs affect experts may be concerned. For example, the house- project returns.The proper pricing of inputs and out- hold survey might yield statistics on fuelwood usage puts can be seen as a way of designating property that indicates that local forestry stocks are rapidly rights or extending existing property rights, as can being depleted. The data can help decisionmakers regulatory measures such as environmental quality answer two basic questions: what is the condition that standards. Sound pricing policy requires that policy- is causing the concern and how important is it? makers know the value of environmental goods and Tabulating household survey data can show services. If the policymakers fail to include nonmar- whether or not a problem condition exists. For exam- keted environmental services in their calculation of ple, household surveys often collect information on that value, the opportunity costs of development proj- whether or not a household has access to an improved ects will be wrongly defined and measured. For exam- water source such as a private connection or a water ple, if a rural development project involves destroying pump. If members of the household have access to an a tropical forest, it is necessary to factor the forgone improved drinking water source, they are said to be benefits of the tropical forest into the calculation of "covered." Simple "coverage" statistics from household the project's economic rate of return. survey data can be tabulated to calculate the percent- In the past the techniques used to appraise devel- age of a population that has access to an improved opment projects have been criticized because of their drinking water system.2 If a significant percentage of failure to account for environmental values in the the population is not covered, the analysis will have same terms as the costs and benefits of conventional revealed a problem condition to which policymakers development. This can make the costs and benefits of should pay attention. Most environmental policy ana- conventional development seem more concrete than lysts in both industrialized and developing countries environmental gains and losses. What is quantified use household survey data in this way. appears more important than what is not quantified, Many of the questions in non-LSMS household and development costs and benefits are traditionally surveys addressing environmental issues have asked measured in monetary terms while environmental respondents about their household's use of an envi- costs and benefits have not often been valued in such ronmental resource or the damage that members of terms. the household may sustain as a result of various kinds Project appraisal techniques have also been criti- of degradation of the local environment. Analysts can cized for discriminating against the interests of future then use these data to draw inferences about problem generations by discounting the future costs and bene- conditions. However, survey planners should word fits of environment degradation or conservation. This questions on the household's use of resources so that is also a failure of valuation because it implies that responses can be used for more analytical tasks than conventional project appraisal techniques fail to ade- simply describing a problem. quately account for future values. In addition to providing information on resource use, household surveys can collect information on Specific Tasks in Environmental Policy Analysis households' knowledge of, attitudes toward, and prac- Data from household surveys can help analysts per- tices regarding a wide range of environmental condi- form three of the tasks associated with good environ- tions and natural resources. This information can also mental policy analysis: help analysts describe and diagnose problems. In many * Diagnosing and measuring environmental problems. cases very simple questions about the ownership of a * Investigating the causes of environmental problems. natural resource can provide important information to * Designing policy alternatives and evaluating the analysts that helps them explore ways of managing impact that these policies may have on the problems. natural resource problems. (This is in part because This section discusses these three tasks in turn. households' willingness to pay for capital improve- ments often depends upon the households' percep- DIAGNOSING AND MEASURING ENVIRONMENTAL tions of who holds the property rights for the facili- PROBLEMS. Data from household surveys can help ana- ties.) For example, the policy of the government of lysts explore an environmental "problem condition," a Tanzania is to transfer the ownership of most village- 8 CHAPFER 14 ENVIRONMENTAL ISSUES level capital facilities (like water supply systems) back Table 14.2 Respondents' Perceptions of Who Owns the to communities.The government conducted a house- Local Water Source (Tanzania, 1995) hold consultation survey in August 1995, which Perceived holder revealed that households were aware of the govern- of property right Percentage of respondents ment's policy. The respondents appeared to have Central government 3 accepted the notion that their water source was owned District government or council 6 accepted the nohon that thelr water source was owned .........................................l....o......w.........d.. .........o....... ..n.t.co....m. ......t.t.ee........ ............................ Village council or ward development committee 58 by the community; only 9 percent believed that it was ...................................... Those who live near the water source 7 owned by the central or district government (Table Those who use the water source 1 8 14.2). Interestingly, however, in some communities, a An ind.ivi.dual 4 few respondents seemed not to have noticed that the Source: Human Resources Deveopment Pilot Project I,World Bank AF2PH. property rights to water sources had been relinquished Systematic Client Consultation, August 1995. to the state in the first place. supply, are of concern because they are valued or dis- INVESTIGATING THE CAUSES OF THE PROBLEM. valued. Surveys can help understand what is valued Investigating the causes of environmental problems and track changes in those conditions. requires asking a more complex series of questions For example, in a recent survey in Marracuene, than the ones needed to collect descriptive informa- Mozambique (Pinheiro and Whittington 1995), an tion. For example, questions about water use in almost analysis of respondents' social and environmental pri- all household surveys ask respondents what water orities suggested that people were most concerned source their households use. This may be adequate for about the lack of drugs and supplies delivered to (pur- describing a problem, but it is not sufficient for inves- chased by) the hospital. This tells analysts what is val- tigating why a household chooses a particular water ued and suggests possible conditions to track. One way source for a specific purpose. To do this, analysts must to measure impact would be to track drug supplies in be able to model the choices that households make hospitals, which would require data from the hospitals. among existing water sources, for which they need Another possible method would be to measure the information on all of the water sources to which a change in the percentage of respondents that still household has access, not just the source(s) that they reported the supply of drugs was their top priority have chosen. after a policy intervention designed to address this When government policies are the cause of envi- problem was implemented; this method would require ronmental problems, household surveys may not seem household data. to be a particularly useful research tool because Household surveys can collect information that respondents might be reluctant to criticize govern- can be used to assess which among the proposed poli- ment policy. However, asking respondents simple cy alternatives are best. The surveys can support a vari- questions about their preferred way to receive a serv- ety of analytic approaches. For example, analysts trained ice or pay for a service may reveal some underlying in disciplines other than economics, as well as members problems. For example, respondents can be asked to of the public, often define criteria for policy evaluation say whether they would prefer a new community in terms of specific problem descriptions and these water supply system to be managed by the ministry of individuals' understanding of the causal relationships water, a community-based utility, or a private contrac- that relate policy alternatives to outcomes. Such crite- tor. If no one chooses the ministry of water, this may ria tend to be less abstract, more explicit, and, for some reveal past failures on the part of government people, easier to understand than economic criteria. providers. They may include indicators of public health, exposure to pollutants, respondents' subjective perceptions of DESIGNING POLICIES AND EVALUATING THEIR IMPACTS environmental amenities and conditions, and physical ON ENVIRONMENTAL PROBLEMS. Decisionmakers must indicators of access to services such as water supply and choose which among proposed policies are better than sanitation. Surveys can provide noneconomists and others. The criteria on which a choice is judged are members of the public with this kind of information- the valued (or disvalued) features of these policies.The information that these individuals find useful in policy conditions measured, such as coverage of the water discourse about policy alternatives. 9 DALE WHITTINGTON Surveys can also provide data that economists vidual may consider many factors including the size need for a formal economic appraisal of the costs and and age of the house, its proximity to schools, shop- benefits of different policy alternatives. Economists ping, and place of employment and possibly also the typically propose that an all-encompassing criterion of air quality in the neighborhood. The value of increased human well-being (or preference satisfac- improved air quality can be estimated by doing a care- tion) be used to evaluate different policy alternatives. ful analysis of such transactions in the housing market. The valuation task is to determine how much better This surrogate market method is also known as the or worse off individuals are (or would be) as a result of "hedonic property value model." Other surrogate a change in environmental quality or public health market methods are the travel cost model, the hedonic conditions. There are four approaches analysts can use wage model, and avertive behavior technique. to estimate economic values of environmental quality All of these surrogate market methods rely on the changes; all four require data from household surveys. behavioral trail left by individuals as they make deci- sions that affect their lives, since people are assumed to THE CONTINGENT VALUATION APPROACH. This reveal their preferences through their behavior. The approach involves simply asking people how much estimates of the value of nonmarket goods are based they would be willing to pay to ensure that a particu- on information about what people actually did and on lar improvement in environmental quality is made. a set of assumptions about why they did what they This is known as the "stated preferences" or "contin- did, rather than on what the people said they would gent valuation" method.3 It has also been called the do under a set of hypothetical conditions. "direct approach" because people are directly asked to This second approach has its own disadvantages state or reveal their preferences. This approach meas- and limitations. For example, it is not feasible to use ures precisely what the analyst wants to know-the surrogate market methods to estimate the value of a strength of an individual's preference for the proposed new good or service or of a change in environmental change-and could be used not just for nonmarket quality that has not yet been experienced. This is goods and services but also for market goods. This because there are no previous instances in which peo- approach would be ideal if analysts could be certain ple have been offered a new level of environmental that the respondents would fully understand the quality and revealed their preferences for it. Even when change in environmental quality being offered and a nonmarket good or service has been available for that they would answer the questions truthfully. some time, there may not yet have been any significant However, analysts worry whether the intentions that changes in its quality, making it impossible to infer how respondents state in contingent valuation surveys people in an area would value a change in environ- before a change will accurately describe their actual mental quality. Finally, to implement any of the surro- behavior after the change when they face no penalty gate market methods, analysts must interpret and value or cost associated with a discrepancy between the two individuals' decisions within a theoretical framework. behaviors. As a result, their estimates of value will depend upon a series of assumptions that remain largely untested. THE SURROGATE MARKET APPROACH. Because of these Information from household surveys is essential concerns about the "contingent valuation" method, for the implementation of surrogate market approach- economists have traditionally used a second approach es. Data must be collected on respondents' behavior for measuring the value of nonmarket goods and serv- and the determinants of their choices. For example, ices such as those provided by many environmental using the hedonic property value model requires data assets (for example, air quality). This is known as the on the following variables: respondents' rent or the surrogate market approach. In this technique, econo- value of their houses, the characteristics of respon- mists try to find a good or service that is sold in mar- dents' housing units (such as square footage, lot size, kets and is related to the nonmarket service. Thus the quality of construction, and access to infrastructure), individual reveals his or her preferences regarding both and neighborhood characteristics (such as crime rate, the market and nonmarket service when he or she quality of local schools, distance of respondents' purchases the market good. For example, when decid- dwellings from public facilities and environmental ing what house to buy or apartment to rent, an indi- amenities, and local air quality). I0 CHAPTER 14 ENVIRONMENTAL ISSUES THE DAMAGE FUNCTION APPROACH. In order to esti- estimates of value for the same or similar goods or mate the value of changes in environmental quality services that have already been derived in other loca- that reduce individuals' well-being, an analyst can tions, then transfer these estimates into their analysis, attempt to determine the damages an individual will possibly after adjusting them to match the prevailing suffer. A deterioration in environmental quality can circumstances in the location in question. Analysts can cause a loss of productive assets or a loss of earning transfer estimates of value that were made using any of power.An individual can be restored to his initial state the approaches described above. This approach is of well-being by being compensated in either money termed "benefit transfer." The data from an LSMS- or other goods or services, by the amount of the loss. type survey can be used in a benefit transfer calcula- This is termed the "damage function" approach. The tion, but such data are not sufficient in themselves to damage function and surrogate market techniques are conduct such calculations, and the benefit transfer both termed "indirect" valuation approaches because approach is not discussed further in this chapter. neither relies on people's direct answers to questions about how much they would be willing to pay (or Data Requirements for Environmental Policy accept) for a change in environmental quality. Analysis In the damage fuinction approach analysts seek to determine the monetary losses that an individual Household surveys cannot gather all of the informa- incurs in terms of both expenditures made to cope tion that may potentially be relevant for environmen- with the damage and residual losses after such expen- tal policy analysis.While open-ended questions can be ditures have been made. Information from household very useful for identifying environmental priorities surveys is often necessary but usually not sufficient for and attitudes, LSMS-type surveys are not the appro- implementing the damage function approach to valu- priate place for such questions. And because LSMS ation. Household surveys may not be able to provide samples are usually nationally representative with only all the information required because respondents may a few households in each of many different neighbor- not have sufficient knowledge about the health dam- hoods, watersheds, or ecological zones, such samples ages associated with different kinds of environmental are not appropriate for investigating local problems. It pollution. Therefore, LSMS-type surveys are not an can be difficult to design questionnaires that contain appropriate tool for collecting the data to implement all of the questions pertinent to each different setting, this valuation approach, although household data can and there are often not enough sample households in play a partial role in supporting such analysis. In addi- each area to yield reliable data. tion, a thorough implementation of the damage func- Nevertheless, LSMS and similar survey data can tion approach requires information on pain and dis- be used in the analysis of a wide variety of environ- comfort that is difficult to collect in a general mental issues.Thus it is important to have a clear sense household survey. For example, a case of malaria caus- of the priorities for data collection using LSMS-type es an individual to suffer three main types of losses: surveys. This section of the chapter presents conclu- monetary expenditure on medicine, lost wages or sions about how best to use LSMS-type surveys to reduced productivity from not being at work, and pain support environmental policy analysis. and discomfort. It is relatively straightforward to use In urban areas of developing countries, the main household surveys to collect information on monetary environmental policy questions are "What is the opti- expenditures for medicines and lost wages. However, mal (appropriate) level of environmental quality?" and asking a respondent to put a monetary value on pain "How stringent should environmental standards be?" and discomfort requires questions that are much more In most urban areas in developing countries, the most carefully designed. important environmental standards are those for the quality of air, water, and sanitation. THE BENEFIT TRANSFER APPROACH. The fourth In rural areas the most pressing environmental approach to obtaining estimates of the value of envi- issues are often in the water and sanitation sector, and ronmental goods and services takes a somewhat differ- are often location-specific.The main "green" problems ent tack. Rather than developing new estimates of the in developing countries-deforestation, soil erosion, value of environmental goods or services, analysts find and desertification-are found in rural areas. The I I DALE WHITrINGTON destruction of habitats and loss of biodiversity fre- both urban and rural areas it is recommended that pri- quently arise from deforestation, especially when ority be given to collecting information on water and farmers encroach on protected areas. Siltation of rivers sanitation practices and household fuel use. These and streams and loss of agricultural productivity often activities are relevant in almost all locations and cir- accompany soil erosion. In general, the study of these cumstances, and it is relatively easy to collect data on problems is best done with samples designed to study them in household surveys. Also, having information the local area only. Therefore it is not recommended on these activities is vital for analysts to develop a crit- that national LSMS-type surveys be used to study such ical understanding of environmental conditions. topics. The draft modules in Volume 3 include three In both urban and rural areas, LSMS-type surveys modules covering households' use of natural resources are most useful as a source of information about: that household survey designers can include in their households' environmental priorities and perceptions, survey if interest in the topic is significant: one on households' use of natural resources, and the monetary water use (module 4), one on sanitation practices benefits to households of pollution control and infra- (module 5), and one on fuel use (module 6). Modules structure provision. 4 and 5 are based on extensive field experience and have been well tested. Module 6, on fuel use, has not Measuring Environmental Priorities and Perceptions been field tested and is included here for purposes of LSMS and sinilar surveys can be an effective tool for illustration. Survey designers will need to carefully measuring households' perceptions and rankings of gen- revise and pretest module 6. It would be natural for eral and specific environmental problems. Many previ- survey designers to put these modules next to the ous (non-LSMS) surveys have asked people what they housing module; if very short versions of these mod- considered the most important environmental problems ules are used, survey designers might even include in their neighborhood, their province, and their country. them in the housing module. Respondents have often been asked to rank a list of Modules 4 and 5, on water and sanitation, appear problems or issues from most to least important. quite long, but in the large majority of cases it will The environmental draft modules provided in take 20 minutes or less for interviewers to complete Volume 3 include two modules that can be used in both modules. These water and sanitation modules are LSMS surveys to examine households' general envi- substantially different from the modules on the same ronmental priorities: module 1 (for fielding in urban topics that were used in previous LSMS questionnaires areas) and module 2 (for fielding in rural areas). Both and in almost all other multitopic household surveys of the modules contain a list of problems that can be that have been used in developing countries. In the found in many locations. Because the purpose of these current draft modules questions are included on each modules is to identify which environmental problems possible water source and each sanitation practice, are most important in the view of the respondents, it whether or not they are used by the household sur- does not matter if some of the items on the list are not veyed. (In fact, in many locations households often use of particular concern in a specific survey situation. two or three sources of water.) (The list of issues will generally be easier to standard- There are two principal reasons for this approach. ize for urban than for rural areas.) If certain items of First, household water use practices in developing importance in the country surveyed are missing, countries are often much more complex than such designers can easily add these items to the modules. practices in industrialized countries. In situations in Another draft module, module 3, illustrates how which a significant number of households use multi- LSMS surveys can be used to measure environmental ple water sources for different purposes in different attitudes and perceptions. This module focuses on seasons, asking questions about each water source is urban air quality but could be adapted to cover other the most systematic, accurate way to determine actual environmental issues. conditions and practices. Analysts may know little about existing water use practices, and initial assump- Measuring Households' Use of Natural Resources tions often turn out to be wrong. LSMS-type surveys can also be used to collect infor- A second reason for this systematic approach is to mation on households' use of natural resources. In better understand why different households choose 12 CHAPTER 14 ENVIRONMENTAL ISSUES different water sources for different purposes. To naire will not necessarily reflect the choices available to model households' water source choices, analysts need the survey's sample households. to know the attributes of the water sources that were It should also be noted that by placing quality not chosen as well as the attributes of the sources that questions in the community questionnaire, survey were chosen. Modules 4 and 5 collect information on designers make the implicit assumption that all house- both sets of attributes. holds' evaluation of the quality of water is the same- The designers of many LSMS-type surveys will that what looks good to one individual looks good to not be able to include such a lengthy water use mod- all and that what tastes bad to one tastes bad to all. ule due to budget limitations or due to fears that the The fourth approach to shortening the module questionnaire will become too unwieldy. However, would be to have it consist only of the questions about there are several options for shortening the water use the principal water sources (for example, private con- module. nections and public taps) and the summary section at First, if the distinction between rainy and dry sea- the end. sons is not pertinent to the country studied, the sepa- In cases in which the goal is very limited descrip- rate sets of questions for each season may be omitted. tive analysis of living standards, very limited informa- The wording may then be changed in the remaining tion on resource use can be placed in other sections of set. For instance, Questions 2-5 could become "how the questionnaire; water and sanitation questions can would you judge..." rather than "in the rainy season, be included in the housing module, and fuel use ques- how would you judge..:" and the corresponding ques- tions can be included in either the housing module or tions on the dry season (Questions 6-10) could be the consumption module. To illustrate these options, dropped. Similarly, Questions 21-23 could be reword- resource use questions are presented in all three of ed to be more generic, and Questions 24-26 could be these modules. In any real survey the questions should dropped. be coordinated. For example, questions on use of and Second, in some instances the module can be expenditure on electricity are in the housing module shortened by dropping any potential water sources introduced by Chapter 12 rather than at the end of that no one seems to use (though this might be diffi- module 6C of this chapter (which would be another cult in a national survey). In many developing coun- alternative for their placement). tries, every kind of water source is used somewhere in the country. Nevertheless, few households have access Placing a MonetaryValue on Changes in Environmental to many different sources. For example, modern, well- Quality to-do neighborhoods in cities may use only piped The problem with using LSMS-type surveys to value water, whereas urban slumdwellers may have access nonmarket environmental costs and benefits is that only to public wells, taps, or vendors and poor rural respondents in the sample are likely to be drawn from residents may have to rely on springs or surface water. throughout the country where the survey is being If only a few sources pertain to each household, the fielded, whereas many environmental problems are amount of time needed to administer the module will location-specific.Thus for an LSMS-type survey to be be much shorter than if each household uses many useful it must put a value on outcomes that are com- sources. mon to many locations. Third, there is the possibibty of moving some of On the other hand, one advantage of LSMS and the quality and price questions, such as Questions 2-10 similar surveys is that extensive data are collected on and 13-16, to the community questionnaire. This will the socioeconomic characteristics of households. work best when the community questionnaire is filled These and other potential covariates can be used to out for each primary sampling unit or community and model the determinants of the value that households when these units are geographically compact. If a unit put on environmental improvements and to model is defined as the few blocks around an urban house- how environmental considerations influence various hold, the water sources available to each household in choices that households make.4 the unit will tend to be very similar. In contrast, if the The fact that environmental problems and the way unit is defined as the entire city, the answers to ques- people react to them are often location-specific can be tions about water sources in the community question- a problem when using household survey data in 13 DALE WHITTINGTON revealed preference models (such as the travel cost and reliable or accurate estimates of economic value (see the hedonic property value models) to value changes in Box 14.1). environmental outcomes. Therefore, it is difficult to In practice there are two main ways to use a stat- design a standard set of questions on the different ways ed preference approach.The first is to ask respondents in which households adapt their behavior in accor- to value a program or policy that was designed to dance with environmental stresses that can be used uni- improve environmental quality. (What is meant by a formly for a nationwide sample of households. program is an activity or set of activities designed to However, there is one simple type of question that alleviate a problem. For example, a program to can be asked in all household interviews in a nation- improve air quality might include a variety of regula- wide sample, and the answers to questions of this type tory and investment actions such as prohibiting the use can be used to estimate the economic value of envi- of lead in gasoline, installing stack gas scrubbers on ronmental improvements. Rather than analysts certain industrial facilities, establishing a system of attempting to infer the effect on housing values of, say, tradable emission permits, and improving mass trans- improved water service or improved air quality, inter- portation.) The survey interviewer would describe the viewers can ask respondents about these matters program or policy and its consequences to the respon- directly. The following questions are examples of how dent, then ask the respondent to value the program or this could be done: policy. A variety of elicitation procedures are available * For households that rent and have a private water for survey designers to choose from in phrasing the connection: How much would this apartment/ questions. flat/house rent for if it did not have a private water The second way of using a stated preference connection? approach is to ask a respondent to value certain envi- * For households that rent and do not have a private ronmental outcomes without providing specifics water connection: How much would this about the policies or programs that will be used to apartment/flat/house rent for if it had a private achieve these outcomes. For example, a respondent water connection? may be asked to value the public health benefits of * For households that own their dwelling unit and improving the quality of drinking water without the have a private water connection: How much would interviewer explaining the treatment technologies that this apartment/flat/house sell for today if it did not would be used to do this. have a private water connection? Both approaches have advantages and disadvan- - For households that own their dwelling unit and do tages. If respondents are asked to value specific policies not have a private water connection: How much or programs, it is not clear that they will understand or would this apartment/flat/house sell for if it had a believe that the program will achieve the benefits private water connection? described. Also, the respondents' answers to such ques- However, the author believes that the most prom- tions reflect not only their willingness to pay for the ising way in which household survey data can be used benefits that they receive from the program but also for environmental valuation is to use a stated prefer- their willingness to pay for the benefits that others ence (or contingent valuation) approach to measure receive. Some policy analysts would prefer to have data both households' willingness to pay for improved on the former but not the latter. water and sanitation services and the value that house- On the other hand, questions designed to ask holds place on the improvements in health outcomes respondents to value outcomes directly (without resulting from improvements in the environment (for knowing about the policies or programs that will be example, urban air quality). From the author's per- used to achieve these results) may appear more abstract spective, the potential benefits of including contingent to respondents and thus be more difficult to answer. valuation modules would far outweigh the costs Also, this method prevents policymakers from learning involved, and this is an intriguing new avenue for sur- whether respondents perceive the specific institution- vey designers to consider. It must be emphasized, al and financing arrangements for improving environ- however, that the contingent valuation method is con- mental quality as likely to be effective and fair. troversial within the economics profession; many In fact, both ways of using a stated preference economists feel that this method is unlikely to yield approach are likely to work well in the case of 14 CHAPTER 14 ENVIRONMENTAL ISSUES Box 14.1 Criticisms of the ContingentValuation Method Over the last decade the contingent valuation method has dents will answer contingent valuation questions as if they become the most widely used nonmarket valuation tech- were faced with a real budget constraint when in fact they are nique in the world. Perhaps in part because of its popularity, not. Proponents of the method have not offered a systemat- this method has no shortage of critics. Probably the majority ic theory of how respondents answer contingent valuation of professional economists view the contingent valuation questions, so there is no convincing theoretical reason to method with considerable skepticism if not downright hostil- believe that respondents answer accuratelyThis criticism calls ityThere have been three broad types of criticism of the con- attention to the need for empirical evidence on whether tingent valuation method; I will briefly summarize each in turn. respondents answer contingent valuation questions accurate- ly and reliably. However, the fact that there is no theory to I. Estimates of economic volues based on the contingent vol- explain why respondents would answer accurately does not uation method are unreliable and inaccurate, mean that they do not do so. The most common criticism of the contingent valuation Respondents in the third category are confused by the method is that it yields unreliable and inaccurate estimates of way the words in the contingent valuation survey are used and the economic value of gains and losses (Diamond and by the context given to the good or commodity described in Hausman 1994; Diamond and others 1993). Some people the contingent valuation scenario (Kahneman and Knetsch argue that respondents do not give accurate answers to con- 1992; Kahneman, Knetsch, and Thaler 1990). Respondents tingent valuation questions, giving three main reasons: often need a frame of reference or a basis for comparison in * Respondents intentionally give inaccurate answers to val- order to value the quantity of the good or service offered in uation questions. a contingent valuation scenario. Critics of the contingent valu- * Respondents have no incentive to answer contingent val- ation method are correct that people can have considerable uation questions accurately and thus do not do so. difficulty judging the quantity of a good described without * Respondents are easily confused by the way contingent being given a context for assessing whether the amount is scenarios are crafted and the kinds of questions typically large or small.This is likely to be a particularly serious difficul- asked, and thus give inaccurate or widely varying answers ty when respondents are asked open-ended valuation ques- depending on the context. tions such as, "What is the maximum amount you would pay These reasons for inaccurate responses are not in a formal for good x?" Respondents' insensitivity to the quantity of a sense mutually exclusive, and many critics of the contingent good or service offered in a contingent valuation scenario has valuation method freely cite them all as justification for reject- become known as the "embedding problem." ing contingent valuation estimates of economic values. But it There is also considerable evidence from a wide range should be noted that each of these reasons is based on of psychological experiments that people's answers to ques- somewhat different notions of a typical respondent. tions are influenced by the initial quantities or prices offered. Respondents who intentionally give inaccurate answers There is little reason to think that contingent valuation sur- can carefully discem the purpose of the contingent valuation veys would be free of such anchoring effects, and indeed it is survey, and quickly develop a strategy for answering questions common to find such evidence. that will further their self-interest and thus foil the contingent There is little doubt that these embedding and anchor- valuation researcher Their responses are thus subject to ing problems exist in at least some contingent valuation sur- "strategic bias," which critics believe renders these responses veys, and careful contingent valuation work requires that essentially useless as measures of economic value. Despite practitioners test for the existence of such phenomena. If the widespread belief among economists that respondents they are found, such phenomena add considerable uncertain- answer contingent valuation questions strategically, there is lit- ty to an assessment of the accuracy and reliability of the data, tle empirical evidence that they do so. and certainly need to be discussed. In some cases it is possi- Respondents in the second group again perceive their ble for a contingent valuation researcher to adjust estimates self-interest, but in this case quickly determine that it is not of economic value to account for such biases in the data. worth the energy or mental effort to think carefully about the questions being asked. They thus act rationally and preserve 2. Contingent voluation researchers often ask the wrong their mental powers, waiting until a more important problem question. comes along that is worthy of their full attention.This line of The second line of criticism is that the contingent valuation criticism bears careful consideration. It is certainly true that method as practiced has been used to answer the wrong respondents do not have strong incentives to answer contin- question.There is wide recognition within the economics pro- gent valuation questions accurately Critics of the contingent fession that the conceptually correct measure of a welfare valuation method are thus quite right to ask whether respon- (Box continues on next page.) 15 DALE WHITTINGTON Box 14.1 Criticisms of the Contingent Valuation Method (continued) change may be either an individual's willingness to pay or will- includes all types of household surveys. This line of criticism ingness to accept compensation, depending upon whether comes largely from sociologists, anthropologists, and other the change is viewed by the individual as a loss or a gain in advocates of more participatory approaches to development well-being (Cohen and Knetsch 1992; Knetsch 1990; planning. It is rarely heard from economists because if one Rutherford, Knetsch, and Brown 1 998; Vatn and Bromley accepts the criticism, it cuts with even more force on other 1 994).There has not, however been a similar acceptance of nonmarket valuation techniques such the travel cost method, the overwhelming empirical evidence that has accumulated hedonic property value models, and the marginal productivi- over the past two decades that the difference between will- ty approach. Indeed, the basic thrust of this criticism applies ingness to pay and willingness to accept compensation meas- to all formal economic analyses of individuals' demand for ures is large people value losses 2-4 times more than goods and services, including cost-benefit analysis and project gains-and is not the result of income effects or transaction appraisal. Economists typically rely on survey and experimen- costs. Economists still often assume that willingness to pay tal approaches for gathering data and use theoretical models and willingness to accept compensation are (or should be) of human behavior to interpret such data rather than con- practically equivalent. sulting with people themselves about the policy and project Many contingent valuation researchers have used will- matters at hand. ingness to pay valuation questions when in fact willingness Proponents of participatory techniques also make the to accept compensation questions would have been the argument that ifthe powers that be do not want to relinquish conceptually correct approach. Some contingent valuation their authority, and want data on individuals' demand for researchers have done this because they believed there goods and services for planning purposes (as opposed to should be no difference between willingness to accept actively involving households in the planning process and compensation and willingness to pay; others have favored empowering them to make their own decisions), a variety of willingness to pay questions for reasons of practicality and participatory techniques will provide more accurate data on expediency (It is often quite difficult to obtain meaningful demand than do the contingent valuation method or other answers to questions that ask a respondent about the min- survey methods. This is in line with the second class of criti- imum levels of compensation he would accept.) However, cism above. There have been few direct empirical compar- neither of these is an adequate rationale for asking the isons of demand estimates using both the contingent valua- wrong question (and thus estimating the incorrect concept tion method and more participatory data collection of economic value), and critics of the contingent valuation techniques; the few that have been done suggest that the two method are quite right to call attention to this mistake in data collection approaches can give quite different results, the way the contingent valuation method is sometimes although they are inconclusive as to which yields the most practiced. However, this is not strictly speaking a criticism of accurate and reliable estimates (Davis and Whittington 1998; contingent valuation itself This problem could be addressed Davis 1998). if contingent valuation practitioners used willingness to pay It is important to note that the criticisms of the contin- measures where these were the conceptually appropriate gent valuation method do not all suggest that the contingent measure of economic value, if the practitioners were able valuation data will overstate or inflate the economic value of to adjust or benchmark willingness to pay measures to con- gains and losses. Critics of the widespread use of the willing- vert them to reasonable estimates of willingness to accept ness to pay structure of contingent valuation questions in fact compensation values in situations where willingness to contend that contingent valuation practice understates eco- accept compensation was the correct measure of econom- nomic values. Advocates of participatory data collection ic value, and if the practitioners were able to successfully methods make no claim about the direction or magnitude of measure willingness to accept compensation when appro- the mistakes they believe the contingent valuation method priate. will make. Most of the critics who argue that contingent val- uation will yield inaccurate and unreliable results seem to 3. The use of the contingent voluation method is ethicoaly believe that the contingent valuation method will yield over- inoppropriote. estimates of demand. But even here the direction of alleged The third type of criticism of the contingent valuation method bias is often inconclusive. For example, if respondents answer is that it is an ethically flawed method of data collection contingent valuation questions strategically, they might either because it treats respondents as subjects rather than as active understate or overstate their willingness to pay for a good or participants in the planning process. From this perspective, service, depending on how they believed their answer would contingent valuation is just an example-albeit an egregious affect the provision of the good or service by the government one-of a class of 'extractive" data collection methods that or another authority. 16 CHAPTER 14 ENVIRONMENTAL ISSUES improved water supply services, because private water improved water or sanitation service must typically be connections and reliable water services are largely pri- phrased according to the level of service that the vate goods in the sense that their benefits accrue prin- respondent already has.This means that the interview- cipally to the households that receive them. Moreover, er must determine the respondent's housing tenure these benefits are more immediate and obvious than, and water and sanitation service level in order to skip for example, the effect of some pollutants on the to the appropriate valuation question. lungs. It is likely that both kinds of contingent valua- In the first section of this chapter it was noted that tion approach will become easier and more effective as the way environmental policies and programs were they are used more often in LSMS and similar surveys. appraised in the past has been criticized on the Volume 3 contains five environmental modules grounds that it failed to take into account these poli- that illustrate the ways in which a stated preference cies and programs' future costs and benefits, thus dis- approach could be used in an LSMS-type survey. criminating against the interest of future generations. Modules 7 and 8 are designed to measure households' However, little information is currently available on willingness to pay for improved water services in individuals' own rates of time preferences. Module 11 urban and rural areas. Module 9 is designed to meas- uses a simple contingent valuation approach to ask ure households' willingness to pay for improved sani- questions that will yield the information necessary to tation services, and Module 10 measures households' impute individuals' rates of time preferences (Cropper, willingness to pay for improved health and other out- Aydete, and Portney 1994).This module has been used comes from improved urban air quality. Module 11 successfully for this purpose in Uganda, Mozambique, illustrates how a contingent valuation approach can be Indonesia, Bulgaria, Venezuela, Ethiopia, and Ukraine used to collect data for estimating individuals' rates of (Poulos and Whittington 2000). time preferences for nonmarket goods. Even more If any of these stated preference modules are than Modules 1-6, Modules 7-11 will probably have administered in a specific location within a country, to be adapted to suit local or regional conditions. analysts can use benefit transfer methods to estimate These environmental valuation modules need environmental values in other locations. However, not be used for all respondents in an LSMS-type because benefit transfer methods have only recently survey. Survey designers may use a module designed begun to be used to value nonmarket goods, the pro- to estimate the value that households place on cedures for dealing with common problems have not improved water services only in one region of a been standardized. country where water problems are known to be par- ticularly acute. Similarly, the air quality module Estimating Physical Effects of Environmental Degradation should generally be restricted to large urban areas and Physical Benefits of Pollution Control and perhaps even to specific urban locations. In this In industrialized countries household surveys have context, how urban areas are defined will depend often been used to implement a damage function upon the extent and magnitude of the environmen- approach to valuing changes in the quality of the envi- tal problem; multiple definitions might be used in a ronment, including measurement of the impact of air single survey. For example, sanitation problems can pollution on people's health. In developing countries be severe in small communities, while air quality household surveys have been used to measure the problems are most likely to be a serious concern in impact that a lack of drinking water and sanitation much larger cities. facilities has on people's health. The question arises An important complication in the design of stan- whether LSMS and similar surveys are an appropriate dardized modules for valuing improved water and san- vehicle for either of these purposes. In the opinion of itation services is that these values depend upon what the author of this chapter, the answer depends on alternative sources of supply exist and on how much which aspects of environmental health analysts wish to they cost. Therefore, the survey questions must usually study and on how the sample for the LSMS-type sur- be designed to reflect specific housing and water and vey is drawn. sanitation conditions. For example, valuation questions For example, household surveys are not an effi- must generally be phrased differently for homeowners cient technique for measuring the impact of envi- than for renters. Similarly, questions about the value of ronmental pollution on a particular health outcome, 17 DALE WHITTINGTON particularly when this health impact is a rare event (for example, mortality). The problem is that too Box 14.2 Environmental Policy Issues and Household many resources must be spent in order to find too Survey Data few cases in which the environmental pollution has Environmental poaicy issues that can be addressed using caused a health effect.When the health outcome in household survey data question is more common, two primary questions Urban and rural water and sanitation remain: whether the sample will provide enough Urban air quality variation in the environmental variable studied (for Household fuel use example, ambient air pollution levels) and whether the environmental variable can he measured at a Ioxv Information relevant for environmental policy analysis that can be collected using household surveys cost. it is not generally advisable to use householdThe damages that households suffer from environmen- Thus it is not generally advisable to use household tal degradation of air water and land resources surveys for environmental epidemiology. For measur- * Household use of local renewable resources such as ing the impact of water and sanitation interventions forests, fisheries, and groundwater on waterborne disease, for example, a case controlled Households' priorities for environmental improvements study is usually preferable to a household survey unless Households' willingness to pay for some kinds of envi- the sample size of the survey is very large. Some ronmental quality improvements household surveys have been used to study air pollu- Households'rates oftime preferences * The extent of problem conditions (such as households' tion; the most successful of these have been prospec- access to improved water supplies) tive studies in which each household kept a diary of any respiratory problems and illnesses suffered by its Environmental issues that are difficult to analyze using house- members. However, LSMS-type surveys are not gen- hold survey data erally organized in a way that permits respondents to * Global environmental issues about which households keep such diaries.Another problem is how to measure may have little understanding the ambient air quality experienced by respondents during the period when the diaries are kept. One way Environmental issues that can only be analyzed using special- that analysts could use this kind of data from a house- Property rights regmes for local environmental and hoold survey would be to perform a cross-sectional natural resources (such as ownership of natural study in which they correlate symptoms indicative of resources) chronic respiratory problems with air pollution levels. Localized natural resource management issues Such studies can be useful, but require that either indoor or outdoor air pollution be measured for all people in the sample. households' environmental attitudes and preferences Box 14.2 summarizes some ofthe main points dis- that can only be captured through the use of open- cussed thus far in this chapter. The principal environ- ended questions, and environmental health problems mental policy issues that can be addressed in household that occur only rarely in a population. surveys include urban and rural water and sanitation Box 14.3 summarizes some cautionary advice quality, urban air quality, and household use of natural about which of the draft modules presented in this resources such as fuel. Considerable information can be chapter are new and untested. Household survey collected in LSMS-type surveys that is relevant for designers are again warned that the contingent valua- environmental policy analysis, including the damages tion modules (modules 7-11) need to be carefully that households suffer from environmental degrada- adapted to country conditions and also need to be tion, households' attitudes toward and priorities for the pretested. Implementing the contingent valuation environment, households' rates of time preferences, and modules also requires more extensive training of inter- the economic value of environmental quality changes. viewers. The household fuel use module (module 6) However, Box 14.2 also points out that some impor- has not been pretested. tant environmental issues are difficult to analyze using On the other hand, the modules on general envi- household data. These include global issues about ronmental priorities (1 and 2) and on household water which households may have little understanding, and sanitation use (4 and 5) have been used extensive- 18 CHAPTER 14 ENVIRONMENTAL ISSUES Who Should Be Interviewed? Box 14.3 CautionaryAdvice It is recommended that only one respondent per household answer the questions in the environmental How many of the draft modules are new and unproven? modules.The respondent should be either the head of None of the environmental modules have yet been used as part of an LSMS survey.The household fuel household or the spouse of the head of household. use module is untested and is likely to require careful Particularly for the contingent valuation modules, the field testing and subsequent modification. Module 3 interviewer should speak to someone who has the (environmental attitudes toward and perceptions of authority over the finances of the entire household. urban air quality) has only been used once in a large It is suggested that the interviewer should simply survey and may require extensive modification in ask to speak to the head of household or the spouse of other locations. the head of household-whichever of these two people * How well have the modules worked in the past? The agrees to be interviewed becoming the selected respon- modules on general environmental priorities (modules aees t ie nterview in the seletd e n- I and 2) have been tested in numerous surveys dealing with households' water and sanitation conditions in the spouse of the head of household are interviewed many developing countries.They have worked very well together and answer the questions jointly.) In some and should present no major problems. The modules countries this approach may result in most respondents on household water use and sanitation have also been being male heads of household. However, in practice it thoroughly tested in many countries.They may need to is usually relatively easy to obtain a sizable percentage of be modified to fit a particular country situation (for female respondents. In most surveys that have used this example, in a region where the distinction between procedure there has been close to a 50/50 splt between rainy and dry seasons is not important). Nevertheless, prcdr, teehsbe ls oa5/0shewe survey plannedryseares uliky oto experi. enermanylr, male and female respondents. This has allowed analysts survey planners are unlikely to experience many prob- lems with these modules, either in their long or short to test whether the gender of respondents affects the versions. Different variations of the valuation modules answers they give in the environmental module. (modules 7-1l ) have also been used in household sur- veys, though not in LSMS surveys, in many countries. How Should Urban Areas Be Defined? * Which parts of the modules most need to be customized? This chapter has recommended that the contingent The contingent valuation scenarios and elicitation proce- valuation modules for air quality be fielded in urban dures in the contingent valuation portion of the valua- tion modules (modules 7-I I) have not been tested in LSMS-type surveys. In particular, referendum elicitation Box 14.4 List of Standard Environment Modules procedures and split-sample experiments have not been tried in LSMS-type surveys, and interviewers will need to * Module I General environmental priorities (urban) be given special training in these methods. However, sin- * Module 2 General environmental priorities (rural) gle-purpose contingent valuation surveys have become * Module 3 Environmental attitudes and perceptions about increasingly widespread in developing countries during urban air quality the past I O years.Thus there is nothing novel about ask- * Module 4 Household water use-attitudes and prac- ing respondents in developing countries such hypotheti- tices (include in housing module) cal questions about environmental goods and services. * Module 5 Household sanitation-attitudes and prac- tices (include in housing module) * Module 6 Household fuel use attitudes and practices ly all over the world and should work well with only (include in housing module) minor modifications to fit local circumstances. * Module 7 Households' willingness to pay for improved water services (urban) Questionnaire Modules * Module 8 Households' willingness to pay for improved water services (rural) Box 14.4 lists the 1 1 environment modules introduced * Module 9 Households' willingness to pay for improved Boxi 14.4nlist thisesect l of these modules are troduted isanitation services (urban) in this section; all of these modules are presented in * Module 10 Households' willingness to pay for heafth Volume 3. Two general implementation issues are also outcomes from improved urban air quality discussed in this section: who should be interviewed * Module I I Stated preference module for imputing indi- and how urban areas should be defined in the context viduals' rates of time preference for nonmarket goods of the environmental modules. 19 DALE WHTTrINGTON rather than rural areas. The definition of "urban" in recreation, tourism, and national parks (Grandstaff and this context needs to be an operational one-not a Dixon 1986; Shyamsundar, Kramer, and Sharma 1993; definition used only by a country's statistical bureau. Menkhaus 1994; Hadker and others 1995). However, Most (but not all) cities with severe air pollution are the areas of application are growing rapidly and now larger than the statistical office's standard definition of include surface water quality (Choe,Whittington, and an urban area. Survey designers need to be careful to Lauria 1996), health (Swallow and Woudyalew 1994; field the contingent valuation module on air quality Alberini and others 1995; Whittington, Pinheiro, and only in urban areas that have an air pollution problem Cropper 1996), and biodiversity conservation (Moran rather than in all urban areas of a certain standard pop- 1994).6 ulation size. Even some large cities do not have an air Some contingent valuation analysts believe that it pollution problem, and it would be a waste of is easier to administer high-quality contingent valua- resources to field this module in such cities. tion surveys in many developing countries than it is in In contrast, the urban sanitation modules will be industrialized countries. Response rates are typically pertinent in smaller urban areas. Special codes may be very high in developing countries, and respondents required on the household identification page of the will often listen to and consider the questions posed to questionnaire to designate the sorts of urban areas to them. Also, interviewers are inexpensive in developing which the different modules apply.The wording of the countries relative to their cost in industrialized coun- first questions in the modules presented here would be tries. Thus the costs of a contingent valuation survey changed to match that nomenclature. administered in a developing country are typically an order of magnitude lower than the costs of a survey Comments on the ContingentValuation with a similar sample size in an industrialized country. Modules This means that larger sample sizes and more elaborate experimental designs can be used in contingent valu- The most innovative aspect of the proposed modules ation surveys in developing countries. for the environment are the contingent valuation There are also likely to be less data on the bene- questions.5 These portions of the questionnaire are fits of different kinds of environmental policies in designed to determine the economic values that developing countries, which means that the marginal households assign to environmental quality or infra- value of data obtained from contingent valuation sur- structure changes. Ten years ago only a handful of very veys is likely to be large. Therefore it seems both fea- rudimentary contingent valuation studies had been sible and desirable to use the contingent valuation conducted in developing countries; at the time, con- method in developing countries to evaluate a wide ventional wisdom was that such studies simply could range of environmental projects. However, this does not be done. The problems associated with posing not mean that conducting contingent valuation sur- hypothetical questions to low-income, possibly illiter- veys in developing countries is easy.There are numer- ate respondents were assumed to be so overwhelming ous issues that arise in conducting contingent valua- that it was not even possible to try to pose such ques- tion surveys in developing countries that must be tions. However, these days many environmental and given careful attention to ensure that high-quality data resource economists and policy analysts working in are obtained. It may be worth considering hiring an developing countries assume that contingent valuation experienced contingent valuation consultant to help surveys are straightforward and easy to do. to prepare modules, since implementing the contin- Bilateral donor agencies and the international gent valuation method requires expertise beyond the development banks increasingly use contingent valua- usual know-how needed to formulate and administer tion techniques in project and policy appraisals as part the other modules of the questionnaire. oftheir everyday operations work.The contingent val- This section of the chapter discusses some of the uation method was initially applied in developing issues that have arisen and some of the lessons that countries primarily in two areas: water supply and san- have been learned during the past 10 years of admin- itation (Whittington and others 1988, 1990, 1993; istering contingent valuation surveys in developing McConnell and Ducci 1988; Briscoe and others 1990; countries. The discussion covers five basic issues: Altaf and others 1993; Singh and others 1993) and explaining a contingent valuation study to nonecono- 20 CHAPTER 14 ENVIRONMENTAL ISSUES mists, interpreting respondents' answers to contingent One particularly common source of confusion valuation questions, setting referendum prices, con- relates to the distinction many noneconomists make structing joint public-private contingent valuation between a household's willingness to pay and ability to scenarios; and addressing ethical problems involved in pay. It is important that interviewers clearly under- conducting contingent valuation surveys. This list is stand that the purpose of a valuation question is to not meant to be exhaustive, but it will hopefully pro- determine what respondents would do if they had to vide survey designers with insights into some of the make a real economic commitment. In other words, issues involved in including contingent valuation the objective of a valuation question is to determine modules in an LSMS-type survey. how much respondents are both willing and able to pay. Explaining a Contingent Valuation Study to Noneconomists The classification scheme presented in Table 14.3 The first difficulty that survey designers may face can be used to clarify this point. As shown, the total when fielding a contingent valuation survey in a population of respondents can be envisaged as four developing country is to explain the contingent val- groups.The respondents of cell 1 say they would make uation method to government officials and survey a real economic commitment if the consequences of interviewers. The concepts of economic value and the contingent valuation scenario could be deliv- "maximum willingness to pay" (or minimum com- ered-and have sufficient income to make such a pensation that a respondent is willing to accept) are commitment.The respondents of cell 2 are able to pay often difficult to transmit to noneconomists. In but not willing to do so-presumably because they order to include open-ended willingness-to-pay prefer to spend their money on other things. questions in the survey questionnaire, survey design- The respondents of cell 3 would like to purchase ers need to ensure that the language used captures the commodity but cannot afford to do so. It is this the notion of the maximum amount an individual is group that typically causes noneconomists the most willing to pay. confusion.The argument is often made that individuals Unfortunately, this can be particularly difficult to in this third group would like to purchase the proposed translate. For example, in a contingent valuation survey good or service if their income was higher but, in their conducted in Haiti (Whittington and others 1990), a current financial circumstances, they are not able to pay. respondent reacted to an early version of an open- Noneconomists often want to classify these respondents ended contingent valuation question by asking one of as "willing to pay," but contingent valuation researchers the interviewers, "What do you mean the maximum I must emphasize that for the purposes of the study such would be willing to pay?You mean when someone has individuals must be categorized as not willing to pay (in a gun to my head?" In fact the interviewer was trying other words, not willing and able). to determine the maximum amount that the respon- The respondents defined by cell 4 consist of indi- dent would be willing to pay for the proposed (or viduals who cannot pay and say that even if they could hypothetical) good or service in the context of the pay for the hypothetical good or service, they would existing institutional regime within which individuals not.These people should be classified as not willing to are free to allocate their personal financial resources. accept the contingent valuation scenario. Contingent valuation analysts would like to measure The important point to recognize is that demand the amount of income that the household could give for a proposed good or service is not likely to be a up after obtaining the goods and services from the function solely of income. Increases in households' project that would leave the household just as well-off income may at times have a negligible effect on the as it would have been had the project never been households' willingness to pay for a specific good or implemented. service. Table 14.3 Willingness and Ability to Pay Respondent is willing to pay Respondent is not willing to pay for the hypothetical good or service for the hypothetical good or service Respondent is able to pay for the hypothetical good or service Willing and able (Cell I) Able but not willing (Cell 2) ............a..b....................... ...f.o..... t....... ...h.. y.. p............ t......................d...............r ........ce .................n ..... ....u .......n........a..b................l.....3......... ............ ...o.... ..a........ ....a.... ....n....... w...... I.i................. l--4.).... .. Respondent is not able to pay for the hypothetical good or service Wlilling but not able (Cell 3) Not able and not wiling (Cell 4) Source Author's examples. 21 DALE WHITrINGTON Interpreting Respondents'Answers to Contingent Valuation ways in which a respondent might answer the valua- Questions tion questions by saying "yes, but ..." but essentially One of the reasons why many economists and analysts meaning "no." have been skeptical about the feasibility of conducting Table 14.4 presents this list of different ways to say contingent valuation surveys in developing countries "no" and the number of times respondents gave each is these analysts'presumption that it will be difficult to "yes, but" answer to the valuation question about understand and interpret respondents' answers to whether the respondent's household would want to be abstract (or hypothetical) questions. Such worries are connected to the new water and sewer lines if a spec- often well founded, and care is needed when drafting ified monthly tariff were charged. For example, of the these questions. 164 answers that were recorded as "no," 52 respon- Analysts had problems in interpreting preliminary dents (32 percent) answered "Yes, but I cannot afford responses to the valuation questions in a contingent it." Another 18 percent said, "I agree, but the costs are valuation survey conducted in Semarang, Indonesia too high." These "yes, but" responses-50 percent of (Whittington and others 2000).The contingent valua- the total number of"no's"-seem clearly negative and tion survey was designed to determine whether a correctly classified as "no." household would vote in favor of having water and However, another 30 percent of the respondents sewer lines installed in its neighborhood if everyone in said, "I need to know others' opinions about the pro- the community had to pay a specified assessment fee gram before I decide." The interviewers assured the (whether or not they connected to the new lines). In survey designers that this was again simply a polite way addition, the survey was designed to determine of saying "no," but the designers thought that this whether the household would choose to connect to answer seemed reasonable. The respondents might such lines if a given monthly tariff were charged. simply have needed time to think about their decision, After the first couple of days of the pretest of the and discussing the matter with their neighbors would contingent valuation questionnaire, the survey design- have been a reasonable way to analyze the pros and ers discovered that everyone was saying "yes" to every cons of the proposed project (Whittington and others question, regardless of the assessment fee or monthly 1992). Thus it was less clear that this response should tariff offered them. So they stopped pretesting and be assigned to the "no" category than it was in the case held a meeting with the team of interviewers to try to of the previous two types of answers.7 Other answers find out why this was happening. During the course listed in Table 14.4 also seemed ambiguous and uncer- of a lengthy discussion it became clear that respon- tain. Therefore, the designers came to believe that the dents were in fact answering "yes, but ...... then giving proportion of respondents they had placed in the "no" many different qualifications to their answer. The category for this valuation question was probably too interviewers informed the survey designers that, in high. Although they had followed their interviewers' Indonesia, these were all polite ways of saying "no." guidance in coding the answers, they subsequently The designers then developed a coded list of the many came to believe that their analytical results probably Table 14.4 Description and Frequency of Different "No" Responses, Semarang, Indonesia Description of response Number of times recorded Percentage of responses I cannot afford it 52 32 ................................................................................................................................................................................................................................... I need to know others' opinions about the program 49 30 i agree. hot the costs are too high 30 18 ........ ~ -e......... t's.................. 'd............................................................ *................... ................................................................................. .....7..................... Yes, if the costs are reduced I 1 7 ..........................................................................................*....................................................*.................................................................................... Ihave many expenses, chiloren. . . (and so on) 8 5 I agree, but the current situation is satisfactory 6 4 I agree, but I do not want to pay in advance 4 2 ........"t e'......... "m.................... p . ...i'd ... ..'s ... ..xt"""ded.................................................................2.........................................................................................I.............. Yes, if tne payment period is extended 21 ...................... o"n.... is......................'y.................................................................................................................................................................<...I.............. Yes, if participation is mandatory < I ...............................................*................................................................................................................................................................................... I can pay, but I want to avoid rumors about my wealth i < I .............................. ........................................................................ -4..........................................................................................0........ ...... Woral number of verboim responses t164 100 Source: Whitt ngton and others 2000. 22 CHAPTER 14 ENVIRONMENTAL ISSUES underestimated the number of households that would happening. Although the merits of referendum-type have agreed to be connected to the water and sewer questions are still being debated,8 most contingent val- lines. uation practitioners favor using only one or two such This example illustrates how careful analysts must questions in order to reduce the possibility of eliciting be in interpreting respondents' answers to valuation biased responses. questions in a cross-cultural context and how important This split-sample technique is routinely used in it is to pretest a contingent valuation questionnaire. U.S. surveys. While professional interviewers in the United States are familiar with the use of this tech- Setting Referendum Prices nique, this is not the case in many developing countries, One commonly used way of asking valuation ques- and the interviewers in such countries will want to tions in a contingent valuation survey is the referen- understand the reason for the split-sample experiment. dum elicitation procedure. In this method the ques- In the past when referendum-type elicitation pro- tions are often couched in terms of voting, as in a cedures have been used in contingent valuation surveys referendum; for example: "If the improvement in air in developing countries, the designers of these surveys quality that I have described were to cost your house- have often made the mistake of specifying too narrow hold $50 in higher taxes, would you vote for it?" To a range of prices. They have tended to set the highest use this procedure the survey sample is randomly split referendum price too low and the lowest price too into several different subgroups.The interviewers pres- high, making it difficult for analysts to estimate "good" ent the respondents in each subgroup with a different valuation functions. This tendency to use too narrow a hypothetical price for the good or service in question. price range is understandable because extremely high Thus a person in one subgroup may be told that the and extremely low prices often lack credibihty. If the price of a service would be $50, while a person in amount that an interviewer mentions to the respon- another subgroup is told that the price would be $100. dent lacks credibility, the respondent is unlikely to Analysts use the responses to construct values for the answer the question on the basis of the price asked. whole sample. In order to increase the credibility of the contin- Many contingent valuation researchers feel that gent valuation results, it is generally advisable that the the referendum approach is the best way to ask a highest price used be rejected by 90 to 95 percent of respondent for information about his or her willing- the respondents. If the frequency distribution of ness to pay because this approach presents the respon- respondents' values is known, using such a high price dent with a realistic, easy-to-answer question. is not efficient (Alberini 1995a, 1995b; Kanninen Moreover, it is not obvious how a respondent would 1995), but in developing countries it is useful to show answer a referendum question if he or she wanted to that setting prices high enough will cut off demand for give a biased answer in order to influence the results the good or service. Nevertheless, survey designers of the study (or advance personal goals). For example, tend to be reluctant to set the highest referendum if the maximum a respondent would pay for an envi- price at a high enough level to do this, partly because ronmental quality improvement were $10 per year and it is embarrassing for interviewers to have to mention the referendum question asked the respondent if he or such an unrealistically high price to respondents. she would vote for a plan that would cost households Respondents often take the contingent valuation sce- like his or hers $5 per year, the respondent would not nario very seriously, and if the interviewer implies that have any obvious incentive to give an untruthful the hypothetical good or service will cost the highest answer. If the respondent indicated an unwillingness to referendum price, they may be acutely disappointed vote for the plan, the plan's chances of being imple- that the good or service would be so expensive. mented might be reduced. It is in this sense that some This problem is exacerbated in countries with a contingent valuation researchers term the referendum highly unequal income distribution. Interviewers approach "incentive-compatible." A respondent might often complain that asking about such a high price is hope to have the plan implemented and not have to silly because everyone knows that the sample house- pay even the $5 per year, but it is not obvious why holds cannot afford to pay such a high price, which answering "no" to a referendum question that posed a makes the interviewers look insensitive or unin- price of $5 per year would increase the chances of this formed.Thus in some situations a very high price may 23 DALE WHITTINGTON not be plausible to respondents and may thus cause whether or not its members are willing to share some them to doubt the credibility of the entire scenario. of the capital costs of a project. For example, consider This problem is compounded if there is a tendency of an investment in sewer lines. If it could be assumed respondents to say "yes" to whatever question the that all households in a particular neighborhood interviewer asks ("compliance bias"). However, in would connect or could be forced to connect to new these circumstances it is even more vital to prove that sewer lines if they were installed, it would not be nec- demand will be cut off if a sufficiently high price is essary to elicit a collective decision on whether the charged. lines should be installed. However, if this could not be Survey designers often set the lowest price too assumed, a fiscally responsible sewer authority could high because the agency funding the survey often not bear the financial risk of installing such expensive wants to use the results to help it set the prices it will infrastructure without some form of payment guaran- charge and may not be interested in learning about the tee. This means that the authority would need some extent of demand for the good or service at prices assurance that, if the sewer pipe were laid in a neigh- lower than those it intends to charge. Thus, for the borhood, households would pay a predetermined funding agency, asking part of the sample about a very amount for this infrastructure improvement, whether low price may seem like a waste of resources.9 or not they agreed to be connected. From the agency's In addition, like prices set too high, prices set very financial perspective, each household in the neighbor- low can make interviewers seem uninformed and hood would be required to pay a share of the sewer undermine the credibility of the contingent valuation network installation costs (whether or not the house- scenario. One implication of this is that the survey hold obtained a connection) because the value of its needs to include questions that aim to find out why property would increase simply by having the option respondents would reject a very low price. Another to connect in the future. implication is that survey designers should explicitly Second, a household must decide whether it will recognize that many projects have negative effects on connect to such infrastructure if it is installed. Because some people; thus designers should expect some many infrastructure projects have positive externalities respondents to be unwilling to pay for the services that and public good characteristics, it is plausible that a a project will (hypothetically) provide. Some goods or household would vote in favor of a project and agree services described in contingent valuation scenarios to pay some share of the capital costs even if it decid- may have little or no value to some respondents. An ed not to use the service immediately. Because these improved water supply system will threaten the business two decisions are conceptually interrelated, the con- of water vendors; such respondents will never accept the tingent valuation scenario needs to inform the respon- contingent valuation scenario, no matter how low a dent about the terms and conditions of both public price is hypothetically charged for the service. and private components of the deal to enable the respondent to make an informed choice. In practice Constructing Joint Public-Private Contingent Valuation this means that a lot of information may need to be Scenarios conveyed to respondents-typically necessitating the Many of the contingent valuation surveys conducted use of photographs and drawings. in developing countries have been concerned with Also, respondents are likely to have numerous ques- estimating the demand for infrastructure services. In tions about the proposals. Thus it is vital to use highly one important respect, the contingent valuation sce- trained, well-informed interviewers who can easily narios required for such surveys are considerably more respond to questions from respondents. It is also gener- complex than those used in contingent valuation sur- ally inadvisable to allow interviewers to deviate from veys about environmental quality improvements in the script of the questionnaire in an ad hoc manner.The industrialized countries. In order to understand house- interviewers should be instructed to give the informa- hold demand for infrastructure services such as tion provided in the questionnaire script in a different improved sewers or piped water supply, it is often nec- form if the respondent does not initially understand it. essary to model two household decisions jointly. In some cases survey designers may need to develop First, a household must decide whether to support contingencies that help interviewers deal with particu- the collective decision of a community regarding lar inquiries that are rarely raised by respondents. 24 CHAPTER 14 ENVIRONMENTAL ISSUES Ethical Problems in Conducting Contingent Valuation short time (generally two to three hours). In this way Surveys respondents would supposedly have little time to dis- Two ethical issues arise in the implementation of the cuss their interviews with one another before they had contingent valuation method; neither of these issues all been conducted. However, in one community the has received the attention it deserves.The first issue neighborhood leader dropped in on an early interview concerns the use of a referendum elicitation proce- unannounced and heard the referendum price offered dure. Because this method entails giving different sub- the respondent.This price happened to be the highest groups of respondents in the survey sample different of the four prices used, and the neighborhood leader prices for the same hypothetical good or service, it became concerned. He quickly spread word through- may sometimes cause confusion and spread misinfor- out the neighborhood to answer "no" to the valuation mation about the real costs of addressing a problem question because he felt that the improved water and that may be of great public concern. sanitation program offered in the contingent valuation For example, a referendum question was used in scenario was simply too expensive. two recent contingent valuation studies conducted for Obviously, this problem arose partly because the the World Bank (Pinheiro and Whittington 1995; field supervisor and the interviewer were unable to Whittington and others 2000). In a contingent valua- exclude the neighborhood leader from a supposedly tion survey conducted in November 1994 that was private interview (although, in fairness to them both, designed to estimate households' demand for improved this is not a easy thing to do in Indonesia). However, water services in a small town in Mozambique, five dif- it also illustrates how quickly information can spread ferent prices were randomly assigned to different sub- in a close-knit urban community, how seriously some groups of respondents. In June 1995 the study team community members may take the information pre- returned to the town where the survey was conducted sented to them in a contingent valuation scenario, and to brief a group of local government officials and com- how easily a community can be confused by the use of munity leaders on the results of the contingent valua- different prices and by other aspects of split-sample tion survey. During the ensuing group discussion, one experiments. neighborhood leader said that he had talked to many Contingent valuation experts may argue that any respondents after their interviews and that he did not such misinformation is the fault of the survey design- understand why different households were asked to pay er, who is supposed to craft language that informs different prices. He said that it did not seem fair or nec- respondents that the choice is "just" hypothetical. essary to charge one household more than another for Respondents are usually told to "suppose" or "imag- a water connection. The mistaken impression that dif- ine" the choice described, and that the choice is not ferent households in the community would be charged actually or necessarily going to be offered.This nuance different prices for a water connection seems to have is often lost in translation; in some cases the condi- been one outcome of the contingent valuation survey. tional subjunctive may not even be translatable. The Thus the use of a referendum approach with different interviewers' disclaimer may also be false in the sense prices may have increased public uncertainty and con- that a project is actually being considered and is thus fusion about the costs of improved water services in not hypothetical at all. this small Mozambique town. A good contingent valuation scenario is designed In July 1995 a contingent valuation survey of a to be realistic and taken seriously by respondents. In few hundred households was conducted for the World practice, the more seriously respondents take the Bank in three areas of Semarang, a city of 1.2 million choice put to them, the less hypothetical the scenario people on the north coast of Java, Indonesia. The sur- is likely to seem to them. This is particularly true for vey was administered in three districts of the city.The goods and services with large use values that are com- leader of each neighborhood unit had to be informed monly provided by the government, such as water about the survey by higher-level community leaders supply services. The less hypothetical the provision of before the survey could take place. After the leaders' the good or service described in the contingent valu- permission was secured, a team of interviewers and a ation scenario, the more likely the different referen- field supervisor was sent into the neighborhoods to dum prices will confuse serious public discussion of interview all of the sample households in a relatively the issue at hand. 25 DALE WHIrrINGTON Contingent valuation researchers generally scenarios used in such surveys may not be strictly assume that they will sample large populations and hypothetical. If the donors and governments that fund thus that there will be little chance that one respon- the contingent valuation surveys judge the results to dent will talk with another, so any misinformation be credible, the findings will likely be used in policy communicated to a relatively small number of respon- decisions. This movement from hypothetical to "real" dents about a hypothetical good or service will not be contingent valuation scenarios raises a host of ethical widely discussed and will not influence public debate. concerns. However, in small towns, villages, or urban neighbor- From a theoretical perspective, it is not possible to hoods in developing countries, this assumption is often value a project independent of how it is to be paid for unwarranted. Even in large capital cities, a sample of or independent of the institutional regime assumed to 1,000 to 2,000 households is not too small to avoid be in place when the project is implemented. As long the discussion of the contingent valuation survey by as people have preferences xvith regard to various many people, including some who may be knowl- aspects of how a project is carried out, such prefer- edgeable about the problem addressed in the survey or ences need to be taken into account. The fact that val- who will be influential in deciding how it should be uation estimates are context-specific has nothing to do solved. with the contingent valuation method itself, although The issue of spread of misinformation arises not the contingent valuation method does give survey only with the prices used in the referendum elicitation designers substantial control over what assumptions method but also many with other features of split- they can make about the institutional arrangements sample experiments, including the scenarios used. A for the delivery of the hypothetical good or service. contingent valuation survey in support of the State of Revealed preference valuation approaches generally Alaska's case against Exxon in the Exxon-Valdez oil accept the prevailing economic, political, and institu- spill is one of the finest, most professional contingent tional context within which the data were generated. valuation surveys conducted to date (Carson and oth- ers 1992). In this survey a contingent valuation sce- Notes nario was crafted that described an oil spill prevention program with two main components: the requirement The author would like to acknowledge Maureen Cropper at the that oil tankers be accompanied by escort ships while University of Maryland for her significant contributions to this in the Valdez Straits (to reduce the chances that the chapter. tankers inadvertently drifted onto nearby rocks) and 1. Of course, not all environmental problems are caused by the use of an oil spill containment technology called a poorly defined property rights. Many analysts have found govern- "Norwegian sea fence" that could be used in the high ment failure itself to be a pervasive force causing environmental seas of Prince William Sound. Respondents in the sur- degradation. vey were asked whether they would vote for or against 2. "Improved" water sources are the result of modifications to a rapid response oil spill containment force in Prince the natural source to increase the quahtv or quantity of water.They William Sound that would deploy these escort ships do not, however guarantee that the water is clean. Piped water, for and the Norwegian sea fence if the implementation of example, is always considered an improved source, but the wvater that the plan would cost their household a specified pipes carry may need additional treatment before it is safe to drink. amount of money. Although the "Norwegian sea 3. The contingent valuation method is a survey technique that fence" technology did exist, it was not as large or attempts to ehcit information about individuals' or households' pref- effective as was indicated by the contingent valuation erences for a good or service. Respondents in a survey are asked a scenario.The survey interviewers told respondents that question (or series of questions) about how much they value a good the Norwegian sea fence was more effective than it or service. The technique is termed "contingent" because the good actually was so that the respondents would not reject or service is not necessarily going to be provided by the interviexver the scenario as implausible. or analyst.Thus, the provision of the good or service is hypothetical. A second ethical issue concerns how honest one The contingent valuation method can be used to obtain values should be about the institutional regime contemplat- of pure public goods, goods wvith both private and public charac- ed for delivering the "hypothetical" goods or servic- teristics, and private goods. It is often used to assess households' es. In developing countries, the contingent valuation preferences for goods or services for xvhich a conventional market 26 CHAPTER 14 ENVIRONMENTAL ISSUES does not exist. For a brief introduction to the contingent valuation the benefits received by households may serve little purpose from method see Pearce and others (1994), chapter 7. For more in-depth the client's perspective and may even be deemed wasteful. presentations see Mitchell and Carson (1989) and Cummings, Brookshire, and Shultze (1986). References 4. LSMS surveys typicaDly provide a rich set of possibilites for covariates, and in practice only a few such covariates are usually need- Alberini, Anna. 1995a. "Efficiency versus Bias ofWillingness-to-Pay ed for the analysis of the contingent valuation data.The specific vari- Estimates: Bivariate and Interval-Data Models." Journal of ables used to explain a respondent's answers to contingent valuation Environmental Economics and Management 29 (2): 169-80. questions depend on the theoretical demand model employed. Most . 1995b. "Optimal Designs for Discrete Choice Contingent models would use the following minimum set of explanatory vari- Valuation Surveys: Single-Bound, Double-Bound, and ables: household income, household assets, and respondent's educa- Bivariate Models." Journal of Environmenttal Econotnics and tion, gender, and occupation (and perhaps rehgion). MVanagement 28 (3): 287-306. 5. The material in this section is drawn fromWhittington 1998. Alberini,Anna, Maureen Cropper,T.T. Fu,Alan Krupnick,J.T. Liu, 6. See Georgiou and others (1997) for additional references and D. Shaw, and Winston Harrington. 1995. "Valuing Health an annotated bibliography. Effects of Air Pollution in Developing Countries: The Case of 7. Assigning "I need to know other people's opinions" answers Taiwan." Resources for the Future Discussion Paper 95-01. to the "no" category is consistent with the recommendations of the Washington, D.C. report of the U.S. National Oceanic and Atmospheric Altaf, Mir Anjum, Dale Whittington, Haroon Jamal, and V Kerry Administration's Expert Panel on the ContinentValuation Method Smith. 1993. "Rethinking Rural Water Supply Policy in the (Arrow and others 1993). This assignation practice is followed in Punjab, Pakistan." Water Resources Researcih 29 (7): 1943-54. many large contingent valuation surveys conducted in the United Arrow, Kenneth J., Robert Solow, Edward Leamer, Paul Portney, States. For further discussion and a theoretical treatment of Roy Radner, and Howard Schuman. 1993. "Report of the ambiguous or"don't know" responses in contingent valuation stud- NOAA Panel on ContingentValuation." United States Federal ies, see Wang (1997). Register 58. 8. The principal alternatives in contingent valuation surveys Briscoe,John, Paulo F de Castro, Charles Griffin,john North, and would be to ask a single open-ended question-"How much Orjan Olsen. 1990. "Towards Equitable and Sustainable Rural would you be wiling to pay?"-or to ask a series of questions that Water Supplies: A ContingentValuation Study in Brazil." World honed in on a particular answer. For example, a respondent would Bank Economic Reviewv 4 (2): 115-34 be asked, "Would you be willing to pay $500?" If he said no, he Carson, Richard, Robert C. Mitchell,W. Michael Hanemann, Ray would be asked: "Would you be willing to pay $100?" If he said no J. Kopp, Stanley Presser, and Paul A. Ruud. 1992. "A again, a price of $50 might be mentioned. If the respondent said yes Contingent Valuation Study of Lost Passive Use Values to $100, he would be asked if he would pay $300. The questions Resulting from the Exxon-Valdez Oil Spills. A Report to the would be asked three or four times until the price the respondent Attorney General of Alaska." would pay xvas bracketed by a higher price that he would not pay Choe, KyeongAe, Dale Whittington, and Donald T. Lauria. 1996. and a lower price that he would pay "The Economic Benefits of Surface Water Quality Improve- 9. A contingent valuation expert is generally engaged by a client ments in Developing Countries: A Case Study of Davao, organization to estimate both the benefits of a project and how Philippines." Land Economics 72 (4): 519-37. these benefits would change if different prices were charged. At the Cohen, David, and Jack L. Knetsch. 1992. "Judicial Choice and time that a contingent valuation survey is undertaken, the Disparities between Measure of EconomicValues." Osgood Hall researcher typically does not know the actual cost of the project, LawvJournal 30: 737-70. either because the cost analysis is being done simultaneously or Cropper, Maureen, S. Aydete, and Paul Portney. 1994 "Preferences because several different kinds of project or levels of service are for Life Savings Program: How the Public Discounts Life being considered. Thus the results of the contingent valuation sur- Savings Programs."Journal of Risk and ULicertainty 8: 243-65. vey may be used to inform the design process. Cummings, Ronald G., David Brookshire, and Wiliam Schulze, If the client organization has already decided which specific eds., 1986. Valuing Environmental Goods: .4n Assessment of the project will be implemented, it may also have decided in general Contingent Valuation Metliod. Totowa, NJ.: Rowman & terms what it will charge. In this case the client may use the results Allanheld. of the contingent valuation survey to get an accurate prediction of Davis, Jennifer. 1998. "Assessing Conununity Preferences for revenues. Using low referendum prices to gain accurate estimates of Development Projects: Does Mode Matter?" Ph.D. diss. 27 DALE WHITTINGTON University of North Carolina, Department of Environmental National Park, Bharatpur, India." Institute of Economic Sciences and Engineering, Chapel Hill, N.C. Growth, Delhi. Davis, Jennifer, and Dale Whittington. 1998. "Participatory Mitchell, Robert C., and Richard T. Carson. 1989. Using Surveys to Research for Development Projects: A Comparison of the Value Public Goods: The Contingent Valuation Miletlsod. Community Meeting and Household Survey Techniques." Washington, D.C.: Resources for the Future. Economic Development and Cultural Change 47 (1): 73-94. Moran, Dominic. 1994. "Contingent Valuation and Biodiversitv Diamond, Peter A., and Jerry A. Hausman. 1994. "Contingent Conservation in Kenyan Protected Areas." Biodiversity and Valuation: Is Some Number Better than No Number?"Journal Conservation 3. of Economic Perspectives 8 (4): 45-64. Pearce, David, Dale Whittington, Steven Georgiou, and David Diamond, Peter A., Jerry A. Hausman, Gregory K. Leonard, and James. 1994. Project and Policy Appraisal: Integrating Economics and Mide A. Denning. 1993. "Does ContingentValuation Measure Environment. Paris: Organisation for Economic Co-operation Preferences? Experimental Evidence." In J. Hausman, ed., and Development. Contingent Valuation:A Critical Assessment. Amsterdam: Elsevier Pinheiro, Armando, and Dale Whittington. 1995. "Introducing a Science Publishers. Demand-Side Approach to Rural Water Investment in Georgiou, Stavros, Dale Whittington, David Pearce, and Dominic Mozambique: A Rapid Appraisal of Household Demand for Moran. 1997. Economic Values and the Environment in the Improved Water Services in Marracuene." A Report to the Developing World. Cheltenham, U.K.: Edward Elgar. National Program for Rural Water Supply. Government of Grandstaff, S. andJohn Dixon. 1986. "Evaluation of Lumpinee Park Mozambique, National Directorate forWater, Maputo. in Bangkok, Thailand." In John Dixon and Maynard Poulos, Christine, and Dale Whittington. 2000. "Time Preferences Hufschmidt, eds., Economic Valuation Techniques for the for Life-Savings Programs." Envirotinmental Science and Teclhnology Environment: A Case Study WVorkbook. Baltimore, Md.: Johns 34 (8): 1445-55. Hopkins University Press. Rutherford, Murray B.,Jack Knetsch, and Thomas C. Brown. 1998. Hadker, N., S. Sharma, A. David, T. T Muraleedharan, S. Geetha, and "Assessing Environmental Losses: Judgments of Importance P Babu. 1995. Are People in Developing Countries Willing to Payfor and Damage Schedules." Harvard Environmental Law Review 22: Natural Resource Preservation? Evidencefrom a Contingent Valuation of 51-101. the Borivli .National Park, Bombay Bombay, India: Indira Gandhi Shyamsundar, P, Randy Kramer, and N. Sharma. 1993. Does Institute of Development Research. Contingent Valuation Work in NVonmarket Economies? Center for Kahneman, Daniel, Jack L. Knetsch, and Richard Thaler. 1990. Resource and Environmental Policy Research. Durham, N.C.: "Experimental Tests of the Endowment Effect and the Coase Duke Universitv. Theorem."Journal ofPolitical Economy 98 (1): 1325-48. Singh, Bhanxvar, Radhika Ramasubban, Ramesh Bhatia, John Kahneman, Daniel, and Jack Knetsch. 1992."Valuing Public Goods: Briscoe, Charles Griffin, and C. Kim. 1993. "Rural Water The Purchase of Moral Satisfaction." Journal of Environmental Supply in Kerala, India: Hoxv to Emerge from a Low-level Economics and AMianagement 22: 57-70. Equilibrium Trap." Water Resources Research 29 (7): 1931-42. Kanninen, Barbara J. 1995. "Bias in Discrete Response Contingent Swallow, B.M., and M.Woudyalew. 1994. "EvaluatingWillingness to Valuation." Journal of Environmental Economics and Management Contribute to a Local Public Good: Application of Contingent 28 (1): 114-25. Valuation to Tsete Control in Ethiopia:" Ecological Economics 1 1. Knetsch, Jack L. 1990. "Environmental Policy Implications of Vatn,Arild, and DanielW Bromley 1994. "Choices Without Prices Disparities Between Willingness to Pay and Compensation Without Apologies." Journal of Environmental Econonuics and Demanded Measures of Value:" Journal of Environmental Mlanagement 26 (2): 129-48. Economics and Alanagement 18: 227-37. Wang, Hua. 1997. "Treatment of 'Don't Know Responses' in MacRae, Duncan,Jr., and Dale Whittington. 1997. Expert Advicefor Contingent Valuation Surveys: A Random Valuation Model:" Policy Choice: Analysis and Discourse. Washington, D.C.: Journal ofEnvironmental Economics and Mvanagement 32 (2): 219-32 Georgetown University Press. Whittington, Dale. 1998. "Administering Contingent Valuation McClelland, Elizabeth. 1997. "The Use ofAlutiundinal Indicators in Surveys in Developing Countries." World Development 26 (1): Contingent Valuation Research: A Test of Validity Theoretic 21-30. Compatibility." Ph.D. diss. University of North Carolina, Whittington, Dale, Armando C. Pinheiro, and Maureen Cropper. Department ofCity and Regional Planning, Chapel Hill, N.C. 1996. "The Economic Benefits of Malaria Control: A Menkhaus, S. 1994. "Measurement of Economic and Other Contingent Valuation Study in Marracuene, Mozambique." Benefits of Wildlife Preservation: A Case Study of Keoladeo World Bank Draft Discussion Paper,Washington, D.C. 28 CHAPTER 14 ENVIRONMENTAL ISSUES Whittington, Dale, John Briscoe, Xinming Mu, and William and Sanitation for Health Project Field Report 246, US Barron. 1990. "Estimating the Willingness to Pay for Water Agency for International Development,Washington, D.C. Services in Developing Countries:A Case Study of the Use of Whittington, Dale, Donald T. Lauria, Albert Wright, KyeongAe Contingent Valuation Surveys in Southern Haiti." Economic Choe, Jeffrey A. Hughes, and Venkateswarlu Swarna. 1993. Development and Cultural Change 38 (2): 293-311. "Household Demand for Improved Sanitation Services in Whittington, Dale, Jennifer Davis, Harry Miarsono, and Richard Kumasi, Ghana:A ContingentValuation Study." Water Resources Pollard. 2000. "Designing a "Neighborhood Deal" for Urban Research 29 (6): 1539-60. Sewers: A Case Study of Semarang, Indonesia." Journal of Whittington, Dale, V Kerrv Smith, Apia Okorafor, Augustine Planning and Education Research 19 (3). Okore, Jin Long Liu, and Alexander McPhail. 1992. "Giving Whittington, Dale, Mark Mujwahuzi, Gerald McMahon, and Respondents Time to Think in Contingent Valuation Studies: KyeongAe Choe. 1988. "Willingness to Pay for Water in A Developing Country Apphcation."Jourmal of Environrmental Newala District,Tanzania: Strategies for Cost Recovery."Water Economics and MlSanagement 22 (3): 205-25. 29 , - ~~Fertility 1 5) lndu Bhushan and Raylynn Oliver Data from Living Standards Measurement Study (LSMS) surveys have been used extensively to study the determinants of fertility, contraceptive use, and child mortality and to inform policy on these issues. Data on fertility and related topics are often also useful for analyzing labor market and schooling decisions. Among past LSMS surveys there is considerable vari- for empirical analysis of these issues.The third section ety in the methods used for collecting fertility data and presents short and standard draft fertility modules, and in the criteria used for selecting respondents. Most the fourth section contains annotations to those mod- LSMS surveys have included a fertility module to col- ules and guidance to survey designers on how to cus- lect information on fertility, child mortality, contra- tomize modules to the circumstances in the country ceptive use, and reproductive health.' In the LSMS where their survey is to be fielded. surveys that did not contain a fertility module, the designers often included a few questions on these Policy Issues issues in the health module.Thus the fertility data col- lected in LSMS-type surveys can vary from just a few Public policy can influence fertility and child mortal- questions-usually about the number of children ever ity outcomes by affecting prices, resource constraints, born to the respondent-to detailed information on and even individuals' preferences. Some factors that the respondent's experience of maternity, child mor- influence fertility and child mortality-the availability tality, marriage, contraceptive use, and use of repro- of health services, and health conditions in the com- ductive health services. This chapter presents short and munity-are beyond the control of the household. standard fertility modules based on the cumulative However, most fertility and child mortality outcomes experience of using data from LSMS surveys and are determined by individuals reacting to opportuni- Demographic and Health Surveys to analyze fertility ties and to changes in their environment by delaying and child mortality. marriage, using contraception, and spacing births. The chapter provides an overview of fertility and Data from an LSMS-type survey can yield nation- child mortality issues and makes recommendations al estimates on levels, trends, and distribution of fertility about how to collect the necessary data in an LSMS or and child mortality outcomes. Analyses of the data can similar multi-topic survey. The first section discusses provide policymakers with key information on the the key policy issues related to fertility and child mor- determinants of fertility and child mortality and enable tality outcomes.The second identifies the data needed comparisons of various policy options. In particular, the 31 INDU BHUSHAN AND RAYLYNN OLIVER data can be analyzed to quantify relationships between Policy Issues Conceming Fertility policy variables and important fertility, child mortality, Policymakers need to address two different types of and contraceptive use outcomes. In conjunction with problems related to fertility. The first problem is that information on the costs of various policy alternatives, women may not bear the number of children they these quantified relationships can be used to identify want. A woman may bear more children than she wants policies that will most effectively achieve desired out- if she lacks basic reproductive knowledge, if she has no comes. Also, data gathered in the fertility module can be access to modern contraceptives, or if her status within used as control or explanatory variables in the analysis the household is too weak for her to assert her will. A of other aspects of household behavior, such as labor woman may bear less children than she wants due to force participation (Behrman and Wolfe 1984) or restrictive population policies or infertility. human capital investments (Pitt and Rosenzweig 1990). The second problem is that even if a woman has Information about the levels, trends, and distribu- the number of children that she desires, that number tion of fertility and mortality variables across socioe- may be too large (or too small) from the point of view conomic, geographic, and ethnic groups can be used of society as a whole. This may happen if there are to quantify demographic outcomes and thus set poli- externalities from childbearing decisions such that cy priorities. (See Box 15.1 for a list of important fer- other members of society bear costs (or reap benefits) tility and child mortality variables.) For example, when women bear children.2 while child mortality may not be high in a given In many countries throughout the world high pop- country as a whole, it may be a problem that needs to ulation growth is a significant concern. Therefore, in be addressed among certain socioeconomic groups or recent years most discussion and research has focused on in certain geographical areas. devising effective policies to remove the barriers that There can be substantial differences in the impor- prevent a woman from having fewer children and to tance of the many issues involving fertility and child create incentives for women to reduce their fertility. For mortality variables across regions of the xvorld (Table example, policies that encourage the schooling of girls 15.1). In many Sub-Saharan African countries high and increase income-earning opportunities for women fertility and child mortality rates are a major concern. increase the cost of a woman's time-possibly leading to In many countries in East Asia a preference for sons is reduced fertility. The impact of schooling levels and an important issue. And different countries within a employment opportunities on women's fertility region may have different policy priorities. depends on the prevailing economic and social condi- tions in a country (Cochrane 1979; Ainsworth 1989; Ainsworth, Beegle, and Nyamete 1995; Pitt 1995). In many countries with high fertility rates policy- Box 15.1 Fertility and Child Mortality Variables makers are especially interested in the impact of con- traceptive use. High levels of unmet needfor contraception The following fertility and child mortality variables are important to policymakers and can be collected in LSMS- m should me more thatnt-orientednnBngaarts type surveys: ~~~~~~es should become more client-oriented (Bongaarts * Family size, and Bruce 1995; Robey and others 1996). If contra- * Total fertility rate. ceptives are expensive, of poor quality, or unavailable, * Birth rate. introducing policies that affect the price, availability, * Contraceptive use. and quality of modern contraceptive methods can * Unmet need for contraception. remove an important obstacle to reducing fertility * Age at first marriage. rates (Feyistan and Ainsworth 1994; Thomas and * Age at first childbirth. Maluccio 1995; Oliver 1995; Beegle 1995). * Abortion rate. Another aspect of fertility that often interests pol- * Use of prenatal care. Another aspectertilt thoen terestsehol- * Breastfeeding practice. icymakers is the interaction between the household's * Spacing of births, decisions about family size and its other decisions, * Infant mortality rates (deaths before one year of age). including those about human capital investments such * Child mortality rates (deaths before five years of age). as expenditures on children's health and schooling. Such interactions are the subject of studies by Pitt and 32 CHAPTER 15 FERTILITY Table 15.1 Variations in Demographic Outcomes by Region, 1995 Total Infant Life Child fertility mortality expectancy mortality Population rate rate at birth rate Areas of potential Region (millions) (per woman) (1,000 births) (years) (1,000 births) policy concern Sub-Saharan Africa 583.3 5.7 92 52 157 High fertility * Large unmet need for contraception (spacing) * High-risk births * Reproductive health (adolescents) * High chiid/infant mortality * High adult mortality due to AIDS the Pacific . Strong preference for sons South Asia 1.,243.0 3.5 75 6.1 1 06 * High fertility * Large unmet need for contraception * High-risk births * High child/infant mortality * Strong preference for sons Central Asia * Low contraceptive use Middle East and 272.4 4.2 54 66 72 * High fertiiity North Africa * Large unmet need for contraception * High-risk births * High child/infant mortality Latin America and 477.9 2.8 37 69 47 Inequalities in access to family planning the Caribbean and health care Sourme:World Bank 1984, 1993 1997a and 1997b. Rosenzweig (1990), Montgomery, Kouame, and Reducing infant and child mortality is an objec- Oliver (1994), and Benefo and Schultz (1994). tive of virtually every developing country government. One fertility issue that has been difficult to influ- The policy options for achieving such a reduction ence through public policy is the preference of some include implementing appropriate medical and public parents (mostly in parts of Asia, the Middle East, and health interventions, improving water and sanitation North Africa) for male children.A distorted gender ratio services, and disseminating accurate information both is emerging in many East Asian countries, probably due within and outside the education system. Research on to strong parental preference for sons and a decrease in the determinants of infant and child mortality at the family size. Technologies that can identify the sex of the household level can reveal other policy interventions fetus and the consequent wide prevalence of sex-selec- that would reduce infant and child mortality, such as tive abortions have contributed greatly to this trend mother's schooling (Benefo and Schultz 1994). (Park and Cho 1995).The implications of this distortion in the gender ratio are only beginning to be understood. Policy Issues' Implications for LSMS Surveys The multisectoral nature of LSMS surveys makes them Policy Issues Conceming Mortality suitable for modeling the determinants of fertility and Since 1960 infant and child mortality rates have child mortality and related behavior. (The fertility and decreased steadily in developing countries sometimes mortality issues that can be analyzed using data from even in the absence of improvements in nutrition, LSMS surveys are summarized in Box 15.2.) Many housing, and income. Much of this decrease is attrib- other demographic issues, such as the abortion rate, uted to better medicines (particularly antibiotics), female genital mutilation, and child fostering, are also public health interventions such as immunization, important areas for policy research in certain coun- diarrhea control programs, and safe motherhood ini- tries. Questions to cover these issues are not included tiatives (World Bank 1993). However, levels of infant in the draft questionnaire module presented here, but and child mortality are still high in many countries, the addition of some specific questions would enable and most developing countries have high infant and an LSMS survey to collect the data necessary to ana- child mortality rates in specific population groups. lyze these issues. However, including such questions in 33 INDU BHUSHAN AND RAYLYNN OLIVER countries where they are not appropriate or relevant household characteristics and behavior to enable ana- could upset respondents and unnecessarily lengthen lysts to study the underlying causes of these levels and the interview. trends and how policy can influence these outcomes. Some limitations of LSMS data are described in Also, few Demographic and Health Surveys collect Box 1 5.3.The relatively small sample size of LSMS sur- data on community-level variables and services, though veys (2,000-5,000 households) makes it difficult to col- the inclusion of community questionnaires in these lect enough data for analysts to explore issues that are surveys is under consideration. In contrast, LSMS-type not common throughout the population. Dis- surveys collect data on a much broader set of explana- aggregating demographic rates by region or socio- tory variables at the household level, including land economic status and measuring rare events such as holdings, productive assets, mother's height, and par- maternal mortality require samples much larger than ents' schooling. They also collect community-level those typically used in LSMS surveys. Another limita- exogenous variables that are influenced by policy, such tion of LSMS data is that they usually do not contain as the availability, quality, and price of public services. enough variation in macroeconomic and policy vari- Thus LSMS survey data are suitable for analyzing ables to analyze how those variables affect fertility. the determinants of demographic outcomes and for During the planning stage, designers should con- informing policies. Analyzing demographic outcomes sider other existing data sources. A recent national using LSMS data can provide insights into household census or demographic and health survey can often behavior that will help policymakers decide on appro- provide very precise national (and sometimes region- priate policy interventions. al) estimates of fertility outcomes. Existing data on health and family planning facilities may make the col- Data Requirements for Policy Analysis lection of such data unnecessary. Even in countries where a Demographic and Analyzing the relationship between policy variables Health Survey has been carried out, it is usually still and demographic outcomes is complicated by the fact appropriate to include a fertility module in an LSMS that those outcomes are not directly chosen by indi- survey. While demographic and health surveys yield viduals. People make decisions about contraceptive use data on levels and trends for important demographic and immunization that affect the probability of a birth indicators, they do not provide enough information on or a death, but the actual number of births and deaths Box 15.2 Fertility and Child Mortality Issues That Can Be Analyzed Using LSMS Data Determinants of fertility and contraceptive use: specifically, the Effectiveness of various components of programs-such as out- roles played by reach services-relative to clinic-based services. * Schooling. * Income. Impact of cost recovery on the use of fomily planning and repro- * Price, availability, and quality of contraceptives. ductive health services. * Labor market conditions. * Information, education, and communication programs. Determinants of infont and child mortolity: specifically, the roles * Other socioeconomic factors. played by * Parents' schooling. Determinants of age at first childbirth and age at marriage. * Income. * Other socioeconomic variables. Quality-quontity tradeoff in demand for children * Public health interventions. * Interaction between child health or mortality and family * Water and sanitation programs. size or birth interval. * Interaction between child schooling and family size or Existence of gender gops in infont and child mortality. birth interval. Prevalence of breastfeeding. Efficient targeting of government programs through identification of underserved areas and groups. Other issues: abortion, female genital mutilation, child fostenng 34 CHAPTER 15 FERTILITY Epidemiological research can throw light on the Box 15.3 IssuesThatAre Difficult to Analyze with first question, the relationship between a behavioral LSMS-Type Data choice (such as an immunization) and a demographic * The impoct of mocroeconomic policy cnanges on demo- outcome (such as child mortality). However, appropri- graphic behavior, With LSMS data it is difficult to achieve ate economic models and econometric methods are enough variation in the macroeconomic variables for a needed to help policymakers understand the determi- given sample to measure macroeconomic policy nants of behavioral choices and the policies that may impacts. affect those choices. Unfortunately, in practice, * Levels of and trends in demographic variables for small geo- research is beset with conceptual and data-related graphical or socioeconomic categories of the population. problems. Some of these problems can be avoided by LSMS samples are too small to measure these levels and trends. National census or demographic and health sur- linkin polic re ctly to demographicou- vey data may be more appropriate sources for this comes. While this relationship, known as a reduced- information. form relationship, does not show exact causal mecha- * The impact of regulatory and legal reforms on demo- nisms, it can indicate the net effect of a given policy. graphic behavior. With LSMS data it is difficult to achieve To analyze household behavior and outcomes, econ- enough variation in the variables representing legal and omists use behavioral models that assume that households regulatory reforms across a sample. make rational decisions. (See Chapter 26 on economet- * The causes and dynomics of parents' preferences for sons. rics for a detailed explanation of this economic model More qualitative and anthropological research is and reduced form relationships.) Most economic models required to analyze this issue. * How living arrangements may hove changed due to dec/in- of fertility and mortality are based on a model in which ing fertility, chong'ng inheritance patterns, and migration to households are both producers and consumers, following urban areas. Analyzing these issues would require time- the seminal work of Becker (1960). In this framework series data over a long period of time. children and health are "produced" by the household using the time of its members and other inputs such as purchased food and health services. Households are is also influenced by chance and other factors beyond assumed to allocate their limited time and economic the control of individuals or households. Infertility is resources to maximize their welfare or utility.The house- one example; high prevalence of childhood diseases in hold derives utility from children and health just as it the community is another. does from other goods or services. For a given set of pref- Analysts can inform public policy by attempting erences (in other words, a utility function), prices, and to provide data that answer the following three ques- income, household production theory explains how tions (DaVanzo and Gertler 1990): households decide the number of children they want, the * How do specific behavioral choices affect fertility level of health status they want to maintain, and the bas- and mortality outcomes? ket of inputs they will use to achieve these goals. The * What determines these behavioral choices? household's demianid for children, healtlh, and inputs * How can these choices be influenced by policy? would change if changes occurred in market prices or in To design effective policies it is important to household preferences, income, or other household vari- know the answer to each of these questions. For ables.This model guides the analyst in selecting the vari- example, to address child mortality policymakers ables required for analysis. must first understand how the use of a particular medical technology-say, child immunization- Data Needs for FertilityAnalysis affects child mortality.Then they need to know what Public policy can be informed by the answers to the makes individuals decide to take advantage of related three questions shown above. In theory it should be interventions. For example, if knowledge about the possible to answer each question separately using availability of immunization services is found to be household survey data. In practice most research significantly related to use of these services, then a involves estimating reduced-form relationships for publicity campaign may be more effective than an which the fertility variable of interest is determined by expansion of service in increasing immunization a set of explanatory variables. This approach raises the rates. following questions: 35 INDU BHUSHAN AND RAYLYNN OLIVER * What fertility variables are of interest to policy- previous years. Second, cumulative fertility may be less makers in the country being studied? sensitive to policy changes than recent fertility. * Which explanatory variables should be included in Therefore, measures of recent fertility, such as the num- the estimation of the reduced form relationship? ber of children born in the previous five years, whether This section proposes a set of variables that should be the respondent is currently pregnant, or whether she collected as part of the fertility module of LSMS and has given birth in the previous vear, are often more similar multi-topic surveys. The variables are summa- useful for policy analysis. However, collecting complete rized in Box 15.4. birth histories from respondents increases the accuracy and completeness of the data and increases the analyt- FERTILITY VARIABLES OF INTEREST. The central phe- ical possibilities of the resulting data set.4 Moreover, a nomenon of interest in fertility research is fertility few questions are usually enough to find out the num- itself Fertility can be measured in two ways: cumula- ber of children ever born to a respondent. tive fertility and recent fertility. The most frequently Many women use modern contraceptive methods used variable for cumulative fertility is the number of to control their fertility. Because the most frequently children ever born.3 Although cumulative fertility can analyzed proximate (direct) determinant of fertility is be an appropriate dependent variable for some analyt- contraceptive use, this variable is of great interest to pol- ical purposes, there are two problems with this meas- icymakers. The use of contraceptive methods is under ure. First, cumulative fertility is influenced by socioe- the control of a woman (or another household mem- conomic factors through all of the respondent's ber such as the woman's husband) and is, therefore, an reproductive years prior to the survey, and data on endogenous variable. Several measures of contracep- many explanatory variables (such as household income, tive use can be investigated, including whether the community wage rates, and the availability of health woman has ever used contraception, whether she is and contraceptive services) may not be available for currently using contraception, and whether she has Box 15.4 ImportantVariables for Fertility Analysis Fertility outcomes Explanatory individual and household characteristics Cumulative fertility Mothers schooling * Children ever born Mothers age Recent fertility Household income and other household resources * Children born in the past five years Ethnicity and religion * Current pregnancy Schooiing of other household members * Birth in the past year Explanatory community characteristics Other fertility-related endogenous variables Region and degree of urbonization Contraceptive use Prices * Current use (method-specific) Local wage rates (for adults and children) * All current and previous use (method-specific) Food and non-food prices Age at first marnoge * Formal and informal interest rates Age at first childbirth Information, public education, communications Numnber of abortions Availability ofTV, radio, and newspapers Breastfeeding practice Use of media for family planning messages * Duration of breastfeeding for last birth Family planning informational campaigns * Current breastfeeding status Health services Length of postpartum abstinence * Price, availability, quality Current pregnancy status Family planning services . Price, availability quaii y Fertility preferences Schools Desire to hove additional children Price, availability, quality Desired fertility (ideal) Unmet need for contraception 36 CHAPTER 15 FERTILITY used contraception within the past 12 months. Also, makers, but these variables should not be considered depending on the policy options being analyzed, it exogenous explanatory variables in regression analysis. may be desirable to estimate the use of a specific method, the use of each method separately. or the use FERTiLITY PREFERENCES. Many economic models of of modern methods relative to traditional methods or fertility propose that parents' desired fertility-how no method at all. many children they would like to have-is an impor- Which variable is the dependent variable depends tant determinant of fertility. It is useful to collect data on on the objective of the analysis. For example, if the a household's desired fertility so that analysts can fore- aim is to examine how the availability of family plan- cast medium-term changes in fertility and measure the ning facilities affects contraceptive use, current use prevalence of unwanted births. Recently, fertility pref- may be the most appropriate dependent variable erence data have most often been used to analyze the because accounting for all current and previous use much-debated question of how family planning pro- might include contraception used before the facility grams affect fertility, focusing on the magnitude of in question existed. In countries where a variety of unwanted fertility and the unmet need for family plan- methods are available and rates of usage are rather ning (Pritchett 1994). However, whether and how high, use of individual methods may be analyzed. On desired fertility can be measured accurately are con- the other hand, use of any modern method could be tentious issues. Early attempts to measure fertility pref- estimated to examine contraception use in countries erences used questions about ideal family size, such as: with low utilization rates. It may be of interest to "If you could start over again, how many children measure households' "knowledge of" various meth- would you like to have? How many sons and how many ods if publicity campaigns have been part of recent daughters?" Such questions have a number of theoreti- public policy. cal and empirical shortcomings.The responses may be Again, contraceptive use data can be gathered in a ambiguous, as respondents may be reluctant to name an few questions or in a complete set of questions on indi- ideal family size smaller than the size of their existing vidual methods. Gathering more data requires more family. Furthermore, the risk of child mortality is not interview time but also provides analysts with more made explicit in these questions; presumably respon- options for research. Contraceptive prevalence and the dents did not factor children's deaths into their ideal most important policy issues in the country will deter- family size, although in practice they may need to bear mine the level of detail of data needed. more children than their desired family size because of In countries facing an AIDS epidemic, it may be the high rates of child mortality in many developing worthwhile to expand the questionnaire to gather data countries. In practice, many respondents provide non- on men's and women's sexual behavior-including numerical answers such as "It's up to God," which are number of sexual partners-as well as adult mortality. not useful in empirical work. Finally, respondents may This would be a substantial expansion of the ques- have compositional preferences (for example, wanting at tionnaire and is not, therefore, discussed further here. least one boy or at least one child of each sex) that affect For more information see Ainsworth (1992), family size but are not captured in questions about Dehenesse, Carael, and Noumbissi (1996), Cleland and desired family size. Demographic and Health Surveys Ferry (1995), and Filmer (1997). See also the discus- typically ask questions about whether the respondent's sion of the health module introduced in Chapter 8. last child or current pregnancy was wanted, but the Other variables related to fertility analysis are age at answers to these questions involve many of the same first marriage, current breastfeeding practice (or the problems. Also, these questions yield no information length of time for which the last child was breastfed), and about women who have never been pregnant. Another number of abortions. Demographers identify these vari- approach is asking about the respondent's desire to have ables as proximate determinants of fertility. While it is additional children, which may yield information on tempting for many analysts to use these variables direct- potential demand for contraception or for future fertil- ly to explain fertility, they are clearly the result of house- ity. However, the responses would still be subject to hold decisions. Analysts may wish to estimate their many of the problems listed above. determinants in order to explain various aspects of fer- The unmet need for contraception is the differ- tility that may of themselves be of interest to policy- ence between women's stated fertility preferences and 37 INDU BHUSHAN AND RAYLYNN OLIVER their actual contraceptive behavior. In the demo- some LSMS surveys collect household income data, graphic literature women who are sexually active and but all LSMS surveys collect household expenditure fertile are categorized as having an unmet need if they data, and total household expenditure can serve as an are not currently pregnant or amenorrheic, do not indicator of household income. (Collecting and using want any children, and are not using contraceptives. expenditure and income data are discussed in detail in Women who are currently pregnant with an unwant- Chapters 5 and 17.) A household's income is the result ed pregnancy and amenorrheic women who did not of its decisions about the labor market participation of want their last child are also classified as having an its mem1bers, which are made jointly with fertility unmet need.To measure unmet need as defined above, decisions. To use household income as an explanatory it is necessary to find out three aspects of fertility pref- variable, it may be necessary to use data on nonlabor erences: the respondent's preference concerning her income of household members or other instrumental filture childbearing, whether her last child was want- variables that are exogenous to the fertility decision. ed, and whether her current pregnancy is wanted. (Instrumental variables and related econometric issues These questions suffer from the weaknesses listed are explained in Chapter 26.) above. In addition, it is important to stress that when a Parents' birthplace or place of residence may be woman has an unmet need for contraception, it can- correlated with their fertility desires.Variables that sig- not necessarily be assumed that she has an unmet nify ethnicity and religion can also be included to demand for contraceptive services. Even if she has a control for the effect of household preferences on the strong desire to have no more children, a woman may demand for children. It is important to collect infor- not want to use contraceptives because of financial, mation on the religious affiliation and ethnicity of religious, or other constraints. households. The schooling of household members may also influence their preferences for number of EXPLANATORY VARLABLEs. One of the most important children. policy variables in fertility analysis is mother's school- An important community-level variable is degree ing, as this variable represents the opportunity cost of of urbanization. This is an indicator of the general cost having children-in other words, the potential house- of living, availability of services, economic activity, and hold income lost during the time that the woman prevailing cultural mores. Local prices of essential spends caring for her children. In addition, years spent goods and local wage rates may also be important in school are usually years that are not spent bearing determinants offertility outcomes. However, to be use- children. Many studies suggest that schooling may also ful, the data on prices must not vary with the quantity have a socializing influence. Different rneasures of demanded by the consumer; otherwise the data will be schooling, such as years in school or highest level com- contaminated by consumers' purchasing choices pleted, may capture the different ways schooling affects (Schultz 1984). It is best to collect information about fertility decisions. Which measure of schooling to wages and prices directly in community-level surveys, choose depends on the objective of the analysis and on although if necessary wage and price information can the data available. Diplomas or literacy, rather than be obtained indirectly by aggregating household-level years in school, may make an important difference in information. (See Chapter 13, on community and people's labor market decisions and income opportu- price data, for details about how this information nities. The highest level of schooling attained by a should be collected.) woman may reflect the socializing influences of Perhaps the most important variables for policy schooling better than the number of years spent in analysis are those pertaining to the price, availability, school. (See Chapter 7 for a detailed discussion of how and quality of schools, health services, and family plan- to measure schooling.) ning services. Prices have a direct effect on how much Obviously, the number of children ever born the services are used by the household. The availabili- depends in part on mother's age. Since it is standard ty and quality of services also affect how much they practice to collect age data for all household members in are used because these factors help determine servic- LSMS surveys, this variable should always be available. es' costs and benefits to a household. Data on prices, In economic models, household income is an availability, and quality should be gathered in facility important determinant of household decisions. Only or community questionnaires. To be useful for analy- 38 CHAPTER 15 FERTILITY sis, the data must be available for all households, not use information from men. Collecting this informa- just those that use the services. (Chapter 13, on com- tion from men can be problematic for several reasons. munity and price data, describes ways of collecting this In some countries men frequently are not members of information and illuminates difficulties involved in the same household as their wives, and the fertility doing so. Chapters 7 and 8 present questionnaires preferences and contraceptive practices of a husband about schools and health care facilities, respectively.) may not be directly related to the fertility of his wife. One problem with using community-level data to Demographic and Health Surveys sometimes understand the determinants of fertility outcomes is include questions about male knowledge, attitude, and that community services may be affected by unob- use. If it is important for LSMS data to be comparable served community characteristics that also directly with Demographic and Health Survey data, it may be affect fertility outcomes. For example, communities necessary to administer the fertility preference and that emphasize the quality of their children's lives (as contraceptive use sections of the LSMS fertility mod- opposed to the quantity of children) may better attract ule to men as well as to women. This would signifi- public or private medical or family planning facilities cantly increase the length of the fertility module and because of higher demand or greater willingness to should only be considered when there are obvious pay. This would overestimate the impact that these benefits from comparability. In countries with a high facilities have on fertility; the true impact in this case prevalence of AIDS, condom use is an important issue would be the other way around-the effect of reduced independent of fertility issucs; it may be useful in thcse fertility on the attraction of such facilities. Correcting countries to gather information on condom use from for this problem is difficult and often requires panel all sexually active adults. data. (For examples of such corrections see Rosenzweig andWolpin (1986) and Pitt, Rosenzweig, Data Needs for Child Mortality Analysis and Gibbons (1993).A more detailed discussion of this The data requirements for analyzing mortality policies general issue is found in Chapter 23 on panel data.) depend on which population group is of interest to Nevertheless, LSMS-type surveys should attempt to analysts and policymakers. Some studies may focus on gather community-level data on the range and quality infant mortality while others may focus on child mor- of the available services, including the number of tality.While each of the three general questions posed hours and days during which the services are available, above about economic demography research can be the quality and type of services provided, the qualifi- answered separately using household survey data, cations of the service providers, inventories of drugs almost all relevant policy questions can be answered by and medicines, and the number of years that the serv- estimating reduced form relationships for which the ices have been operating (see the health care facility mortality variable of interest is determined by a set of questionnaire introduced in Chapter 8). exogenous explanatory variables. A final set of policy-relevant community-level data As with fertility, this approach raises two ques- that can be collected in LSMS surveys relates to infor- tions: mation, public education, and communications activi- * What mortality variables are of interest to analysts ties, including variables such as the availability of TV, and policymakers? radio, and newspapers in the community; the use of * Which explanatory variables should be included in media for fanily planning messages; and the information the reduced form relationship? available on family planning. Public information cam- This section proposes a set of variables that deter- paigns operate on several levels. Politically, they help set mine mortality-variables that should be collected as the policy agenda and elicit the commitment of politi- part of the fertility module of LSMS and similar multi- cal leaders. lnstitutionally, they can train service providers topic surveys.5 The variables are summarized in Box and provide individuals with information about available 15.5. services and changes in public attitudes and preferences. Some analysts are also interested in men's knowl- CHILD MORTALITY VARIABLES OF INTEREST. Even in edge, attitudes, and practices regarding reproduction countries with high levels of infant and child mortali- and contraception. However, most previous LSMS ty, mortality is usually a rare event.Thus it is important surveys have not gathered fertility and contraceptive to understand how to measure it accurately. It is usu- 39 INDU BHUSHAN AND RAYLYNN OLIVER ally necessary to analyze cumulative mortality over a relatively long time. As a result, child mortality experi- Box 15.5 ImportantVariables for Mortality Analysis enced in the entire lifetime of- a mother (either in Mortality outcomes terms of the total number of her children who died or Number of children thot died (under one year, under five years) the proportion of all her children who later died) is Proportion of children ever bom thot died (under one year often used as the dependent variable. Age of children under five years) at death is crucial in defining the dependent variable Age of children at death for mortality analyses.The definition of the dependent variable depends on whether neonatal, infant, or child Other mortality-related variables mortality is to be analyzed. Inaccurate recording of the High-risk births (very young or old mother short birth interval) age at death, in particular the "heaping" of observa- Low birth weight Use of and expenditure on maternal and child health care tions at one month and one year, makes it difficult to Water source, use of woter expenditure on water measure infant and child mortality. Health practices (woshing hands, storing food, use of drugs, Variables related to mortality includc: high-risk alcohol, and tobocco) births (births to very young or very old mothers and Delivery circumstonces births for which the interval between births was Child immunization short), low birth weight, use of and expenditure on maternal and child health care, delivery circumstances Explanatory individual and household characteristics (such as where the baby was delivered, what medical Household composition Age, sex, and schooling of household members personnel attended the birth, and xvhether there were Household income and other household resources any complications during the delivery), and child Ethnicity and religion immunization. High-risk births can be measured from Housing characteristics %water source, toilets, sanitary condibons) the data on mother's age and birth history that are col- lected in the fertility module. Fertility modules can Selected community characteristics also collect information on the use of prenatal health Region and level of urbanization care and breastfeeding, but only for the most recent Environment births.6 Data on respondents' use of and expenditure . Climate (rainfall, temperature, and altitude) o Prevalence of major diseases on health care are often collected in the health mod- Water supply ule, but only for the past 4 weeks or the past 12 General sanitary conditions (sewage facilities and waste months, and not for children who have died. Data on disposal) the circumstances of the delivery could also be col- Government programs lected in the fertility module, but, again, only for the * Spray for malaria most recent births. Immunization data could be col- * Purfication of water sources lected for all children in the birth history section. Information, public education, communicotions Hoxvever, again, the accuracy of data is questionable Availability ofTV, radio, and newspapers ' . ~~~* Use of media for health messages for children born more than three years prior to the UsCommuiation campaigns ; r *~~~~ Communication campaigns survev-especially for children who died young. Pnces * Local wage rates EXPLANATORY VARIABLES. Virtually all of the variables Food and non-food prices that determine fertility also determine mortality, Formal and informal interest rates including age, sex, and schooling of household mem- Heaith services bers; household income and other household * Availability, type, quality, and fees resources; and ethnicity and religion. Additional explanatory variables for mortality include health would not be difficult to include questions on both of practices in the home and households' water sources. these topics in the housing and health modules.7 While questions about sources of drinking water have In general, variables related to environment are often been collected in the housing module of previ- not under the behavioral control of household mem- ous LSMS surveys, questions about households' health bers unless the members migrate. Environmental vari- habits have rarely been included. In future surveys it ables include climate (rainfall, temperature, and alti- 40 CHAPTER 15 FERTILITY tude), water supply, prevalence of major diseases, and tial for estimating the determinants of infant and child general sanitary conditions (sewage facilities and waste mortality: disposal). These kinds of data can be gathered in the a How far an individual has to travel to obtain serv- community questionnaire, although some may have to ices (distance and travel time). be collected from other sources, such as meteorologi- * How frequently services are available (days per cal records. (For details see Chapter 13, on the com- month, hours per day). munity and price questionnaires.) * What types of services are available (primary care, Environmental variables usually do not vary across maternal and child health care). small geographical areas. However, they may vary con- * What types of procedures can be performed at the siderably over time. Therefore, environmental data service center (for example, major surgery or cae- should be collected each month or each season, and sarian sections). perhaps over more than one year. At a minimum the . Quality of service providers (qualifications and community questionnaire should ask about two differ- experience, behavior toward patients). ent seasons, such as the wet season and the dry season. * Quality of the facility (cleanliness, electricity, water However, the structure of the LSMS-type surveys and quality, refrigeration). the limits of empirical work make it difficult to * Whether emergency transport is available. include environmental data that correspond to the * Whether essential drugs and supplies are available time and place of each child's life and death.This lim- and properly stored. itation must be borne in mind when interpreting the While some of this information, such as travel estimated impact of environmental variables. times and hours that a facility is open, can be collect- Government programs that provide health servic- ed in the community questionnaire, most of the es or subsidize the prices of medicine and services information can only be collected in a health facility affect household behavior by changing the cost to questionnaire. (See Chapter 8 for a draft facility ques- households of using these services. Government- tionnaire and a detailed discussion of how to use designed public education programs can also affect information from such a questionnaire in empirical behavior, encouraging people to behave in ways that work.) are good for their health by changing perceptions of Data on public health programs (such as cam- the benefits of various actions. paigns to encourage childhood immunization, diar- Policymakers are often interested in the impact that rhea management, and safe motherhood) are also these government programs have on household behav- important policy variables for mortality analyses. ior. But measuring this impact is complicated by the Although these programs are implemented through fact that such programs are rarely located randomly existing health care facilities, data about them are best throughout the community but instead have been collected in the community questionnaire; further placed in specific areas-for example, in areas where information can be collected in a health facility ques- need appeared to be greatest. If the programs are locat- tionnaire. ed in places with the worst health problems, analysts may see a negative correlation in the general population Draft Modules between health and the existence of government pro- grams. If services are concentrated in affluent neighbor- This section provides two draft modules for collecting hoods for political reasons, a measured correlation may data on fertility and child mortality in LSMS surveys: capture the effect of high income on health and thus a standard module and a short module. Due to the overstate the impact of the health services. To accurate- multisectoral nature of the surveys, the fertility mod- ly measure the impact of government programs, analysts ule cannot be as detailed as it might be in a single pur- need to have some information about the criteria used pose survey.The recommendations in this section have to distribute government services. (For a more detailed been made with the following strictures in mind: discussion, see Chapter 8 on health, Chapter 26 on * The design of the fertility module should ensure a econometrics, and Chapter 23 on panel data.) minimum level of comparability with data from The following data on the health services provid- previous surveys in the country, especially previous ed by both the public and the private sector are essen- LSMS and Demographic and Health Surveys. 41 INDU BHUSHAN AND RAYLYNN OLIVER * The information collected should be relevant for of the questionnaire. If any of the latter three modules policy analysis. are abridged significantly, questions may need to be * Information unlikely to be used often in analysis added to the fertility module to ensure that the neces- should not be collected. sary information is collected for women of childbear- * The module and its methodology should be ing age. The reader may refer to the list of explanato- designed to ensure that the data collected are of ry variables in Boxes 15.4 and 15.5 to verify that the high qualitvy other modules in the questionnaire collect the infor- The standard module is organized in three sec- mation needed to analyze fertility and mortality. tions: maternity history, reproductive health, and con- In most previous LSMS surveys the fertility module traceptive use. The short version omits the questions has been administered to women ages 15 to 49 (inclu- about breastfeeding and abortion and the detailed sive). Survey designers can modify these criteria for the questions on methods of contraception. fertility module if conditions in the field allow only mar- The length of the questionnaire is an important ried women to be interviewed or if it is clear that there consideration in survey design. Table 15.2 presents esti- is substantial sexual activity and childbearing among girls mates of the number of questions per woman for the younger than 15. In many previous LSMS surveys the short and standard versions.8 Estimates are also present- fertility module has been administered to one randomly ed on the questions per household depending on selected woman of childbearing age in each household. whether one woman or all eligible women are surveyed. Doing this reduces interview time but also results in In most cases the module will need to be modi- women in large households being underrepresented in fied to ensure that the questions asked and the infor- the resulting data.9 Alternatively, the fertility module mation provided are appropriate to the prevailing cir- cumstances in the country where the survey is to take Box 15.6 CautionaryAdvice place. Box 15.6 presents cautionary advice on the extent to which the draft module has not been proven How much of the draft module is new and unproven? in the field and on which sections will require the Most of the draft module has been used in the major- most customization. The following section of this ity of past LSMS surveys. Part B on Reproductive chapter provides detailed comiments and explanatory Health is a slight expansion of what was usually includ- notes to gnide survey designers in modifying the draft ed (most previous LSMS surveys gathered data only on notesrtoi gwdelsurveyidesignersym module.g the draft each woman's most recent birth. The questions on fertility module. desired fertility (Questions 21, 25 and 32-34 in Part A) have not been used in previous LSMS surveys, but they Explanatory Notes on Design of the Draft have been used in many Demographic and Health Fertility Module Surveys. How well has the module worked in the past? The data The household roster of the LSMS survey is organized from the fertility module have been used widely. so that all xvomen in the household can be linked to Including all births in the past three years in the repro- the head of the household. Each woman can also be ductive health section should increase its usefulness. linkehed tof her housbandond, Eahen toancard vrsion bData on breastfeeding is difficult to use and interpret linked to her husband and, when the standard version for reasons discussed in the final section of the chapter of the household roster is used, to her children if they Which parts of the module need to be customized? are also members of the household. In addition, data Throughout the module, care must be taken to ensure about women that are gathered in the fertility section that the language used in the questions is sufficiently can be linked to data about the same women collect- precise to elicit accurate responses without making the ed in the education, health, and employment modules respondent unnecessarily uncomfortable. Questions 26 and 27 on first sexual intercourse are the most Table 15.2 Number of Questions in Fertility Module obvious example. Similarly, in some situations it may be unnecessary or impossible to ask questions on abor- Questions One woman All eligib e tion (Questions 30 and 31). The questions on contra- Version per woman per household women ceptive use will require the most customization. Standard 26620.0 28.5 Methods and sources should include only those avail- Short 16.9 12.7 8.2 Source: Authors ca culations us ng data derived from Macro Internatonal nc. able in the country. I992a and 992b. 42 CHAPTER 15 FERTILITY could be administered to all women of childbearing age questions are really "no." It is conceivable that inter- in a household. To do this, multiple copies of the fertility viewers might purposely record a negative response for modules should be included in the questionnaire or addi- these questions in order to avoid administering the sec- tional fertility modules should be made available to inter- tions that follow. Therefore, supervisors should pay spe- viewers. In practice, interviewing two or three eligible cial attention to checking these questions. women per household would address sampling consider- ations and be much simpler than including all eligible A8. The question asks for the name of the child even women. Survey designers might wish to refer to a recent if the child is dead, since the association of a name census to estimate the number of copies of the module makes it easier for the child's mother to recall the that would be required for the average household. details asked for in the questionnaire. In addition, it is The draft module presented in the previous sec- easier for the interviewer to refer to a child by name tion was designed to be administered to women. If in the later parts of the questionnaire. It is recom- data on knowledge about, attitudes toward, and use of mended that the maternity history should be record- contraceptives are also to be collected from men, this ed starting with the firstborn child. Mothers find it can be done by duplicating Part C of the module, on easier to recall various details if the maternal history is contraceptive use, and perhaps the fertility preference discussed from the first birth to the latest birth, rather questions from Part A of the module. than starting with the latest birth and going back to The rest of this section describes the different the first birth (Shyrock and Siegel 1976). parts of the draft fertility module and clarifies the design and purpose of many of the questions. All. Supervisors should instruct interviewers never to leave the question on date of birth blank. Even if the Part A: Matemity History respondent does not remember the child's exact date This section contains questions about each birth, of birth, interviewers should help them recall the year, including the child's name, sex, birth date, survival sta- and hopefully the month, by providing some impor- tus, and age at death if the child died. The section also tant reference dates and by asking them if they collects the ID code of all children living in the house- remember in what season of the year the birth took hold and the highest level of schooling for all children place. Survey designers should prepare a list of refer- who are not household members. (Schooling for chil- ence dates relevant to the country for which the sur- dren who are household members can be obtained vey is being planned.l" from the education module.) It is important to acquire a complete maternity A13. This question collects information on the age at history because this provides analysts with information which children died. In order to increase the accuracy not only about a woman's cumulative fertility but also of measurement of infant and neonatal mortality, the about her recent fertility and infant mortality. Survey time units used vary according to the age at death. designers need to ensure that: * All live births and deaths are identified. A16. This question on the highest level of schooling * The number of children ever born can be recon- attained by the child is included in the birth history to ciled with the answers given in the maternity his- ensure that this information is collected for children tory section. who do not live in the household.This information is * The date of each birth is recorded. essential for analyzing issues related to the "quality- * The age of death for each child that died is recorded. quantity tradeotf." The list of applicable schooling codes should reflect the local education system and A4-A7. These questions determine whether a woman should be the same as those used in the education should answer the maternal history and reproductive module. For children who do not live in the house- health questions. In order to ensure that all pregnancies hold, the ID code can be used to obtain the same are counted (even those that ended in an early mtiscar- information from the education module. riage) and all births are counted (even if the child lived only a short time), interviewers should be trained to A19. After recording the maternal history, interview- probe in depth to find out whether the answers to these ers should count all of the recorded births and deaths 43 INDU BHUSHAN AND RAYLYNN OLIVER and use question A20 to confirm that no birth is issues, questions about these issues have been exclud- missed. ed from the short module. In some countries, respon- dents will consider any questions on abortion inap- A21, A25, AND A32-A34. These questions collect propriate; in these countries such questions should be information on women's fertility preferences regard- excluded from the questionnaire. ing their last child, last pregnancy, and future fertility. Analysts need the answers to these questions to esti- Part B: Reproductive Health mate unmet need for contraception. However, some This section should be administered only to women or all of the questions could be excluded depending who have had at least one live birth in the previous upon the policy priorities of the country where the three years. It contains questions on prenatal care, assis- survey is to be fielded. This is a section of the ques- tance during delivery, the place of delivery, and breast- tionnaire that will require extensive field testing to feeding. In most previous LSMS surveys, such ques- vernfy that the questions are easily understood and tions were asked about the last live birth before the worded sufficiently delicately. If"I don't know" or "It's survey.The draft module presented earlier gathers data up to God" are common responses, these questions on up to three births and on the woman's utilization may not elicit useful information, and including them of prenatal and postnatal care services-allowing ana- will only frustrate both interviewers and respondents. lysts to study the effectiveness and utilization of those services. Information is collected only for births in the A23. This question deals with a woman's current preg- previous three years in order to limit the interview nancy status, which is an important piece of information time, and because it may be harder for respondents to for many types of demographic analyses. For example, in remember such details about earlier births. studying the determinants of current contraceptive use, analysts may want to exclude pregnant women. B3-B4. These questions ask about prenatal care. In order to gather the information requested in this sec- A27. This question has been included to determine the tion, interviewers should be trained in the definition age at which the woman began having sexual relations. of prenatal care so, if necessary, they can clearly explain The exact wording of this question should reflect exist- this concept to respondents. ing cultural norms in the country. In some cultures mar- riage does not take place until after the birth of the first B5-B6. For questions on the place of birth and who child, so "age at marriage" would not be the relevant assisted at the birth, the lists of responses should be question. Field testing will reveal the appropriate way to modified to reflect the full range of local options. phrase the question. It may appear inconsistent to ask questions about marital status near the end of the mod- B7-B9. These questions deal with breastfeeding, ule, after having asked about childbearing. However, which not only contributes significantly to infant experience has shown that collecting information about health but also delays the return of the mother's the respondents' age at the time of their first marriage at menses. The information collected in questions B7- the beginning of the survey may embarrass respondents B9 is useful for determining the prevalence of breast- and thus hamper the smooth flow of the interview. feeding. However, an accurate measure of the nutri- tional value of breastfeeding would require A28-A31. These questions on miscarriages and abor- information on weaning foods and more detailed tions, can yield useful insights into these aspects of fer- information on breastfeeding. Moreover, deducing tility. Questions about abortions are included because the contraceptive effect of breastfeeding is very diffi- abortions reflect women's fertility preferences. In some cult and would require much more detailed informa- countries, survey designers may wish to include more tion. And the quality of data on breastfeeding is sus- abortion-related questions, such as the type of facility pect because it may be difficult for respondents to where the abortion took place, the type of provider, recall exactly when the child was completely weaned. the length of the pregnancy before the abortion, and It may be advisable to drop these questions from the how much the procedure cost. However, because in module. This entire section has been removed from many countries these will not be the most important the short module. 44 CHAPTER 15 FERTILITY Part C: Contraceptive Use 4. Survey designers may also wish to collect sucl informsation This section should be administered to all respondents. from women who are past childbearing age. This issue is discussed It should gather information on each woman's knowl- later in this chapter. edge of, current use of, lifetime use of, source of, and 5. In general, past LSMS surveys have not included a separate payment for contraceptives. This section will almost mortalit module. Data on infant and child mortality are best col- certainly need to be modified to reflect the conditions lected in the same place as data on fertility, since analyses of fertil- in each particular country. As the draft module collects ity should include children who died at an early age. Deaths among information about contraceptive use by method, the school-age children are very rare, so rare that the relatively small list of methods included should reflect those that are sample sizes of LSMS-type surveys effectively preclude using these used and available within the country where the sur- data to analyze mortality of school-age children. vey is to be fielded. In countries where contraceptive 6. Information is collected only for the most recent births use is exceptionally low, it will not be useful to ask all because, generally, respondents' ability to recall details of all births is of the questions about each method. In some coun- very limited. tries it will suffice to ask about a woman's knowledge 7. See Chapters 8, 12, 13, and 14 for detailed discussion of the and use of modern versus traditional methods. In oth- information collected in the health, housing, community, and envi- ers it will be sufficient to ask about the woman's ronment modules of the survey. knowledge and use of any method, with a follow-up 8. Estimated number of questions is calculated using the question asking her to specify which, if any, method responses to similar questions in the Egypt and Ghana she uses or has used. If contraceptive use is widespread Demographic and Health Surveys.There are fewer questions in the and policymakers wish to know more about it, the "one women per household" column because some households survey may need to collect more detailed information have no women of reproductive age. on, for example, the length of time for which a 9. Evidence from Sub-Saharan African countries shows that woman used each method, her complete contraceptive randomly selecting one woman from each household reduces the use history, or more detailed information about her interview time by about 30 percent without biasing the descriptive source of, expenditure on, and willingness to pay for analyses. Howvever, it considerably reduces the size of the sample, contraceptives. especially in the case of young women, for whom the reduction Questions relating to a woman's knowledge of may be as large as 50 percent. Thus, if resources permit. it is best to contraceptives may require interviewers to prompt interview all eligible women in each household. respondents. In some cultures women may be embar- 1o. In AnnexV, Grosh and Munoz (1996) present an example rassed to acknowledge that they know about contra- of a list of historical events of local importance to which inter- ceptives. Also, some women may not understand ques- viewers can refer to help respondents identify the date of a partic- tions about contraceptives. The evidence from ular event. Demographic and Health Surveys is that reported knowledge of contraceptive methods is significantly References higher when respondents are prompted. Ainsworth, Martha. 1989. Socioecontomic Determinants of Fertility in Notes Cote d'lvoire. Living Standards Measurement Study Working Paper 53.Washington, D.C.: World Bank. 1. Data collected in a standard LSMS survey are not appropriate for . 1992. Mleasuring the Impact of Fatal Adult Illness in Sub- analyzing adult or adolescenit mortality See Aiisworti (1992) for a Salharan Africa: An Annotated Houselhold Questionnaire. Living description of an LSMS survey on adult mortality conducted in Africa. Standards Measurement Study Working Paper 90.Washington, 2. If high population rates put pressure on public resources, the D.C.: World Bank. number of children desired by households may be higher than is Ainsworth, Martha, Kathleen Beegle, and Andrew Nyamete. 1995. socially optimum. On the other hand, the desired number of chil- The Inmpact of Female Schooling on Fertility and Contraceptive Use: dren be less than is socially optimum. For example, pension A Study of 14 Sub-Saharatu Countries. Living Standards schemes may become insolvent if the ratio of future pensioners to Measurement Study Working Paper 110. Washington, D.C.: future workers is high. World Bank. 3. Children ever born is the number of children born live to a Beegle, Kathleen. 1995. The Quality and Availability of Fanmily wooman, including children who died after childbirth. Planning Services and Contraceptive Use in Tanzania. Living 45 INDU BHUSHAN AND RAYLYNN OLIVER Standards Measurement Study Working Paper 114. Family Planning Situational Analysis Study. New York: The Washington, D.C.: World Bank. Population Council. Becker, Gary S. 1960. "An Economic Analysis of Fertility" In Grosh, Margaret, and Paul Glewwe. 1995. A Guide to Living National Bureau of Economic Research, ed., Demographic and Standard MTeasurement Study Surveys and Their Data Sets. Living Economic Changes in Developing Countries. Princeton, NJ.: Standards Measurement Study Working Paper 120.Washington Princeton University Press. D.C.:World Bank. Benefo, Kofi D., and T. P. Schultz. 1994. Determinants of Fertility and Grosh, Margaret, and Juan Munoz. 1996. A Mlanualfor Planning and Chlild Alortality in Cdte d'Ivoire and Ghana. Living Standards Implementing the Living Standards Measurement Study Survey Measurement Study Working Paper 103. Washington D.C.: Living Standards Measurement Study Working Paper 126. World Bank. Washington D.C.:World Bank. Behrman, Jere R., and B. Wolfe. 1984. "Labor Force Participation Macro International Inc. 1992a. "Demographic and Health Survey: and Earnings Determinants for Women in the Speci3l Egypt." Conditions of Developing Countries." Journal of Development . 1992b. "Demographic and Health Surveys: Ghana." Econormics 15 (1-2-3): 259-88. Montgomery, Mark, Aka Kouame, and Raylynn Oliver. 1994. The Behrman, Jere R., Mark R. Rosenzweig, and Paul Taubman. 1994. Tradeoff between iNumbers of Children and Child Schooling: "The Endowment and the Allocation of Schooling in the Evidence from Cite d'lvoire and Ghana. Living Standards Family and in the Marriage Market: the Twin Experiment." Measurement Study Working Paper 112. Washington, D.C.: Journal of Political Economy 102 (6): 1131-74. World Bank. Bongaarts,J., and J. Bruce. 1995."The Causes of Unmet Need for Oliver,Raylynn. 1995. Contraceptive Use in Ghana:The Role of Service Contraception and the Social Content of Services." Studies in Availability, Quality, and Price. Living Standards Measurement Famiily Planning 26 (2): 57-75. Study Working Paper 11l.Washington, D.C.:World Bank. Cleland, J., and Benoit Ferrn, eds. 1995. Sexual Behaviour and AIDS Park, Chai B., and N. Cho. 1995. "Consequences of Son Preference in the Developing World. London: Taylor & Francis. in a Low-Fertility Society: Imbalance of the Sex Ratio at Birth Cochrane, Susan. 1979. "Fertility and Education: What do We in Korea." Population and Development Review 21 : 59-84 Really Knowv?" World Bank Staff Occasional Paper 26. Pitt, Mark. 1995. Women's Schooling and the Selectivity of Fertility and Washington, D.C.. Child Alortality in Sub-Saharan Africa. Living Standards DaVanzo, Julie, and Paul Gertler. 1990. "Household Production of Measurement Study Working Paper 119. Washington, D.C.: Health: A Microeconomic Perspective on Health Transitions." World Bank. A RAND Note. RAND Corporation, Santa Monica, Cal. Pitt, Mark, and Mark Rosenzweig. 1990. The Selectivity of Fertility Dehenesse,J., M. Carael, and A. Noumbissi. 1996. "Socioeconomic and the Determinants of Human Capital Investments: Parametric and Determinants of Sexulal Behaviour and Condom Use: Analysis Sem iparametric Estimates. Living Standards Measurement Study of the WHO-GPA Surveys." In Martha Ainsworth, Lieve Working Paper 72.Washington, D.C.:World Bank. Fransen, and Mead Over, eds., Confronting AIDS: Evidencefrom Pitt, Mark, Mark Rosenzweig, and Donna M. Gibbons. 1993. "The the Developinig World-Selected Background Papers for the World Determinants and Consequences of the Placement of Batik Policy Research Report Confronting AIDS: Public Priorities in Government Programs in Indonesia." The World Bank Economic a Global Epidemic. Washington, D.C.:World Bank. Review 7 (3): 319-48. Feyisetan, Bamikale J., and Martha Ainsworth. 1994. Contraceptive Pritchett, Lant H. 1994. "Desired Fertility and the Impact of Use and the Quality, Price, and Availability of Family Planning in Population Policies." Population and Development Review 20: 1-56. Nigeria. Living Standards Measurement Study Working Paper Robey, B., J. Ross, and I. Bhushan. 1996. "Meeting Unmet Need: 108. Washington, D.C.:World Bank. Newv Strategies." Population Reports Series J-43. Johns Filnier, Deon. 1997. "Socioeconomic Correlates of Risky Hopkins School of Public Health, Baltimore, Md. Behaviour: Results from the Demographic and Health Rosenzweig, Mark R., and Kenneth Wolpin. 1986. "Evaluating the Survey" In Martha Ainsxvorth, Lieve Fransen, and Mead Over, Effects of Optimally Distributed Programs!" American Economic eds., Confronting AIDS: Evidence from the Developing World- Review 76 (3): 470-82. Selected Background Papers for the World Bank Policy Research Schultz, T. Paul. 1984. "Studying the Impact of Household Report Confronting AIDS: Public Priorities in a Global Epideniic. Economic and Community Variables on Child Health." Washington, D.C.:World Bank. Population and Development Review 10: 215-35 (Supplement). Fisher, Andrewv, B. Mensch, R. Miller, I. Askex,x A. Jain, C. Ndeti, L Shyrock, H.S., and J. S. Siegel. 1976. The Methods and Materials of Ndhlovu, and P Tapsoba. 1992. Guidelines and Instruments for a Demography. New York: Academic Press. 46 CHAPTER 1 5 FERTILITY Thomas, Duncan, and John Maluccio. 1995. Contraceptive Choice, Westoff, Charles F., and A. Bankole. 1995. Unimet Need: 1990-94. Fertility, and Public Policy in Zimbabwve. Living Standards Demographic and Health Surveys Comparative Studies 16. Measurement Study Working Paper 109. Washington, D.C.: Columbia, Md.: Institute for Resource Development-Macro World Bank. International. Todaro, Michael P 1969. "A Model of Labor, Migration and Urban Westoff, Charles F, and L.H. Ochoa. 1991. Unmet lNeed and the Unemployment in Less Developed Countries." American Demandfor Family Platning. Demographic and Health Surveys Economic Review 59: 138-48 Comparative Studies 5. Columbia, Md.: Institute for Resource United Nations. 1995. "Population Trends and Population- Development-Macro International. Related Issues: The Need for International Assistance." World Bank. 1984. florld Development Report 1984. New York: Economic Commission for Europe, Population Activities Oxford University Press. Unit, New York. . 1993. 14'orld Development Report 1993: Investinig in Health . Westoff, Charles F 1991. Reproductive Preferences-A Comparative NewYork: Oxford University Press. View Demographic and Health Surveys Comparative Studies .1997a. 10orld Development Indicators 1997. Washington, D.C. 3. Columbia, Md.: Institute for Resource Development-Macro . 1997b. 14'orld Development Report 1997: The State in a International. Chtanging World. NewYork: Oxford University Press. 47 ' *t ~Migration J 6 Robert E. B. Lucas Collecting migration infornmation has not been a high priority in past LSMS surveys.Yet both internal and international migration have pervasive effects throughout most economies, with pol- icy implications for many issues that are central to LSMS and similar multi-topic surveys-issues regarding labor markets, income generation, consumption smoothing, the environment, educa- tion, and the provision of community services and facilities. Government policies affect migration flows by alter- 1985-90; this figure did not include seasonal migrants. ing the factors that people take into account when Three LSMS surveys in the past decade-in Ghana, they decide to relocate. In addition, the economic Pakistan, and Vietnam-also show a high percentage effects of government policies-including effects on of people in developing countries moving from one both efficiency and income distribution-depend on location to another. The survey in Vietnam revealed migration patterns. that 23 percent of Vietnamese adults live in a place other than the place where they were born. Equivalent Policy Issues Concerning Migration figures from surveys in Ghana and Pakistan were 35 percent and 53 percent, respectively. Several important policy issues arise from the interac- tion between policies and migration outcomes. This MIGRATION AND EMPLOYMENT. Much of economists' section scans these issues, for both internal and inter- interest in migration has focused on labor migration: national migration, and identifies the kinds of infor- reducing the supply of labor in one labor market while mation policymakers need to make well-informed increasing the supply in another. Normally, workers policy decisions. move from a place where wages are low to a place where wages are high.To the extent that higher wages Internal Migration reflect higher worker productivity, migrants who suc- Internal migration is common in many developing ceed in increasing their wages have also enhanced economies. According to the October Household their productivity. However, many factors limit the Survey conducted in 1994 by the Central Statistical geographical mobility of workers, including the Office of South Africa, 2.3 percent of South Africa's expense of moving, a preference for staying at home, adults moved from one place to another in a single and limited information about job alternatives. year. The 1990 census for Thailand showed that about How responsive are population movements to 8 percent of the country's population moved during earnings and employment opportunities? Answering 49 ROBERT E. B. LUCAS this question can provide valuable insights into a num- bers in town already or people in rural areas located ber of important issues: close to a town? And does accepting an interim job, * If the geographical dispersion of productive activi- such as casual work in the informal sector, reduce a ties changes-as a result of, say, trade liberalization person's chance of finding a good long-termjob? Data or economic growth in the absence of reform- on these largely neglected issues would be very useful will large wage increases be necessary to induce to policymakers. workers to move to the new jobs? * If a job creation program is undertaken in urban MIGRATION, POVERTY, AND INCOME DISTRIBUTION. areas, how many migrants can be expected to move Migration has the potential to alter income distribu- for every job that is created? tion in a number of ways. Over time migrants may * If a rural development program succeeds in become more socially mobile than nonmigrants. enhancing rural earnings, will this limit the number Migrants' departure from some areas and arrival in of people who move out of rural areas? Or is young others may affect the income distribution within those people's attraction to urban areas strong enough areas. The distribution of resources within a family that they will move anyway? may be altered by the departure of some members; * How far are workers willing to move to obtain a indeed, family units-among which income distribu- job in a rural works program? tion is frequently measured-may be greatly trans- • What are the important constraints on people's formed by migration. geographical mobility? Could certain policies alle- A number of key questions must be answered to viate these constraints? understand the effects of migration on poverty and If workers are unwilling to relocate elsewhere, income inequality. even for much higher wages, policymakers may wish To what extent is social mobility enhanced by to consider moving the jobs to the workers. On the migration? To what extent do migrants' earnings other hand, it is strongly suspected, at least in some rise more rapidly than the earnings of people who industries, that firms benefit greatly from being locat- do not relocate (including people native to the ed close to other firms-particularly firms in the same place of migration)? Do migrants succeed in mov- line of production. And significant economies of scale ing up the job ladder? can be achieved where infrastructure is provided in * Are migrants drawn disproportionately from low- one place to serve a group of companies. income, middle-income, or high-income families? It might be assumed that the poor are more hkely to MIGRATION AND JOB SEARCH. Workers may be able to migrate because they have few prospects at home, yet conduct a job search more effectively if they have the wealthy may be in a better position to pay the already moved to the area where they are looking for expenses of moving (and of acquiring education- a job. This simple fact can have profound implications which makes it easier to find work elsewhere). for the economic efficiency of migration.To illustrate * Do migrants send remittances to their families, or these implications it is useful to review the Harris- vice versa?' Are poor families likely to receive more Todaro (1970) framework. or fewer remittances from their migrant members Suppose urban wages are relatively high and open than rich families? Migrants from poor families urban unemployment is rife, while wages in rural areas might be expected to have an interest in supporting are low but workers can find productive work. For a their family members, but migrants from wealthy rural worker to find a more lucrative urban job, he or families may have a greater vested interest in pleas- she must generally move into town and remain unem- ing their parents, because these migrants may hope ployed while searching for a job. Any policy initiative to inherit their parents' wealth. that creates urban jobs may attract so many more * To the extent that migrants from poor families do workers into town that the unemployment rate may not send remittances home, does this leave the fam- actually rise, while total GDP may fall as workers are ily members who stay behind (and who may be pulled away from productive rural employment. elderly or incapacitated) in poverty? Does the But just how essential is relocation for finding a departure of migrants compel remaining family job? Is it less necessary for people with family mem- members to work longer, harder hours? 50 CHAPTER 16 MIGRATION • How does migration affect the wages in areas from * Do migrants send more remittances when condi- and to which migrants move? Migration generally tions in their original household are temporarily alters the incomes of people in areas of both in- worse? Conversely, do migrants receive more migration and out-migration-nonmigrants as well remittances from their original household when as migrants. However, these effects are not straight- they suffer a temporary setback? forward. The departure of skilled people presum- * Are families with members who have moved away ably boosts the earnings of locals with skills similar from home better able to smooth their consump- to those of the migrants. And in the long run, this tion during periods of adversity? trend encourages the next generation of locals to * In times of economic adversity, are remittances acquire these skills. On the other hand, the depar- smaller and less frequent when insurance is available ture of skilled people from an area can either help or other transfers (such as social security) are avail- or hurt wages of less skilled people, depending able? upon the nature of production in the area. * How is the process of trickle-down development MIGRATION AND INFRAsTRucTuRE. Migrants may be shaped by migration? The creation of industrial jobs attracted to certain places by the existence of infra- in urban areas may or may not benefit poor rural structure and facilities. However, this can cause the families, depending on who is induced to migrate to facilities to become overcrowded. In such circum- town, whether they send remittances to families in stances policymakers may build new facilities or rural areas, and how these factors affect local wage- improve existing facilities in areas with high population setting. densities; this is especially probable when economies of To understand the above questions, policymakers scale will lower the costs of providing many services. need information about social mobility, the income On the other hand, the existence of improved facilities classes from which migrants are drawn, patterns of may attract even more migrants, leading to more over- remittances and their relation to poverty, and the crowding.Thus it may be better to improve facilities in effects of the arrival and departure of migrants on local less crowded areas, encouraging migrants to move labor markets. there rather than to high-density areas. When policymakers establish a new town, expand MIGRATION AND RiSK. Migration may be one impor- a settlement, or promote a settlement scheme, they tant way for families to mitigate inherent economic make important decisions about where to locate facil- risks. By sending a family member to a place where ities.The policymakers must consider:Are these strate- times of economic misfortune do not normally coin- gies cost-effective? How much migration is there to cide with such times at home, the family can spread such areas and from which population groups? out its risk over two different sets of circumstances, To answer these questions it may be valuable to thus diminishing total risk. A household located in a find out how important the existence of public facili- frequently drought-stricken rural neighborhood ties is for people's migration decisions. If public facili- might send a migrant member to an urban area that is ties are important factors, analysts need to account for not drought-stricken but has an uncertain job market. this when evaluating the impact of the facilities on The concept of migration to mitigate risk raises a households' living standards. If sick migrants are number of questions: attracted to a village with a health clinic, and sick * Is out-migration more frequent from areas where inhabitants of this village choose not to leave, the pop- inherent risk (such as drought, flood, and disease) is ulation of the village may end up with worse health especially common? than the population of a village without such facilities. * Do migrants tend to move to locations where eco- The relationship between migration and transport nomic misfortunes are unlikely to occur at the facilities is complex and not yet well documented. It is same time as they do at home? For example, does not clear whether the existence of cheap and easy migration as a result of marriage occur especially transportation between towns and the countryside frequently between places where economic misfor- promotes or curtails out-migration from the rural sec- tunes are unlikely to occur at the same time? (See tor. The existence of reliable transportation makes it Rosenzweig and Stark 1989.) easier for rural dwellers to market their products in 51 ROBERT E. B. LUCAS town, but it also heightens competition by making may lead to a kind of "brain drain," any evaluation of goods from town available in local rural areas. It is the returns to rural education should also take into therefore not clear how easier transportation affects account the benefits to those whose education enabled the relative prices of local goods, whether such trans- them to migrate from these areas. portation encourages the production of labor-inten- sive goods, and what the consequences are for rural SOCIAL IMPLICATIONS OF MIGRATION. Social implica- employment. Moreover, the existence of reliable trans- tions of migration are often important to policymak- portation encourages migration by reducing the costs ers.Three aspects of migration's social implications are to a migrant of both moving to a town and making mentioned here. subsequent visits home. However, reliable transporta- First, migration can separate married couples, tion also allows rural dwellers to commute to urban possibly resulting in social tension and reduced fertil- areas to take advantage of urban employment oppor- ity. Separation of a husband and wife may also impov- tunities and facilities-while continuing to reside in erish a spouse left at home, although this depends the country. greatly on remittances received by the remaining Improving transportation in rural areas may also spouse and on production conditions at home. make it easier for workers to move around and change Policymakers are interested in knowing how often jobs within the rural sector.This can affect rural-urban families relocate as one unit, how often migration movements in a number of ways. If it is cheap and easy results in families being reunified, and what the eco- for workers to move from one village to a different nomic circumstances are for family members left village, the arrival of migrant workers in the second behind when household heads or other key members village may cause wages there to decline-encourag- of households migrate. ing subsequent migration to urban areas.The ability of A second set of issues arises from the role played workers to move between rural locations may offer by social and kinship networks in facilitating migra- families the opportunity to insure themselves against tion. Having kin in town may make it much easier for economic shocks by locating members in two differ- a migrant to find an urban job by providing him or ent rural locations as well as by sending some mem- her with a place to stay on arrival and generally mak- bers to urban areas. And improving transport between ing the new context less alien.Thus, when analysts are rural areas can increase returns to rural capital (notably examining the role played by other factors in promot- on trucks or other vehicles)-possibly encouraging ing or constraining migration, they should control for workers to migrate from rural areas to towns so they the existence of social and kinship networks. can save money to invest in their home area. Third, in many countries one of the principal rea- For policymakers to understand the interrelation sons for relocating is local violence-war, political of migration and transport patterns, they need com- unrest, or crime. Documenting such causes of dis- munity-level information on the availability, cost, and placement can identify the people who most need quality of transport (preferably for goods as well as relief, and lead to measures-such as enhanced law people), as well as household-level information on enforcement-that will address the underlying causes migration patterns. of this violence. Finally, the strong link between migration and educational facilities is worthy of separate mention. International Migration People may migrate in order to have access to better To a large extent, international migration and internal schools. In turn, the level (and possibly content) of the migration are driven by similar forces, forces that yield education achieved by migrants and nonmigrants may similar consequences. However, international migra- affect migration patterns, both by shaping their atti- tion raises several major policy issues not raised by tudes and by presenting them with employment internal migration. opportunities. Enhancing education in rural areas is an While immigration is subject to controls in vir- important way of enabling rural workers to land jobs tually every country in the world, the efficacy of in urban areas. But because tertiary education is remu- these controls varies enormously. Therefore, one nerated far better in towns than in rural areas, few col- major set of policy issues involves the imposition, lege graduates tend to return to rural areas.While this efficacy, and nature of immigration controls. Does 52 CHAPTER 16 MIGRATION making controls stricter simply raise the rate of ille- policy issues. Some of these data must be gathered gal immigration? Are penalties against the employers from other modules of LSMS surveys. of illegal immigrants effective or do they simply encourage discrimination against all aliens? Does Collecting Migration Histories: Methodological Issues trade protection encourage industries that employ The United Nations manual on measuring internal many immigrants, whether legal or not? Do immi- migration defines a migrant as "a person who has grants undercut the local wage? Do foreign workers changed his usual place of residence from one migra- catch up with locals in terms of their career trajecto- tion defining area to another . . . during [a given] ries, and what is the role of language in influencing migration interval" (United Nations 1970).To under- this? Do foreign students stay on to work in the stand what this means in practice, three concepts must country where they studied? Do family members of be defined: "usual place of residence," "migration immigrants work, or do they receive income from defining area," and "migration interval." Depending state aid and public expenditures? how these terms are defined, migration can mean any Fortunately, very few governments attempt to stay away from home, from a visit with a relative in the impose direct emigration controls. Indeed, some gov- same village to an irrevocable break with the migrant's ernments actively seek to export workers and have home in which he or she moves to another region of implemented programs providing training, informa- the country or world. tion, and even credit for workers interested in emi- In most contexts the "migration defining area" is grating. Several countries have offered exchange and taken to be an administrative unit such as a province, interest rate incentives, along with tax breaks, to district, county, township, or village. Anyone who, encourage emigrant workers to send remittances back within a specified time, changes his or her "usual resi- home (through legal channels). It is important to dence" across the boundary of such a unit is defined as examine the efficacy of such programs. a migrant. The key elements that help survey design- Finally, there is the issue of the "brain drain" of ers decide which administrative unit to use as the educated and skilled workers from developing coun- migration defining area in a given analysis are the het- tries to developed countries. Since the gap in earnings erogeneity of each administrative level and the focus of between most developing countries and developed the analysis. If the province were chosen as the "migra- countries is very large, it is unrealistic to expect devel- tion defining area," this would mean that any rural- oping countries to generate sufficient wage incentives urban migration within each province would not be to keep these workers from emigrating.Thus the prin- recorded. Thus, if the purpose of the analysis is a study cipal policy issue at stake is the design and financing of of rural-urban migration and if provinces are hetero- the educational system. Policymakers will likely want geneous (in other words, contain both rural and urban to know about the educational background of emi- areas), it is inappropriate to choose the province as the grants, who financed their education, and whether migration defining area. Analysts and policymakers such emigrants return home or continue to send may even be interested in knowing about changes in remittances back to their families at home. residence within the boundaries of a given city (for example, from a squatter area to a planned neighbor- Data Requirements hood) so they can analyze access to different facilities. In this case, the migration defining area would need to By no means are there straightforward answers to all be the city sector. On the other hand, village-to-vil- of the policy questions regarding migration, even lage migration may not be particularly relevant to a when ideal data exist with which to analyze them. labor market study if the rural labor market in a par- But a well-designed household and community sur- ticular province is well-integrated. vey can provide useful insights into many migration In the draft migration module it is suggested that issues. survey designers should adopt movement between vil- This section discusses the methodology involved lages, towns, or other similar units as the basis for col- in collecting migration histories in large-scale house- lecting migration information. This is often called hold surveys like LSMS surveys and outlines the data place-to-place migration. Using this definition has the that must be gathered to analyze specific migration advantage that it generally also permits analysis of 53 ROBERT E. B. LucAs movement between broader administrative units (such In this chapter, the first approach will be called the res- as districts or provinces), provided the location of each ident migration history approach and the second will place is appropriately identified in the data set made be called the absentee approach. Both approaches have available to analysts. On the other hand, using place-to- some strengths and some limitations. place migration does rule out the possibility of analyz- Perhaps the greatest limitation in using the resi- ing residential mobility within towns or villages. dent migration history approach is that it relies on The draft migration module suggests defining the respondents' knowledge and recollection of the "usual place of residence" as any place where someone households they have left. Another weakness of this lived (meaning slept and ate) for three months or more approach is that no information is collected about at one time. This avoids the complications that can emigrants even though such information can be of arise when individuals, having a sense of allegiance to great interest to analysts and policymakers. A major their original home, give that location as their "usual strength of the resident migration history approach is residence" even though they neither slept nor ate the ability to collect information about the migrant's there during the reference period. The choice of experiences directly from the migrant. On the other "three months or more at one time" is admittedly hand, using the absentee approach also relies on the arbitrary but is likely to rule out short-term visits for respondents' knowledge and recollection-in this case social, business, religious, or vacation purposes. about who has left the household. Another disadvan- A "migration interval" that is perhaps the most tage of the absentee approach is that when entire fam- commonly used in surveys is time since birth. In sur- ilies migrate, the new occupants of their old dwelling veys using this interval, respondents are asked where are unlikely to possess much information about them. they were born and are defined as migrants if this However, a major strength of the absentee approach is "migration defining area" differs from the place where that information about the household and communi- they are living when interviewed for the survey. Since ty from which the person migrated will be more accu- major moves are comparatively rare in the lives of rate than any comparable information collected using most people, using time since birth has the advantage the resident migration history approach. of maximizing the chances of observing a migration. The draft migration module presented in this On the other hand, it has the disadvantage that it does chapter is designed to allow survey designers to adopt not establish a rate of migration per unit time. Thus either (or both) of the approaches. In the absentee there are good reasons to collect information for at approach, a person who has migrated from the sam- least two points in a person's migration history-his or pled dwelling is defined using a set of people associat- her initial location and his or her location at some ed with the household. In particular, three sets ofthese fixed point in time. "associates" are defined: Note that the initial location does not necessarily * Nonresident surviving parents of each household correspond with the person's place of birth, which is member (in section B of the standard household the location used by some other surveys. When the roster) and nonresident children of household birth takes place in an urban hospital but the mother members (in section C of the standard roster). and infant soon return home to their village, the place Additional people identified on the extended ros- of birth is less relevant for most purposes than the ter, where this is applied (see Chapter 6). baby's initial residence. For this reason, the draft migra- * The head of household. In the draft standard roster tion module asks about the baby's initial residence introduced by Chapter 6, the head of household is rather than about his or her place of birth. always defined as a household member even if he or she has been absent for the whole of the previous Two WAYS TO COLLECT MIGRATION INFORMATION. 12 months. If the household head has been absent There are two different ways to approach the process for at least 6 ofthe previous 12 months and has not of collecting migration information. Either the inter- been present for the previous 7 days, then the head viewer can ask each person at the sampled dwelling is also treated as a household associate. about his or her migration history, or the interviewer If a particular associate is not present at the time can ask the resident members of the household of the interview, information about that associate whether others have migrated from their household. should be collected from his or her spouse, parent, or 54 CHAPTER 16 MIGRATION adult child or from the household head. (Note that the migrations that older people may have made in their associate may be present even if he or she is not actu- youth. ally a household member, if he or she has been living As a result, almost all surveys that include a migra- with the family intermittently.) Absentee migrants tion module have used a person's entire lifetime as the from a household are associates who at some point recall period. This raises the question of the reliability lived continuously at the dwelling for three months or of migration data recalled over a long period of time. more but who now live in another place. A recent study by Smith and Thomas (1997) addressed An associate might have migrated from another this question. The authors compared the recalled dwelling where he or she lived with the family sur- migration histories of individuals each of whom was veyed, after which time the family moved to its cur- interviewed in two rounds of the Malaysia Family Life rent residence. In such a case the associate might be Survey, with 12 years having passed between the considered a migrant from this family even though he rounds.The study did not aim to find out whether the or she has never lived in the family's current dwelling. initial information given by the respondents about A question covering this eventuality is included in the their migration histories was correct; instead it aimed draft standard migration module. to discover whether information they gave in the later round of the survey corresponded accurately with the PANEL DATA. The fact that analysts and policymakers information they had given 12 years earlier. The are often interested in information about migrants and authors concluded that "respondents tend to remember their households both before and after migration sug- salient moves, those linked with other important life gests that panel data may be appropriate. However, it events such as the start of a marriage, the birth of a can be difficult to trace individuals when-as in child, change in a job and moves that lasted for a long LSMS-type surveys-dwellings, not individuals, are time. In contrast, migrations that dim in memory as the units of study. When out-migration occurs time passes are typically short duration or local moves. between rounds of the survey, either the migrants must . . . When collecting complete lifetime histories, it be traced or data about the migrants must be collect- would seem prudent to focus on longer-term moves, ed from remaining residents to maintain the original leaving shorter-duration and circular moves to be cap- panel of individuals. tured in a supplemental module on all migrations that Although tracing is not ruled out as a possibility have taken place in, say, the last year or two" (Smith and in Chapter 23 on panel data, tracing can be an expen- Thomas 1997). Thus this study has at least two impor- sive process and is often unsuccessful. If tracing is tant implications for the design of migration surveys. rejected as a possibility in the design of a survey, it may First, the study suggests that people's memories of their be critical in later rounds of the panel survey to col- main migration events do not deteriorate substantially lect information on absentees. If individuals who do over fairly long periods of time. Second, the study sug- not appear in later rounds have migrated from the gests that it might be more reliable to gather partial household, the economic situation of the household migration histories in which data are collected on a can be compared, say, before and after the migration. person's major moves than to attempt to glean this per- Similarly, absentee information might be used to com- son's complete migration history. pare the employment of individuals before and after they migrate. However, as Chapter 23 recommends, it PARTIAL VERSUS COMPLETE MIGRATION HISTORIES. In is better to allow several years to pass between the very peripatetic societies, collecting a complete migra- rounds of the survey; otherwise very few migrations tion history can become cumbersome, even if infor- will have occurred between rounds. mation is only collected on long-term moves as Smith and Thomas suggest. This is particularly true since, RECALL PERIOD. Since major moves are comparatively although many people never move, those who move rare events for most people, using a short recall period once often move several times. In consequence, the to collect migration information is not likely to yield draft migration module is designed to collect infor- many instances of migration. Moreover, most migra- mation on only a limited number of major migrations. tion occurs among young adults, which means that The short and standard versions of the module using a short recall period would fail to capture the focus on the most recent move, the first move (if dif- 55 ROBERT E. B. LUCAS ferent), and the location of all the residents of the sam- circumstances (age, education, family background) of pled household five years previously. If a resident was migrants are contrasted with the circumstances of born abroad, this resident is asked when he or she ini- nonmigrants. Most econometric analyses of the causes tially arrived in the country of the survey.The expand- of migration use some form of discrete regression ed version of the module includes additional questions analysis (such as probit or logit) to relate whether or about whether the resident ever lived abroad and not a person has moved to a list of potential explana- where the resident lived just prior to marrying. The tory variables.2 first question of the expanded module, which asks about any past time living abroad, is designed to per- MOVERS AND STAYERS. Migration can be studied using mit analysis of the post-migration experience of partial migration histories. For example, it is possible to returned emigrants. study rural-to-urban migration by examining a sample Establishing where a person lived before he or she of people who initially lived in rural areas and distin- got married makes it possible for analysts to examine guishing those who have stayed from those who have such issues as risk-spreading through marriage (marry- moved to urban areas.3 However, an alternative distinc- ing into a family from a place where times of eco- tion is occasionally made between those who have nomic misfortune are unlikely to coincide with such stayed in a particular location and absent household times at home) or peer-group learning about contra- associates who are reported by other household mem- ception (in which a marriage partner's information bers to be in an alternative "migration defining area," and attitudes about contraception may have been say, in a town or abroad. In either case it is possible to shaped prior to marriage by people from his or her distinguish moves to various locations-abroad, to a original home). See Rosenzweig and Stark 1989, city, to a town, within the rural sector-instead of Munshi and Myaux 1997. focusing on a simple mover-stayer dichotomy (see The expanded version of the module additionally Lucas 1985, Falaris 1987, Pessino 1991, andVijverberg asks for a brief migration history of each household 1995). Other possibilities include distinguishing moves associate, using essentially the same categories of across administrative boundaries, although it is less moves as the categories used in the short version of apparent how this would be useful to policymakers. the questionnaire for household members. A person's initial location is usually interpreted to be their location at birth or shortly after birth. The Causes of Migration However, "initial" can also refer to some fixed point in There are two approaches to collecting information time-say, five years before the survey. As will be seen about why people migrate. The first and simplest in subsequent sections, specifying a short time interval approach is to ask migrants why they moved. Many has some advantages in terms of measuring explanato- surveys have used this approach, and some questions ry factors. However, choosing a short time interval along these lines are incorporated in the draft migra- means that only a small number of migrations will be tion module. However, this approach has some critical able to be studied. In less mobile societies, this small shortcomings. Migrants may have had to weigh several sample may make it impossible to perform any effec- different factors in deciding to move. Asking them to tive analysis of migration flows. identify only one (albeit the most important) may cause analysts to miss other contributing factors. On MEASuRES AFFECTING MIGRATION DECISIONS: GENERAL the other hand, migrants may find it difficult to rank CONSIDERATIONS. The theoretical and empirical litera- several contributing factors in order, let alone to ascribe ture suggest several key factors that are likely to influence weights to each reason. Moreover, why migrants people and households in deciding whether or not to moved is only part of the question; at least as important migrate.4 Some of these key components are: is why nonmigrants did not move.Yet such questions * Personal attributes, such as age or gender, that influ- are rarely, if ever, posed-partly because they require ence attitudes toward moving. respondents to be extraordinarily introspective. * Differences between earning opportunities and job Given these limitations, most economists have prospects at home and in alternative locations. preferred to rely on a second approach, known as the * Prior movement of family members (and possible "revealed preference" approach. In this approach, the reunification of a family). 56 CHAPTER 16 MIGRATION * Marriage. PERSONAL ATTRBUTES. It is well established that the * Distance and cost of relocating. propensity to migrate varies systematically with cer- * Access to information and relocation networks. tain personal attributes such as age, gender, and educa- * Ability to finance costly moves. tion. To some extent, it is likely that these patterns * Possession of assets that are difficult to transfer. reflect differences in employment opportunities. For * Family strategies to minimize economic risks. example, it is frequently supposed that young people * Availability and quality of facilities at home and in migrate more often because they have a longer lifes- alternative locations.5 pan over which they can reap the benefits of finding a * Economic inequality and relative standing in the new job and that better educated people possess supe- community. rior information about job opportunities. However, * Incidence of violence, disease, or disasters. available evidence suggests that personal attributes also * Migration controls and incentives, especially on play an independent role in influencing migration international migration. decisions. Thus it is likely that young (and single) peo- Ideally, analysts would like to have information on ple are simply more footloose and that education tends all of these components, if only because omitting some to widen people's horizons. measures might suggest patterns of association that are It should also be noted that the interpretation of actually spurious.Yet largely because of a lack of data, some of these factors depends upon which migration no existing study incorporates all of these elements.At measure is in question. If migration is measured rela- one extreme, the revealed preference approach to tive to a person's birthplace, the cumulative chance of studying the causes of migration uses a multivariate having migrated will generally rise with the person's approach to relate the migration outcome to as many age. On the other hand, if migration is measured by potential explanatory variables as are available in the any location changes in the previous five years, the data, while bearing in mind the dangers of omitted chance of migration will probably rise with the per- terms. At the opposite extreme, it is possible to do son's age until his or her late twenties, after which it much simpler cross-tabulations that relate the migra- will decline. Whatever the interpretation, these per- tion outcome to specific measures from the above list. sonal attributes are important explanatory factors that This would yield instructive insights into what moti- should generally be included in any analysis of migra- vated people to migrate, although it would not pro- tion. Data on these attributes are available from both duce results that could be safely interpreted as causal. the household roster and the education modules. The other modules in volume 3 contain many of the elements listed above, although not necessarily in EARNINGS AND EMPLOYMENT STATUS. It is necessary to an ideal form for studying migration. This is not measure job and earning opportunities at home and in because of inadequacies in the draft modules but alternative locations in order to study the role that rather because of two main problems in understanding these opportunities play in migration decisions. migration. First, when the migration occurred some- However, measuring these factors involves at least two time in the past, the relevance of information collect- fundamental difficulties. ed at the present time to understand that past outcome First, the draft employment module introduced by is debatable.6 Second, since migration often implies a Chapter 9 does not inquire into what employment change of household and always involves a change of opportunities individuals perceive will be available to community, collecting information on a migrant's cur- them if they relocate elsewhere. The reliability of rent household and community will not yield infor- responses to such questions would be very dubious so mation on any factors that prompted the move. this should not be seen as a shortcoming of the These two problems need to be borne in mind in employment module. The most common-if still the following discussion, which will cover each of the rare-technique for addressing this problem is to sim- explanatory categories listed above. In some instances ulate earnings (and, less frequently, employment questions in the draft modules will yield plausible opportunities) in alternative locations.This simulation proxies for factors that analysts would like to measure; can be done by performing a regression analysis of in other cases the data need to be collected within the earnings on personal characteristics within various migration module. categories of location such as rural areas, capital cities, 57 ROBERT E. B. LUCAS and other urban areas.7 The results of this analysis can The time dimension of this problem suggests that be used to project what earnings each person would panel data may be a solution. However, as has already have received in the alternative locations based on his been noted, it is not possible to use panel data to track or her characteristics.8 One major drawback to this respondents' employment histories unless the migrants approach is that the same personal attributes may can continue to be included in the sample in later influence not only migration decisions but also earn- rounds of the survey, perhaps through tracing. In prin- ings and employment status, making it difficult for ciple, an alternative would be to collect information analysts to distinguish among these effects. on the employment of absentee household associates The second fundamental difficulty with this in later rounds of the survey. However, it seems unlike- approach is the problem mentioned above regarding ly that interviewers could glean much reliable infor- the tinme frame for measuring employment opportuni- mation on the earnings of absentee household associ- ties. Sjaastad (1962) viewed migration as an irreversible ates from the remaining family members, although the investment decision and argued that the appropriate remaining family members may at least know whether measure of employment opportunities was the dis- or not the absentee is employed. counted stream of future lifetime earnings that migrat- In summary, how earnings differentials and the ing may bring to the migrant. At the opposite probability of finding employment (formal or infor- extreme, the decision to migrate may be seen as mal) affect the rate of migration is an important factor instantaneously reversible, in which case the only rel- in several areas of policymaking.These effects are cru- evant employment opportunities are those available at cial for evaluating the geographical integration of the the moment the decision is made.Yet neither of these labor market. And knowing the rate of migration may two approaches has been taken in the literature on help analysts study the paradoxical possibility that migration decisions. Instead, most studies have made a urban job creation programs may create unemploy- very strong, tacit assumption: that information about ment by prompting more migrants to move to the the labor markets at the time of the survey is a rea- place where the jobs are being offered than the pro- sonable proxy for the employment opportunities peo- gram can accommodate (see Harris and Todaro 1970, ple will consider when deciding whether to move. Stiglitz 1969, Corden and Findlay 1975, Smith 1983, When labor market conditions in the relevant loca- and Fields 1989). However, in practice, the technical tions are relatively stable, this assumption is somewhat problems of simulating employment alternatives mean plausible. But in most situations this assumption that it can be difficult to obtain reliable and precise becomes less reasonable as more time passes-indicat- estimates of whether and how people migrate in ing that it is preferable to analyze only recent migra- response to employment opportunities. tions in relation to current employment opportunities. LSMS-type surveys are designed to collect certain PRIOR MIGRATION OF FAMILY MEMBERS, MIGRATION pieces of information that can enrich this area of TO REUNITE A FAMILY, AND RELOCATION AFTER investigation in a couple of ways. First, the draft stan- MARRIAGE. The location of an individual's family dard migration module asks questions about a person's members may play several different roles in influenc- employment just before and after his or her most ing his or her decision to migrate. First, if other fam- recent move. Second, the recall period in the retro- ily members have already moved to a new area, they spective section of the expanded version of the draft may be able to give prospective migrants useful infor- employment module introduced by Chapter 9-five mation about what opportunities are available, help years-is the same recall period that was used in ques- them make contacts in their attempt to find a job, and tions about respondents' location in the draft migra- provide them with a cheap place to stay (Carrington, tion module. Moreover, the short version of the draft Detragiache, and Vishwanath 1996). Second, it is migration module asks respondents how long they common for family members or a spouse to move to have lived in their current place of residence. This join a migrant in a new area in order to reunite the means that it is possible to use data on time since the family or couple. Third, in many places it is common last move in a simulation of earnings, in recognition of at the time of a marriage for at least one partner to the fact that migrants may take some time to realize migrate to live with the other (Rosenzweig and Stark their full earning potential in a new location. 1989). 58 CHAPTER 16 MIGRATION It is important to control for these forces or ana- At least three elements are commonly assumed to lysts may attribute a person's decision to migrate to the be associated with distance: transport costs, the avail- prospect of employment opportunities in a new place ability of information about job opportunities, and the rather than to the real reasons for migration. However, extent to which a migrant feels estranged from home there is still little literature dealing with these factors. (in other words, the psychological costs of moving). As a first step, it may be desirable to include among the Policymakers are mainly interested in the role of trans- explanatory variables whether a spouse, spouse-to-be, port costs, which will be examined later when the role or other close relative (perhaps a child, parent, or sib- of facilities as a cause of migration is discussed. For ling) already lives in the contemplated destination of now, it may be noted that reductions in the costs of the prospective migrant.9 This information can be transport (and communication) can have an impact on gathered in two places in the draft migration module. the other two correlates of distance by giving the In the case of a person who has not yet migrated, migrant more access to information and by diminish- the issue is whether this person has family members ing the psychological costs of relocation. Isolating living elsewhere. This will be ascertained through these other two correlates of distance can be quite dif- questions about the location of household associates, ficult, and their policy implications are less obvious. In questions that are even in the short version of the some places governments have attempted to promote migration module. In the case of migrants, what mat- migration by providing information on certain desti- ters is whether a relative preceded them; this will be nations, although these initiatives appear to have had found out from the migration histories of any of the little effect.'0 The idea that the further away a person migrant's relatives who may live in his or her new is from a place, the less he or she is likely to know household.The possibility exists that a relative preced- about it has been offered as an explanation for the ed the migrant but has since moved on, or does not common phenomenon of step-migration (first from a live in the same dwelling as the migrant.Thus the stan- village to a town and later from the town to a metro- dard version of the draft migration module incorpo- pohtan area). However, this theory remains untested rates some specific questions about family and friends and its policy implications are not obvious (Pessino, who already lived in the migrant's destination prior to 1991). the migrant's move. In principle, it ought to be possible to find out To supplement this approach, migrants are asked if from respondents what information is available to they migrated because they got married or because them about potential migration destinations.The cor- their parents moved. In addition, in the standard ver- relation between this information and whether there sion of the questionnaire, migrants are asked with are family members or groups of a similar ethnic ori- whom they stayed on arrival at the migration destina- gin at these destinations might then be explored; how- tion, who helped to pay their settling-in costs, and how ever, it is not clear what policy purpose such analysis they found their initial job. The lack of comparable would meet. Another approach might be to collect measures for nonmigrants precludes the use of these information in the community questionnaire on measures as explanatory terms in regressions to study whether there has been previous migration to specific migration outcomes. It is envisioned that the answers destinations from the community studied. The prob- to these questions will simply be tabulated to present lem with this is that it is dangerous to use past migra- frequencies of the different responses that were given. tion to explain current migration; both may be shaped by the same underlying factor, in which case there is DiSTANCE, INFORMATION NETWORKS, AND THE COST OF no direct casual connection. It may be as well to col- RELOCATING. Many studies have found that long dis- lect data on both distance and the availability of trans- tances between the present location and the potential port without attempting to disentangle the other cor- migration destination are negatively correlated with relates of distance. the incidence of both internal and international migra- This raises the question of how to measure dis- tion (Schwartz 1973; Lucas 1975; Molho 1995; tance. There are several issues involved. The first issue Greenwood 1997). Thus it is important to control for is between which two points distance should be meas- distance when analyzing other causes. However, dis- ured.The solution depends on what kind of migration tance can also be a proxy for several underlying causes. is analyzed. In the case of rural to urban migration, the 59 ROBERT E. B. LUCAS distance that needs to be measured is the distance from not yet been satisfactorily proven (Cornelius and the migrant's rural place of origin to the town. For the Martin 1993; Faini and de Melo 1994; Hatton and migrants themselves, however, the key question is how Williamson 1994; Lucas 1999).What is clear is that the far they moved. To calculate this distance, analysts ide- income class of migrants is well worth studying, as it ally need information on migrants' place of origin. For has profound implications for how migration will this reason (among others), the standard version of the affect income distribution. draft migration module includes questions about the It is important to collect information on the com- migrant's specific place of origin and not just about his position of a household's assets as well as on the total or her province of origin. On the other hand, for rural value of these assets, because owning certain assets can inhabitants who have not moved to a town, the rele- sometimes give individuals a disincentive to migrate. vant distance might be the distance from their present There are several reasons why families or individuals dwelling to the nearest town of a certain size. may find it difficult either to take specific assets with The second issue to be settled is the difference them when they move, or to rent or sell these assets. between measuring the distance between two places as For example, the owner of an asset may possess specif- a straight line and measuring it in terms of the route ic information that makes the asset more valuable to that must be taken to travel from the first place to the this person than to others, yet he or she may not be second. Depending on the topography of the country, able to share this information (Manove, Papanek, and there may be a substantial difference between these Dey 1987).A landowner may know how best to oper- two measurements. Generally, it makes sense to meas- ate his own land and find it difficult to explain this ure distance in terms of the actual travel route, information to others (Rosenzweig andWolpin 1985). although it may be more difficult to measure distance It may be difficult for an owner to supervise a vulner- this way than to measure it as a straight line. able asset from far away, and the asset may also lack a A third issue will arise if confidentiality consider- rental or sale market.12 ations dictate that analysts cannot be given data on the There may also be other explanations for observed places where the interviews were conducted. In these correlations between asset composition and migration circumstances, only the staff of the statistical office will outcomes. Measurement errors in evaluating assets be able to calculate such measures as the distance to could generate such patterns. Finally, the policy impli- the nearest town or how far migrants moved from cations ofanalyzing the composition ofassets as poten- their original places of residence-since statistical tial explanatory variables for migration are not clear office staffare the only people with access to the con- other than the possibility that, for some, government fidential information on specific interview sites. policy could affect the relative price of assets or indi- cate the absence of a credit market (Morrison 1994). ASSETS AND FINANCING MIGRATION COSTS. Two key An important issue to be considered is the appro- aspects of wealth and unearned income are worth priate measure of wealth to include in an analysis of considering: the ability of migrants to finance costly the causes of migration. For most migrants the rele- moves and how people's migration decisions are vant information is likely to be the wealth of the fam- affected by owning assets that are difficult to transfer ily with whom he or she lived prior to migrating.13 In geographically. the case of nonmigrants what matters is the wealth of Clearly, well-off families find it easier to pay the their current household. The wealth of a migrant's expenses of moving than do poor families. However, if previous household is not measured elsewhere in these well-off families have high levels of unearned LSMS surveys, but the wealth of the current house- income, they may feel less pressure than other families holds of both migrants and nonmigrants is thorough- to relocate to gain access to jobs that pay more than ly measured in the agricultural, household enterprise, they earn already. Therefore, several authors have sug- and savings modules. Another alternative is to classify gested that the propensity to migrate may initially rise a household's wealth in terms of its level of consump- with the wealth (or unearned income) of a family and tion as collected in the survey's consumption module. subsequently decline at higher income levels.ii This A distinction must then be made between analyz- hypothesis has attracted particular attention in the ing respondents' migration histories and analyzing context of more costly international migration, but has information about absentee migrants. While the 60 CHAPTER 16 MIGRATION wealth and/or income of the absentee migrant's new gathered at the household level and measures that family can be well documented, it is unlikely that a should be gathered at the community level. migrant who has left his or her parents' home will be Many decisions made by households affect their able to provide much detail about his or her parents' exposure to risk. Risk may be reduced by investing in assets or income at the time when the migrant left the assets that can be liquidated if it becomes necessary to original household.'4 Even if the migrant were head smooth the household's consumption, while risk may of the household at the time he or she migrated, it is increase if a household adopts new agricultural tech- unreasonable to expect him or her to remember nologies or installs private irrigation schemes. If a details about the assets the family owned at that time, household has taken risks, it has a considerable incen- particularly if the move occurred many years previ- tive to insure itself by sending some household mem- ously. Therefore, the expanded version of the draft bers to live and work elsewhere. This can also work in module is designed to collect only some very broad reverse; sometimes households may take risks precise- indicators of family wealth at the time of migration. ly because their migrant members offer some degree These include whether the migrant's family owned of insurance. Also, some families may take less care to land or operated a household enterprise and whether avoid negative outcomes when they feel they can rely the family was relatively well-off, about average, or on their migrant members for insurance-a phenom- poor at the time the migrant left.To gather compara- enon known as moral hazard. In addition, those ble measures for nonmigrants, data from the agricul- households that adopt risky strategies may care less ture, household enterprise, and consumption modules about risk than those that do not and may, thus, be less could be used. interested in insurance. These complexities are part of the reason why FAMILY STPRATEGIES TO MINIMIZE ECONOMIC RISKS. It analysts have yet to examine family risk-taking as a used to be the case that analysts of internal migration cause of migration. Nevertheless, the various modules concentrated on studying the risks involved in trying presented in this book collect the data necessary to to find a job in town (Todaro 1969). However, these carry out this analysis. The agricultural module con- days it is increasingly recognized that in many con- tains questions on irrigation, on what agricultural texts, living in the rural sector involves even greater technologies a household has adopted, and on what economic risks (Stark and Levhari 1982). Families may agricultural assets it has that can be used to smooth attempt to spread their risk by sending some of their consumption. The household enterprise module con- members to various other locations to minimize the tains questions on other kinds of household assets. In chances that an economic downturn will happen in all some surveys the consumption module may contain of these different places at the same time. In this way, questions that make it possible to analyze actual con- the different household members can insure one sumption smoothing. (Surveys with panel data are an another against economic shocks. example.) At least in principle, examined in conjunc- There are two important policy reasons to include tion with data on the incidence of absentee migrants risk minimization in analyses of the causes of migra- from these households,'s these data allow analysts to tion. First, omitting this factor may create false impres- study whether households that have exposed them- sions of the role played by other policy-influenced selves to higher economic risk encourage more of variables. Second, the extent to which households are their members to migrate. exposed to risk can be affected by government poli- At the community level, there are also a number cies. For example, irrigation projects, crop and animal of factors that influence the economic risks faced by disease programs, and efforts to induce farmers to households and may thereby encourage household adopt risky new technologies can each alter the extent members to migrate. Section 8 of the draft communi- to which those who work in agriculture are exposed ty module introduced by Chapter 13 suggests collect- to risk. Social insurance (typically confined to towns) ing data on several of these factors, such as the inci- and relief efforts can mitigate risks. dence of floods, droughts, earthquakes, epidemics, and In order to explore such questions, it is necessary crop diseases in the previous five years. Using the data to obtain some measures of economic risk. A distinc- on absentee migrants, it should be feasible to examine tion should be made between measurements properly whether more migrants leave communities where 61 ROBERT E. B. LUCAS such incidents are common than leave other commu- ing the role that the lack of such facilities plays in nities."6 This analysis is facilitated by the fact that, in encouraging people to out-migrate, they must exam- contrast to household-level risk factors, most of these ine the data on absentee associates. In both cases, the community-level risk factors are highly unlikely to be analysts can enrich their analysis using data on the age affected by migration. of the education and health facilities from the draft In principle, analysts might also like to know the community questionnaire; such age data reveals if the incidence of risks in the various locations to which a facilities were available at the time of the migrant's potential migrant might consider moving and to know departure from the community. whether disasters occurred in these locations at the However, in neither case can analysts draw on data same time as in the potential migrant's home area.The on differences between facilities located at origin and problem with this is that the community survey would destination. Partly because of this lack of data, the draft have to be administered in communities other than migration module includes a question asking migrants those in which the sampled households were located, whether gaining access to more and/or better facilities which is not feasible. However, in some instances, use- influenced their decision to relocate. Again, secondary ful measures can be assembled from secondary sources. sources may supply useful data on the schools or For example, if meteorological records exist both for health facilities that are available in various places; pro- the communities where the sampled households are vided that specific places (as opposed to just the dis- located and for other communities, variability in trict or the region) are coded, this data can be merged recorded rainfall can be used as an alternative measure into the data set after the survey. Where place names of weather risk. are not coded, secondary data can still be merged, by using averages for the relevant district or region.These FACILITIES. Another community-level variable that can averages could be derived from the relevant commu- affect migration is the availability of various facilities nity questionnaires administered within each region. such as schools, health clinics, and transportation. To However, using such averages is clearly inferior to hav- examine this relationship, analysts need community- ing data on specific places, as the averages are a less level data on the existence of these facilities. precise measure.The averages can also be particularly Additional useful information would include the aver- misleading when the districts or regions in question age distance that people must travel to get access to are very heterogeneous. (For example, some villages these facilities, how much is charged for the use of may have secondary schools while other villages in the these facilities, and the quality of the various amenities same region have no schools at all.) These arguments that the facilities provide. Sections 3, 8, and 10 of the in favor of coding specific places (which is done in the draft community questionnaire introduced by Chapter standard version of the draft migration module) also 13 are designed to collect information on schools, apply to merging data on rainfall or other risk com- health clinics, and transportation-three kinds of facil- ponents into the data set, as was discussed above. ities that are likely to affect migration. As noted earli- er, the role that good rural transport plays in promot- ECONOMIC INEQUALITY. Several recent studies have ing or discouraging migration is complex and may emphasized the possibility that migrants' decisions to depend in part on whether improving transport relocate are affected by their relative economic stand- encourages household farms or enterprises to adopt ing in the community (Stark and Taylor 1991). cash crop production and small-scale manufacturing. Investigating this possibility requires community-level The section of the draft community questionnaire on data on prevailing economic inequality, which would employment opportunities collects information on permit analysts to establish where a potential migrant's these elements.17 household appears on the spectrum. Given a sufficient As was mentioned above, the community ques- number of household observations within each pri- tionnaire is administered only in communities where mary sampling unit, the relative standing of a given the sampled households are located.As a result, analysts household can be estimated from the household-level can only examine whether the existence of these facil- data. However, analysts must also take into account the ities attracts migrants by studying the respondents' household's likely standing in the community to which migration histories. If analysts are interested in study- it is contemplating moving."S This is not possible with 62 CHAPTER 16 MIGRATION a typical LSMS survey. At best, LSMS and similar sur- policy question is whether, all other things being veys provide analysts with data on the present standing equal, the recipients of these benefits are more likely of each household in the sample. This can be regarded to emigrate than other people. The expanded version as a factor that prompted absentees to leave or per- of the draft migration module includes a few questions suaded migrants to move in, but not as both a push and about absentee emigrants that might be useful in cir- a pull factor in any individual's decision. cumstances where such incentive schemes are impor- tant. These questions deal with whether the absent DISPLACED PERSONS. Although the number of dis- emigrants received state-funded training, how their placed persons is estimated to be very high globally, relocation costs were financed, and whether they were only a few economic studies of the causes of migration recruited by the private or the public sector.Tabulating have taken into account the factors associated with this these data should make it clear who benefits from such phenomenon (Schultz 1971; Barkley and McMillan programs where the programs exist. However, even 1994; Morrison 1993).This could easily lead analysts to with these data, it is not possible to discern how much make erroneous assumptions about other causes. In additional emigration these programs may have caused addition, if high out-migration rates are observed in since there is no information on whether people who places devastated by violence and crime, there may be did not emigrate were denied access to emigration some potential for stemming this migration by rein- incentives.20 forcing law and order. To begin to tackle this subject, analysts need information at the community level on Analyzing the Labor Market Implications of Migration the rate of crimes, violence, or incidents of civil unrest. It is crucial in any study of migration to analyze labor Invaluable information of this type is often available market implications. Important issues include from official records or from NGOs. Such secondary migrants' absorption into the labor market, the effects data can best be merged with the survey data if the spe- of migration on the labor market outcomes of non- cific place of origin-not just the region of origin-of migrants, employment patterns of immigrants, each migrant is coded. This will make it possible for migrants' skills, commuting to work, and job searches. analysts to study the extent to which migrants move from unstable places to more tranquil places. ASSIMILATING MIGRANTS INTO THE LABOR MARKET. The dynamics of migrants' absorption into the labor MIGRATION CONTROLS AND INCENTIVES. Most countries market have attracted particular attention in migration impose immigration controls. Ideally, analysts would literature. Analysts of international migration have like to know the legal status of each immigrant-both examined whether the earnings of immigrants are to check the efficacy of the controls and to examine the lower than the earnings of natives with comparable experience of undocumented immigrants in using facil- experience and education and, if so, whether the gap ities and finding jobs in the labor market. A few coun- closes over time and whether immigrants eventually tries have attempted to impose internal migration con- overtake natives.2' In countries from which many trols."9 In these countries analysts may be interested in guest workers migrate overseas, critics often express finding out how often people are granted or refused a concern about how returning emigrants can be assim- permit to relocate. However respondents may not wish ilated back into the labor market of their home coun- to answer such questions truthfully, and attempting to try and whether their sojourn abroad enhances their collect such information may compromise the remain- productivity on their return. However, these issues der of the survey by alienating respondents in the sam- have not yet been thoroughly researched. pled households.As a result, the draft migration module Analysts of internal migration have demonstrated does not address this issue at all. considerable theoretical interest in the question of On the other hand, some countries have offered whether rural-to-urban migrants tend to make a tran- incentives to their workers to encourage them to emi- sition from informal sector employment to formal sec- grate-usually as guest workers (such as migrant tor employment after working in town for a while workers in the Persian Gulf). Such incentives can (Todaro 1969; Harris and Todaro 1970; Fields 1975; include state-funded training, subsidized relocation Mazumdar 1981). Employment opportunities in the costs, and state-sponsored recruiting. An important formal sector (and workers' concomitant upward 63 ROBERT E. B. LUCAS social mobility) are a key incentive for people who live extent, wages adjust to these movements, although the in rural areas to move to the city. However, there are incidence of unemployment may change instead. Thus few empirical studies that address the issue of transi- important questions arise about what effect in- tion into the formal sector (Banerjee 1983; Lucas migration and out-migration have on the wages of 1985; Vijverberg and Zeager 1994; Marcouiller, de other workers and whether the level of unemployment Castilla, and Woodruff 1997). changes both in the place from which people migrate Issues of migrants' employment transition and and in the places to which they move.These questions assimilation can be studied using data on migrants' are relevant both to international migration (which employment histories, employment data on a cross- may or may not depress the wages of native workers) section of migrants who have arrived in the urban area and to internal migration (which, in the case of rural- or from abroad at different times, or panel data on to-urban migration, could increase the number of the employment. There are tradeoffs involved in using the urban unemployed). What kinds of workers migrate first two kinds of data. Employment histories can con- may influence the answers to these questions. Skilled tain recall errors. On the other hand, employment his- immigrants may enhance the productivity of natives tories are preferable to comparisons among a cross- and, hence, raise their earnings. And the departure of section of individuals, for which unobserved skilled people from a community may prompt other differences may be the real cause of employment dif- community members to invest in their own training. ferences. Using panel data can circumvent both of To understand how migration affects wages and these difficulties, although there is always a danger that unemployment, analysts need to understand how in a very mobile population the panel will become wages are determined in various labor markets and biased-unless migrants are traced or data on the how the demand for workers changes in response to employment status of absentees are collected in later changes in wages.They also need to know what deci- rounds. sions local workers make-decisions about the num- Whatever kind of data are collected, it is possible ber of hours they choose to work, the extent of their for analysts to relate the earnings and employment sta- participation in the labor force, and whether they will tus of migrants and the probability of them having take additional training courses in response to wage changed their employment status (unemployed to changes (Chapter 9). employed or informal to formal employment) to the A few empirical studies examine the effects of amount of time since the migrants moved, their biog- migration on changes in the labor supply and the raphical information (gender, age, education, country implications of such changes for wage formation (for or region of origin, and mother tongue), and, often, a review see Friedberg and Hunt 1995). Some of these their prior job experience.22 Regardless of whether studies use time-series data to examine the evolution analysts use data on migrants'employment histories or of mean wages as migration varies over time. compare individuals at different stages since migrating, However, such time-series analyses can suffer from the it will be necessary to collect some retrospective fact that the observed mean wages include the earn- employment information, to measure migrants' prior ings of migrants (after their arrival at their destination employment experience, and to establish whether the or before their departure from their original location). migrants moved from unemployment to employment This makes it impossible to discover what effect or from the informal sector to the formal sector. The migration has on natives' earnings (Greenwood, draft migration module collects this information; addi- Ladman, and Siegel 1981; Garcia-Ferrer 1980; tional questions on previous employment experience Salvatore 1980; Lucas 1987; Faini and Melo 1994). are included in the draft employment module Cross-sectional studies of this issue have examined the (Chapter 9). wages of natives and related them to intercity varia- tions in the number of in-migrants. However, a dis- THE IMPLICATIONS OF MIGRATION ON THE LABOR torted picture can arise in these studies simply because MARKET OuTcoMEs OF NONMIGRANTS. Much of econ- migrants presumably prefer to locate in high-wage omists' interest in migration focuses on how labor cities. A very few studies have attempted to adjust for migration reduces the supply of labor in one labor this "reverse causality" by relating present in-migration market and increases the supply in another. To some flows to in-migration flows in the past. But such cor- 64 CHAPTER 16 MIGRATION rections are not entirely satisfactory since in the past which commuting represents a viable alternative to people may have migrated to cities that today offer migration. Commuting from a village to a town is not exceptionally high wages (Altonji and Card 1991; the only kind of commuting with importance for pol- LaLonde and Topel 199 1; Pischke and Veiling 1994; icymaking; policymakers must also understand the dif- Hoddinott 1996). Another approach is to examine the ficulty of reaching work from remote urban slums. earnings of natives according to whether their occu- Moreover, understanding the dynamics of commuting pation employs a large number of in-migrants; how- within rural areas may be crucial for establishing the ever, such studies may be very sensitive to the degree extent to which rural public works projects can create of heterogeneity in each occupation (Friedberg 1997). employment. The draft employment module present- Data for cross-sectional studies (either of the ed introduced by Chapter 9 contains relevant ques- intercity or interoccupational type) can be collected in tions on commuting to work, the answers to which the kinds of surveys presented in this book. The draft will enable analysts to compile some cross-tabulations migration module in this chapter identifies individuals on who is commuting and who instead works within as migrants or natives; the remaining data necessary for their own community. It might be particularly inter- analyzing labor market issues can be collected in the esting to compare this information with data from the employment module (Chapter 9). community questionnaire on the availability of trans- port. Researchers may also be interested in exploring PATTERNS OF IMMIGRANT EMPLOYMENT. Policymakers whether a worker's decision to seek or accept a job are interested not only in understanding the dynamics outside of his or her immediate environment depends of the employment of immigrants but also in finding on whether it is possible for the worker to commute out immigrants' sectoral employment patterns. As or whether he or she will have to relocate. To date noted in the first section of this chapter, if the sectors there seems to be no well-developed framework for that employ significant numbers of immigrants receive analyzing this. effective government protection from import compe- tition or receive any form of government subsidy, this JOB SEARCH. The process of searching for a job and is tantamount to encouraging immigration.23 A simple how migration is linked to this process is poorly doc- cross-tabulation of employment by sector and by resi- umented and little understood. (For what documenta- dence status (immigrant or native) would clarify the tion exists see Banerjee 1983, 1984b, 1991; Banerjee picture of these sectoral patterns.24 and Bucci 1994, 1995; Fields 1989; Lucas 1985.) Yet the ability to conduct an effective urban job search MIGRANTS' SKILLS. The first section of this chapter while still in a rural area affects both production effi- mentions several reasons why policymakers are often ciency and income distribution-in important ways. interested in migrants' skills. How much natives earn One of the very important implications of the Harris- depends on the skills of people who migrate into their Todaro (1970) model is that urban job creation and area, and the "brain drain" effect relates directly to the wage increases may cause people from rural areas to skills of emigrants. An LSMS-type survey measures move to cities without having a job, under the several dimensions of skill, including the quantity (and assumption that it will be easier for them to look for a sometimes the quality) of migrants' education, how city job once they are already located in town. On the much and what kind of formal training they have other hand, there is a considerable amount of anec- received, their prior job experience (from their dotal evidence-if little systematic information- employment history), and-in surveys that incorpo- showing that many rural-to-urban migrants find an rate aptitude tests-their inherent skills.25 Descriptive urban job before they relocate (Banerjee 1991). In data on these measures for migrants (including inter- practice, it seems likely that some migrants find a job national as well as internal migrants) relative to non- before migrating, while others migrate to urban areas migrants could be very useful to policymakers. in search of a job. It is useful for policymakers to understand what COMMUTING TO WORK. Commuting is of special inter- differentiates these two groups. Are people with con- est in the study of migration, because policymakers tacts in town more likely to find a job before moving? may be able to influence the circumstances under Are better educated people more likely to find a job 65 ROBERT E. B. LUCAS before moving, either because they have access to more dard of living can be computed from data gathered in job opportunities or because having more education the consumption module (Chapter 5). The standard enables them to access and process relevant information version of the draft migration module presented in more effectively? Are people from villages near towns this chapter includes questions about whether each more likely to find a job before moving than people household associate ever lived with the sample house- from more distant villages (especially since they may be hold, the present location of each associate, and each able to commute into the city to search)? Do people associate's principal current activity.When these sets of who find a job before moving conduct their urban job measures are available, it is possible to cross-tabulate search while being unemployed in the village?26 the type of migration (and employment status) of peo- To help analysts study these issues, the standard ple who have left a particular family with that family's version of the draft migration module includes ques- standard of living. tions that ask inigrants how long after migrating they One weakness of this analysis is that the family's started working and whether they were already offered standard of living may have been affected by the their job before moving. It would then be possible to departure ofthe migrant. If panel data are available, the relate this information to the migrants' education, family's living standard before the person left can be whether they had family members or other acquain- measured directly using data from the first round of tances already in the toxvn, the distance to their previ- the survey. In the absence of panel data, it is useful to ous home, whether they visited the present town to find out what the family's standard of living would look for a job prior to moving, and their employment have been had the person not left.27 This requires an status (especially, whether they were unemployed) understanding of how the living standard of the prior to migrating. remaining family members is altered by the departure of migrants. Analyzing the Implications of Migration for Income Distribution THE EFFECT OF THE MIGRANT'S DEPARTURE ON THE To perform a complete analysis of migration's effects FAMILY'S LIVING STANDARD. There are several aspects to on income distribution, a wide range of links must be the question of whether the remaining members of a analyzed. While computable general equilibrium migrant's family experience a lower or higher standard models are sometimes used to simulate the interac- of living as a result of the migrant's departure. While tions arising from some of these links (Adelman and living with the family, the migrant may have consumed Robinson 1978), household survey data alone can cer- more or less than he or she contributed to the family tainly yield useful insights into a number of the key income. In addition, if the migrant sends money back elements involved. to the family, this means that not all of the migrant's income is lost to the household. Remittances from the THE ECONOMIC STATUS OF MIGRANTS' FAMILIES. Are migrant (or simply the implicit insurance offered by migrants more likely to come from rich or poor fam- migrant members) may enable the family to make ilies? Does this pattern differ depending on the type of extra investments-increasing the income-generating migration involved (such as rural-to-urban migration, potential of the remaining family members. rural-to-rural migration, or international migration)? On the other hand, the migrant's departure could How is the employment status of migrants influenced cause remaining members of the household to lose by the income class of their home family? In particu- access to certain productive assets. For example, the lar, are rural-to-urban migrants from poor households migrant might rent out some land that he or she owns. more likely to be unemployed than migrants from Alternatively, women left behind in a household may xwealthier households? The answers to these questions be denied access to communal assets because social are critical for building a picture of the effects of norms decree that only men can use them. The migration on income distribution, yet the information remaining members may be induced to work harder that has been collected to date, in previous LSMS sur- than they did before the migrant left-either to offset veys or elsewhere, is very limited. the loss of income previously generated by the So what sort of analyses are feasible using house- migrant or because the migrant's departure would hold survey data? Measures of each household's stan- otherwise leave productive assets idle. However, if the 66 CHAPTER 16 MIGRATION departure of the migrant means that a greater burden families sometimes send their members to places of child care or other non-income-earning activities is where their potential for generating income will not laid on remaining household members, they may not be affected by local fluctuations in income levels. If have time to do this extra work. this insurance strategy is effective, analysts should The various modules of LSMS surveys are expect to find that families with migrant members liv- designed to gather data that can be used to calculate ing elsewhere are better able than other families to the net outcome of these effects directly, even with a weather local fluctuations in earnings and thus main- single cross-section of data. However, having panel data tain their consumption levels during both good and would be particularly helpful in this respect. If analysts bad times. One key factor in a family's capacity to do had panel data on consumption, they would be able to this is whether the migrant member is sending them find out if consumption per person fell more in fami- money-an issue that will be discussed in the next lies from which a migrant departed between panel subsection. In the meantime, the question is how rounds than in families from which no one left.28 household survey data can be used to explore whether Indeed, it would be possible to find out in which kinds households with migrant members are better able to of families (rich or poor) this happened, and for which smooth their consumption than other households. kinds of migrant (for example, household heads or Panel data are probably the best kind of data for adult children of elderly parents).Without panel data, exploring this issue, but some analysts have used data analysts can only establish whether per capita con- from surveys that have interviewed households more sumption is lower among households that a migrant than one time during the year to examine how families has left. Considerable care is then necessary to address smooth their consumption across different seasons of the the statistical pitfalls of reverse causality-the chance year (Paxson 1993). Nevertheless, the information about that the person's migration was actually caused by the household associates that the draft migration module family's low consumption level.29 provides is sufficient for analysts to relate any consump- Even if the consumption levels of remaining family tion fluctuations to the number of migrants who have members stay the same, these members may still be departed from the family, as well as to the migrants'rela- worse off if since the migrant's departure they have had tionship to the household and current place of residence to work harder to sustain their consumption level; work- (Kochar 1995). In forming this relation, it is important load must be considered an aspect of living standards. to consider other key factors that may also affect con- The employment, household enterprise, agricul- sumption smoothing (see Chapter 5) as well as the ture, and time-use modules in this book are designed potential for reverse causality-that excessive fluctua- to collect data on both work outside the home and tions in consumption actually prompt migrants to leave. work in household enterprises and family-run agri- cultural enterprises. But for the present purpose, the Analyzing Remittances standard version of the employment module offers Remittances between migrants and their original sufficient information on the employment activities of households have several policy implications (some of household members. which were mentioned in the first section of this chap- Having panel data from the employment module ter). Policymakers need to know if the creation of high- would be very useful because it would allow analysts paying urban jobs benefits the rural poor because the to relate any changes in the amount of time that each workers send remittances to their families in rural areas family member allocates to various employment activ- (Stark, Taylor, andYitzhaki 1986, 1988). Also, they need ities to the migration of a household member. With to know whether families insure themselves against only one round of data, analysts are restricted to com- local income shocks by sending family members to live paring time allocation in families that a migrant has in areas unlikely to be affected by the same shocks left with time allocation in families that no one has left (Lucas and Stark 1985, Hoddinott 1994). Another issue (while controlling for other factors likely to affect of crucial interest to policymakers is whether the exis- these time allocation decisions).30 tence of government and private transfer programs causes migrants to send less money back to their fami- DEALING WITH A TRANSITORY INCoME Loss. The earli- ly, in the belief that the family will receive help from er discussion of the causes of migration noted that these programs instead (Cox andjimenez 1997). 67 ROBERT E. B. LUCAS Remittances sent by migrants to a household with ates (including absentee migrants). Answvering this which they used to live or from that household to question requires measuring the household's prosperi- the migrant-are only one of many different kinds of ty.The family's current consumption level presumably private interhousehold transfers (one example being depends, to some extent, on the level of transfers that remittances between individuals and families who have it receives, which means that using current consump- never lived together). A large number of these other tion as an indicator of the family's prosperity would be transfers may also need to be considered in analysis. To misleading. If household income data are available, an address this issue, it seems practical for analysts to alternative measure of the family's prosperity might be examine the transfer relationships between a household total household income minus any transfers from asso- and the entire set of people who have been identified ciates (see Chapter 17 on the relative merits of col- as household associates. Some of these household asso- lecting income data). But even this measure can be ciates (possibly including former family members) will misleading, because receiving transfers may affect deci- send nothing to the household, and the household will sions family members make about working. send nothing to some of its associates. Analysts need to A third alternative is to measure what assets the know not only what kinds of associates send transfers family has that can be used to generate income. These to a household but also what kinds do not. This is why assets can include not only physical assets such as land the transfer section of the miscellaneous income mod- or household enterprise assets but also human capital ule is designed to collect data on transfers-both in assets such as level of educational attainment of adult cash and in kind-between the family and its house- members of the household (which influences these hold associates (see Chapter 11). members' earning potential). Data on these income- generating assets are collected in the agriculture, TRANSFERS AND INDIVIDUAL HOUSEHOLD ASSOCIATES. household enterprise, and education modules.31 This Several factors influence the likelihood of associates third alternative could be extended to incorporate sending transfers to a particular household. These fac- components of household income unlikely to be tors include whether the associates ever lived with the affected by current transfers from associates, such as family and how long they have been away, how close- pensions or social security payments. Incorporating ly the associates are related to household members, these components would enable analysts to explore whether the associates have established their own fam- the hypothesis that families receiving benefits from ily elsewhere, and how much the associates earn. Data government and private transfer programs are less like- on a migrant's relationship to the members of his or ly to receive transfers from migrant relatives. her original household should be available from the roster and extended roster. However, data on the other TRANSFERS AND RISK. Analysts may also be interested in factors are not available elsewhere in typical LSMS- relating transfers from associates to whether the family type surveys. Therefore, the standard version of the has a temporary loss of income for a reason beyond its draft migration module includes some questions about own control. For example, does the family receive more household associates, such as whether they are now transfer income from migrant relatives when the major married and living with a spouse and/or children, earners in the household become ill? This can be stud- whether they ever lived at the family dwelling and ied by relating data on net transfers to data on workers when they left it, whether they are employed, what being ill and unable to work. (See the health module kind of work they do, and where they currently live. introduced by Chapter 8 and the labor force participa- These data (together with information from the roster tion section of the employment module introduced by and education modules about each associate's age, Chapter 9.) Similarly, if a community suffers from gender, and education) can be used to estimate an drought, flood, or crop disease, agricultural income may associate's earnings-making it possible to analyze decrease but transfers may increase. (See the history and transfers in relation to earnings. development section of the community questionnaire introduced by Chapter 13.) TRANSFERS RECEIVED BY RICH AND POOR FAMILIES. Policymakers may want to know whether rich or poor INVESTMENTS AND TRANSFERS. One issue that often families receive more transfers from household associ- arises regarding transfers is whether the recipient fam- 68 CHAPTER 16 MIGRATION ily invests the money it receives. However, this ques- One recent study noted that when panel data on tion is irrelevant.32 If part of the cash transferred was households are available, it is possible to find out spent on an investment item, this does not imply that whether high-income or low-income households are the transfer has caused a net increase in the family's more likely to split up-either through the migration of investments; the family may have been prepared to some members or through the establishment of a sec- make that investment anyway using an alternative ond household in the same location (Foster and source of cash. In addition, even if the migrant does Rosenzweig 1996). It is important for analysts to know not send any transfers the family may invest more or whether high-income or low-income households are adopt riskier productive activities in the knowledge more likely to split up because if some of the members that the migrant's location elsewhere provides the of rich households leave and take assets with them, these family with a financial safeguard. And if a family households may not appear to become much wealthier invests its transfer income-for example, by buying over time, but the total wealth of the original group of land-this does not imply that investment in the household members may have increased substantially. As whole economy has risen; investment in the whole important as this line of analysis may be, the methodol- economy is affected by how the seller of the land ogy for performing it is very much in its infancy. spends the money that he or she makes from selling it. Perhaps the most obvious way people join a fam- If analysts nevertheless wish to obtain cross-tabu- ily is through marriage. In many cases marriage lations of households' agricultural investments and involves the migration of one or both partners.Yet it adoption of advanced agricultural technologies against is too simple to view marriage as the cause of migra- the level of transfers received by the household, they tion. The decision to marry and the decision to can do so using LSMS-type survey data. (See Chapters migrate, while difficult to disentangle statistically, 18 and 19 on household enterprises and agriculture.) should be viewed as both discrete and interdependent (Behrman andWolfe 1985). REMITTANCE INCENTIE PROGRAMS. There are at least As mentioned in the first section of this paper, the two major difficulties involved in asking whether emi- importance of reunifying families is a concern of social grants took advantage of tax breaks or exchange and policymakers. This concern has certainly shaped interest rate incentives to send more money back to immigration policy in a number of countries. their home country. First, information may not be reli- Conjugal separation resulting from migration may be able; collecting information about taxes is notoriously of particular concern to policymakers, as may the difficult, and it is not possible to ask respondents effects of parental absence on child care. The draft whether they brought in money through the black migration module collects data on the location of market. Second, it can be difficult to analyze whether household associates. Combined with information on those who take advantage of the incentive programs family members currently living in the household, would have remitted more anyway.33 Therefore, the these data can be used to study the likelihood of mar- draft migration module does not contain questions ried couples living apart, along with such issues as the about whether absentee emigrants took advantage of likelihood of small children having absent, migrant remittance incentive programs. parents. Descriptive data on these phenomena can be informative, even though there is not yet a specific Analyzing Migration and Family Structure. framework to more comprehensively analyze these The composition of a cohabiting family can be altered factors (see, however, Banerjee 1984a). not only by births and deaths but also by the arrival and departure of members. When these arrivals and Using Household Survey Data to Address Policy Issues. departures come from migration (and not merely a The extent to which satisfactory answers can be found change of residence within a town or village), migra- to questions about migration and migration policy tion and family composition become intimately asso- varies considerably, depending in part upon the state ciated. Since almost all existing measures of income of current methodology. The extent to which data distribution and poverty are based on the cohabiting from the draft migration module can address migra- family, these measures are influenced by migration's tion issues depends heavily on whether certain data are effects on family composition. gathered in other modules in the survey. (Such data 69 ROBERT E. B. LUCAS will be discussed in the next subsection.) Box 16.1 accurate than coding specific places.The averages can summarizes questions that can be addressed using a be particularly misleading when districts or regions are well-designed LSMS-type survey as well as issues that heterogeneous with respect to the measure merged. cannot be tackled as effectively.This box has been pre- Table 16.1 also broadly indicates how well specif- pared under the assumption that necessary data from ic migration issues can be analyzed given the collec- other modules will be available. tion of appropriate data from migration and other modules. Ranking the analysis potential of these Links with Data from Other Modules. issues, from "excellent" to "not possible," is intended to Migration affects and is affected by many aspects of reflect the remaining limitations of the data, the diffi- individual and family behavior. As a result, the analysis culties involved in analyzing each issue, and how of migration is extremely dependent on links with meaningful the results will be for different areas of other modules of the surveys. Which modules are understanding. Separate rankings are provided for the included in a given survey and which are excluded short, standard, and expanded versions of the migra- determines the aspects of migration that can be ana- tion module. The rankings differ where significant lyzed with resulting data. Links with other modules additional material is incorporated into the standard are summarized in Table 16.1. In addition,Table 16.1 and/or expanded versions that is not included in the itemizes the questions within the migration module short version. that make it possible for specific issues to be analyzed. The relative importance of data from the other mod- New and Unexplored Areas of Analysis ules is categorized as follows: A number of the areas of possible analysis mentioned * Required. A link without which analysis is impossi- in this section are not yet well-developed-with some ble. still very much in their infancy. Nevertheless, collect- * Recommended. A link that is extremely desirable for ing enough data in these areas to construct cross-tab- analysis but not utterly indispensable. ulations can be very helpful to policymakers. And hav- * Other. A link that would be useful, typically for ing such data may eventually help analysts develop extended or complete analysis. more rigorous approaches to some of these little- Most of the table's references to other modules explored issues. Such issues include: should be self-explanatory. However, three references * Whether the prior migration of family members, deserve separate mention. "Roster (B+C)" refers to family reunification, and marriage are causes of sections B and C of the standard household roster; migration. these sections collect information on the nonresident * Whether information networks influence the deci- parents and children of household members. sion to migrate. "Additional roster questions for associates" refers to * Migration costs and their effect on migration deci- the questions in the draft migration module that are to sions. be included in the roster section so as to gather infor- . The link between economic risks and migration mation on household associates. And "secondary data" decisions. refers to data on communities in which the commu- * The influence of facilities on migration decisions. nity questionnaire has not been administered. * The effects of violence and displaced persons on Secondary data about such communities-data on, say, migration. available facilities or incidence of drought-must be . The role of commuting as a possible alternative to gathered from secondary sources (such as administra- migration. tive data from government ministries) after the house- * The link between migration and job search. hold survey has been conducted. * The income class of migrants' families. The key to merging secondary data with data on * The effects of out-migration on family living individuals is the coding of specific place names men- standards. tioned in the migration histories.Where place names * Consumption smoothing and migration. are not coded, secondary data can still be merged, for * The relationship between migration and invest- example using averages for the relevant district or ment and transfers. region. However, using such averages is much less * Migration and family structure. 70 CHAPTER 16 MIGRATION Box 16.1 Policy IssuesThat Can and Cannot Be Analyzed Using Household Survey Data Issues that can be addressed with household survey data Migration and economic risk Migration pattems . Is out-migration more common from areas with more sub- What are people's most recent moves? stantial inherent risks (such as drought, flood, and disease)? * What are people's five-year place-to-place histories? * Do migrants send more remittances when conditions at * What are people's lifetime place-to-place histories? their original home are temporarily worse? * Is there step-migration? What is it like? * Are families with migrant members living away from home * Is there return migration? What is it like? better able to maintain consumption levels through difficult * What are the patterns of immigration? times than families without such migrant members? * What are the patterns of emigration? * Do migrants reduce the amount of remittances they send back to a home facing economic adversity when insurance Causes of migration or other transfers (such as social security) are available? * How do earnings and employment opportunities affect population movements? Social issues and migration * What are the important factors that limit geographical mobil- * To what extent does migration resuft in conjugal separation? ity? * How important is local violence as a cause of migration? * Can constraints on mobility be addressed through policy * How is family structure affected by the arrival and depar- actions? ture of migrants? * Does having wealth enable more family members to migrate by making migration more affordable? Brain drain and emigration * Are migrants more likely to leave a community in which * What is the educational background of emigrants? their relative economic standing is low? * Who finances their education? * Are migrants attracted by good infrastructure facilities? * Do emigrants return home? If not, do they continue to * How important are local violence and unrest in inducing send transfers? departure? * Do good transport opportunities promote migration or Immigration relieve the necessity to migrate? Under what circum- * Does trade protection encourage industries that are stances? major employers of immigrants? * To what extent is the propensity to migrate less in com- * Do immigrants undercut local wages? munities far away from migration destinations? * Do foreign students stay on to work in their host countries? * How important are family reunification, the prior migra- * Are family members of immigrants employed? tion of family members and friends, and relocation upon * Do family members of immigrants receive state aid and marrying as determinants of migration? spending? * How is the decision to migrate affected by the existence of considerable economic risks in the initial location and Employment and migration limited economic risks in the destination location ? * How do immigrants and internal migrants find their first job in a new place? Social mobility and migrotion * Do immigrants and internal migrants improve on their ini- * How great are the economic retums of migration? tial employment (better pay, greater stability, full-time)? * Do the incomes of immigrants eventually catch up with * What are the effects of immigration and internal in-migra- the incomes of natives? tion on the earnings and job prospects of local nonmi- * Do the incomes of internal migrants eventually catch up grants? with the incomes of local nonmigrants? Issues for which household survey data are of limited use Poverty incidence, income distribution, and migration * What are the relative costs of moving jobs to workers * Do migrants come from richer or poorer families? rather than moving workers to jobs? * Do the remaining members of migrants' families work * Do strict controls increase the rate of illegal immigration? harder? * Are penalties on the employers of illegal immigrants effec- * Are the remaining family members enriched or impover- tive, or do they simply encourage discrimination against all ished by the departure of migrant household members? aliens? * Do migrants transfer more money to richer or to poorer * How effective are policy incentives to encourage transfers families? from abroad? 71 ROBERT E. B. LUCAS Table 16. la Migration Module: Requirements and Links with Other Modules, ShortVersion Migration Additional module roster question Links with other modules Abi ity to Aspect questions for associates Requ red Recommended Other analyze Patterns of migration individuul dota Most recent move 1-9 Roster Excellent 'l.......... ...............................................................................*......................................................................................................................................... Five year pace-to-place history 1 13 Roster Excellent Lifetime place to-place hist ory 7 Roster Excelent Generoi trends Step-migration 1-13 Good Return migratFon 13 Excellent .... ......... ....... .............................................................................................................................................................................................................. Immigration 1-4, 14 Roster Excellent Emigration -6 Roster (B+fC) Extended roster Excellent ................................................................................................................................................................................................................................... Causes of migration Methodoiogy Se f-reported causes Not possible Mover- st ay defnisons 1-13xExcellent Cause Persona attributes Roster Education Excellent Earnings and employment Roster Employment Education Good Family and marriage .6 Roste B+C Extended roster Far Distance and transport 8 9 Community Fair Wealth and finance Consumption Agriculture, Poor Household enterprise , Risks 8 9 Community Secondary data Consumption, Fair Agriculture, Household enterprise Facilities 8-9 Communty Secondary data Fair Economic inequai ty Consumption Poor Vioence 8 9 Communty Secondary data Fair Contro a Not posil Labor market implications Effects on natives 1-9,14 Roster, Employment Household enterprise Fair Assimi ation of migrants 1-4, 4 Roster, Employment Househo d enterprise Exce lent immigrant empioyment patterns i-4, 4 Roster Employment Exte lent M grant skills i- 4 Roster Educat on Employment Excei ent Brain drain i-6 Roster (B--C) Extended roster Poor ..................... ................................................................ ....... .............................................................................................. ................................ Commuting I-14 Community Em..y ment ...............Et Good job search and transtion Not........pos...........ble... Income distribution Income class of migrants' families 1-6 Consumption Miscellaneous ncome, Good Employment ................................................................................................................................................................................................................................... Effect of migration on family I 6 Roster (B+C), Employment, Good living standards Consumption Extended roster ............................................................................................................................................................................. ............ 11........ Consumption smoothing 1-6 Roster (B+C), Extended roster Good Consumption, .. .................. ................... Remittances and famniy income in6 Miscellaneous income, Extended roster Agriculture, Ea Roster (B-C), Household .....................................................*.................................................*.......................................................................................................................... Consumption 72 CHAPTER 16 MIGRATION Table 16.1 a Migration Module: Requirements and Links with Other Modules, ShortVersion (continued) Migration Additional module roster question Links with other modules Ability to Aspect questions for associates Requ red Recommended Other analyze Remittances and risk 1-6 Miscellaneous income, Extended roster Good Community, Employment. Roster (B+C) Remittances and individual migrants 1-6 Miscellaneous income, Extended roster Good Roster (B+C) Policies to promote remittances Not possible Investments and remittances Miscellaneous income, Extended roster Fair Agriculture, HH Enterprise, Roster (B+C) Marriage, migration, and family structure 1-6 Roste, Roster (B+C) Extended roster Fair Source: Authors evaluation of the migration module Table 16.1 b Migration Module: Requirements and Links with Other Modules, Standard and Expanded Versions Migration Additional Ability to analyze module roster question Links with other modules Standard Expanded Aspect questions for associates Required Recommended Other module module Patterns of migration Individual data Most recent move 1-11 Roster Excellent 1-10 Roster (B+C) Extended roster Excellent Five year place-to- 1-1 1, 36-40 Roster Excellent place history iIi0 Roster (B+C) Extended roster Excellent Lifetime place-to- 1-6 Roster Excellent place history 1-10 Roster (B+C) Extended roster Excellent ................................................................................................................................................................................................................................... General trends Step-migration 1-1 1, 36-40 Roster Good Return migration 1-1 1, 36 40 Roster Excellent ..........................................................................*................................................................................................................................................ Immigration 1-1 1, 41 Roster Excellent Emigration 1-10 Roster (B+C) Extended roster Excellent ................................................................................................................................................................................................................................... Causes of migration Methodology Self-reported 12-14 Fair ...................................................................................*............................................................................................................................................... Mover-stayer defined 1-11,36-41 Excellent ................................................................................................................................................*.................................................................................. Couse Personal attributes Roster Education Excellent Earnings and employment Roster Employment Education Good Family and marriage 5-17 1-7 Rost er (B+C) Ex ended ros er Excellent Distance and transport I0 Community Good Wealth and finance 27-35 Consumption Agriculture, Poor Good Household enterprise Risks i0 Community Secondary data Consumption, Excellent Agriculture, Household enterprise .......................... I: ......... .................................................. on... i ...........................................................................................*............................ Facilities IC Community Secondary data Excellent Economic inequality Consumption Poor Violence IC Community Secondary data Excellent ..............s...........................*..................................................................... ........................................................................................N... ot... possible................. Controls Not possible ................. *...................................................................................... .......................................................................................................................... (Table continues on next page) 73 ROBERT E. B. LUCAS Table 16.1 b Migration Module: Requirements and Links with Other Modules, Standard and Expanded Versions (continued) Migration Additional Ability to analyze module roster question Links with other modules Standard Expanded Aspect questions for associates Required Recommended Other module module Labor market implications Effects on natives I -I I, 41 Roster Employment Household enterprise Fa r Assimilation of migrants I I 1, 41 Roster Employment Household enterprise Exce lent .................................................................................................................................................................................................................................... Immigrant employment patterns I-I 1,41 Roster, Employment Excellent ....................................................................................................................................................*.............................. "'",.....................---,--......... Migrants' skills I - I1, 41 -47 Roster, Education Employment Excellent .......... ... ..u"''ng........................................................................................... ... ..u"'ity................................................*............................................""d..................... Brain dra n 12-16 Roster (B--C) Extended roster Good Excellent Commuting Community, Good Employment Job search and transition 18-35 Roster Excellent . ............................................................... ...................................................................................................................................................................................................................................................................................... Income distribution ncome class of migrants' families 1 10 Consumption Miscellaneous income, Good Excellent Employment Migration effect on family I -.IC0 Roster (B+C), Em ployment Good living standards Consumption Extended roster .....................................................................................................................................,............................................,........................................ Consumption smoothing 1 10 Roster (B+C), Extended roster Good Consumption " .. .............................................................................................................................................................................................................................. Remittances and family income 1-10 Miscellaneous income, Extended roster Agriculture, Fair Roster (B+C), Household Consumption enterprise Remittances and risk I 10 Miscellaneous ncome, Extended roster Good Excelient Community, Employment, Roster (B+C) Remittances and ndividuai migrants I -IC Miscellaneous income, Extended roster Excel ent Roster (B+C) .................................................,- - - ,- ,................................. ................ ..................................... ........................... Policies to promote rem ttances Not poss ble ..... .................................................................................. *............................................ *............................................................................................... Investments and remittances I 0 Miscellaneous income, Extended roster Fair Agriculture, HH Enterprise, Roster (B+C) ..... ................................................................................................................................................................................................................... Marriage, migration, and family structure I-10 Roster Roster (B+C) Extended roster Good Soircer Authors evaluation of the m gration module. The Migration Module * Whether the migrant had family or other contacts in the new area prior to his or her most recent Three draft versions of the migration module are pre- move. sented in this section. Notes pertaining to a few spe- * The educational attainment of household associates cific questions in these draft questionnaires and some (for brain drain analysis). general issues of definition in relation to the question- . The current principal economic activity of house- naire are contained in the notes on the migration hold associates. module (the final section of this chapter). * The family members with whom associates live. Principal additions to the standard version of the * Whether associates lived with the household in a module not included in the short version are: previous dwelling. * Job search and the employment transition associat- The main suggested additions to the expanded ed with the migrant's most recent move. version of the module not included in the standard * Coding of the migrant's place of birth and place of version are: previous residence. * Whether each household member ever lived * The reasons for the migrant's most recent move and abroad. the role of amenities in affecting the migrant's deci- * The location of a member's residence prior to mar- sion to move. rying. 74 CHAPTER 16 MIGRATION Box 16.2 CautionaryAdvice How much of the draft module is new and unproven? The * How well has the module worked in the past?The data gath- draft migration module differs substantially from the migra- ered on migration in previous LSMS surveys have not tion modules that were included in previous LSMS surveys. been studied very extensively but this may be because Even the short version of the module is designed to col- LSMS surveys to date have collected only extremely lim- lect more information on respondents' migration histories ited information on the subject. than previous surveys, which usually collected only a por- tion of a migration history (such as the migrant's province * Which parts of the module most need to be customized? of birth and previous residence). Many censuses and sur- Whether the list of household associates should include veys have collected information on migration histories, and people on an extended roster (as opposed to just non- there is at least some evidence that recall data on major resident parents and children of household members) migration moves is reasonably reliable. Collecting informa- may depend upon the specific social setting as well as tion on household associates is new to LSMS-type surveys, on how these data are likely to be used in analysis. although data on nonresident parents and children of Where analysts are planning to study interhousehold household members have frequently been collected with- transfers, the list of associates may need to include in the roster portion of previous LSMS surveys.Also, other absent spouses of household members and, in many kinds of surveys (such as the Botswana National Migration societies, members' siblings. In contexts where emigra- Survey) have successfully collected migration information tion is a comparatively rare event, the questions on on this "absentee" basis. Indeed, most of the components brain drain could easily be omitted. In places where of the standard version of the draft migration module have internally displaced persons and international refugees been successfully adopted in other contexts. However, a are of particular concern, it may be appropriate to add few aspects of the expanded version are new, including the some related questions to the draft migration module questions on the wealth of the migrant's original family and outlined here (as well as to the community the migration histories of nonresident associates. questionnaire). - The wealth of the family with which the member of guest workers overseas, survey designers may want lived prior to his or her most recent move. to include the expanded version of the emigration * More details on the migrant's employment before section. They may also wish to include the questions and after his or her most recent move. in the expanded module about whether the person - More details on emigration (for brain drain analy- ever lived abroad-making it possible for analysts to sis). study the experiences of emigrants who have * The migration history of household associates. returned. * Whether associates ever lived in the place where the interview is taking place. Notes on the Migration Module * More details on the current employment status of associates. It is important to clarify some issues about the draft The standard version is the form recommended in module, relating to both general definitions and spe- most contexts. The expanded version incorporates cific questions. some additional questions that are specific to particu- lar contexts or useful for extended analyses.The short General Definitions version may be used in surveys for which migration is A "place" normally refers to a concentration of a low analytical priority. Appended to each version is dwellings. For migration purposes an entire metropol- a set of questions to be included in the household ros- itan area, including its immediate suburbs, might be ter module (which is discussed in detail in Chapter 6). considered one place, even though this definition These questions gather data about household associ- would rule out the possibility of analyzing migration ates that can extend the analysis of migration in essen- within metropolitan boundaries. Similarly, an entire tial ways. village (and its associated land area) is normally con- Which version of the module designers include sidered a place. However, where settlement is very dif- in their survey should depend on prevailing local fuse, survey designers will need to define the concept circumstances. In countries that send large numbers of a place very carefully prior to fielding the migration 75 ROBERT E. B. LucAs module; preferably this definition will follow some Short Version notion of a community. (See Chapter 13 for a discus- The comments in this section explain questions in the sion of the concept of communities.) short, standard, and extended versions of the draft To analyze place-to-place migration effectively, module. the place of current residence must be identified as precisely as possible. At the very least, analysts need to 2. If the interviewee has lived in this place for more know a comparatively small administrative area (such than one period of time, interviewers should establish as a district or county) in which the place of inter- and record how long the person has lived there since view is located, as well as whether this is an urban or he or she last stayed in another place for three months rural place.Without such information, analysts cannot or more. observe patterns of migration among administrative areas or between urban and rural areas. Where confi- 3. The purpose of this question is to find out where the dentiality rules permit, it is desirable for the name of person first lived. This does not necessarily refer to the the town or village where the interview was con- person's place of birth; the birth may have occurred in ducted to be revealed. Knowing the town name a hospital or while the mother was traveling. allows more effective merging of secondary data, more accurate recording of distances, and more 12. It is best to reword this question so it refers to a detailed examinations of place-to-place migration. In well known national event that occurred around five most previous LSMS surveys, it has been possible to years earlier. release the names of the places where the interviews were conducted. Standard Version Throughout the draft migration module, to "live" 2 AND 3. Questions 2 and 3 are the same as questions somewhere means to eat and sleep in this place for 2 and 3 of the short version; see notes above. three months or more. In some societies people dis- tinguish between "staying" in a place-meaning eating 5. Place names will usually need to be coded after the and sleeping there-and "living" in a place-meaning initial interview. The main purpose of question 5 is to owing allegiance to that place. In these cases the more merge information about specific places, such as dis- appropriate term to use is "staying." tance to the current place of residence, amenities avail- The term "province codes" is used throughout the able, incidence of disasters, or variability of rainfall. If a draft migration module. These codes should refer to country can be divided into a small number of small areas. In some countries "county" or "district" provinces or regions and the above measures vary little may be a more appropriate term. Interviewers must be from place to place within the "province," the code for provided with guidelines about what constitutes an each province can be printed immediately below the "urban" area in a country. question, and no coding will need to be done after the "Household associates" are all the living people interview. If the place is not coded until after the inter- identified in Section B and C of the household roster: view, it is useful to obtain a set of codes that are disag- that is, all living parents and children of household gregated to the level of the district or the county. members who are not themselves members of the household. The additional questions for household 8. Interviewers should be careful to enter the person's associates should also be asked about a head of house- age, not the year in which his or her move occurred. hold who has been absent for more than three months, even if this person is still considered a household mem- 10. Question 10 is similar to question 5 of the standard ber. If an extended roster is applied-identifying spous- version; see notes above. es, siblings, or other relatives of household members who are not themselves members-these people 12. Enter the single most important reason for moving should also be included on the list of household asso- to the current place of residence. ciates.The ID number referred to in the associates sec- tion of the migration module is the ID number these 32. If the answer to question 32 is more than 60 hours or associates were given in the household roster module. less than 10 hours, the interviewer should make sure the 76 CHAPTER 16 MIGRATION respondent has understood that the time period being 4. See Lucas (1997, 1998) for surveys in the context of internal examined is one week (not one month or one day). migration in developing countries. 5. The term "facilities" is used in this chapter to refer to a set of 34. This question refers to the number of people who amenities that are available in a given place, such as elementary worked for the interviewee's firm or employer-not schools, health clinics, or services providing clean drinking water or the number of people who worked at his or her plant electric power. or job site. 6. This issue probably becomes more severe the further back in time the analysis goes, which is a good reason to analyze more 8 (QUESTIONS TO ADD TO ROSTER). The head of the recent migration decisions-assuming there are enough recent household (and sometimes even the entire household) migration outcomes to make such an analysis feasible. may have lived in another dwelling within the previ- 7. These earnings and employment opportunities should ous five years. If this is the case, question 8 of the addi- include the informal sector as well as wage employment (Fields tional questions for household associates should estab- 1975; Mazumdar 1981). This means that data from the household lish whether the associate lived with the household enterprise module are needed to estimate the labor component of (head) at that time. informal self-employment earnings (see Chapter 18). In addition, it is conceivable that personal attributes have quite different effects on Expanded Version earnings in the informal sector. If so, there is a need for a separate When the expanded migration module is used, two study of these earnings patterns as well as of the forces that affect questions need to be inserted into the standard version distribution of workers between the formal and informal sectors. of the household roster module immediately follow- 8. In the more sophisticated approaches, it is explicitly recognized ing the question on marital status. The first question, that the choice of location is endogenous to this process; hence, an "Where did you live during the three months before attempt is made to correct for positive or negative selection. In other you were first married?" should use answer codes for words, migrants may differ in ways that are not reflected in the meas- provinces and foreign countries. The second question, ured attributes alone. See Falaris (1987) and Pessino (1991). "Was the place where you lived just before you were 9. It may also be interesting to control for whether potential first married an urban area?" should use the answer migrants have other friends and family in the location to which codes "1" for yes and "2" for no. If a respondent has they are considering moving. However, collecting this information been married more than once, the second question is more complex-partly because of the difficulty of being precise refers to his or her first marriage. about "knowing" someone. Information about friends and more distant relatives is not collected in the draft module. Notes 10. See Fuller, Lightfoot, and Kamnuansilpa (1985) for descrip- tion and analysis of such an attempt in Thailand. The author is very grateful to Gary Fields, John Harris, Julie 11. Some analysts have hypothesized that precisely the opposite Schaffner, and the members of the LSMS authors' workshop for holds: that the propensity to migrate is lowest in the middle- substantive comments on earher drafts. Fiona Mackintosh provided income range. See Connell and others (1976), Baneijee and Kanbur invaluable editorial contributions. Special thanks are due to Paul (1981), and Stark and Taylor (1991). Glewwe and Margaret Grosh for the major and patient role they 12. Social assets (such as local networks) may also tie individu- played in shaping this chapter. als to a specific location, although to date this has not been tested 1. In this chapter the term "remittances" refers to private inter- in a very systematic way. See Jagannathan (1987). household transfers of money or goods. 13. This distinction has been neglected in the few empirical stud- 2. Such relationships are often referred to as micro migration ies that have attempted to examine the role of wealth in promoting equations, in contrast to macro migration studies-which (often or limiting migration. An even more complex issue arises when one using cross-tabulations of census data) analyze the proportion of a takes into account the wealth of the migrant's extended family-from population that has migrated. which migrants borrow in some societies. See Ilahi andJafarey (1995). 3. One weakness of adopting such a simple dichotomy is that 14. Note, however, that panel data on households can help some people may have moved into town and then returned to the bridge this gap by providing data on wealth in the early round and rural area prior to sampling. Note also that if migration abroad is to data on subsequent migration in later rounds. be studied on this basis, emigrants who are identified as household 15. For residents who have a migration history, the appropriate associates need to be included in the sample. measure is really the economic risk faced by the family from which 77 ROBERT E. B. LUCAS they migrated.The migration draft module does not collect data on entirely feasible if the migrant received a disproportionate share of this measure because of the likelihood that respondents will have consumption. difficulty remembering such information accurately 29. The problem of reverse causality is not entirely removed by 16. Note that this is not true for data on resident migration histo- the availability of panel data; people may be persuaded to migrate ries, where the appropriate community data xvould refer to the com- by the prospect of their consumption falling in the future. munity that the migrant left behind, on which data are not collected. 30. For references and a discussion see Chapter 9 on 17. Using data from the price questionnaire, it may be possible employment. to study the indirect effect of better transport (and hence easier 31. Indeed, there is an additional interest in relating transfers access to external food sources or markets) on prices (such as food from household associates to the availability of such assets, as it is grain prices); the resultant price changes can also affect migration. possible that people send transfers in the hope of eventually inher- 18. There is some difficulty in defining the relevant reference iting the assets, although in the long run even the possession of such group for social standing-particularly in larger setdements (Stark assets can depend on transfers too. and Taylor 1991). 32. See Lucas (1988) for a discussion in the context of guest 19. This has been the case in China and Vietnam, and was the worker emigration. case in South Africa under apartheid. 33. Note that this difficulty is not entirely removed in a panel sur- 20. In places where studying and evaluating emigration incen- vey when a change in policies occurs between rounds of the survey tive programs is important, additional questions might be inserted It is still possible that those who plan to increase their remittances the into the migration module asking non-emigrants vhether they most are more likely to take advantage of the new incentives. applied for specific forms of state emigration aid and whether these applications were successfuil. References 21. This issue has received a lot of attention in the United States, where it has played an important role in debates over immi- Abowd, J., and R. Freeman, eds. 1991. Immigration, Trade and time gration policy (Borjas 1987, 1994). Labor Mfarket. Chicago, Ill.: University of Chicago Press. 22. A remaining problem is that migrants with unsuccessful Adelman, Irma, and Sherman Robinson. 1978. "Migration, employment experiences may choose to return home or move on Demographic Change, and Income Distribution in a Model of elsewhere, vhich wvould leave analysts with a distorted impression a Developing Country" In Julian L. Simon, ed., Research in if they stadied only the remaining migrants. It may be possible to Population Economics, Vil 1. Greenwich, Conn.:JAI Press. correct for this distortion using data from the migration histories; Altonji, Joseph, and David Card. 1991. "The Effects of Immigration however, this issue has not yet been systematically addressed. on the Labor Market Outcomes of Less-Skilled Natives. In J. 23. Studies in the United States and Europe have shoNvn that Abowd and R. Freeman, eds., Immigration, Trade and the Labor heavily protected industries tend to be the major employers of Market. Chicago, Ill.: University of Chicago Press. immigrants. This is not quite the same as testing whether protec- Banerjee, Bissvajit. 1983. "The Role of the Informal Sector in the tion encourages inmnigration, but it is part of the picture. See Migration Process: A Test of Probabilistic Migration Models Abowd and Freeman (1991) and Faini andVenturini (1993). On and Labour Market Segmentation for India." Oxford Economic Malaysia see Martin (1994). Papers 35 (November): 399-422. 24. In contrast, it seems unlikely that sanctions imposed on . 1984a. "Rural-to-Urban Migration and Conjugal employers can easily be examined using household survey data, Separation: An Indian Case Study" Economic Development and partly because of a lack of variation in the coverage of the legisla- Cultural Chanige 32 Uuly): 767-80. tion but also because respondents are unlikely to answer questions . 1984b. "Information Flow, Expectations, and Job Search: about illegal employment practices honestly. Rural-to-Urban Migration Process in India." Journal of 25. See Chapter 7 for a discussion of measuring skills. Development Economics 15 (May/August): 239-57. 26. Herzog, Schlottmann, and Boehm (1993) studied data from . 1991. "The Determinants of Migrating with a Pre-Arranged industrialized countries on migration and job searches; these data Job and of the Initial Duration of Urban Unemployment: An included an analysis of the net effect of migration on the duration Analysis based on Indian Data on Rural-to-Urban Migrants." of the migrant's period of unemployment. Journal of Development Economics 36 (October): 337-51. 27. There have been some studies along these lines. See Stark Banerjee, Bissvajit, and Gabriella Bucci. 1994. "On-the Job Search and Taylor (1991). After Entering Urban Employment: An Analysis Based on 28. Note, however, that consumption per capita can fall even Indian Migrants." Oxford Bulletin of Economics and Statistics 56 though each remaining person may be consuming more. This is (February): 33-47. 78 CHAFrER 16 MIGRATION 1995. "On-the-Job Search in a Developing Country: An Activity in LDCs." Journal of Development Economics 2 June): Analysis Based on Indian Data on Migrants." Economic 165-87. Development and Cultural Change 43 (April): 566-83. . 1989. "On-the-Job Search in a Labor Market Model: Ex Banerjee, Biswajit, and S.M. Ravi Kanbur. 1981. "On the Ante Choices and Ex Post Outcomes."Journal of Development Specification and Estimation of Macro Rural-Urban Economics 30 (March): 539-58. Migration Functions: With an Application to Indian Data:' Foster, Andrew D., and Mark R. Rosenzweig. 1996. "Household Oxford Bulletin of Economics and Statistics 43 (February): 7-29. Division, Inequality, and Rural Economic Growth." University Barkley, Andrew P., and John McMillan. 1994. "Political Freedom of Pennsylvania, Philadelphia, Penn. and the Response to Economic Incentives: Labor Migration in Friedberg, Rachel. 1997. "The Impact of Mass Migration on the Africa, 1972-1987." Journal of Development Economics 45 Israeli Labor Market." Population Studies and Training (December): 393-406. Center Working Paper 97-11. Brown University, Providence, Behrman, Jere R., and Barbara L. Wolfe. 1985. "Micro R.I. Determinants of Female Migration in a Developing Friedberg, Rachel, and Jennifer Hunt. 1995. "The Impact of Country: Labor Market, Demographic Marriage Market, and Immigrants on Host Country Wages, Employment, and Economic Marriage Market Incentives." In Schultz and Growth." Journal of Econotnic Perspectives 9 (Spring): 23-44. Wolpin, eds., Research in Population Economics, Vol. 5. Fuller, Theodore D., Paul Lightfoot, and Peerasit Kamnuansilpa. Greenwich, Conn.: JAI Press. 1985. "Toward Migration Management: A Field Experiment in Boijas, George J. 1987. "Self-Selection and the Earnings of Thailand:" Economic Development and Cultural Change 33 Immigrants." American Economic Review 77 (September): (April): 601-21. 531-53. Garcia-Ferrer, Antonio. 1980. "Interactions between Internal - 1994. "The Economics of Immigration." Journal of Migration, Employment Growth, and Regional Income Economic Literature 32 (December): 1667-717. Differences in Spain."Journal of Development Economics 7 June): Carrington, William J., Enrica Detragiache, and Tara Vishwanath. 211-29. 1996. "Migration with Endogenous Moving Costs:" American Greenwood, Michael J. 1997. "Internal Migration in Developed Economic Review 86 (September): 909-30. Countries." In Rosenzweig and Stark, eds., Handbook of Connell,John, B. Dasgupta, R. Laishley, and Michael Lipton. 1976. Population and Family Economics. Amsterdam: North Holland. Migration from Rural Areas: The Evidence from Village Studies. Greenwood, Michael J.,JR. Ladman, and B.S. Siegel. 1981. "Long- Delhi: Oxford University Press. Term Trends in Migratory Behavior in a Developing Country: Corden, W Max, and Ronald Findlay. 1975. "Urban The Case of Mexico." Demography 18 (August): 369-89. Unemployment, Intersectoral Capital Mobility, and Harris, John R., and Michael P. Todaro. 1970. "Migration, Development Policy." Economica 42 (February): 59-78. Unemployment, and Development: A Two-Sector Analysis." Cornelius, WA., and P.L. Martin. 1993. The Uncertain Connection: American Economic Review 60 (March): 126-42. Free Trade and Mexico--US. Migration. San Diego, Cal.: Center Hatton, T.J., and J.G. Williamson. 1994. "What Drove the Mass for US-Mexican Studies. Migrations from Europe in the Late Nineteenth Century?" Cox, Donald, and Emmanuel Jimenez. 1997. "Coping with Population and Development Review 20 (September): 533-59. Apartheid: Inter-Household Transfers over the Lifecycle in Herzog, HenryWJr.,Alan M. Schlottmann, andThomas P. Boehm. South Africa." Boston College, Chestnut Hill, Mass. 1993. "Migration as Spatial Job-Search: A Survey of Empirical Faini, R., and J. de Melo. 1994. "Trade Liberalization, Employment, Findings." Regional Studies 27, Special Issue: 327-40. and Migration: Some Simulations for Morocco." Paper pre- Hoddinott, John. 1994. "A Model of Migration and Remittances sented to the Organisation for Economic Co-operation and Applied to Western Kenya:" Oxford Economic Papers 46 July): Development Workshop on Development Strategy, 459-76. Employment, and Migration, Paris. . 1996. "Wages and Unemployment in an Urban African Faini, R., and A. Venturini. 1993. "Trade, Aid, and Migrations." Labor Market." EconomicJournal 106 (November): 1610-26. European Economic Review 37 (April): 435-42. Ilahi, Nadeem, and Saqib Jafarey. 1995. "Guest Worker Migration, Falaris, Evangelos M. 1987. "A Nested Logit Migration Model Transfers, and the Extended Family: Evidence from Pakistan." with Selectivity." International Econonmic Review 28 June): Paper presented to the Seventh World Congress of the 429-43. Econometric Society, Tokyo. Fields, Gary S. 1975. "Rural-Urban Migration, Urban Jagannathan,Vijay. 1987. Tie Logic of Unorganized Markets. Oxford: Unemployment and Underemployment, and Job-Search Oxford University Press. 79 ROBERT E. B. LUCAS Kochar, Anjini. 1995. "Explaining Household Vulnerability to . 1994. "Capital Market Imperfections. Labor Market Idiosyncratic Income Shocks." Anierican Economic Review 85 Disequilibrium, and Migration: A Theoretical and Empirical (May): 339-71. Analysis." Economic Inquiry 32 (April): 290-302. LaLonde, Robert, and Robert Topel. 1991. "Labor Market Munshi, Kaivan, and Jacques Myaux. 1997. "Social Effects in the Adjustments to Increased Immigration." In Abowd and Demographic Transition: Evidence from Matlab, Bangladesh." Freeman, eds., Immigration, Trade and the Labor l1arket. Chicago, Boston University, Economics Department, Boston, Mass. Ill.: University of Chicago Press. Paxson, Christine H. 1993. "Consumption and Income Seasonality Lucas,Robert E.B. 1975."The Supply of Immigrants' Function and inThailand."Journal of Political Economy 101 (February): 39-72. Taxation of Immigrants' Incomes: An Econometric Analysis." Pessino, Carola. 1991. "Sequential Migration: Theory and Journal of Development Economics 2 (September): 289-308. Evidence from Peru." Journal of Development Econoniics 36 - 1985. "Migration amongst the Batswana." The Economic JUly): 55-87. Journal 95 June): 358-82. Pischke, Jorn-Steffen, and Johannes Velling. 1994. "Wage and Em- 1987. "Emigration to South Africa's Mines." American ployment Effects oflmuigration to Germany:AnAnalysis Based Economic Review 77 June): 313-30. on Local Labor Markets." Massachusetts Institute of Technology - 1988. "Guest Workers, Circular Migration, and Transfers." Economics Department Working Paper 94-8. Boston, Mass. In D. Salvatore, ed., World Population Trends and their Impact oni Rosenzweig, Mark R., and Oded Stark. 1989. "Consumption Economic Development. Westport, Conn.: Greenwood Press. Smoothing, Migration, and Marriage: Evidence from Rural - 1997. "Internal Migration in Developing Countries." In India."Journal of Political Economy 97 (August): 905-26. Mark R. Rosenzxveig and Oded Stark, eds., Handbook of Rosenzweig, Mark R., and Kenneth I. Wolpin. 1985. "Specific Population and Family Economics. Amsterdam: North Holland. Experience, Household Structure, and Intergenerational - 1998. "Internal Migration and Urbanization: Recent Transfers: Farm Family Land and Labor Arrangements in Contributions and New Evidence." Background Paper for Developing Countries." Quarterly Journal of Economics 100 [Torld Developmitenit Report 1999/2000. Washington, D.C.: (Supplement): 961-87. World Bank. Salvatore, Dominick. 1980. "A Simultaneous Equations Model of - 1999. "International Trade, Capital FloNvs, and Migration: Internal Migration with Dynamic Policy Simulations and Economic Policies toward Countries of Origin as a Means of Forecasting: Italy 1952-76."Journal of Development Economics 7 Stemuning Immigration." In A. Bernstein, ed., .M1igration and June): 231-46. Refugee Policies: The International Experience. London: Cassell Schultz, T. Paul. 1971. "Rural-Urban Migration in Colombia." Academic Publishers. Review of Economics and Statistics 53 (May): 51-58. Lucas, Robert E.B., and Oded Stark. 1985. "Motivations to Remit: Schwartz, Aba. 1973. "Interpreting the Effect of Distance on Evidence from Botswana." Journal of Political Economy 93 Migration."Journal ofPolitical Economy 81 (October): 1153-69. (October): 901-18. Sjaastad, Larry A. 1962. "The Costs and Returns of Human Manove, Michael, Gustav F Papanek, and Harendra K. Dey 1987. Migration." Journal of Political Economy 70 (October), "Tied Rents and Wage Determination in Labor-Abundant Supplement: 80-93. Countries." Boston University, Boston, Mass. Smith, James P. 1983. "Income and Growth in Malaysia." Report Marcouiller, Douglas,Veronica Ruiz de Castilla, and Christopher R-2941-AID. RAND Corporation. Santa Monica, Cal. Woodruff. 1997. "Formal Measures of the Informal-Sector Smith, James P, and Duncan Thomas. 1997. "Remembrances of Wage Gap in Mexico, El Salvador, and Peru." Economic Things Past:Test-Retest Reliability of Retrospective Migration Development and Cultural Change 45 January): 367-92. Histories." RAND Corporation and University of Californina Martin, Philip L. 1994. "Augmenting the Labor Force: The Role of at Los Angeles, Santa Monica. Migration."World Bank,Washington, D.C. Stark, Oded, and David Levhari. 1982. "On Migration and Risk in Mazumdar, Dipak. 1981. The Urban Labor M11arket and Income LDCs." Economic Development and Cultural Change 31 Distribution: a Study of Mfalaysia. Oxford: Oxford University (October): 191-96. Press. Stark, Oded, and J. Edward Taylor. 1991. "Migration Incentives, Molho, Ian. 1995. "Migrant Inertia, Accessibility, and Local Migration Types: The Role of Relative Deprivation." The Unemployment." Economica 62 (February): 123-32. EcononmicJournal 101 (September): 1163-78. Morrison, Andrew R. 1993. "Violence or Economics:What Drives Stark, Oded, J. Edward Taylor, and Shlomo Yitzhaki. 1986. Internal Migration in Guatemala?" Economic Development and "Remittances and Inequality." The Economic Journal 96 Cultural Chanzge 41 July): 817-31. (September): 722-40. 80 CHAPTER 16 MIGRATION 1988. "Migration, Remittances, and Inequality: A United Nations. 1970. Methtods ofMeasuring Internal Migration. New Sensitivity Analysis Using the Extended Giri Index."Journal of York. Development Economics 28 (May): 309-22. Vijverberg,Wim PM. t995."Dual Selection Criteria with Multiple Stiglitz, Joseph E. 1969. "Rural-Urban Migration, Surplus Labour, Alternatives: Migration, Work Status, and Wages." International and Relationships between Urban and Rural Wages." Eastern Economic Revieuw 36 (February): 159-85. Africa Economic Review 1 (December): 1-27. Vijverberg, Wim P.M., and Lester A. Zeager. 1994. "Comparing Todaro, Michael P. 1969. "A Model of Labor Migration and Urban Earnings Profiles in Urban Areas of an LDC: Rural-to-Urban Unemployment in Less Developed Countries?" American Migrants versus Native Workers." Journal of Development Economic Reviewv 59 (March): 138-48. Economics 45 (December): 177-99. 81 7 Should the Survey Measure Total Household | s Income? Andrew McKay Income is clearly a variable of critical importance in the household economy, as it provides the resources to finance current consumption and to undertake any savings. Households can derive their income from many different sources, which can be classified into factor income (payments received by households or their members in return for supplying factors of production that they own, such as labor or land) and non-factor income (net transfers received from sources outside the household that do not need to be repaid).Total household income is the sum at the house- hold level of these diverse sources, and represents the total purchasing power available to a house- hold in a given time period. This chapter reviews the importance of being able to measure total household income in LSMS and similar surveys in circumstances where consumption data are also being collected (including all circumstances described by this book).The relative importance of measuring total household income clearly has substantial implications for questionnaire design. When full LSMS surveys have been fielded in the past, sumption can be measured more accurately and com- they have collected sufficiently detailed information prehensively than household income. Second, much of on the dimensions of household living standards to the analysis of household surveys has focused on living enable analysts to estimate the level and composition standards and poverty, and there are widely accepted of both total income and total consumption expendi- theoretical reasons for using consumption-based ture. In principle, having estimates of both total con- measures rather than income-based measures to ana- sumption and total income is useful for measuring and lyze living standards.The validity of these arguments is analyzing households' living conditions, both in their considered in the first section of this chapter as well as own right (for example, for studying poverty) and as in Chapter 5 on consumption. explanatory variables for other important characteris- Given that consumption data are always collected tics of households and their members (for example, in LSMS surveys, is it worthwhile to also try to esti- nutritional outcomes). In practice, analysts have used mate total household income, bearing in mind the the consumption data gathered in LSMS surveys far inevitable extra costs involved in collecting the addi- more often than the income data. There seem to be tional data? The version of the questionnaire used in two reasons for this (Lipton and Ravallion 1995; the original LSMS surveys in Cote d'Ivoire and Peru, Deaton 1997). First, most analysts believe that in which collected enough information to enable ana- developing and transition economies, household con- lysts to estimate total household income, was consid- 83 ANDREW McKAY ered too expensive or not necessary in some places, information for immediate policy purposes (including where a "stripped down" version was used instead. In officials in the statistical office and elsewhere in the these cases (which have included the surveys in government) to those who require more detailed Bolivia, the Kyrgyz Republic, and Nicaragua), the information in order to conduct careful, in-depth resulting data set is not sufficiently detailed to enable research that will later have policy implications. analysts to construct reliable (or, in some cases, any) Because users in the latter category need different types estimates of total household income.1 Although all of information than users in the former category, it is previous LSMS surveys have collected data on the important for survey designers to find out in advance income of those in wage employment, several surveys which of these groups is most likely to use the data have not gathered data on household income from from the survey; this will influence the level of detail agricultural or nonfarm enterprises. In these circum- and precise content of the data that the survey collects stances analysts have inevitably had to measure living (which will vary from country to country). For the standards using consumption data. Thus the question purposes of this chapter it is assumed that the data will arises: what is lost by not being able to estimate total be used by the research community both within and household income, given that estimates of total house- outside the country and also that the data will meet the hold consumption are available? short-term needs of government policymakers. In considering the issue of whether or not to As stated above, total household income is of measure total household income, two general points interest both in its own right and as an explanatory should be borne in mind. First, if survey planners have variable for other household- or individual-level vari- already decided to include specific modules on house- ables. In other words, total household income can be hold agriculture and nonfarm enterprises, the margin- thought of either as an output measure for the house- al costs of collecting the additional information hold (for example, in studying poverty) or as an input required to measure total household income are likely measure (for example, as one of the determinants of a to be very small. Second, even if it is judged in any sit- household's willingness to send its children to school). uation that relatively little is lost by not being able to In either case, having a measure of aggregate house- estimate total household income, it can still be very hold income is useful primarily because of its close useful for analytical purposes to collect information relationship to the household's standard of living. For on some specific components of household income. some analytical purposes knowing the composition of The following five sections consider in detail the total household income can be as useful as knowing its issue of whether or not LSMS surveys should aim to level, but the implications for collecting data are the measure the level and composition of total household same in either case. income. In the first section, the benefits of estimating There are at least three main arguments for col- total income are discussed.The second section consid- lecting the data needed to estimate total household ers what is involved in measuring total income. The income: total household income can be used to meas- third section considers experience from recent LSMS ure households' standards of living; it can be used to surveys and other sources. The fourth section consid- understand determinants of poverty; and it can be used ers the costs involved in measuring total household to estimate household savings. These arguments are income. The fifth section makes an overall evaluation considered in the following three subsections, after of the evidence and indicates what conclusions can be which other uses for income data are considered. drawn. Total Household Income Can Be Used to Measure What Are the Benefits of Being Able to Standards of Living Measure Total Household Income? When analysts have a measure of a household's total income, this provides a means of measuring the house- In assessing the benefits of being able to measure hold's standard of living. This may seem to be the most household income, the first question to address is who obvious reason for collecting data on household is likely to use this information in their analysis. In the income because a household's standard of living is case of most surveys, there is a wide range of potential closely associated with the income it receives. users ranging from those who need rapid, descriptive However, as was noted above, total household con- 84 CHAPTER 17 SHOULD THE SURVEY MEASURE TOTAL HOUSEHOLD INCOME? sumption can also be used to measure living standards, If these points are valid, and they are quite com- and this can be calculated relatively straightforwardly pelling, they constitute a strong argument in favor of from the data gathered in an LSMS-type survey. Thus using consumption-based measures of standards of liv- two questions arise. Is there any reason to prefer ing instead of measures based on income, at least in income-based standard of living measures to con- developing and transition countries. However, even sumption-based measures? If not, do income-based after accepting this, are there any reasons for analysts to measures provide sufficiently useful complementary use income-based measures of standards of living as information to the consumption-based measures to well? One reason is that analysts can use the income- justify estimating them as well? based measure to check the accuracy and validity of Regarding the first question, it is widely held that the consumption-based measure, although this is only consumption-based standard of living measures are useful if the two measures are consistent within a rea- preferable to income-based measures (Deaton 1997; sonable order of magnitude.A second reason is that for Lipton and Ravallion 1995), at least for developing some households current consumption level may not and transition countries. Two justifications are com- be an accurate measure of long-term sustainable stan- monly given for this. The first (equally valid in devel- dard of living, even if current consumption level is an oping and industrial countries) is that a household's accurate measure of current standard of living. A consumption level is more directly and closely associ- household facing a major reduction in income may ated with its current standard of living than is its cur- respond by selling some of the assets it needs to gen- rent income. Income is certainly the means of financ- erate income-a move which may be costly to reverse ing consumption, but it is consumption that provides and may reduce its future income levels. If the house- utility, the economist's measure of a person's welfare. hold's income does not return to its previous level, this According to this view, income can be thought of as sort of response is not sustainable indefinitely. In these an input while consumption is more closely associat- circumstances, the household's current income might ed with the output that is being measured (although be as good a measure of its standard of living as its cur- Sen 1985 shows that consumption also has limits as a rent (unsustainable) level of consumption. Analysts standard of living measure). Moreover, current income often wish to study such households that may not cur- is often volatile from one year to another, being sub- rently be poor but may become poor in the future, but ject to significant shocks.This is especially the case for they cannot easily identify these households unless households that are engaged predominantly in self- comparable information is available on both income employment activities or that are very reliant on trans- and consumption. fers from either public or private sources. According to A third and more complex reason for using permanent income or life-cycle models, current con- income-based standard of living measures is to distin- sumption is usually significantly more stable than cur- guish between transitory and chronic poverty-an rent income, given that it can be smoothed to some important distinction for studying the dynamics of extent by saving and dissaving/borrowing.2 As a result, poverty and directing assistance to those who most current consumption bears a closer relationship than need it. Making this distinction requires that data exist current income does to a household's permanent for two different points in time. Panel data are not income or long-term standard of living, even when always available and can be problematic to use even current income is reliably estimated. This remains true when they are available (see Chapter 23 on collecting even when households cannot borrow to smooth their panel data and Ashenfelter, Deaton, and Solon 1986), consumption because of the lack of effective credit but, to some extent, this distinction can be considered markets. using data from repeated cross-sectional surveys. Using The second justification (see, for example, Deaton panel data from South India, Chaudhuri and Ravallion 1997) is the perception that, in developing countries, (1994) argued that consumption-based standard of liv- estimates of total household consumption tend to be ing measures do not necessarily identify the chronically more accurate than estimates of total household income, poor more accurately than income-based measures. a view based both on empirical evidence and a priori They argued that although current incomes of house- assumptions.3 The empirical evidence for this view will holds tend to be more volatile over time than current be discussed in the third section of this chapter. consumption levels, if current incomes display sufficient 85 ANDREW McKAY co-movement (positive correlation) across households of a household's income and on the household's pover- in the sample then observations of current income in a ty status, they can identify which income sources are single cross-section might more accurately give the characteristic of poor households and they may be able appropriate ranking of households according to their to discover some possible reasons for this relationship. long-term standards of living.This was indeed so in the For example, they might find that poor households South Indian case studied by Chaudhuri and Ravallion; have low economic returns to the activities in which in that case current income data were more effective in they are engaged (possibly reflecting the nature of the distinguishing the long-term or chronic poor from the household or the nature of the activity) or that they are transitory poor. This evidence is no more than sugges- not engaged in any economic activity (for example, tive; the validity of the argument will certainly depend due to unemployment or nonactivity) and thus rely on on the extent and nature of any measurement error in interhousehold transfers. Without data on total house- the income estimates. hold income, it is hard for analysts to find out which of In summary, a standard of living is a complex con- these situations applies and to identify the underlying cept that cannot satisfactorily or comprehensively be reasons. Having data on household consumption alone measured by a single indicator. For most purposes, would not provide this information. Also, having data consumption-based standard of living measures are on the time household members devote to different probably superior to income-based measures, at least economic activities (available from the employment in developing countries. However, if income can be module) is not enough to inform analysts about the measured with sufficient accuracy, income-based economic returns to those activities and hence their measures offer analysts information about household importance as sources of household income. While welfare that goes beyond what is offered by the con- having these time use data is desirable, it is not a sub- sumption measures. This can be particularly important stitute for knowing the income derived from these dif- for studying the dynamics of poverty. ferent activities. That some types of activities have lower economic returns than others is a central factor Total Household Income Con Be Used to Understand the in understanding poverty. Determinants of Poverty Of course, it is possible to model the determinants In looking at living standards, analysts are interested of living standards by looking at the relationship not only in who is poor and who is not, but also why between a household's standard of living, as measured some households are poor and others are not. While through consumption, and the characteristics of that consumption data are probably the best way to answer household. A number of studies have done this. For the first question, they can only partly address the sec- example, this was done in regression-based studies of ond question. Ultimately analysts also need to know the determinants of household living standards by about the source of a household's purchasing power- Glewwe (1991) using the results of the 1985 Cote in other words, its income-and the determinants of d'Ivoire Living Standards Survey, and by Coulombe this. Thus data on total household income and its and McKay (1996) using the results of a similar Living components can be very helpful in understanding the Standards Survey in Mauritania. This approach makes determinants of poverty. Data on household income it possible to identify which factors (demographic fac- also allow analysts to gauge the likely effects on house- tors, the education level of the economic head, and so holds of policy changes, many of which affect house- on) are associated with households with a high stan- hold income directly. dard of living. However, many of the factors that influ- An important starting point is to find out from ence a household's standard of living do so by affect- where households derive their income. Many house- ing its income level. Moreover, certain factors are holds in developing or transition countries receive likely to be more important for some types of income their income from more than one source (for example, than for others; for example, a person's level of educa- from a combination of agricultural income, wage tion may have a greater influence on his or her wage income, and interhousehold transfers).This means that income than on his or her self-employment income. analysts need to know not only a household's income Also, the point remains that the relationship between level but also the sources from which this income is consumption-based standard of living measures and derived. When analysts have data on the composition the types of variables that analysts might want to con- 86 CHAPTER 17 SHOULD THE SURVEY MEASURE TOTAL HOUSEHOLD INCOME? sider as determinants of standard of living is indirect, section of this chapter for further discussion of this). operating through income. Where possible, modeling Second, calculating household savings as the difference the determinants of different categories of income (for between two large figures (household income and example, wage income and agricultural income) household consumption), each of which may be sub- would enable analysts to have a clearer understanding ject to significant measurement error, means that the of the causes of poverty. estimate of savings is liable to be affected by the meas- To examine this relationship. analysts need data on urement errors contained in both estimates. This each of the major components of household income becomes even more serious if the income and con- because the factors that influence the levels of the dif- sumption estimates are subject not just to random ferent sources of income are typically different. For measurement error but also to bias. In this case, the example, wage income is likely to be influenced by differential in the bias between the income and con- human capital variables, while income from agricul- sumption estimates will affect the estimate of savings. ture is influenced by (among other things) the use of Worse still, the magnitude of the bias and of any ran- inputs to production (including land size and quality) dom measurement error might vary systematically and the price and composition of outputs.When data according to the type of household, which would be a are available on the sources of household income, ana- problem for analysis, econometric or otherwise. lysts can make a much closer mapping of the welfare Despite these problems, modeling the determi- level of a household and the individual determinants nants of savings at the household level is a highly relevant to each of the household's income sources. attractive goal and has been attempted, for example, by This presupposes that sufficiently accurate estimates of Deaton (1992) based on data from the Cote d'lvoire household income are available. Living Standards Survey. Nevertheless, significant It is crucially important that analysts and policy- questions need to be raised about whether the esti- makers understand the determinants of a household's mates of household savings obtained in the manner standard of living to guide them in choosing the most described above are sufficiently reliable to make effective policy measures to raise the living standards econometric or other analysis of them meaningful. of particular poor or vulnerable groups and to protect The answer is likely to vary from case to case. these groups from the adverse effects of a recession. It is true that estimating total household income is not strictly necessary for estimating the flow of Total Household Income Can Be Used to Estimate household savings, as these data can be collected in Household Savings other ways-for example, by asking respondents a A reliable estimate of total household income enables direct question about how much their savings the estimation of household savings as the difference changed in the previous year. (This was done in the between household income and household consump- third round of the Ghana Living Standards Survey.) tion over a given period of time.4 This is clearly desir- Another way (which may yield more accurate infor- able, as there is relatively little reliable information in mation than asking respondents directly) is to collect many developing and transition countries about the data on the changes in assets held by the household. levels of household savings, much less their determi- This option is discussed in some detail in Chapter 20 nants. Many existing estimates of household savings on savings in this volume. However, in order to use are derived from the national accounts at an aggregate this approach to estimate household savings, the ques- level and are often of highly questionable reliability. tionnaire must be designed to collect comprehensive The critical issue here is whether estimates of the data on the household's acquisition and sales of a full flow of household savings derived in this way from range of assets. As argued in Chapter 20, it is very LSMS and similar surveys will be sufficiently reliable. unlikely that this approach will yield more accurate For instance, it must be ensured that the estimates of estimates of household savings than if this measure is both total household income and total household computed by deducting total consumption from total consumption relate to the same period of time and income. Therefore, if analysts want to be able to use that inflation does not make them incomparable (as it the survey to measure flows of household savings, it is might if the recall period were different in each case necessary for the survey to measure total household and if the inflation rate were quite high; see the third income. 87 ANDREW McKAY Data on the Income of Household Members Can Be Used analysis-for example, to estimate the contributions of to Study Intrahousehold Issues different members to household income, the different Having data on total household income is also useful economic activities in which different members are for a number of other analytical purposes. Among involved, and the different returns to these activities. these, having data on the income earned by different However, there are some limitations to what kinds of household members can be useful for analyzing intra- analysis can be done with these data. Neither children household issues. The simple neoclassical model of the nor old people within the household can be included household assumes that the household's total income in such analysis because they are unlikely to earn is shared equally among all of its members and that the much, if any, income. Also, as was mentioned above, it household makes consumption choices according to a can be difficult to attribute to any one individual the single, well-defined set of preferences, but it is widely income from household activities that involve more recognized that the reality is likely to be different (see, than one member-although this might be attempted for example, Haddad, Hoddinott, and Alderman if the analyst knows which household members are 1997). There is a class of important policy questions engaged in a given self-employment activity and the that cannot be addressed within this aggregated number of hours they devote to it. Such information household framework and that require information on is generally collected in the employment module (pre- what is happening within the household (for example, sented inVolume 3). the distribution of income and consumption). To Some analysis of this type could probably be con- address such issues it can often be highly desirable to ducted without an estimate of total household collect consumption data at the individual level, but income. However, if analysts want to look at the dis- collecting such data tends to be difficult and also very tribution of purchasing power among household expensive. For these reasons (and as argued in Chapter members (and the extent to which income pooling 5 on consumption), LSMS surveys tend not to collect takes place), they clearly need data on all of the com- extensive individual-level consumption data. ponents of the household's income, with as many of Therefore, consumption data are not very useful for these data as possible at the individual level. analyzing intrahousehold issues in this instance. Anthropometric data are commonly used to Income Data Can Be Used to Measure Economic Activities examine intrahousehold dynamics in practice. As Not Adequately Covered by Existing Statistics argued in Chapter 10 on anthropometrics, it is highly Data from LSMS surveys can be used to address statis- desirable that anthropometric data be collected as part tical issues beyond those for which they were explic- of LSMS-type surveys. These data are useful for ana- itly designed; in particular they can help improve and lyzing intrahousehold issues, even though some ana- develop existing aggregate and sectoral statistics. In lysts feel that they are only trustworthy for young chil- many low-income developing countries, a serious dren (see discussion of this issue in Chapter 10). dearth of statistical information on important facets of Irrespective of this, it can be useful to collect informa- their economy (such as the informal sector and subsis- tion on income at the individual level (as far as is pos- tence agriculture) raises questions about the quality of sible given that some household production is joint national accounts and sectoral statistics, with obvious and is not necessarily easily attributable to individual implications for analysis. In such circumstances the household members) to supplement the anthropomet- income data available in LSMS and similar surveys ric data in studying some intrahousehold issues. may be better than those currently used to measure Hoddinott and Haddad (1995) have used individual- such activities. The LSMS data may be used to level income data from the Cote d'Ivoire Living improve existing aggregate and sectoral data, as well as Standards Survey to investigate whether the share of to construct or update social accounting matrices. household income earned by female household mem- This is not by itself a sufficient argument for seek- bers affects the consumption pattern of the household. ing to measure total household income in a household Other than this, the individual-level income data survey. However, it may be a valuable side-product gathered in previous LSMS-type surveys have not where such data have been collected, one that has been widely exploited, but there is a lot of potential been underexploited in the past. One example of for using them in both descriptive and more in-depth LSMS data being used to improve government statis- 88 CHAPTER 17 SHOULD THE SURVEY MEASURE TOTAL HOUSEHOLD INCOME? tics is in measuring and studying the informal sector Summary in Ghana using data from the Ghana Living Standards A number of potentially important uses for income Survey (Coulombe, McKay, and Round 1996).While data have been set out in this section. Some of them household survey data may not be the ideal source, require reasonably reliable estimates only of the level and while it is necessary to consider the possibility of of total household income, while others require accu- income underestimation in such surveys, the resulting rate information on the composition of household estimates in this instance still appeared to be better income. Some uses are more important than others. than those that had been used previously. Comparable Ultimately, though, the most compelling argument for instances are likely to arise elsewhere. having data on the level and composition of total household income, assuming it can be estimated with Data on the Composition of Household Income Can Be sufficient reliability, is the information these data can Used to Define Socioeconomic Groups yield for understanding poverty and how poverty is In analyzing standards of living and related questions, it affected by policy changes. is invariably desirable to disaggregate households into groups according to various criteria in order to com- What Is Involved in MeasuringTotal Income? pare the different groups with one another and explore the reasons for any variations within and between each This section considers the question of what precisely group.The obvious and economically relevant criterion measuring total household income is likely to involve. by which households can be disaggregated is the main First it addresses the conceptual question of what ana- economic activity of the household, which, for conven- lysts are trying to measure. Then it discusses what ience, is referred to here as the socioeconomic group of measuring household income involves in practice and the household. There are various ways to define the what problems can arise. main economic activity of the household. One com- mon procedure is to classify households according to Conceptual Issues Involved in Measuring Household the activity of the economic head of the household Income (raising the issue of how the economic head should be A useful framework for thinking about measuring total defined). Another more direct and probably more household income (as well as household consumption) meaningful criterion is the household's main source of is offered by a system of household accounts that sum- income.While this is often the same as the activity of marizes the production, consumption, and accumula- the economic head, it is not always, and, when it is not, tion activities of households in three separate but relat- it is probably the more appropriate criterion to use. The ed accounts. (These accounts are production, current, most obvious way to find out the household's main and capital accounts, respectively; see Johnson, McKay, source of income is to collect data on a household's and Round 1990, as well as UNSD 1993 and Ruggles income by its main components (appropriately defined) and Ruggles 1986 for analogous concepts at the macro and identify which source is most important. For this and aggregate levels.)6 purpose, what matters is that the data on the composi- tion (more than the level) of the household's income is DEFINING HOUSEHOLDS. The issue of how to define a reasonably accurate. A third, less attractive alternative is household is discussed in detail in Chapter 6 on the to identify the type of activity to which household household roster. However, in the context of measur- members devote the most time, using information nor- ing household income, it is appropriate to recognize mally collected in the employment module of the ques- that in developing and transition countries, many tionnaire (see Chapter 9 on employment).5 households, such as agricultural households and Again, the fact that data on total household households that run household enterprises, may be income and its composition can be used to define production units as well as consumption units; fur- meaningful socioeconomic groups is not by itself suf- thermore, the best way to define a household as a pro- ficiently compelling to warrant collecting data on all duction unit may not be the best way to define it as a household income. Clearly, though, where these data consumption unit.This can be an important difficulty are available, it would be sensible to use them to define in practice, as Devereux 1992 shows based on field- socioeconomic groups. work in northern Ghana. It is clearly important that a 89 ANDREW McKAY consistent definition of the household is applied tant in estimating total household income, which may throughout the survey. If this definition is based on the change the analyst's perceptions of poverty and consumption unit, the income from the household's inequality among households.9 production activities needs to be related to the same However, in practice, household production is unit, which is not a trivial exercise. generally defined less broadly than Hill's definition because too many measurement problems can arise AN OPERATIONAL DEFINITION OF INCOME.7 As dis- with this ideal concept.Thus, while own-account pro- cussed above, in broad terms households and house- duction of goods within the household is usually hold members earn "income" by supplying factors of included in a practical definition of production, own- production that they own to productive activities and account production of services within the household by receiving current transfers. With regard to factor (such as childminding) is usually excluded. This is the income the fundamental issue that arises is the defini- case in the latest System of National Accounts and is tion of productive activities, which can be particularly also consistent with previous LSMS surveys. problematic in developing countries given that markets Therefore, this chapter, in accordance with the con- are often underdeveloped and so much economic sumption chapter (Chapter 5), assumes that it is activity takes place outside the market. Thus, while it is impractical to try to cover service activities conducted possible to start from Hicks' (1971) definition of pro- within the household in the estimation of total house- duction as "any activity directed to the satisfaction of hold income (or consumption) because of the practi- other people's wants through exchange:' this must be cal problems involved in valuing these activities.iO interpreted sufficiently broadly to include various Finding ways to overcome such problems is neverthe- forms of nonmonetized exchange such as barter and less an important priority for future research. wage payments in kind. Even this expanded interpreta- Factor income can be defined as the payment tion of Hicks' definition of production fails to take into received by a household for supplying factors that it account household own-account activities in which owns (such as labor, land, and capital) to a productive the household produces a good or service that is a close activity. This definition holds whether this productive substitute for one available on the market but con- activity is carried out by the household or by an insti- sumed by the household itself. The most obvious tution or individual outside the household and example is subsistence agricultural production, whether the payment is made in cash or in kind.There although some nonagricultural self-employment activ- are three main types of factor income: wage income, ities also include an element of production for the rental income, and self-employment income. (These household's own consumption-for example, the con- categories can obviously be disaggregated further in struction of furniture or sewing garments for the specific cases.) Wage income is received in return for household's own use. Natural habitat utilization- the supply of labor services; rental income is received collecting and gathering items from natural sources- in return for the supply of land, capital, and other should also be allowed for in a measure of total house- assets; and self-employment income is typically a hold income where this is important.8 Note that where return both to labor supplied by household members each of these imputations is included in a measure of and to other factors these members own, such as land total household income, they must also be included in or capital. the measure of total household consumption. Transfers can be received from various sources, An extended definition of production used by including firms, government agencies, other house- Hill (1979) considers production to be any activity holds, and nongovernmental organizations of various that can be carried out with comparable results by an types (any of which may be domestic or foreign). Only economic unit other than the one that actually carries current transfers (such as interhousehold transfers of it out. Under this definition, not only subsistence pro- cash or food) and not capital transfers (such as inheri- duction but also a whole range of additional service tance of land or receipt of a loan) should be regarded activities performed within the household can now be as income.11 In practice it can be difficult to distin- regarded as productive, including childminding, food guish between current transfers-a source of income preparation, and fetching firewood or water. for the household-and capital transfers. For example, Moreover, such activities can be quantitatively impor- when a household receives a transfer from another 90 CHAPTER 17 SHOULD THE SURVEY MEASURE TOTAL HOUSEHOLD INCOME? household, this may or may not involve an obligation production themselves. Given that this will be difficult (implicit or explicit) to repay. Also, a dowry may fall for respondents to do, they are not told which valua- into either category depending on what form it takes. tion principle to use. However, since they are also asked to quantify their own consumption, the result- VALUATION. For market transactions, goods are usually ing data set will contain the information necessary to valued in accordance with prevailing market prices- compare the implicit unit value of the household's in other words, those prices actually paid or received consumption of its own production with data from by the household.12Valuation is much more problem- the agriculture and consumption modules on the unit atic in the case of imputed transactions. When the value of sales and purchases of the corresponding household's productive output is exchanged by barter, commodity in the market. This is important because if the most appropriate way to value the goods received different respondents used different principles to value in return is presumably to use local market prices. A their consumption of their own production, estimates similar principle would hold for wage income of total household income or consumption (which received in kind. would include these data on the household's con- However, in both cases, the question arises sumption of its own production) would not be com- whether the consumer (purchasing) or producer (sell- parable without being adjusted to account for the use ing) price should be used. (The same question also of these different valuation principles. Since unit value arises in the valuation of nonmarketed household pro- information would be available in the data set, this duction such as goods produced by the household for adjustment could be made if necessary. its own consumption; see the discussion of this issue in Chapter 5 on consumption). Using the consumer SUMMARY. A theoretically valid concept of household price will typically yield a different valuation than income should include an appropriate and consistent using the producer price. Using the consumer price valuation of factor income from all productive activi- may be more appropriate for measuring welfare (given ties, whether they are carried out entirely within a that an analogous good purchased on the market household or for another institution outside the would be valued at its market price), but inconsisten- household. Such a concept should also include a valu- cies can arise. If some of the output of the household's ation of all current transfers received by the house- production were sold in the market, it would be val- hold, whether in cash or in kind. Table 17.1 reflects the ued at the producer price. Yet a higher valuation is above discussion in setting out the types of income placed on exactly the same output when it is con- components on which information should be collect- sumed by the household. Moreover, there is the prob- ed in the survey questionnaire. Some of these compo- lem that the household's own produced food is unlike- nents involve imputation; where they do, exactly anal- ly to be a perfect substitute for that purchased on the ogous imputations are required to estimate total market. household consumption. (Table 17.1 provides details Similarly, if both food that a household produces of this.) If analysts are interested in intrahousehold for its own use and food that it produces to sell on the behavior, it may be desirable to collect information on market were valued at the producer price (which which individuals are involved in each activity and to would be sensible from a production point of view), what extent, on who receives the income, and on who the household's consumption of its own production controls its use. would be valued differently from the same goods pur- chased by the household on the market, which would Estimating Total Household Income in Practice be valued at the consumer price. This appears incon- Many household surveys have aimed to collect the sistent from the point of view of measuring welfare. information necessary to estimate total household The consumer and producer prices can be thought of income, including the LSMS surveys conducted in as representing upper and lower bounds, respectively. Cote d'Ivoire, Ghana, Nepal, Pakistan, Peru, and Neither valuation is clearly superior to the other. For Vietnam. However, the difficulties involved in measur- reasons discussed in more depth in Chapter 5 on con- ing household income should not be underestimated. sumption, the proposed consumption module asks Even if respondents are willing to answer all of the respondents to value their consumption of their own questions asked and do so as honestly as possible, the 91 ANDREW McKAY Table 17.1 MeasuringTotal Household Income Income component Data that must be collected income from wage employment Wage income in cash Wage income in kind * Bonuses Household agricultural income Revenue from the sale of crops Revenue from the sale of processed crop products Revenue from the sale of animal products Consumption of self-produced food * Minus Expenditure on inputs for crop cultivation Expenditure on inputs for producing processed crop products5 Expenditure on livestock inputs Depreciation of agricultural capital equipment Nonfarm self-employment income Revenue in cash from sale of output Revenue in kind from sale of output Consumption of own produced output (where appropriate) * Minus Expenditure on inputs Depreciation of capital equipment .............................................................................................................. ....................................................................................................... .............. Imputation for commodities obtained Food commodities I from natural sources* Nonfood commodities (where not otherwise included) * ........................................................................................................................................................................ .........................................*............ .. Actual and imputed rental income Income from renting out household assets Imputed rent of owner-occupied dwellings * ............................................................................-................................................................................................................................................. ...... Income from private interhousehold transfers Income from private interhousehold transfers in cash and kind (where no repayment is expected) Other income Various miscellaneous income (income from pensions, unemployment benefits) * Elements that shou d a so be nc uded in an estimate of total household consumption a. Excepting products supplied by the househo d tself Source: Data compi ed by author myriad of income sources households can have makes (The sources excluded may not be very important it difficult to be sure that all income sources for a given overall but may be very important for a small number household have been identified. For example, some of households.) Another important issue is that where sources of income may be very casual or infrequent, respondents are asked to make imputations, such as and, therefore, the respondent might not think to placing a valuation on wage income received in kind, mention them in response to questions about wage it may be very difficult for them to do so. employment or nonfarm enterprises, either of which Estimating total household income in accordance may be taken by the respondent to refer to more for- with Table 17.1 involves collecting information on at mal or more sustained activities. least the following four elements: income from wage Respondents may genuinely not know their employment; agricultural income; nonfarm self- income from certain activities, especially self-employ- employment income; and nonlabor income. Given the ment activities for which they often do not keep very different nature of each of these income sources, accounts. Indeed, the concept of income in an the data necessary to estimate them will almost cer- accounting sense or in the sense used by economists tainly be collected in different modules within the may be quite foreign to respondents.This need not be questionnaire. The natural place to collect the data a problem, but it means that it will generally be nec- necessary for estimating income from wage employ- essary to estimate some components, such as income ment is the employment module. Data on income from self-employment activities, indirectly. For income from household agriculture and nonfarm enterprises from self-employment activities it will probably be can naturally be collected in the modules devoted to necessary to collect information on revenues and these topics, while data on nonlabor income can be input expenditure separately (as reflected in Table gathered in one or more short modules designed with 17.1). The wide variety of forms in which households the principal aim of identifying such income and esti- can receive transfer income may mean that some mating its magnitude.This is essentially the model that sources are simply not included in the questionnaire. was followed in the early LSMS questionnaires as well 92 CHAPTER 17 SHOULD THE SURVEY MEASURE TOTAL HOUSEHOLD INCOME? as in many other household surveys that have collect- hold members is very small compared with the cost of ed the information needed to estimate household adding whole modules on household agriculture and income. nonfarm enterprises, which makes this seem an attrac- When survey designers are deciding how to col- tive option in some cases. lect these income data, they need to recognize the However, how accurate can respondents be in potential complexity of the household economy, reporting their self-employment income? The infor- which is relevant to all four of the major components mal nature of most self-employment activities means of household income outlined above. Household that the vast majority of respondents probably do not members may be engaged in more than one wage keep accounts for these activities. Moreover, since employment activity, simultaneously or sequentially, most self-employment activities involve both expendi- during the reference period or periods. The house- ture on inputs and revenue from the sale of outputs, hold's agricultural activities may involve cultivating both of which may include imputed as well as cash several different crops as well as keeping livestock- elements, it is very difficult to see how a respondent and may thus involve various inputs. Households may can answer a single question about their self-employ- also run more than one nonfarm enterprise. Thus a ment earnings from a given activity in a particular ref- large and complex questionnaire is needed to ensure erence period (for example, the previous seven days). that all of this information is collected, even though The situation is even more complex when the self- large parts of the questionnaire may be irrelevant for employment activity involves more than one house- many households (for example, because they do not hold member working together and members are farm), which will make the interviewers' burden less asked individually about their earnings from the self- daunting than it appears at first sight. employment activity; in such cases it will be problem- This means that a questionnaire collecting total atic to attribute the total profit of the activity between income data must contain both an agriculture module them. However much time is saved by estimating self- and a nonfarm enterprise module.When survey plan- employment income this way, the reliability of the ners already intend to include both of these modules responses obtained must be open to serious doubt, because the information they contain is of interest in more so than most other questions in the question- its own right, the data collection implications of aim- naire. (This view is consistent with the opinions of the ing to estimate total household income are modest. authors of the employment, agriculture, and nonfarm The proposed standard versions of the employment, enterprise chapters.) The author of this chapter nonfarm enterprise, and agriculture modules (intro- strongly recommends against estimating income from duced in Chapters 9, 18, and 19) collect information self-employment this way, whether or not the ques- on income from wages, nonfarm enterprises, and agri- tionnaire includes the agriculture and nonfarm enter- culture. Therefore, measuring total household income prise modules. simply requires that a module on transfers and other Another important practical issue is the recall nonlabor income is included in the questionnaire (see period used to measure the different components. Chapter 11).13 However, if survey planners do not Because household income is complex, data should be include the standard agriculture and nonfarm enter- gathered component by component in such a way that prise modules, analysts will not be able to measure analysts will subsequently be able to add the different total household income. Another possibility is to try to income components together to compute a house- collect information on self-employment income hold's total income, to analyze the implications of the directly in the employment module. The employment composition of a household's income, or both. module in many previous LSMS surveys asked indi- However, the most appropriate reference period to use viduals about their cash earnings from the self- for each component of income is not necessarily the employment activities they reported. Combining this same. (See the specific chapters for discussion of information with data on the household's consump- appropriate recall periods for estimating each compo- tion of its own production could yield an alternate nent of income.) It is appropriate to collect wage estimate of income from self-employment activities. income data on a weekly recall basis, as the short recall The cost of adding a couple of extra questions on period will hopefully make the responses more accu- earnings from the self-employment activities of house- rate. However, a weekly recall period is clearly inap- 93 ANDREW McKAY propriate for collecting data on transfers and other purposes (see the first section of this chapter). The nonlabor income income, which may be received by problem of a lack of comparability is most likely to only a small number of households, and may also be arise in countries that are experiencing high or mod- received infrequently. For these components a longer erately high inflation or that are affected by large sea- recall period, perhaps 12 months, may be more appro- sonal variations in prices throughout the year. In these priate. A more complex issue is the appropriate recall countries, respondents would most likely value trans- period for the self-employment income components, actions that took place within a 12-month recall peri- especially for agriculture. In agriculture, revenues and od at the prices prevailing at the time of the transac- input expenditure are likely to be made at different tions, whereas they would probably value transactions points in time throughout the agricultural season. that took place within a recall period of the previous Moreover, the agricultural season is the natural refer- week at current prices.While the values of both kinds ence period during which respondents can be expect- of transactions can be expressed on a comparable basis ed to supply information about their revenues and (say, monthly or annually), there is a potential source input expenditure. This can cause a number of diffi- of error in adding them up because of the different culties in practice. The agricultural season does not valuations used due to the difference in recall periods correspond neatly with the calendar year.The agricul- from component to component. tural season may vary from one part of the country to An additional problem in comparing total income another or from one crop to another. And while it may and total consumption in a high-inflation economy is be desirable to interview all agricultural households at that the "average recall period" used for the income a similar point in the agricultural season (for example, modules is probably longer than the one used for the after the harvest), this is likely to cause problems in the consumption modules. Correcting for this problem is interview schedule. Chapter 19 on agriculture discuss- not straightforward, as it requires detailed information, es these questions in more detail. ideally by locality, on the variation of prices during the The key point is that data on different compo- period when the survey is conducted and the preced- nents of income are usually collected using different ing year (or whatever is the longest recall period used reference periods, which analysts need to take into in the survey). It may be especially difficult to make account when adding the data together to calculate this correction for self-employment income, given total household income. Of course, a similar problem that sales revenues have probably been received and arises in estimating total household consumption, input expenditures incurred by the household at given that it is desirable to use different recall periods different-and usually unknown to the analyst-inter- for collecting data on consumption of different items vals throughout the reference period. Any measure- to reflect different frequencies of consumption. ment of this kind is difficult in economies with high However, the problem is more serious in measuring inflation rates or intrayear variability in prices, and this total income, in particular because of the desirability of applies to all monetary variables, including consump- using the agricultural season as the reference period tion as well as income. for collecting data on income from this source. The agricultural season cannot easily be converted into the How Successfully Has Income Been Measured calendar month or year basis on which the other in Developing and Transition Countries? components of income can be computed, and the appropriate conversion factor may vary from situation How successfully have household surveys in develop- to situation. ing and transition countries measured total household The need to use different reference periods for income? There is a wealth of experience on this ques- gathering data on different components of both tion; this section will focus on recent experience in income and consumption can make it difficult to developing and transition countries, particularly in compare estimates of total income and total consump- instances where data from LSMS surveys have been tion. Making this comparison is absolutely essential for used to calculate total household income. measuring household savings (by subtracting total As noted above, there is a widespread perception household consumption from total household that estimates of total household income derived from income) and is necessary for many other analytical household surveys conducted in developing and tran- 94 CHAPTER 17 SHOULD THE SURVEY MEASURE TOTAL HOUSEHOLD INCOME? sition countries are often unreliable-and certainly A third point is that it may be more difficult for a less reliable than estimates of total household con- survey to identify and cover all prospective sources of sumption derived from the same source. This could be a household's income comprehensively than to cover because the questions relating to income are less accu- consumption comprehensively. This may be due to a rately answered than those relating to consumption; it reluctance on the part of respondents to report certain could also be because information on consumption is types of income or to the fact that the questionnaire more fully and more comprehensively collected than does not prompt respondents to provide information data on household income. If income information on certain sources of income, so the respondents do were not collected as comprehensively as consump- not remember or report them. Respondents may not tion information, income is likely to be underestimat- feel the need to report their most casual or infrequent ed relative to consumption. If the problem were inac- income-earning activities and will obviously be very curate answering, this might or might not lead to an reluctant to report any of their activities that may be understatement of income. of dubious legality (or downright illegal). Respondents may also be reluctant to reveal their Why Income Appears More Difficult to Calculate Than receipt of transfers and other nonlabor income. To Consumption some extent this syndrome is inevitable, but survey There are a number of reasons why this perception is planners should ensure that it does not reflect prob- so widespread. First, income is a much more sensitive lems in the design of the questionnaire or in its imple- topic to ask people about than expenditure. mentation. A poorly designed questionnaire may fail Respondents may have an incentive to understate to ask about (or fail to prompt the respondent suffi- their income in a survey interview, especially if they ciently about) a wide enough range of informal activ- fear that the information may be used for tax purpos- ities, or the interviewers may ask some of the ques- es (notwithstanding assurances to the contrary).This is tions in such a way that the respondent feels obviously a problem in developed countries as well as uncomfortable answering them. in developing and transition countries, but it may be The questionnaire cannot include questions on more acute in the developing and transition countries every conceivable source of household income, just as given the significantly greater importance in these it cannot include questions on every type of con- countries of self-employment income (which is easier sumption expenditure. However, when the sources of to understate than wage income). household income are more diverse than the cate- Second, as has previously been observed, respon- gories of household consumption, as is the case in dents may genuinely not know how much income they most developing or transition countries, it is invariably make, especially in their self-employment activities. easier to collect comprehensive consumption data Measuring self-employment income is difficult even in than comprehensive income data. Having said that, developed countries, and higher proportions of house- many key difficulties, such as the identification and holds in developing and transition countries are gener- valuation of nonmarket transactions, arise in both ally engaged in work for themselves. However (aside cases. from the above incentive to underreport income), the Consequently, the difficulties involved in calculat- situation is further complicated in developing and some ing total household income, especially self-employ- transition countries by the general absence of written ment income and transfers, should not be underesti- accounts for household production activities. As noted mated. So how much confidence can analysts have in above, it may be necessary to take an indirect approach the estimates of income derived from household sur- by asking about as many financial details as possible and veys? Are there any good practices that would increase computing an income figure from these bits of infor- the reliability of these estimates? At the outset it mation. However, this is likely to require collecting should be recognized that there are few objective tests quite a lot of information. Moreover, the accuracy of of the reliability of income (or consumption) data estimates computed as the difference between total rev- derived from household surveys. Even comparing enues and total input expenditure must be open to them with apparently similar data from national some doubt as both of these estimates may contain sig- accounts can be fraught with difficulty, as discussed in nificant measurement errors. detail in Chapter 5 on consumption. Conceptually, 95 ANDREW McKAr household income and consumption rarely have an can generally be expected to be of similar magnitudes. exact equivalent in national accounts. Even when esti- While the household sector may save (or dissave) in mates from national accounts and estimates from any given time period, the magnitude of such savings household surveys are of similar orders of magnitude, or dissavings relative to income or expenditure can be this in itself does not prove the accuracy of the house- expected to be relatively small. hold survey estimates. Clearly, when they are highly Table 17.2 provides summary information on the dissimilar, this suggests that the estimate from at least magnitude of estimates of total household income and one source is seriously inaccurate, but this does not consumption derived from a number of recent house- help analysts to identify which is the more accurate. It hold surveys in several countries. Table 17.3 provides is certainly not appropriate to assume that the nation- similar information (in a different format) for a 1995 al accounts estimates are necessarily more accurate. income and expenditure survey in Belarus (not an LSMS survey). Evidence on Accuracy of Income Estimates The information presented in these tables has Some criteria based on household survey data alone been summarized greatly, so it should be interpreted can give clues as to the possible accuracy of estimates with great care. There is a different reason for the of household income. One straightforward issue is the apparent success or failure of each survey to yield the relative consistency of estimates of household income data necessary to estimate total household income, and and consumption.While individual households save or it is impossible to do proper justice to each of these dissave in any given time period, estimates of total stories here. The surveys from which the estimates of household consumption and total household income household income and consumption have been com- Table 17.2 Comparing Estimates of Income and Consumption, Selected Household Surveys Correlation Mean total Ratio of Mean per capita between per Mean total annual per capita income annual savings capita annual annual income consumption to per capita (income minus income Country Survey per capita per capita consumption consumption) and consumption Bulgaria Integrated Household Survey 1995 * 46.02 50.32 0.915 -4.30 0.178 Cote d'lvoire Enquete Permanente aupres des menages (EPAM) 1985 * 186.5 294.8 0.633 -108.3 0.688 EPAM 1986 * 242.8 276.8 0.877 -34.0 0.849 EPAM 1987 * 236.6 286.7 0.825 -50.2 0.800 EPAM 1988 * 220.9 246.2 0.897 -25.2 0.600 Gnana Ghana Living Standards Survey .............. Round I (GLSS 1), 1987-88 * 61.0 87.0 0.701 -26.1 0.406 GLSS 2, 1988-89 69.2 107.9 0.641 -38.7 0.598 GLSS 3, 1991-92 118.8 208.9 0.569 -90. 0.540 Jamaica Jamaica Survey of Living Conditons 1993 * 26326 28308 0.930 -1982 0.412 Pakistan Pakistan Integrated Household Survey 1991 * 7682 7871 0.976 -188.8 0.151 Peru Encuesta nacional de hogares sobre medicion de niveles de vida, 1994 * 2423.4 2176.2 1.114 247.2 0.675 South A rica Sout Africa Integrated Household Survey, 1993 * 9903.8 6961.4 .423 2042.4 0.229 Venezuela, R. B. de Encuesta do Presupuestos Familiares, 1988-89 19.63 32.49 0.604 Encuesta Social, 1991-92* 96.33 69.68 1.382 - * LSMS surveys. -Not available. Note: Bulgaria: thousands of Lavy; C6te d'lvoire: thousands of CFA; Ghana: thousands of Cedis; Jamaica: Jamaican dollars; Pakistan: Rupees; Peru: Soles; South Africa: Rand. Source: Bulgaria, South Afhca: computed from estimates of household income and expenditure on LSMS Web site; C6te d'lvoire, Ghana, Peru: com- puted by author from raw data; Jamaica: based on Handa 1995; Pakistan: based on estimates of household income and expenditure constructed by the World Bank; Republica Bolivariana de Venezuela: based on Scott 1994. 96 CHAPTER 17 SHOULD THE SURVEY MEASURE TOTAL HOUSEHOLD INCOME? Table 17.3 Average Monthly Cash Income and Cash Consumption Expenditure of Households in Belarus, Ranked by Quintile of Cash Income Quintile of Average cash income Average cash consumption expenditure Ratio of cash income cash income (thousands of Belarussian rubles) (thousands of Belarussian rubles) to cash expenditure (percemt) Lowest 427 700 61 Second 780 1,064 73 ... .....................................................................................................................,....................................................................... 8...2....................... . ..... Third 1,100 1,349 82 ...... ............................................ ......................................................................5,..........................................................................9,....................... . ..... Fourth 1,527 1,725 89 ................................................................................................................................................................................................................................. Highest 2,625 2,795 94 Source: Martini, Ivanova, and Novosyolova 1996, table 3. puted are all different and not all are LSMS surveys. ding estimates of household consumption. In such The various estimates of household income and con- instances it is very unlikely that the shortfall can be sumption have been computed by different authors, explained by dissaving by the household sector because and there may be some differences in the concepts and the magnitude of the difference between income and methods used to derive them (as well as in the survey consumption is much too great. Consequently, in these information on which they are based).'4 When the cases it does appear that there has been a significant data are computed by the author from a primary underestimation of income, a significant overestimation source, the figures reported in the table are all com- of consumption, or both. As suggested above, compar- puted over the same households (those for which esti- ing these estimates with estimates of private consump- mates of both total income and total consumption can tion expenditure from the national accounts does not be computed). It is assumed that this is also the case always help analysts to tell whether income or con- where the information is derived from secondary sumption is more accurately estimated."5 The general sources. In the case of the Republica Bolivariana de arguments set out above suggest that the differences Venezuela, information was not available to compute stem more from underestimating income than from the ratio of average income to average consumption, overestimating consumption, although even if this is the average magnitude of household savings, or the true (an issue that will be considered further below) the correlation between income and consumption. explanations may differ from case to case. However, the average magnitudes of income and con- In several other surveys on which information is sumption are broadly comparable in all instances. presented in Table 17.2-Bulgaria, Jamaica, and the The extent to which the estimates of household 1987 and 1988 C6te d'Ivoire surveys-it was found income and consumption are consistent varies signifi- that, on average, households spent in excess of their cantly from country to country. It is clear that in some income for the year in question but that the implied surveys-Pakistan, Peru, South Africa-the estimates magnitudes of dissavings were within the range of of household income and consumption are broadly plausibility in general. Of course, in these kinds of consistent, which implies credible household savings cases, analysts must always test the credibility of the or dissavings rates. Of course, as noted above, the pos- implied dissavings rate by comparing the survey esti- sibility remains that both household income and mates with other evidence pertinent to the country household consumption expenditure are overestimat- and time period in question. Underestimation of ed or underestimated to similar extents, though this is income and overestimation of consumption expendi- unlikely to be substantial unless there is an overestima- ture may still have occurred in these surveys but if so tion or underestimation of a component common to they occurred to much less of an extent than in the both household income and consumption (for exam- cases of Belarus, Ghana, the 1985 C6te d'lvoire survey, ple, consumption of self-produced food). and theVenezuelan Encuesta de Presupuestos Familiares. By contrast, some of the other surveys-Belarus, In the case of the Republica Bolivariana de Ghana, the 1985 C6te d'Ivoire survey, and the 1988-89 Venezuela, the estimates of household income obtained Encuesta de Presupuestos Familiares or Income and from the Encuesta Social (Social Survey) are significantly Expenditure Survey in the Republica Bolivariana de greater than the estimates of expenditure derived from Venezuela-yielded estimates of total household the same source, even after removing outliers (Scott income that were substantially below the correspon- 1994).This is in sharp contrast to the estimates derived 97 ANDREW McKAY from the 1988-89 Encuesta de Presupuestos Familiares. explore any obvious explanations for these and make However, the questionnaires for the two surveys were consequent recommendations for future LSMS sur- very different, and in the Encuesta Social consumption, veys, it is possible to draw on the work of various ana- the data for which were collected at a relatively aggre- lysts who examined some of these surveys to see if any gated level, appears to have been significant underesti- evidence could be found to suggest that income wvas mated. Therefore, it is difficult to say how accurately less accurately estimated than consumption. Judging household income was estimated in this survey. the relative reliability of estimates of total household The discussion so far has focused entirely on the income and consumption based on information main- mean values of estimates of total household income ly from within the survey is clearly a partly subjective and consumption (or savings). However, mean values exercise. There is no clear objective basis for assessing are significantly affected by outliers and other extreme whether the underestimation of income or the over- values,16 and looking only at mean values implies los- estimation of consumption is predominantly responsi- ing a lot of information. The extent of correlation ble for the apparent underestimation of household between total household income and household con- savings in these instances. However, survey results, sumption also provides useful information, even combined with background information about the though it does not identify cases of underestimation. countries in question, can offer some clues. Household income and consumption are never per- In the case of Belarus, Martini, Ivanova, and fectly correlated across households; most theories of Novosyolova (1996) found that 67.7 percent of house- consumption (for example, the permanent income holds reported income levels below their consumption hypothesis) explicitly suggest this will not be so. levels and that 9.7 percent reported income levels that However, the two criteria should produce a broadly were less than half their reported consumption levels. similar ranking of households, implying that they are Moreover, they found that the relationship between significantly positively correlated.17 For the instances average rates of dissavings and the quintile group to in Table 17.2 where a correlation coefficient can be which a household belonged differed radically computed between total household income and con- depending on whether quintiles were defined by sumption, this indeed appears to be the case, even in income or by consumption.When income was used to cases where household savings appear to have been define quintile groups, households in the lowest quin- significantly underestimated (for example, the 1985 tile dissaved the most, and the average rate of dissaving C6te d'Lvoire survey and the Ghana surveys). fell in each higher quintile group (see Table 17.3).Yet Thus recent experience measuring total house- when consumption was used to define the quintile hold income in household surveys in developing and groups (not reported here), the average proportion of transition countries does not suggest that the objective income dissaved increased significantly with the quin- of measuring total household income using data from tile; those in the lowest quintile actually had a positive multipurpose household surveys should be aban- savings rate. Although total household income and doned. However, the experience is certainly mixed, consumption should both be legitimate measures of and in some of the surveys there is strong evidence of the standard of living, which one of them is chosen underestimation of income, overestimation of con- clearly affects the appearance of the relationship sumption, or both. As discussed above, a priori consid- between savings and the standard of living. (Of course, erations suggest that underestimation of income is this is partly because one of the two variables used in probably a much larger factor than overestimation of measuring savings is also used to define the quintile consumption, but is there any empirical evidence to groups.) back this up? If so, what might be causing this under- However, this does not indicate whether income estimation of household income? or consumption is more reliably estimated. In the case The point has already been made that the expla- of Belarus there are good reasons to assume that nations for problems of this nature may differ from income was significantly underestimated. Martini, case to case as each survey experience has its own Ivanova, and Novosyolova argued that anecdotal evi- storv So what are the possible explanations for the dence and casual empiricism suggest widespread unbelievably big gaps between household income and informal economic activity in the Belarus economy, consumption in the three cases referred to above? To yet in the survey very few people reported having sec- 98 CHAPTER 17 SHOULD THE SURVEY MEASURE TOTAL HOUSEHOLD INCOME? ond jobs, doing occasional work, or running a house- It seems clear that the underestimation of house- hold business.These informal activities may have gone hold savings in the Ghana surveys reflects an underes- unreported due to the way the questionnaire was timation of household income much more than it designed or the way survey interviewers did or did not reflects an overestimation of household consumption. prompt respondents to report such activities. A more There are a number of other arguments that support likely explanation may be that a large number of this view and explain why the underestimation of respondents were unwilling to reveal such informa- income was apparently much more of a problem in tion. If this is the case it is likely to be very difficult to Ghana than in neighbouring C6te d'Jvoire (with the measure total household income with any accuracy. exception of the 1985 survey), where a very similar The example of the surveys conducted in Ghana is questionnaire was administered. As was discussed ear- interesting because having more than one round of data lier, high rates of inflation are likely to lead to an means that the data sets can be compared with one underestimation of income relative to consumption. another, and because the third round was collected using Ghana had an average annual inflation rate of around a different questionnaire design from the one used in the 22 percent between 1988 and 1992 (the approximate earlier rounds. Making comparisons between the differ- period covered by the surveys), while rates of inflation ent rounds can yield useful clues about whether house- in Cote d'Ivoire were much more modest. Moreover, hold income or consumption is more accurately esti- self-employment income, one of the hardest compo- mated.Where similar or identical surveys are conducted nents to measure, is a more important source of close to each other in time, analysts would not expect (in household income in Ghana than in C6te d'Ivoire. the absence of a clear explaining factor) radical changes The estimates of household savings derived from in the composition of income or consumption expendi- the 1988-89 Encuesta de Presupuestos Familiares in the ture or in the nature of poverty. When the first two Republica Bolivariana deVenezuela are clearly under- rounds of data from the Ghana Living Standards Survey estimated, but it is difficult to determine whether (which were collected using identical questionnaires) income underestimation is predominantly responsible were compared, evidence emerged that that the compo- for this. For example, it is very difficult to compare the sition of income is much more unstable than the com- levels of household income and consumption in the position of consumption. More compellingly, the geo- 1988-89 Encuesta de Presupuestos Familiares with those graphical pattern of poverty, which changes gradually of the Encuesta Social conducted in 1991-92 because from one round to another when consumption data are there was a high rate of inflation between these two used to measure living standards, was seen to have dra- periods, because the questionnaires were significantly matically changed when income data were used to different, and because the Encuesta de Presupuestos measure living standards (Coulombe and McKay 1995). Familiares was conducted only in urban areas while the These sharp changes are hard to understand. Income Encuesta Social was nationwide. As in Ghana, the high data suggested that the capital city, Accra, had the high- inflation rate may suggest that income underestima- est incidence of poverty in the country (out of five local- tion is at the root of the problem.18 ities) in the first survey round but the lowest such inci- Thus, in the cases where household savings have dence in the second round (the following year). most obviously been underestimated, the underesti- Coulombe and McKay argue that the apparent geo- mation of household income seems much more likely graphical pattern of poverty based on income data was to be responsible for this than the overestimation of counterintuitive (for example, in its implication that the household consumption. The reasons for the underes- northern savannah region is one of those least affected timation of household income differ from case to case, by poverty) and contradicted most other standard of liv- although high inflation rates (where applicable) may ing measures. What this strongly suggests, therefore, is be a significant common factor. Difficulty in estimat- that the raw data used in measuring total household ing self-employment income is also likely to be a income were significantly less accurate than those used common factor; however, it is worth noting that this is to measure household consumption in this case. Indeed a problem experienced by developed countries as well in this case it appears that the nature of underestimation as developing countries (Atkinson and Micklewright of income varied from one locality to another or from 1983 for the United Kingdom; Branch 1994 for the one component to another. United States). The difference though is that such 99 ANDREW McKAY incomes are often much less important in developed module in the survey for analytical reasons, there are countries than they are in developing and transition likely to be few extra costs involved in adding the ele- countries. This is one reason why developed country ments needed to measure total household income. In experience with measuring household incomes is a survey of this kind, data on income earned from often more successful than that of developing and wage employment both in cash and kind are general- transition countries. ly collected in the employment module (introduced in Overall, however, the empirical evidence Chapter 9). Other data needed to calculate total reviewed in this section suggests that large-scale household income are collected in the standard ver- underestimation of household income is not sions of the nonfarm enterprise and agriculture mod- inevitable. Many surveys have managed to collect suf- ules (introduced in Chapters 18 and 19).19 ficiently accurate data for analysts to measure total Thus all that would need to be added to the ques- household income, which is without doubt a complex tionnaire would be a module to collect data on income variable to measure. It is indeed probable that total from transfers and other nonlabor income (see Chapter household consumption can be estimated with greater 11). The designers of the survey may have decided to accuracy than total household income. But many include a module on transfers and other nonlabor questionnaires have collected consumption data in income in the survey anyway, because these income more detail than income data anyway. Notwith- sources are of interest in their own right. Even if they standing this, there is no basis for general and univer- were not, the costs of including transfers and other non- sal pessimism about the possibility of measuring labor income would be modest. In previous LSMS income, even if doing so is a relatively complex and questionnaires, modules on transfers and other nonlabor risky business. income have tended to account for only about two pages in an approximately 70-page questionnaire. And What Costs Are Involved in Measuring Total the time taken to administer these two pages will gen- Household Income? erally have been proportionately less than the time to administer other sections, because many of the ques- The cost of a survey designed to collect the data need- tions would not have applied to respondents who did ed to measure total household income depends on the not receive a particular kind of income. Even the long amount and level of detail of the information to be versions of a transfers and other nonlabor income mod- collected. Collecting a large amount of information ule (introduced in Chapter 11), which might amount to means using a long questionnaire. A long question- four questionnaire pages, would not take much more naire demands long interviews, which lead to high time to complete; as before, many of the questions costs. And if the survey budget is fixed, a long ques- would not be applicable to a majority of households. tionnaire requires using a smaller sample, which However, when survey designers do not plan to reduces the extent to which resulting data can be dis- include one or both of the agriculture and nonfarm aggregated. An additional disadvantage of having to enterprise modules, it becomes necessary to add at conduct long interviews is that interviewees may least abbreviated versions of these modules to the become tired and bored and, thus, less careful about questionnaire to ensure that the survey yields the data the accuracy of their answers. needed to calculate total household income. This sub- These points are general and apply to all sections stantially increases the cost of fielding the survey. The of the questionnaire, and the reader should bear them short versions of these modules do not yield enough in mind throughout this volume. However, they have information to allow the estimation of income from specific implications for designing a questionnaire to these sources. However, adding the standard versions measure total household income.To illustrate this, it is of these modules simply for the purpose of estimating useful to compare a situation in which the question- income from these sources cannot be justified, given naire is designed to measure total household income the major extra costs that would be involved. When with a situation in which it is not. In broad terms, two these modules are included, it should be primarily different scenarios can be considered. because they are of interest in their own right. When survey designers have already decided to In such circumstances it is easy to see why survey include both an agriculture and a nonfarm enterprise designers may wish to include the questions on 100 CHAPTER 17 SHOULD THE SURVEY MEASURE TOTAL HOUSEHOLD INCOME? income from self-employment in agriculture and non- household consumption expenditure is already a farm activities directly in the employment module. major exercise, and it is probably more difficult to col- The costs of collecting this information are very much lect the data needed to measure a household's income. smaller. However, for reasons set out in the second sec- In some countries respondents may be very unwilling tion of this chapter, data collected in this way may be to supply information on their sources and levels of of dubious value and thus have very limited analytical income, however much assurance of confidentiality interest.Therefore, it is not recommended that data on they are given. In light of this, and since there are real- self-employment income be collected in this way ly no shortcuts in collecting income data, not all sur- despite its low costs (especially as these costs may be vey planners may choose to collect such information. even higher than expected because of the difficulty For example, those planning surveys in countries with respondents can be expected to have in answering limited household survey experience may choose not these questions or providing meaningful responses). to collect the data needed to calculate total household Thus, where survey planners have decided not to income. Alternatively, they may choose to collect such include the agriculture and nonfarm enterprise mod- data only in certain rounds of a sequence of household ules in their surveys, they should abandon the objec- surveys. tive of measuring total household income. For reasons of cost and difficulty, the most basic multipurpose household surveys are unlikely to collect Evaluation and Conclusions the information required to estimate total household income and thus will probably not include agriculture The collection of information needed to measure total and nonfarm enterprise modules and will include only household consumption is regarded as indispensable in a limited employment module. However, there are a questionnaire designed to study the living standards serious limits to the usefulness of these surveys, both of households and their members. The collection of for policy purposes and for in-depth analytical work. comprehensive data on household income is less fun- This is especially so in poorer countries where a damental because it is possible for analysts to compute majority of the population--and an even greater measures of poverty, to relate poverty to the character- majority of the poor-are likely to be engaged in istics of households, and to relate consumption-based agricultural or nonfarm self-employment activities. measures of standards of living to social variables (such Standard LSMS-type questionnaires should collect the as school attendance and use of health facilities), with- data necessary to estimate household income and do out any information on income. so properly. Where the resulting estimates of house- Yet there are serious limitations to the extent to hold income are reasonably accurate, these estimates which it is possible to understand poverty without should be used more widely in analysis than has been data on income. Understanding the reasons for pover- the case to date. ty and understanding its dynamics requires informa- tion not only on the economic activities of household Notes members (and the amount of time they devote to them) but also on the income earned from these activ- The author gratefully acknowledges the helpful comments of many ities. This is the most important reason to collect data people on the issues covered by this paper, in particular William on household income-but it is by no means the only Cavendish, Paul Glewxwe, Margaret Grosh,John Hoddinott,Alberto one. Moreover, the experience of recent household Martini, Jeffery Round, Julie Anderson Schaffner, and participants surveys demonstrates that some surveys appear to have in the World Bank wvorkshops held as part of this project in April successfully yielded the full range of data needed to 1996 and June 1997. calculate income reasonably accurately, despite the fact 1. Many of these surveys do collect information on income from that they generally devote more interview time to col- self-employment activities in their labor modules, which might lecting consumption data than to collecting income appear to be a suitable substitute for the estimation of income from data. household agriculture and nonfarm enterprise modules. However, It is of course riskier than collecting consumption the accuracy of these estimates of self-employment income is open data, as there is no guarantee of the reliability of the to serious question. As will be discussed later in this chapter and as resulting data. Collecting comprehensive data on is accepted in the specific chapters on these topics, this book rec- 101 ANDREW McKAY omimends strongly against trying to estimate household income 10. Information on the amount of time that household mem- from agriculture and nonfarm self-employment in this way. bers devote to such service activities within the household would, 2. Strictly, according to permanent income theories, consumers however, be available from the time use module (introduced in aim to smooth the marginal utility of their wealth rather than their Chapter 22). The difficulty in practice is in placing a meaningful consumption, and this can differ due to, for example, lifecycle value on this time. effects (Blundell and Preston 1994). Thus their consumption may 11. When a household inherits land, this directly and immediate- not be perfectly smoothed. However, consumption is still likely to ly increases the value of assets owvned by the household, and thus is a be a better measure of a household's permanent income and long- capital rather than a current transaction (even though it may lead to term welfare status than is current income. higher income in the future).While loans provide a household with 3. Some authors argue that in developed countries it may be purchasing power in the short term, they also establish liabilties that easier to measure household income than household consumption need to be repaid; thus, loans should not be regarded as income. expenditure (for example, Goodman and Webb 1995 make this 12. It is true that market price valuation will not necessarily rep- argument in the context of the United Kingdom). Even if this is resent "true" welfare valuations when markets fail, but in practice true for developed countries-and this claim is not beyond dis- there are no real alternatives to doing this. In any case, this distinction pute it is unhkely to be true in most developing countries given is of little importance to the household, which is interested in income the much greater importance of self-employment and nonlabor only as a means of financing its present or future consumption. income in these countries and given that their tax systems are gen- 13. In surveys whose consumption module collects information erally less developed, on natural resource utilization, this should also be included in the 4. Some past LSMS surveys have also collected information on measure of income. As explained in Chapter 5 on consumption, the stock of household savings (as opposed to the flow in any one data on natural resource utilization are not collected in the standard year). However, respondents tend to be wary of questions on such consumption module due to the fact that a nationwide, multitopic a sensitive topic, and it may be that these LSMS surveys only yield- household survey may not be suitable for collecting such informa- ed information on savings within the formal financial sector. See tion, wvhich may be highly locality-specific. Chapter 20 on savings. 14. Some of these estimates of household income may have 5. This wvas the problem faced by Coulombe and McKay been estimated using data on estimated household income from (1996) in trying to identify socioeconomic groups using data from agriculture and nonfarm enterprises that wvere yielded by direct the LSMS survey in Mauritania. Because the income data from questions on self-employment income, a procedure not recom- this survey were regarded as very unreliable and not suitable for mended in this book. The estimates for C6te d'lvoire, Ghana, and identifying socioeconomic groups, they instead used data from the Peru, which were computed from the raNv data, did not use the labor module to identify socioeconomic groups based on the eco- answers to these questions about self-employment earnings. nomic activity to which the household devoted most time. This is 15. For example, for the case of Ghana, an attempt to compute clearly not ideal for a number of reasons, but it does offer a possi- a comparable estimate of private consumption per capita from ble xvay to identify socioeconomic groups from survey data where national accounts data produced a figure halfivay between the esti- the survey designers decide not to collect comprehensive income mate of total household income and the estimate of total household data. consumption. Such a comparison is inevitably quite crude and 6. Note, though, that certain marginal differences arise bet ween approximate, given the different definitions applied in the national estimating variables at the micro level, wvhich is of interest here, and accounts and in the household survey. estimating them for macro purposes (for example, in national 16. An alternative procedure would be to look at the mean values accounts). of income and consumption having excluded the highest and loNvest, 7. This discussion draws significantly on Johnson, McKay, and say, 5 percent of values. However, such an analysis could not be con- Round (1990). ducted in all of the cases reported in Table 17.2. In the cases where it 8. For some households, natural habitat utilization can be a sig- was successfully conducted (for example, as done by Scott 1994 for nificant productive activity. Based on data collected in the Shindi the Repubhca Bolivariana deVenezuela, excluding the top and bot- ward in rural Zimbabwe, Cavendish (1999) estimates that the value tom 5 percent of the distribution), the broad conclusions have of commodities gathered from natural sources on average accounts appeared not to change markedly for 35.2 percent of household income. 17. It is true that there is a significant element of the estimates 9. For example, if fetching wvater is regarded as productive and of household income and consumption that is common to both imputed as a household income, it must also be imputed as a house- (for example, consumption of own-produced food). Repeating the hold consumption expenditure. calculations of correlation coefficients for measures of income and 102 CHAPTER 17 SHOULD THE SURVEY MEASURE TOTAL HOUSEHOLD INCOME? consumption excluding these common elements obviously gives Activity in Ghana: Proceedings of a Ghana Statistical Service/Oversears lower correlation coefficients, but the correlation remains signifi- Development Administration ConferenceJanuary 1995. Accra, Ghana. cant in most cases. Deaton, Angus 1980. The Measurement of We!fare: Theory and Practical 18. Altimir (1987) found that the estimates of household Guidelines. LSMS Working Paper 7. Washington, D.C.: World income obtained from a much earlier round of the Encuesta de Bank. Presupuestos Familiares in the Republica Bolivariana de Venezuela . 1992. "Saving and Income Smoothing in Cote d'Ivoire." corresponded quite closely to the national accounts estimates of Discussion Paper 156. Princeton University, Woodrow Wilson household income. However, it is not clear that this applies to the School of Public and International Affairs, Research Program more recent surveys and, as argued in the text, national accounts in Development Studies, Princeton, NJ. estimates are not necessarily an objective standard of accuracyv 1997. The Analysis of Household Surveys: A Microeconometric 19. Where survey designers feel it is important to include Approacih to Development Policy. Baltimore, Md.:Johns Hopkins households' use of natural resources in the estimate of total house- University Press. hold income (and consumption), they are likely to have already Delaine, Ghislaine, Lionel Demeryjean-Luc Dubois, Branko Grdjic, included these issues in the consumption module of the question- Christaan Grootaert, Christopher Hill, Timothy Marchant, naire as necessary elements in the calculation of total household Andrew McKay, Jeffery Round, and Christopher Scott. 1991. consumption. See also Chapter 5 on consumption. The Social Dimensions ofAdjustment Integrated Survey:A Survey to Measure Poverty and to Understand the Effects of Policy Change on References Households. Social Dimensions of Adjustment in Sub-Saharan Africa Working Paper 14.Washington, D.C.:World Bank. Altimir, Oscar 1987. "Income Distribution Statistics in Latin Devereux, Stephen 1992. "Observers are Worried: Learning the America and their Reliability." Revieu' of Income and Wealth 33 Language and Counting the People in Northeast Ghana." In (2): 111-55. Stephen Devereux and John Hoddinott, eds., Fieldwork in Ashenfelter, Orley,Angus Deaton, and Gary Solon. 1986. Collecting Developing Countries. Boulder, Col.: Lynne Rienner. Panel Data in Developing Countries: Does It Make Sense? LSMS Glewwe, Paul. 1991. "Investigating the Determinants of Household Working Paper 23.Washington, D.C.: World Bank. Welfare in Cote d'Ivoire."Journal of Development Economics 35 Atkinson, Anthony B., and John Micklewright. 1983. "On the (April): 307-37. Rehability of Income Data in the Family Expenditure Survey, Goodman,Alissa, and StevenWebb. 1995. "The Distribution of UK 1970-77."Journal of the Royal Statistical Society Series A, 146, Household Expenditure, 1979-92." Commentary 49. Institute Part 1: 33-53. for Fiscal Studies, London. Blundell, Richard, and Ian Preston. 1994. "Income or Grootaert, Christiaan 1983. "The Conceptual Basis of Measures of Consumption in the Measurement of Inequality and Poverty?" Household Welfare and their Implied Survey Requirements." Working PaperW94/12. Institute for Fiscal Studies, London. Review of Income and Wealth 29 (1): 1-21. Branch, E. Raphael 1994. "The Consumer Expenditure Survey: A . 1986. Measuring and Analyzing Levels of Living in Developing Comparative Analysis." Monthly Labor Review 117 (12): 47-55. Countries:AnAnnotated Questionnaire. LSMS Working Paper 26. Cavendish, William. 1999. "The Complexity of the Commons: Washington, D.C.:World Bank. Environmental Resource Demands in Rural Zimbabwe." Haddad, Lawrence, John Hoddinott, and Harold Alderman, eds. Working Paper WPS/99.8. Centre for the Study of African 1997. Intrahousehold Resource Allocation in Developing Countries: Economies, University of Oxford, U.K. Methods, Models and Policy. Baltimore, Md.: Johns Hopkins Chaudhuri, Shubham, and Martin Ravallion. 1994. "HoNv Well do University Press. Static Indicators Identify the Chronically Poor?" Journal of Handa, Ashu 1995. "Employment, Income and Labour Supply: An Public Economics 53: 367-94. Analysis of the 1993 SLC Employment Module." University of Coulombe, Harold, and Andrew McKay 1995. "An Assessment of the West Indies, Department of Economcs, Mona, Kingston, Trends in Poverty in Ghana, 1988-1992." Poverty and Social Jamaica. Policy Discussion Paper 81.World Bank,Washington, D.C. Hicks,John R. 1971. The Social Framework. 4th ed. Oxford: Oxford 1996. "Modeling the Determinants of Poverty in University Press. Mauritania." World Development 24 June): 1015-31. Hill,T. Peter. 1979. "Do ItYourself and GDP." Review of Income and Coulombe, Harold, Andrew McKay, and Jeffery I. Round. 1996. Wealth 23: 321-39. "Estimating the Contribution of the Informal Sector to the Hoddinott, John, and Lawrence Haddad. 1995. "Does Female Ghana GDP" In Ghana Statistical Service, Measuring Informal Income Share Influence Household Expenditures? Evidence 103 ANDREW McKAY from Cote d'Ivoire." Oxford Bulletin of Econonmics and Statistics Ruggles, Richard, and Nancy D. Ruggles. 1986. "The Integration 57 (February): 77-96. of Macro and Micro Data for the Household Sector." Review of Johnson, Martin, Andrew D. McKay, and Jeffery I. Round. 1990. Income and Wealth 32 (3): 245-76. "Income and Expenditure in a System of Household Sen, Amartya K. 1985. Commodities and Capabilities. Amsterdam: Accounts: Concepts and Estimation." Social Dimensions of North-Holland. Adjustment in Sub-Saharan Africa Working Paper 10. World Scott, Kinnon 1994. "Venezuela: Poverty Measurement with Bank,Washington, D.C. Multiple Data Sets." World Bank, Poverty and Human Lipton, Michael, and Martin Ravallion. 1995. "Poverty and Policy." Resources Division,Washington, D.C. In J. Behrman and T.N. Srinivasan, eds., Handbook of UNSD (United Nations Statistics Division). 1993. A System of Development Economics. Volume 3B, Amsterdam: Elsevier. NVational Accounts. NevYork: United Nations. Martini,Alberto P.,Anna Ivanova, and Svetlana Novosvolova. 1996. Vijverberg,Wim P.M. 1991. MWeasuring Incomefrom Family Enterprises "The Income and Expenditure Survey of Belarus: Design and with Household Surveys. LSMS Working Paper 84. Washington, Implementation." Statistics in Transition 2 (7). D.C.:World Bank. 104 8 o Household Enterprises Wim P. M. Vijverberg and Donald C. Mead In most Living Standards Measurement Study (LSMS) questionnaires there is a module explor- ing the dynamics and activities of nonagricultural household enterprises (which, for simplicity, are referred to in this chapter as "household enterprises"). This module gathers information on the portion of a household's income and employment derived from nonagricultural self- employment. More extensive versions of the module have also collected information on the variability of income and employment over time, the impact of the economic environment on household enterprises, and the involvement of household enterprises with credit and commodi- ty markets. The first section of this chapter explores the contri- The Role of Household Enterprises in the Development bution that household enterprises can make to eco- Process nomic development and the ways in which govern- In most developing countries, a very large number of ment policies can enhance that contribution. The people participate in household enterprises. About second links these policy issues to specific data one-half of the households sampled in previous LSMS needs. The third section translates these data needs surveys were found to operate one or more nonfarm into a model questionnaire, and the fourth section enterprises (Moock, Musgrove, and Stelcner 1990; provides some explanatory notes about the model Vijverberg 1992, 1998). Other studies in several coun- questionnaire. tries in Sub-Saharan Africa have indicated that 15-25 percent of adults in these countries are involved in The Role of Household Enterprises in such activities (Mead 1994). In most countries the Economic Development majority of household enterprises are owned and operated by women. In many countries the popula- This section provides an overview of the role of tion is growing faster than the number of new job household enterprises, detailing their prevalence, openings in the public sector and in larger enterpris- their characteristics, and the contribution that they es, so the role of household enterprises is expanding. make to economic development. The section then discusses the determinants of successful household WHAT ARE THE CHARACTERISTICS OF HOUSEHOLD enterprises and the kinds of information policy- ENTERPRISES? Most household enterprises fall into makers need to encourage the growth of these one of two major categories. Many-probably the enterprises. majority-of these enterprises generate only mini- 105 WIM P. M.VIJVERBERG AND DONALD C. MEAD mal income that is barely sufficient to enable their WHAT CONTRIBUTIONS Do HOUSEHOLD ENTERPRISES owners to survive. Such enterprises are sometimes MAKE TO HOUSEHOLD WELFARE? Household enterpris- referred to as survivalist enterprises. These enterpris- es contribute to improving household welfare in two es are often operated on a part-time basis, either year key ways: by generating income for those who work in round for only a few hours a day or full-time during the enterprise (whether they work as owner-operators only certain periods of the year. Sometimes survival- or as employees) and by creating employment. ist enterprises are only one of several income-gener- In the view of many analysts and policymakers, ating activities operated by an individual or the the principal contribution of household enterprises to household, in which case their contribution to household welfare is the generation of income for the household welfare is essentially supplementary. Often household. This includes cash income (cash receipts survivalist enterprises are run by women, who com- minus the cash expenses required to produce these bine them with their other household responsibili- receipts) and in-kind income (the net value of prod- ties. Examples of typical survivalist enterprises are ucts or services produced by the enterprise and con- food preparation, sewing, shoe shining, and street sumed within the household or of products bartered vending. with others). A few key questions need to be Other household enterprises, sometimes referred addressed: to as microenterprises, have a very different role in the * How much income does the enterprise generate development process. These enterprises generate for the household? Is this income sufficient to lift incomes that are substantially higher, and often well the household out of poverty? above the poverty level. While survivalist enterprises * How steady and reliable is this income? Income rely almost exclusively on unpaid family members flows that are highly variable over the year or that (and often consist of one person working alone), can cease unpredictably contribute less to household microenterprises are more likely to use hired workers. welfare than do flows that are regular and reliable. Microenterprises usually have more complex and * Which households receive this income? An income sophisticated production and marketing systems than flow is more or less important as a source of welfare do survivalist enterprises, and are more likely to be the depending on how well-off the recipients are and sole source of income for a household. Examples of what alternative income sources are available to microenterprises are furniture making, manufacturing, them.While survivalist enterprises often yield only and wholesaling. small returns, their contribution can be extremely Survivalist enterprises and microenterprises make important. Since survivalist entrepreneurs are fre- different potential contributions to household welfare quently very poor, even small increases in these and to the development process. As such, the two types people's incomes can contribute tremendously to of enterprises also differ in what kinds of programs their welfare.And since large numbers of people are support them and in how they respond to policy engaged in such activities, survivalist enterprises can changes. mean a great deal for welfare overall. There are many households that for one reason or * How is household enterprise income distributed, another do not operate a nonfarm enterprise. From among both employees of enterprises and different both a descriptive and analytical perspective, it is inter- household members? Does the distribution of esting to examine why some households operate household income differ depending on whether enterprises and others do not. For example, do the the entrepreneur is male or female? How is the households not currently operating an enterprise income spent, and what proportion is saved? Who engage in other, more lucrative activities? If these spends and who saves? All of these questions are households were to start up enterprises, would the important because the existence of a household enterprises earn more or less profit than the ones cur- enterprise can have a significant impact on house- rently in operation? Policies aimed at improving the hold allocation for food, schooling, health care, and performance of current enterprises may also prompt savings-and on which household members bene- other households to start up enterprises. Policymakers fit most from the way money is allocated. should not only be aware of this possibility but also be * In what ways do returns to labor vary by locality, able to quantify the magnitude of this response. sector, and the characteristics of the entrepreneur 106 CHAPTER 18 HOUSEHOLD ENTERPRISES (such as his or her gender, education, experience, working in household enterprises, how many earn an and skills)? income below the minimum wage? How many earn Is income measured per year or per day? For many an income at least twice the minimum wage? If possi- purposes it is important to account for the time ble, such questions should be asked in dynamic terms, dimension of both income received and labor input examining the patterns of income earned in jobs that required to produce this income-and thus to have come into existence during a given period-for express income in terms of hours or days worked. example, the most recent calendar year. An activity that yields high income per hour but Such time-specific analysis is particularly useful for only provides an opportunity to work a few hours exploring the impact of changes in policy or changes each month makes a contribution very different in the macro economy. Having access to this time- from that of a full-time activity that provides a specific wage information would enable analysts to lower income per hour. explore the hypothesis that many new and productive Household enterprises not only contribute to jobs are created in household enterprises at times when household welfare by providing income but also by other sectors are expanding at either the local or the providing a source of employment. The household macro level. Employment in household enterprises enterprise sector employs a substantial portion of the may also grow at times of macroeconomic stagnation, labor force. It is important to know just how many but most of these new employment openings are like- people household enterprises employ and how this ly to yield only marginal returns. Understanding the number changes over time due to sectoral growth as dynamics of household enterprises can be of consider- well as seasonality.The key questions here are: able importance in clarifying what can and cannot be * How many people work in the enterprise? How done by policymakers to promote the growth of pro- much does employment in the enterprise vary by ductive employment among household enterprises. season? Is the work part-time? If so, does this mean working only a part of each day or a part of each The Determinants of Successful Household Enterprises week? Is the work a form of moonlighting? When enough information about household enter- * What are the characteristics of the people engaged prises has been gathered, it becomes possible to pres- in the enterprise? Does the enterprise employ ent a profile of the country's household enterprises, unskilled, semiskilled, or highly skilled labor? including the number of enterprises, the income and * How has the enterprise changed over time? What employment that they generate, and the patterns of are the employment growth patterns for different change in employment. It should be possible to pres- types of enterprises? ent all this information broken down by location, sec- It is important to look not only at the number of tor, enterprise size, and gender of the owner. Such a people who work in household enterprises but also at profile constitutes the raw material for exploring to the characteristics and overall quality of their jobs. An what extent these enterprises create income and jobs. important distinction must be made between owners Using the knowledge gained from this process, policy and unpaid family workers on one hand and paid interventions can be devised that create a favorable cli- employees on the other. In most countries, more than mate for the household enterprise sector. 80 percent of those who work in household enterpris- es are either enterprise owners or unpaid family help. HOUSEHOLD- AND ENTREPRENEUR-SPECIFIC FACTORS. Policymakers need to know how many and which Two sets of factors that can influence the success of an members of households participate in the work of enterprise are the characteristics of the entrepreneur enterprises,just as they need to know who receives the and the characteristics of his or her household.The rel- income these enterprises earn. Policymakers also need evant characteristics of an entrepreneur include educa- to know how many people from outside the household tion, training, experience, and family background (for are employed in the enterprise. Once these things are example, the profession of his or her parents). Relevant known, it becomes clear who is helped and who is hurt household characteristics include what other activities by policy that impacts household enterprises. household members engage in; these activities can Another key issue is the wage levels of people in determine how the enterprise is organized and how the household enterprise labor force. Of all the people many resources household members allocate to it. 107 WIM P. M.VIJVERBERG AND DONALD C. MEAD People involved in household enterprises usually have household might specialize in only one particular other demands on their time (such as child care, cook- fuinction, such as carving backs for chairs; at the other ing, fetching water and firewood, farming, or other extreme, a household might be fully self-reliant, pro- paid jobs). In order to devote time to working for the ducing all its own inputs and making no market trans- enterprise, they must juggle these responsibilities. In actions at all. All enterprises face challenges related to some cases the household enterprise is a seasonal activ- their position along this continuum. ity undertaken only during slack times when other In surveys in several countries, household entre- demands on household labor are less pressing. preneurs have frequently indicated that the most Household enterprises can also affect the range of important problem they face is insufficient markets for other household activities. In some households an their products (Liedholm and Mead 1995). This prob- enterprise is a safe and reliable source of income that lem may arise when an enterprise serves a local mar- allows households to invest more in other activities- ket with too few customers; indeed many of the less activities that might yield higher returns but also carry successful enterprises sell only to neighbors in slow- a higher risk of failure. These kinds of decisions may growing, restricted, and localized markets. If this is the have a gender dimension. To make this point with a case, owners of the household enterprises could caricature of an example, a low-risk household enter- arrange to sell their goods in different locations or prise may be a low-productivity activity (in textiles or through different marketing channels, which is what food) performed by women that generates a low but more dynamic enterprises do. steady income. A household makes investments in Insufficient markets also arise when household riskier but more profitable household enterprises (in enterprises are not producing the exact type of goods manufacturing or wholesale) run by the men of the that consumers want to buy. In such cases owners may household. In other circumstances the household may try changing the design of the product (possibly by use profits from a household enterprise to cover cash incorporating new technologies), improving quality expenses of the household's farming activities, which control procedures, or changing the management of in turn provide food for the household. In this case the the enterprise. Policymakers can provide information household enterprise enables the operation of the about distant markets, organize meetings with distri- farm which, in turn, is critical for the household. bution networks in distant markets, ensure smooth Having data about household- and entrepreneur- operation of transportation networks (Vijverberg specific factors can help policymakers identify the 1998), provide training courses on quality control most appropriate kinds of assistance to support. This technology, and the like. assistance may include general education, vocational Research into markets for household enterprises training, or specialized management training (such as must address a few basic questions: accounting, bookkeeping, or quality control). Having * To what degree do household enterprises partici- information on these characteristics may also help pol- pate in markets? How developed are these markets? icymakers channel assistance to the clients who will * Where do household enterprises obtain capital make the best use of it. equipment, inputs, and hired labor? * Through what channels do household enterprises MARKETS. Participation in markets allows households sell their output? How reliable are these channels? to specialize. This specialization can lead to efficiency * Are enterprises that participate more extensively in and economies of scale for the economy as a whole, markets more profitable? Have the existing policies boosting productivity, income, and standards of living. regarding markets helped improve household However, such specialization is hindered when mar- enterprise performance? kets for inputs (labor, capital, land, raw materials) or outputs (the distribution system) function poorly. It is LOCATION AND INFRASTRUCTlURE. As factors influenc- difficult to operate an enterprise when supplies (either ing the success of household enterprises, location and of inputs or of consumer goods) are sporadic and access to infrastructure are related since some loca- unreliable, when demand for the enterprise's output is tional factors are heavily influenced by the availability sporadic, or when prices (of inputs, outputs, or items of roads or communications facilities. It is important for daily living) fluctuate too much. At one extreme, a to know not only where an enterprise is located but 108 CHAPTER 18 HOUSEHOLD ENTERPRISES also the degree of dynamism in the market at that funding for most household enterprises, it is also location and how effectively that location is served by important to know more about what entrepreneurs do infrastructure linking it with other markets. Access to with their savings. utilities (such as water, electricity, and telephone lines) Once again, it is vital to consider intrahousehold and to adequate workspace are also important factors. dynamics in examining the issue of finance, savings, Questions exploring these issues are normally includ- and credit. Who has responsibility for managing a ed in other modules of the LSMS survey; see Chapter household's financial assets? If profits are earned, how 13 on the community questionnaire and Chapter 12 are decisions made about how they are used? Who on housing. within a household has access to credit? The answers to these questions may depend on the division of FINANCIAL SYSTEMS, SAVINGS, AND CREDIT. There are authority between the genders or between genera- probably more programs to assist household enterpris- tions (for example, between fathers and sons). Some es in the area of credit than in all other areas com- aspects of household dynamics are explored in the bined. It is particularly important, therefore, to know household enterprise module, while others are how household enterprises interact with the financial addressed in other modules of the survey (see Chapter system and how these interactions may be strength- 24 on intrahousehold allocation). ened. Household enterprises are among the actors that exert a demand for credit. Households with accumu- REGULATORY AND LEGAL REFORMS. Some people have lated savings are among the actors that create a supply argued that instituting regulatory and legal reforms can of credit. Financial institutions (for example, banks and lead to a surge of new activity in the household enter- moneylenders) are intermediaries between the prise sector.This line of reasoning is somewhat less pop- demand and supply sides of the credit market. ular today than it was five years ago as research has led Since the LSMS is a household survey, it cannot many to question its significance. However, it remains directly explore issues relating to how financial insti- an important consideration in some countries, particu- tutions operate. For example, it cannot be used to larly in situations where enterprises seek to grow in examine lending decisions of financial institutions and both size and complexity (Mead 1995). It would be why they may choose to give priority in their lending interesting to determine to what degree enterprises to certain categories of borrowers. The household abide by existing rules and regulations and to what enterprise module can, however, throw useful light on degree entrepreneurs feel constrained by these factors. ways in which existing enterprises have financed their original and current stock of fixed and working OTHER ASPECTS OF MACROECONOMIC POLICY. Most capital-whether from an inheritance, from savings, or small enterprises pay few taxes, although they often from loans. It can also be used to find out the extent pay fees, purchase licenses, or pay indirect taxes on the of entrepreneurs' desire and need for credit. inputs they purchase. However, once enterprises grow In recent years new roles and operating proce- beyond a certain threshold, they become subject to dures have been developed for financial institutions substantially higher taxes. The foreign exchange that lend to microenterprises (Otero and Rhyne regime may make imports artificially cheap, but when 1994).When financial institutions have adopted these these imports are allocated administratively rather than new approaches, the flow of credit to household through the market, small household enterprises may enterprises has increased substantially and entrepre- find it hard to compete with larger domestic produc- neurs have had more opportunities both to borrow ers because the latter have better access to (administra- and to save. There is general agreement in the devel- tively allocated) cheap imported inputs. In addition, opment community that information about these new household enterprises may have to compete with approaches should be spread as rapidly as possible imported finished products that are artificially cheap- among financial institutions. In monitoring the effects er. The significance of these and other similar macro- of the new lending programs, it is important to find economic policy constraints appears to differ greatly out who is helped by the programs, to what extent from one country to another (Young 1993). they are helped, and what new binding constraints To analyze the impact of a change in policy, it is emerge. Since own finance is the principal source of necessary to trace the effects of the change over time, 109 WIM P M.VIJVERBERG AND DONALD C. MEAD both in the external environment and in the enter- Second, the products or services of household prise's response to the change. It is possible to trace enterprises are inputs for other activities.The expan- these effects to some extent by asking respondents ret- sion of a household enterprise increases the supply of rospective questions about when specific changes took inputs to other enterprises. If a household enterprise place in the enterprise.This approach has serious lim- sells pesticides, it increases the availability of this input itations, however, since the respondents may have for- to small farmers. Such supply effects are often called gotten relevant details. Here panel surveys, which forward linkages. return to the same set of enterprises on a regular basis Researchers must ask: to ask up-to-date questions, hold a distinct advantage. To what extent does a given household enterprise (The advantages of panel data for studying household produce demand and supply effects (backward and enterprise are elaborated later in this chapter.) forward linkages) that enable other enterprises to Policymakers must know the answers to two key prosper? questions regarding household enterprises and macro- Household surveys can shed light on this question economic policy: only if detailed information is collected about the * How has the success of different types of household nature of products and services bought and sold.' enterprises been affected by the policies and proj- Supply linkages become clearer if entrepreneurs report ects that are in place in particular localities and at in detail what their output is and to whom they sell it. particular points in time? Demand linkages require similar detail on the pur- - How do changes in the existing policies and proj- chases made by entrepreneurs. Also needed is general ects affect the success of different types of house- information about the structure of the local economy hold enterprises? and about ways in which particular household enter- A better understanding of the impact of various poli- prises might contribute to other productive activities cies could make an important contribution to improv- in the region (see Chapter 13 on the community ing program and policy design. The task of program questionnaire). Clearly, the overall data requirements design is beyond the immediate purview of LSMS sur- for analyzing this issue are substantial. The multi-topic veys. However, by throwing some light on the kinds of nature of the LSMS questionnaire, which limits the constraints facing household enterprises, LSMS sur- amount of questions that can be asked on any given veys can be of considerable help to project designers topic, may make it difficult to justify the depth of in the World Bank and other development institutions. inquiry needed to address these questions. LSMS analysis may be country-specific, time-specific, In a cruder way, LSMS surveys may help address a or both, and should take into account such factors as related key question: the dynamism of the local economy, the self-reliance To what extent and in what sectors does a house- of households, the involvement of different household hold enterprise benefit from the establishment of a members in the productive activities of the household, large company in its proximity? the market participation of household enterprises in When a large company (a corporation, parastatal enter- the region, the type of technology used by each enter- prise,joint venture, or multinational corporation) moves prise, and the social culture (in the household or com- into a particular area, it can have a significant impact on munity) within which entrepreneurs make decisions. local household enterprises. This impact can be either positive or negative.The company may displace some of Household Enterprises and Other Sectors of the Economy the demand for these enterprises' products. It may raise Household enterprises have an impact beyond the local incomes and thus increase demand for the prod- household in two key ways. First, household enter- ucts. It may cause local wages to increase. prises use as inputs the products or services of other An additional question might be raised that could enterprises. Thus the expansion of a household enter- qualify answers about a company's impact on local prise creates a new demand for other enterprises' out- enterprises: Why did the company select this particu- puts. When a household enterprise expands its use of lar location? If government officials gave the company small oil seed presses, new markets open up for farm- special incentives to pick this location, why? Was it ers growing oil seeds. Such indirect demand effects are because wages were low, because there was a local often called backward linkages. unfulfilled demand for the company's products, or 110 CHAPTER 18 HOUSEHOLD ENTERPRISES because there was already a thriving private sector? enterprises) it is possible to consider only the impact Might the success of household enterprises near a of policy on small-scale private enterprises. large company signify the company's decision to A number of the issues listed in Box 18.1 are locate near household enterprises that were already descriptive; others refer to a causal relationship. A descrip- profitable? Might the lack of success of household tive analysis of the small enterprise sector is useful enterprises near a large company signify the company's because the amount of reliable information about house- decision to locate where the household enterprises hold enterprises is so small. To understand causal phe- could easily be displaced? The point is that the exter- nomena (including interesting policy variables) goes sev- nal market environment cannot always be taken as eral steps further. Not only are good data needed, but the given. In studying the economic performance of small many causal factors must be identified, accurately meas- enterprises, it is appropriate to ask whether the pre- ured, and related to enterprise performance variables in sumed determining factors (as listed in previous sec- appropriate ways. While economic science has made tions of this chapter) are truly given or are determined progress in these areas, much remains to be learned. simultaneously alongside enterprise performance. Data Issues and Data Needs Summary Box 18.1 summarizes the major policy issues discussed A number of issues must be kept in mind when col- in this chapter. Because LSMS surveys are based on a lecting LSMS data about household enterprises.These sample of households (as opposed to a sample of issues include: Box 18.1 Policy Issues and LSMS Data Issues that can be analyzed using LSMS data Impact of other macroeconomic policies (such as tax poli- * The number of household enterprises in the economy. cies) on household enterprise performance. * Sectoral and locational characteristics of household enter- * Interactions between household enterprise performance prises. and other household activities. * Income generated by household enterprises. * Household enterprise performance over time. * Relationship between poverty and household enterprise income. Issues that are difficult to analyze with household survey data * Variability and seasonality of household enterprise * Impact of the distribution of enterprise income on con- income. sumption patterns within a household. (This requires * Income earned by hired labor in the small-scale private detailed data on the allocation of each kind of income sector among household members and on the role of each * Magnitude and structure of employment in household member in household spending decisions.) enterprises. * Links that enterprises have with other sectors of the econ- * Seasonality in household enterprise employment. omy (This would require an analysis of inter-enterprise * Patterns of employment growth in the small-scale private commodity flows in a local economy.Very detailed house- sector hold enterprise data would be useful but information on * Determinants of household entrepreneurship. the local economic structure would still be needed.) * Determinants of household enterprise income levels. *. Impact of regulatory and legal reforms. (This would require * Patterns of education, training, and experience among cross-country or time-series data to show variations in the household enterprise entrepreneurs. type of regulation across the samples observed.) * Impact of training programs provided by government and *. Impact of macroeconomic variables (such as foreign business organizations (on, for example, product design, exchange availability or market openness) on the small- quality control, marketing techniques, and management) scale private sector (This would require cross-country or on performance of household enterprises. time-series data to show variations in the macroeconom- * Marketing patterns for inputs and outputs and their ic variables across the samples observed.) impact on enterprise performance. Impact of culture, location, and level of development on * Role of locational factors and access to infrastructure in how much income and employment an enterprise gener- enterprise performance. ates. (This would require cross-country or time-series * Financing sources and impact of credit markets. data to show variations in culture and level of develop- * Impact of the regulatory and legal regime on enterprises. ment, both of which are usually constant within regions.) I I I WIM R M.VIJVERBERG AND DONALD C. MEAD * Target population. Sampling Considermtions * Fluctuations in enterprise activity over time. There are several other potential problems regarding * Panel data on enterprises. the randomness of the LSMS sample-problems * Enterprise income, sales revenue, and expenditures. which if not addressed properly could bias the statisti- * Business assets. cal results and policy recommendations derived from the data. Target Population LSMS surveys are designed to collect information SEVERAL ENTERPRISES PER HOUSEHOLD. Some house- about a random sample of households. By using a ran- holds operate more than one enterprise.Typically about dom sample of households, do LSMS surveys neces- 25 percent of the households that operate a nonagri- sarily capture a random sample of enterprises? This is cultural enterprise operate more than one and 5 percent an important question, as it is vital that an analysis of operate more than two.5 It might be tempting to reduce household enterprises be based on such a random costs by letting an interviewer determine on the spot sample. Every enterprise in the population must have which of a household's enterprises is the "most impor- an equal (or at least a priori known) likelihood of tant," and collecting information only on that enter- being selected. The LSMS survey design targets prise.This was the method used by the Peru LSMS sur- households rather than enterprises, meaning that veys of 1990 and 1991. However, if only the every household has an equal probability of being "important" enterprises in multi-enterprise households selected. If the survey collects data on every enter- are surveyed, the smallest family enterprises are likely to prise associated with the households in the survey, the be systematically excluded. To ensure a random sample survey design guarantees that each household enter- the interviewer must either survey all of the enterprises prise also has an equal chance of being selected (as operated by the household or record the number of explained below). enterprises in the household and randomly select one In LSMS surveys, a nonagricultural household or two enterprises from the full list. (Enterprise weights enterprise is defined as a household-operated busi- can be adjusted after the data are entered.) ness that performs any activity for the purpose of There is a compelling reason for surveying all of earning an income, with the exception of sale of the enterprises.To precisely measure household living agricultural crops or livestock products from a standards, all of a household's components must be household farm.2 Because public and parastatal measured, including all of the enterprises operated by enterprises, corporations, and cooperatives are not its members. associated with any one household, they are not included in LSMS samples. Thus LSMS survey JOINT OWNERSHIP OF ENTERPRISES. About 5 to 7 per- samples capture only the private, noncorporate cent of enterprises are owned jointly by two or more sector.3 households. Since the partners in such enterprises live An alternative way to acquire information about in different households, information about the enter- the private productive sector is an enterprise survey. prises could be collected at any one of those house- Because these surveys draw their enterprise samples holds. If the number of partners living in different from registration or address lists,4 these surveys households is n and the probability of selecting any (unlike LSMS surveys) do include larger enterprises household is p, the probability of selecting a particular and corporate, public/parastatal, and joint-venture jointly owned enterprise equals np. In analysis, jointly enterprises. However, they tend to overlook unregis- owned enterprises should be given a weight equal to tered and itinerant enterprises as well as the many 1/n while enterprises owned by a single household are household-based enterprises that do not appear on given a weight of 1. To make this weighting possible, any registration or address lists. LSMS survey samples, it is important that the LSMS questionnaire ask on the other hand, do tend to capture these unregis- respondents for the number of other households in tered private productive activities.Thus neither enter- which other owners of the enterprise live.6 prise nor household surveys yield complete (and hence fully random) samples of the total productive NoNRESPONSE. The sample of enterprises generated in nonagricultural sector. an LSMS survey suffers from several problems of non- 112 CHAPTER 18 HOUSEHOLD ENTERPRISES response. One form of nonresponse occurs when a ences during the recent past-say, the past month or household either cannot be found or refuses to the past 12 months. (This approach is also used in answer. A second form of nonresponse is the collec- Chapter 8 on health, Chapter 9 on employment, tion of household enterprise data from the wrong Chapter 19 on agriculture. Chapter 5 on consump- household member. To collect accurate information tion, Chapter 11 on transfers, Chapter 21 on credit, about an enterprise, an interviewer must try to address and Chapter 20 on savings.) In some cases survey the relevant questions to the member of the house- questions are used to account for fluctuations in the hold who actually operates the enterprise. status of an issue (such as the consumption of occa- How often does nonresponse happen? The 1991 sionally purchased commodities); in other cases survey Pakistan survey provides the best indication since questions are used to establish a trend (such as whether entrepreneurs were questioned in both the first and the household borrowed or saved money). second rounds of the survey. In both rounds, between The income generated by some enterprises follows 5 and 10 percent of households could not be found or seasonal patterns. LSMS surveys should seek to find out refused to answer. According to percentages from the both the level around which the income fluctuates and C6te d'Lvoire (1985-88), Ghana (1987-88), and the amount of variation. Measuring income at only one Vietnam (1992-93) surveys, for about 5 percent of (often nonrandom) point in time during the year is enterprises the respondent was someone other than insufficient because this does not allow seasonal varia- the entrepreneur. tions to be distinguished from the general level of prof- itability. An interview may even occur out of season Fluctuations in Enterprise Activity over Time when an enterprise is not in operation-and lead to the There are two ways to find out if a household is oper- incorrect conclusion that the interviewed household is ating a household enterprise. The first way is to ask not involved in any household enterprise activity. whether the household is currently involved in any It might be interesting to know what kinds of nonfarm activity on its own account with the intent households are most likely to operate enterprises-in to earn an income. This establishes the number of other words, to exhibit signs of entrepreneurship. To household enterprises currently in operation.The sec- study this question, it is necessary to survey all activi- ond way is to ask whether the household is involved ties that a household conducts on its own account, not in any such activity or has been so involved during the only at the time of the visit but also over a recent time past 12 months.This method includes not only house- period such as the past year or the past two years.The hold enterprises currently in operation but also any time factor is important because it can indicate enterprises that were in operation during the previous whether a household's failed business was once a use- year but are not operating currently-which amount- ful source of income or had always been a sinkhole. ed to between 10 and 25 percent of the sample in Either way, a trend exists that cannot be seen by doc- Vietnam, Pakistan, and Ecuador. umenting only the current involvement of a house- These two approaches generate two different hold's members in a farm, in a household enterprise, samples of enterprises, with an important difference or in wage employment. Surveying failed household between them. The second approach captures the enterprises can be as meaningful for documenting the enterprises that have gone out of business during the transitions in a household as is, for example, surveying previous 12 months as well as those that are seasonal changes in household composition. and not operating at the time of the interview. If the activities of an enterprise are seasonal, an Is it necessary to survey enterprises that are not in interviewer should ask when it carries out these activ- operation at the time of the interviewer's visit?Yes, for ities and for how long. This will help researchers and three reasons.While the living standard of a household policymakers understand the fluctuations in local can be measured by evaluating current stocks of labor markets and income and consumption patterns human and physical capital, standards of living are also during the year. When an enterprise is not in opera- measured by flows, particularly of consumption and tion for reasons other than seasonality, it is useful to income.7 Because both stock and flow variables are inquire how long ago the enterprise was last active, affected by temporal and temporary variations, it is why it is not currently in operation, and if and when reasonable to inquire about the household's experi- the entrepreneur anticipates starting it up again. 1 13 WIM P M.VIJVERBERG AND DONALD C. MEAD In any line of questions about past household can rule out all other external influences including enterprise activities it should be noted that in coun- changes in credit markets, output prices, foreign tries where inflation has been high-say, more than 30 exchange rates, and tax rates. or 40 percent per year-an entrepreneur's responses On the other hand, if each round of a panel ques- about monetary values of past revenues and expendi- tionnaire collects information about the wages that tures probably lose some accuracy. household enterprises pay hired workers (or would have to pay them if they hired any), and if these wages Panel Data on Enterprises vary between enterprises (in different areas-say, com- The issue of how enterprises change over time leads munities), one may estimate the impact of any wage to the question of how much can be learned by col- increase on household enterprise hiring practices. An lecting panel data (see also Chapter 23 on panel increase in the minimum wage is merely one example data). Panel data may be useful to track the increas- of a wage increase. ing role of household enterprises over time in many Panel data have value in both static and dynamic countries. Repeating LSMS questionnaires with economic environments. Even in static environments, unrelated samples in various consecutive years may luck or errors of judgment often cause the perform- help researchers examine the role of small enterpris- ance of enterprises to take turns for the better or for es in the economy both in recessionary times and in the worse. While some households continually invest times of rapid economic growth. Yet while using in their enterprises, others let their capital wear out unrelated samples allows one to describe an aggre- until it is no longer very useful. These choices cause gate degree of change in household entrepreneur- changes in income and employment that affect the ship, more information can be gained if the same household. Panel data help analysts see whether households are visited more than once. Researchers enterprises' lean years are temporary or permanent, can then observe which households and household how likely certain survivalist enterprises or microen- members have been more responsive to changing terprises are to succeed, and the long-term impact of external conditions. Has the response been better credit restrictions on enterprises. Panel data can also from poor or wealthy households? From men or be used to explore changes in the universe of enter- women? From better-educated or less well-educated prises-for example, to identify enterprises that have entrepreneurs (Schultz 1975)? closed since the last round of questioning. This may While there is a definite benefit to panel data, make it possible to find the former owner to ask why one must keep in mind that a panel study consisting the enterprise shut down and what the owner has of, say, four rounds generates data at only four points been doing since. And the use of panel data opens up in time. (For that matter, the same is true for repeat- new possibilities for analyzing interactions among dif- ed cross-sectional household surveys.) Between those ferent activities within a household, particularly in four points in time the economic environment terms of profit allocation among different household changes for a variety of reasons. And it is difficult to members and activities. identify which economic factors affect the perform- When panel data are collected, survey costs may ance of the enterprise: does enterprise performance be high. However, these costs can be reduced if the change because of a changing regulatory environ- entire questionnaire is not administered to the house- ment, a macroeconomic policy variable, a structural hold each round (or year).Yet because the household market phenomenon? enterprise module contributes information that is Unless there is a dominant event during the oper- essential for determining a household's living stan- ation of the panel questionnaire, the researcher is able dards, one may wish to administer at least the short to describe responses of household enterprises but- version of the questionnaire to each enterprise, along without other relevant information-not exactly to with questions that gather information about enter- what they respond. Suppose the legal minimum wage prise turnover.And given the flexibility that character- is raised. Household enterprises might be expected to izes good entrepreneurship, it is recommended to reduce their hiring of labor.Yet comparing the hiring gather the full amount of information at least every practices of enterprises before and after a minimum two years. (This will sufficiently account for the lag in wage hike does not show a clear correlation unless one the impact of any policy.) 1 14 CHAPTER 18 HOUSEHOLD ENTERPRISES Enterprise Income, Sales Revenue, and Expenditures model questionnaire in this chapter asks for many Ensuring the accuracy of the data on enterprise details. income is the most challenging part of the enterprise Still, it is useful to have some yardsticks against module. The income yielded by an enterprise can be which to evaluate revenue responses. A simple yard- measured in two ways: directly by asking the owner, or stick might be the following. After inquiring about indirectly by subtracting expenditures from sales rev- recent revenue, the interviewer asks, "Has your busi- enue. The following discussion focuses on the sales ness made more sales since my last visit than in the two revenue and expenditures variables as well as the vari- weeks before that visit?" and gives the respondent the able for enterprise income. choice of three answers-more sales, fewer sales, or about the same number of sales. This question SALES REvENUE. At a minimum, an LSMS question- appeared in theVietnam questionnaire. Indeed, it hap- naire should inquire about recent revenue and revenue pened that some entrepreneurs happily stated that they over the previous 12 months. (For enterprises not had made "more sales" but actually reported a smaller operating when a questionnaire is administered, only figure, or vice versa. This may highlight possible inac- the 12-month question is asked.) This strategy is based curacies in the reported sales, but it is also possible that on two ideas. First, an entrepreneur is likely to give the entrepreneur is not responding correctly to the more accurate answers about his or her enterprise's question itself. It is therefore imperative to incorporate recent economic performance than about its past eco- other yardsticks with which one can examine the nomic performance-so even if the recent period is accuracy of the revenue data. (The section below on not representative of the events throughout the year, enterprise income will return to this issue and propose information provided about this period will be of solutions.) higher quality and therefore be useful to researchers. The questionnaire must distinguish among differ- Second, it is useful to have information about an ent kinds of revenue: cash receipts, in-kind payments enterprise's patterns of revenue over a longer period of for goods and services, and home consumption.9 time, even if the respondent is less than perfectly accu- Typically, fewer than 10 percent of the entrepreneurs rate in his or her recall.8 indicate that they have received in-kind payments. For Comparing figures reported for recent revenues slightly more than 25 percent of the enterprises, with reported figures for 12-month revenue may respondents report some home consumption. reveal wide variations when both recent and 12- month revenue flows are expressed in monthly values. ENTERPRISE EXPENDITURES. To measure enterprise Recent revenue in the Pakistan survey was less than expenditures on inputs, a questionnaire must contain, half the 12-month revenue for 12 percent of the cur- at a minimum, a grid describing expenditures on a rently operating enterprises and more than double the specified list of inputs, with the following kinds of 12-month revenue for 4.9 percent of them. In questions for each input item: "Ql: During the past Vietnam, recent revenue xvas less than half the 12- 12 months, did you purchase ... ?" "Q2: How much month revenue for 7.9 percent of household enter- did you usually pay for ... ?""Q3: How often did you prises and more than double the 12-month revenue pay for .. .?"" These questions do not yield particu- for 18.2 percent. The two measures of revenue occa- larly good data, for three reasons. First, there is sub- sionally differ tenfold or more. stantial variation in how many items, and what types Several factors may explain the difference of items, are listed by entrepreneurs. In existing between recent and 12-month revenue flows. High LSMS data sets most entrepreneurs mention one, inflation may distort people's memory of past mone- two, or three items, and some mention none-with tary values; business cycles may cause fluctuations in wide variation in the items that are mentioned. enterprise performance; respondents (or interviewers) Certain items may be used but not purchased; may be unclear about the length of the reference peri- respondents often mention purchases of either raw od that applies to the questions asked; respondents may materials or items for resale but rarely both, regard- have difficulty recalling a 12-month revenue figure; less of whether their enterprises are in manufactur- and seasonal factors may play a role. In the hope of ing, restaurant services, or the retail trade; and few avoiding these problems, the revenue section of the enterprises report expenditures on electricity, fuel, or 1 15 WIM R M.VIJVERBERG AND DONALD C. MEAD water, even in sectors where they ought to be com- There are four sets of clues. First, the most extreme mon. Of course, differences in production techniques examples are found in Cote d'lvoire and Ghana, may explain some of the variation. where the questionnaires were administered in the The second difficulty with these data is that an mid-1980s. The results from Ecuador, Pakistan, item purchased by an enterprise may well be shared Tanzania (1994), andVietnam are more recent and are with the household and with other enterprises.'1 If more plausible, suggesting that questionnaire design sharing takes place, there is measurement error in both had improved, interviewer training was better, and household consumption expenditures and enterprise interviewers were more alert to potential misreport- expenditures (or in the value of expenditures in the ing. However, the percentage for Peru (1985) is low as two enterprises involved).The third problem with the well. data is that when an enterprise receives inputs (such as Second, Vijverberg (1992) found that in Cote electricity, water, or the use of tools) from a household d'lvoire and Ghana negative profits occurred in enter- or from other household enterprises, the entrepreneur prises regardless of the entrepreneur's education and does not report the use of these inputs because there regardless of whether the interviewer spoke with has been no purchase. For an analysis at the household someone other than the entrepreneur. Third, among level, this is problematic since both the value of house- the seven countries listed in Table 18.1, the percentage hold consumption expenditures and the value of of enterprises with negative profits appears to be lower household income are overstated by the value of the in countries with a higher general education level. It is inputs received by the enterprise. For an analysis at the indeed plausible that general numeracy among the enterprise level, this causes expenditures to be under- population improves the accuracy of household ques- stated while revenue is unaffected-making the enter- tionnaires. Fourth, while it is possible that enterprise prise seem very efficient. income is extremely variable over time, it is hardly The three problems are related, and can be solved plausible that such a large percentage of enterprises in one stroke by ensuring that the questionnaire care- would tolerate negative cash flows for long. fully accounts for the use rather than just the purchase All in all, the clues show that the income figures of inputs. The questions should reflect the fact that are probably skewed toward the negative.The lesson to usable inputs can be acquired by purchasing them, by be learned is that the questionnaire should be designed borrowing them from relatives, friends, or household so that it yields accurate information and also enables members, by picking them up if they are discarded or researchers to cross-check reported revenue and free (such as firewood or packaging materials), or by expenditure figures with the answers to some other receiving them as gifts.Accounting for input use in the direct or indirect questions. questionnaire will also yield more information on One such check is accomplished by including the each enterprise's involvement in formal markets, since following question: "After making purchases for the some of the above methods of input procurement are nonmonetary market transactions that substitute for Table 18.1 Percentage of Enterprises with Negative Profits, purchase in formal markets. Selected LSMS Surveys Country Type of Percentage of enterprises ENTERPRISE INCOME. In principle, enterprise income is of survey profits computed with negative profits defined as the difference between an enterprise's rev- C6te d'lvoirea Food commerce 63.5 Nonfood commerce 38.7 enue and expenditures. In every existing LSMS data Ecuador Last month 19.0 set, enterprise income is highly variable across enter- Ghana' Commerce 63.5 prises, with large outliers at both the positive and the Food manufacturing 56.0 negative ends of the spectrum." Table 18.1 shows the ~ ~ ...................................."t.................................................... 24...9................. negative ends of the spectrum.ii Table 18.1 shows the Pakistan Recent 24.9 percentage of enterprises in various LSMS data sets Normal 21.1 Tanzania Recent (or else normal) 29.2 to 35.5 with negative calculated profits-enterprises that lost Vita nclsvst *13 Vietnam Since last visit I13.7 money even before the values of family labor and Past 12 months 14.9 assets were taken into account. At least in some coun- a. Profits refer to profts received since interviewers' last visit if the enterprise was in operation at the time of the previous interview, during the past 12 tries, these percentages are mplausibly high. Is this a months if it was not. real phenomenon or is measurement error to blame? Source: Vijverberg 1992. 1 16 CHAPTER 18 HOUSEHOLD ENTERPRISES business, is there usually any money left? If so, how use. In terms of actual money values, only 40 percent much?" In recognition of the fact that enterprise and of the enterprises' profit and net revenue values are household monies are sometimes intermingled, it is within 25 percent of each other; for another 20 per- also useful to ask: "Do you use part of the money you cent of the enterprises, one value is less than twice the get from this business for yourself or for your house- other. While some net revenue values are large, they hold? If so, how much?"'3 "Net revenue" is defined as are not nearly as extreme as profits seem to show. And the respondent's estimate of the amount of money net revenue values correlate more closely with busi- taken from the business for household use plus the ness asset values than do profits.'4 value of the household's consumption of the output of Two conclusions may be suggested by the find- the enterprise (Vijverberg 1992). ings. One is that because of the great difficulty in pre- In an ideal data set, net revenue should be the cisely estimating enterprise income from a household same as profits calculated from other data in this mod- survey such as the LSMS, it may be better to save time ule (in other words, total revenues of the enterprise and energy by concentrating on simpler measures- minus total expenses). However, experience with past even while recognizing them to be incomplete and LSMS data is sobering.Table 18.2 shows the dispersion imprecise. A second, opposite conclusion is that of enterprise revenues for Vietnam. Enterprises are because many of the most important policy issues divided into five quintiles, first according to their net relating to household enterprises require accurate esti- revenue and then according to their calculated profits. mates of enterprise income, even more energy should The table shows the cross-tabulation of these two be devoted to collecting the most accurate possible group values. If data measures were accurate and con- estimates of income earned by these enterprises. sistent, each of the diagonal cells (those on row 1, col- If the decision is made to aim for a simpler meas- umn 1; row 2, column 2; and so on) would contain ure that is not fully precise but that is more easily one-fifth of the enterprises. Instead, the lower left and understood (and therefore more Ekely to yield mean- upper right corners show significant numbers of ingful results), this might mean paying only cursory enterprises where one income measure is high and the attention to various types of transfers between an enter- other is low. Three-quarters of enterprises in the first prise and other household activities. It would then be column would run at a loss ifjudged by their account- possible to develop the principal expense and revenue ing profit, even though some of their owners stated categories such that they could be compared during the that they have a substantial sum left over for household course of the interview and presented to respondents to Table 1 8.2a Comparing Net Revenue and Recent Profit,Vietnam Quintile rank for recent profit Rank for net revenue 1 2 3 4 5 Total .! ............. 7.71 5.93 2.71 1.87 1.78 20.00 2 4.25 8.27 4.21 1.96 1.31 20.00 3 2.90 4.44 6.92 3.36 2.48 20.09 ........................................... ..............................I...0.3................................5....6..5................................7...2.9...............................3...3..6................................. 19...5... 4 2.62 1.03 5.65 7.29 3.36 1 9.95 ......*................... .................. ....... *.................................................................1...............................5.....-1............................I.T...................................... 9. '95. ... 5 2.52 0.33 0.51 5.51 1 1.07 19.95 Total 20.00 20.00 20.00 20.00 20.00 100.00 Source: Authors' computat ons. Table 18.2b Comparing Net Revenue and 12-Month Profit,Vietnam __ ---- Q Quintile rank for 12-mnonthLrofit__ _ __ Rank for net revenue 1 2 3 4 5 Total ! 7.93 8.54 2.00 0.71 0.75 19.93 2 3.57 8.86 4.68 2.11 0.82 20.04 3 3.00 1.64 9.93 3.57 186 20.0 ..................................................................... .........................................5j................................................................................................... H~.... 4 2.46 0.54 2.93 10.64 3.46 20.04 5 3.04 0.43 0.46 2.96 13.1 1 20.00 Total 20.00 20.00 20.00 20 00 20 00 i100.00 Source: Authors' computations. 117 WIM R M.VIJVERBERG AND DONALD C. MEAD ensure that, in broad terms, their values are indeed com- an entrepreneur's responses into monthly figures of parable. The standard version of the questionnaire pre- total revenue, total expenditures, and net enterprise sented in this chapter has taken this route. income. The interviewer would immediately be able to Another approach in an abbreviated LSMS might verify these calculations with the entrepreneur and be to ask questions only about net revenue, without make any necessary corrections. However, calculating separating revenues from expenditures. (For further these figures requires a large number of data items, and discussion see Chapter 17 on measuring total house- would be difficult to do manually. To solve this prob- hold income.) While this may seem likely to yield lem, the interviewer could be provided with a laptop enterprise performance information at least as the computer loaded with software that contained a fully accurate as the information yielded by separate gross coded questionnaire. As he or she entered the respon- revenue and detailed expenditure questions, it is prob- dent's answers to the questions into the computer, ably less accurate. Why? In essence, the entrepreneur's another piece of software would automatically and response to a net revenue question is an educated immediately use the data entries to calculate revenues, guess, informed by his or her knowledge about the expenditures, and income of the enterprise in question. revenues and expenditures the enterprise incurs, by a This process is known as "computer assisted personal desired income level, or by his or her perceived con- interviewing." (The questionnaire software would con- sumption expenditures. The response ought to be tain all skip patterns, which would have the additional related to the performance of the enterprise, but there advantage of reducing skip errors as well.) Of course, is no guarantee that it is. computer assisted interviewing has costs as well as the In addition, the detailed revenue and expenditure benefit of increasing data accuracy. Using laptop com- questions provide other important insights about the puters for fieldwork is expensive, makes supervision of entrepreneur's business environment (for example, the interviewers more difficult than in the present system, degree to which the enterprise participates in the and requires interviewers who are more skilled. market) that could be influenced by public policy. Independent of the accuracy of the income estimates, Employment in the Enterprise both the gross revenue/expenditure questions and the A variety of facts are needed to explore employment- net revenue questions contribute useful information related policy issues. First, it is necessary to identify and should be retained. which household members are employed by the It may be possible to check on the accuracy of enterprise and how much work they do. This allows income estimates by asking detailed questions about analysts to relate the characteristics of household inventories.When an entrepreneur states the value of members-such as age, migration status, any illness, his or her inventory, the interviewer can ask how and level of schooling-to the operation and per- many business days this inventory will provide for. formance of the enterprise. Next, details must be col- However, it is important to note that inventories are a lected regarding any paid workers, apprentices, and more meaningful concept when an enterprise is cur- nonhousehold unpaid workers in the enterprise.What rently in operation. A seasonal enterprise or an enter- skills do these workers have? How much are they paid? prise that has ceased operating is less likely to carry an How many are male and how many female? How inventory, and even if it does carry one, the inventory much time do they devote to working in the enter- is unlikely to be reconcilable with any particular rate prise? After these details have been collected, the sur- of production or sales. vey should establish the size of the enterprise's work Another possible check could be to ask questions force over the past few (say, up to five) years-and per- about variations in business sales.The interviewer would haps also past characteristics of the work force ask the entrepreneur to describe both daily variations in (although this is very demanding on the respondents). the sales of the enterprise and how the revenues of the These data will show growth in employment, which enterprise relate to monthly expenditures. In this way can be related to the characteristics of the entrepre- the entrepreneur would be prompted to reveal his or her neur and the household. Finally, while seasonal estimate of the enterprise's cash expenditures. employment patterns are likely to follow seasonal sales The most effective method to collect more accu- patterns, it is nevertheless helpful to ascertain patterns rate data would be a quick, on-the-spot conversion of of employment in the enterprise over one year. 118 CHAPTER 18 HOUSEHOLD ENTERPRISES In order to save time and money, it may be tempt- The standard questionnaire in this chapter con- ing to collect work information about household tains employment (and labor seasonality) questions members in the employment module (see Chapter 9). within the enterprise module. This avoids any match- This strategy, followed in previous LSMS question- ing problems.And if necessary, responses that an entre- naires, has been problematic. In all but two cases sur- preneur gives in this module can be cross-checked veys collected information about hours of family labor against responses that individual household members in the employment module, during the first visit of the give in the Chapter 9 employment module (although interviewer to the household.5 To allow researchers to such a cross-check is subject to the timing and match- link data from the enterprise module with the ing problems mentioned above). The questions are responses of household members to questions about designed to have the same reference period as the their economic activities, many surveys (such as those income information in Part C of the employment in Vietnam and Ghana) have asked entrepreneurs to module. report the names of the household members who The expanded questionnaire additionally asks work in the enterprise; the interviewer recorded the whether and how much the enterprise paid to mem- ID numbers associated with these names. In principle bers of the household for their labor-a line of ques- researchers could use this information to consult the tioning that sheds light on intrahousehold income dis- employment module and extract the relevant enter- tribution. The short version of the questionnaire prise labor information. In practice this is not so easy. reverts to old practice, asking only for IDs of house- It is not immediately clear whether the researcher hold members working in the enterprise. should look at those members' main or secondary jobs during the previous week (or during the previous Business Assets year). What if two or more of a family member's jobs Business assets are an important determinant of enter- match? What if the family member claims to work in prise performance. Enterprise performance can be a different industry? What if the family member does measured not only by labor productivity or by the not report any hours of work in self-employment? If a absolute amount of income generated but also in household operates two enterprises in the same indus- terms of the percentage return to investments in the try, and both entrepreneurs claim that member as hav- enterprise. And an enterprise's start-up and subsequent ing worked for them, with which enterprise should a performance depend heavily on the entrepreneur's given family member be linked? What if other family ability to acquire the assets needed to be competitive members claim to work in self-employment in the in the sector. If one of the purposes of a particular sur- same industry as the enterprise, but the entrepreneur vey is to investigate the credit needs of small-scale pri- does not report employing them? Both conceptually vate enterprises, it is important to collect information and practically, there are likely to be numerous match- about business assets. ing problems that will be very time-consuming to sort Business assets come in two forms: fixed assets and out. inventories. Fixed assets include land, buildings, tools, The questionnaire in this chapter is inspired by the machinery, furniture, and vehicles used by the labor Pakistan and Ecuador formulations. Unlike previous force. Inventories consist of raxv materials, intermedi- LSMS surveys, the Pakistan questionnaire did not have ate goods that need to be further processed, and fin- an employment module. Instead it asked about hours of ished products ready for sale. Finished products are work at the same time that it explored wage employ- especially important for trading enterprises but can ment, farm labor, and work in family enterprises. Both also be significant for manufacturing enterprises. the Pakistan and Ecuador questionnaires required the Current enterprise performance is determined by entrepreneur to list ID numbers and working hours of the business assets in use at the moment.Therefore the all of the household members who work in his or her questionnaire must focus on the typical value of assets enterprise. This prevented many of the matching prob- in use during the reference period. Recent enterprise lems from arising.'6 The household enterprise module income can be analyzed using the current value of in the Ecuador questionnaire asked entrepreneurs for business assets. To analyze income over the past 12 the ID numbers and hours of work of the household months, more information is needed: the value of cur- members who worked for the enterprise.'7 rent business assets as well as sales and purchases dur- 119 WIM P. M.VIJVERBERG AND DONALD C. MEAD ing the past 12 months.8 While it matters when these subject to the same (national) regulations. However, if sales and purchases took place, asking for such dates is the LSMS survey is administered in a country where too burdensome in a multi-topic LSMS survey. local regulations vary across localities, these areas are Rather, assuming that sales and purchases took place particularly valuable targets for information gathering. on average a half year ago, the typical value of business In most cases it is important to know whether entre- assets in use over the past 12 months may approximat- preneurs abide by regulations that should, in principle, ed by apply to them. It may also be important to find out from entrepreneurs whether they are aware of the reg- [current value of assets] + [value of assets sold]/2 - ulations and whether they have sought to act on them. [value of assets purchased]/2. The expanded questionnaire in this chapter contains one example of such questions, regarding enterprise For land and buildings, one might also ask whether the registration. enterprise made any expenditures on improvements; It is equally difficult to measure the impact of these may be counted as assets purchased. Note, macroeconomic variables.The draft questionnaire asks though, that the usual quantity of inventories is diffi- about the incidence of taxes; this question could be cult if not impossible to measure; the questionnaires expanded to fit local conditions. Entrepreneurs could outlined in this chapter ask only for current values. also be asked about what opportunities they have to For many purposes, the most important question purchase foreign currency, their experience in buying about fixed assets is not so much what assets are owned imported commodities (inputs), and the competition by the enterprise but rather what assets it uses. An they face from imported products. entrepreneur may rent, own, or borrow assets from a The community questionnaire should contain neighbor or relative or from another enterprise oper- questions about community characteristics that affect ating in the household. Experience with previous the performance of the enterprise. Such characteristics LSMS data sets indicates that a significant proportion include roads, other infrastructure (such as railroads, (about one-fourth) of household enterprise owners waterways, telecommunication systems, market cen- report owning no assets, and those that do own assets ters, and utility services), and the presence or absence often share them with household members or with of banking and credit organizations. In addition, gov- other household enterprises; this is particularly the ernment and business associations should be asked in case with vehicles.'9 If an asset is shared, it contributes the community questionnaire about the nature of not only to the income of the enterprise that owns it their enterprise assistance programs (both direct and but also to the income of other enterprises or to gen- indirect). For specific examples of such questions see eral household welfare. In light of this fact it is neces- Chapter 13 on the community questionnaire. sary to devise a way to account for the complex Still more data are needed if all the policy ques- sources and uses of business assets; for one possible tions outlined in the first section of this chapter are to method see the expanded version of the questionnaire. be addressed.The amount of exposure that members of the household have had to entrepreneurship may be an Policy Variables important determinant of the household's decision to Variables must be collected that indicate the extent to start up an enterprise. This exposure may have come which enterprises are involved in formal and informal from parents or other kin of household members. It credit markets-both in receiving credit from sources may also have come from friends, but the influence of (such as suppliers and banks) and in extending credit a circle of friends is harder to measure in a standardized to others-and on what terms the credit is provided. manner across a sample. Moreover, causality may be In addition, the questionnaire must ask about the var- difficult to establish; does friendship with other entre- ious forms of (noncredit) professional assistance that preneurs stimulate someone to start an enterprise, or an enterprise might have received-for example, in does the operation of an enterprise cause the entrepre- product design, quality control, management tech- neur to rub shoulders with other entrepreneurs? niques, or bookkeeping. Does receiving an inheritance prompt people to It is hard to measure the impact of regulatory start up an enterprise? To answer this question, it is reform on household enterprises if all enterprises are necessary to ask a question about when a household 120 CHAPTER 18 HOUSEHOLD ENTERPRISES inherited major amounts of wealth-a question that household members. (In any case, it is unlikely that would need to be posed to all households, and not any household survey will ever measure consumption only in the context of the household enterprise mod- expenditures at the level of individual household ule. Data about the receipt of an inheritance could members.) then be analyzed together with the data from the The impact of income earned from household enterprise module about the age of the enterprise. enterprises on household consumption patterns can It would be useful to measure the way an enter- be measured by correlating household expenditures in prise has grown in the past, both in terms of income certain consumption categories with the earnings that (of both the enterprise and its employees) and in terms the entrepreneur and those members of the household of employment. Retrospective information about who are explicitly paid to work for the enterprise employment is less likely to be tainted by rccall errors bring into the houschold. Questions on thcse meas- than retrospective information about income because ures are included in the model questionnaire.To make the employment information is less detailed and easi- this analysis complete, it is necessary to have a full er to recall. However, most enterprises employ only accounting of every kind of income for each house- one or two household members; in such cases, growth hold member, including xvage earnings, farm income, is evident more in their work effort and in the income pensions, remittances, and interest income. For a more that the enterprise generates than in employment cre- elaborate discussion of these issues see Chapter 24 on ation. The draft questionnaire in this chapter asks intrahousehold transfers. about the number of people employed by the enter- prise one and two years previously and also over the Summary previous 12 months (to measure both seasonahty and Table 18.3 summarizes the data requirements implied trend growth). by the policy issues and research questions outlined in Another policy question concerns the impact of the first section of this chapter.The table refers to spe- human capital on household enterprise performance. cific questions by their number in the expanded ver- The human capital of household members is measured sion of the questionnaire. The last column of Table in the education section of the core LSMS question- 18.3 highlights the need for data collected in other naire (Chapter 7); the household enterprise module modules of the LSMS questionnaire. asks which household members work for the enter- The prospects for analysis with data generated by prise. No previous LSMS survey has asked for infor- the short, standard, and expanded versions of the mation about nonhousehold workers beyond what model questionnaire are rated from not possible to number is employed by the enterprise.The draft ques- poor to fair to good. In all cases it must be remem- tionnaire asks how many of an enterprises' workers bered that the sample is of small-scale enterprises, and have reached a certain level of schooling and how excludes corporations, public/parastatal and foreign many of them are "skilled" in the eyes of the entre- enterprises, and most joint ventures. As always the preneur. This "skill" measure is somewhat subjective, quality of the analysis depends heavily on the quality but it is impossible to measure in a more precise way of the data, which is influenced in crucial ways by how given the great diversity among household enterprises well the interviewers are trained and how well they in most developing countries. are supervised in the field. Finally, in order to measure the distributional effects of household enterprises within a household, Draft Module the questionnaire must include questions about who received money from the enterprise and how they (or The expanded version of the household enterprises the household at large) spent it. Some household module contains the most complete but also the most members may be explicitly paid by the entrepreneur; demanding list of questions. The standard version of the others may receive implicit pay when they spend a module, which abbreviates this list, does not permit part of the enterprise's revenue for consumption pur- analysis of some of the policy questions but still attempts poscs, to meet eithcr their owvn or the household's to provide good measures of important data such as needs. It is therefore difficult to get an exact account- enterprise income and employment (see Table 18.2). ing of the distribution of all enterprise income among The short version, even more abbreviated, aims to col- 121 WIM P. M.VIJVERBERG AND DONALD C. MEAD Table 18.3 Household Enterprises Module: Requirements and Links with Other Modules Short version Standard version Expanded version Prospect Prospect Module Prospect Module Data needed Issue for analysis for analysis questions for analysis questions from other sections I. Number of enterprises Good Good B: 1, 3: C:6-7 Good .......................... .... T n ............................................................................................................................................................................................................ . 2. Sectoral and locational Good Good B:3; C:4; E: 1-3 Good characteristics .....................,,................................................................................................................................................................................................... 3. Measuring enterprise income Fair Good C:5-6, 8; D:27-28, Good F: 10- 13, 36-37, 46-47, 53-55; 15-18,22-25 E:4- 13, 19-26, 29, 34, 36-44; F:2-4, 14, 19-21, 26-27, 33-34; G:22-23; H: 1-3, 6-1 1, 31-40 4. Poverty Fair Good All of issue 3 Good Al of Core: household con- issue 3 sumption expenditures S.Variability and seasonality Good Good E27 2 0,9-30, 34, Good 6:28, 31-33 in income 36-35; H:31-33 6. Earnings of hired labor None Fair D:24-28, 32-37, Good D.:29-3 i 43-47 38-40,48-50 7. Employment Fair Fair D:1-6,9-1 1, 14-17, Good D:18,29-31, 21,24-26,32-35,41-45 38-40,48-58 8~~~~~~~.... Sesoa,t ,,,n, employment, ,,, None Poor,,,,,, ,,,,,,,,, Good E:35 9. Employment growvth Poor Poor E:30, 36-39 Fair D:56-57; E:31-33, 35; G:39 ID. Entrepreneurship Poor Poor B:3; C:3-5; E:26-27, Fair A: 1-2; B:2; Community; infrastructure; 29-30 H: 14-16 Core: human capital .............................................................................I...................I.....................-.......................................I........I...................................................... 11. Determinants of income Fair Fair All of issue 3; C:3-4 Good Al of issue 3; Community: infrastructure; G: 1-3, 21-23, 29-33, G:7-20, 24-28; Core: human capitai, size 37-38; H: 1-2, 6-9 H: 17 and value of residence .............................................................................................I.......................................................................I..............................I........................... 12 Education, training, Good Good B:4; C3; D:4, 15 Good A: 1-2 Core: employment history, and experience human capital, training, apprenticeships . .. . ... ... .. .. ... ... . ....................................................... . . ... .... .. .. . .. .. .. I. ... ........................ ......... . .. I ..... ... .a 13. Impact of government Nonc Nore Fair All of Issue 3; training programs G:0- 39; H: 17 .....!....................................................................................................................................................................................................... ................ 14 Marketing patterns None Poor E:27, 30; F: -4, 19-21 Good E:33; F:5-6, 10-18,22-24; H:4-5, 12-13 iS. iocation and nfrastructure Fair Fair All of Issue 3 Fair All of issue 3; Community: infrastructure B:2; H: 14 I 6..Finance and.creditGmarkets None None Good E:14-16; F:7-9; Core: household G:34-36; credit; Community: H:23-30 private sector description; Credit interest rates. individual credit ........ .....................I............................................................."........................................I........................ ............*................................................. 17. Regulaion and enterprise Poor Fair All of issues 3 and 7; Fair All of issues 3 performance F:26-27 and 7; F:28-32; H: 14 i8. Macroeconomic policy and None Fair All of issues 3 and 7; Fair All of issues 3 and 7; enterprise performance D:53-55; F:33-34 D:5 1-52; H: 18-25 ....... ........................................................................................................I......................................... .........................................I....................... 19. Enterprise performance and Good Good All of issues 3 and 7 Good All of issues Depends on issue other household activities 3 and 7 to be studied 20. Enterprise performance Good Good All of issues 3 and 7 Good All of Issues over time 3 and 7 2.intrahousehold interactions Poor Fair All of issue 3 Good All of issue 3; Depends on issue to F: 12-14, 18-25; be studied G7-15, 24-28; H4 I .................................................................................I.........................................I................................................................................... ........*.......... 22. Extrahousehold linkages None Poor G:21-23 Poor F:17;G:16- 8 Community: private sector description ............................................................................................................................................................................................................................ 122 CHAPTER 18 HOUSEHOLD ENTERPRISES Table 18.3 Household Enterprises Module: Requirements and Links with Other Modules (continued) Short version Standard version Expanded version Prospect Prospect Module Prospect Module Data needed Issue for analysis for analysis questions for analysis questions from other sections 23. Legal reform None Poor All of issues 3 and 7; Fair All of issues D:53-55; F:26-27 3 and 7; D:5 1-52; F:28-32 ............................................................................................................................................................................................................................ 24. Macroeconomic variables None Poor All of issues 3 and 7; Fair All of issues D:53-55; F:33-34 3 and 7; D:5 1-52; H: 18-25 25. Culture, location, development Fair Fair All of issues 3 and 7: Fair All of issues Core: urban/rural C:4 3 and 7 residence, ethnicity, religion; Community: culture Source: Author's eva uation of the household enterprise modu e. lect information on enterprise income, value (as meas- mended to study the questionnaire in combination ured by assets), and employment statistics-primarily for with the annotations provided in the last section of the study of other household-related issues. this chapter. Each version of the household enterprise module As a whole, the expanded version is probably too is divided into the same eight parts, which are labeled long to be administered in one LSMS survey. While A through H. The three versions are related and have the expanded version contains questions on every many questions in common. In some cases the core issue discussed above, not all of these questions should questions are restated in the different versions of the all be posed to respondents at the same time. A careful questionnaire. selection of topics of interest should limit the length of The respondent for parts A and B of the module is the questionnaire. the head of the household. For the rest of the module- The standard and expanded versions of the ques- used if the household operates an enterprise-the tionnaire follow the same basic format. The short respondent for each enterprise should be the person in questionnaire follows a somewhat different design, charge of the enterprise or the person most informed making it difficult to generate a customized version about the enterprise. (This person generaly works in that, in length, is somewhere between the short and the enterprise, although in exceptional cases, such as ill- standard versions. ness, he or she may not.) Table 18.5 shows average numbers of questions As was mentioned in the previous section, it is asked in the household enterprise module of an LSMS essential to gather information about all the enterpris- survey.The column labeled "per enterprise" counts the es within any given household. Part B of the proposed average number of questions asked of an enterprise; questionnaire alows up to six enterprises to be listed, the column labeled"per household" shows the average along with the industries in which they operate, the Table 18.4 The Parts of the Household Enterprise Module household members most informed about and/or in charge of their day-to-day operations (often referred Part Respondent Topic to as the "entrepreneurs"), and the sequence numbers A Household head Household exposure to entrepreneurship (1, 2, . . . ) that the interviewer assigns to enterprises B Household head Existence of nonagricultural enterprises within a household. Parts C through H of the module c. Entrepreneur General information about the are used to inquire about the enterprises. These parts enterprise contain a grid for three enterprises; if a household D Entrepreneur Employment of household and nonhousehold labor operates more than three enterprises, the interviewer E ........ nonhsehed labo r must enter information about the fourth, fifth, and F Fntrepreneur ion sc expesire F......E.~n't-re-p-re-ne-u-r...... In"p"ut ..u'se ..a"nd ..e"xp"en"di'tu're ............ sixth enterprises onto another household question- G Entrep-enu; Businessassets naire form.20 H Entrepreneur Inventories: enterprise start-up; Table 18.4 summarizes the content of each part of assistance programs; exposure to th questionnaire. To get a good grasp of the purpose international markets; enterprise debt; the questionnaire. To get a good grasp of the purpose trade credit; enterprise income of questions and skip patterns, it is highly recom- Source. Authors' summary of the household enterprise module. 123 WIM R M.VIJVERBERG AND DONALD C. MEAD Table 18.5 Average Number of Questions in the Household Enterprise Module Expanded version Standard version Short version Part/Type of question Survey items Per household Per enterprise Per household Per enterprise Per household Per enterprise A Exposure Al-A2 5.40 n.a. 0.00 n.a. 0.00n.a. ... ......... .........................................................................................................................................................I................................................... ..... B Enterprise existence B l-B4 2.64 3.00 2.04 3.92 2.04 3.92 .................................................................................................................................................................................................................................. C General characteristics Cl-C9 2.94 5.65 2.94 5.65 2.94 5.65 ................................................................................................................................................ I................................................................................... D Household labor D I-D 19 4.38 8.43 3.52 6.78 1.70 3.28 ................ ..a""or.....................-0.........-0................... 2.....'......................3.....-0......................0.....5,...................... I.....4,.................... 1............................. 2...13 .. . .... Nonhousehold labor D20 D50 2.03 3.90 0.85 1.64 1.11 2.13 Minimum wage D5 I -D52 0.65 1.25 0.00 0.00 0.00 0.00 .......... .................................... ..................... ............................................... ...... .................... ...... ............................................... .... .. .... Social security D53-C55 0.83 1.60 0.83 1.60 0.00 0.00 Labor growth D56-D57 1.04 2.00 0.00 0.00 0.00 0.00 ....................................................................................................................8.93..................... 17... 8 ......................5...20 .................... 10. 22 8 . 4 1 PortC.totol 8.93 17.18 5.20 1Q.02 2.81 5.41 E Type of enterprise EI-E3, E17 2.08 4.00 1.56 3.00 0.00 0.00 T:rading enterpnise E-E 13 5.04 9.70 5.04 9.70 0.00 0.00 Credit E 148E 16 0.43 0.83 0.00 0.00 0.00 0.00 Revenue El 8 E26, E29, 5.15 9.90 5. 5 9.90 4. 8 8.05 E34, E40-E44 ............................................................................................................... .................................................................... ................................................ Interrupted operation E27-E33 0.42 0.80 0.21 0.41 0.00 0.00 . ............................................................................................................................................................. *..................................................................... Seasonal labor E35 6.76 13.00 0.00 0.00 0.00 0.00 Seasonalirevenue E36-E39 7.80 15.00 7.80 15.00 7.80 15.00 Port E toota 27.68 53.23 19.76 38.01 11.98 23.04 F Resources FI-F6, F20-25 23.53 45.25 10.47 20.13 0.96 1.85 ................................................F2"6-F32 .....................2.'22....................... 4...27 ......................0... 60 ......................I... F,..................... 0 00.........*..............0... 0 ............ Credit F7-F9 1.29 2.49 0.00 0.00 0.00 0.00 Registration F26-F32 2.22 4.27 0.60 1.15 0.00 0.00 ............................................................. ~TT4................... .....................F,..................... ....................I...25 ......................0.'0........................0... 0 ............ O)ther taxes F33-F34 0.65 1.25 0.65 1.25 0.00 0.00 .................................................................................................... *......................................... .................................................................................. Port F rotol 27.46 53?26 I (.7 22.53 0.96 1.8 G Business assets G I-G33, 28.46 54.74 18.59 35.76 7.88 15.15 G37-G39 ..... ' 't..................................................G34' G"36................... 0... 31......................0... 60.............. *.......0... 0 ...................... 0... 0 ......................0.OO........................ 0... 0 ............ Credit G34-G36 0.31 0.60 0.00 0.00 0.00 0.00 PortGtotal 28.77 55.33 1859 35.76 7.88 151.5 ........................................................................................................... I........................................................................................................................ H Inventory H I -H3, H6-H I1 2.55 4.90 2.55 4.90 1.04 2.00 ....... ..........................................-4.........'6................... 1.....'......................3.....-0......................0.....0'......................0.....0'..................... 0....0........................0... 0 .. . .... Markets H4-H5, H 12-HI 13 1.46 2.80 0.00 0.00 0.00 0.00 ....... ...... .............* ..............* ........H 17....*....................... 2.60 ......................5... 00......................0.... 00,......................0.... 00,......................0.OO.....*..................0... 0 ............ Start-up H14H H16 1.56 3.00 0.00 0.00 0.00 0.00 ..................................................................... ..................1.'i..................... F 6 .................... ' 6 .................... 6 ................... .6 *............. 6 ........... Assistance H 17 2.60 5.00 0.00 0.00 0.00 0.00 Customer credit H27-H30 1.07 2.05 0.00 0.00 0.00 0.00 ............................ ............ ............................................ ........................................................................................ .... Enterprise income H31-H49 4.52 8.70 4.52 8.70 i.66 3.20 .............. ................................ ................. ...... .......................... ................................... ................ ................. ....... Use of income H4 1 0.52 1.00 0.00 0.00 0.00 0.00 ............... ................*.............................................I 6..............................-2.....5,...................... 7.....7,...................I 3............................ 2.....-0........................5...20 .. . .... Port H totl 16.82 32.35 7.07 13.60 2.70 5.20 Totolforaiiporrs i20.64 220.00 67i3; i29i49 3i.31 60.22 n.a Not appl cable. Soarce: Authors' estimatior based on experience with previous LSMS surveys. 124 CHAPTER 18 HOUSEHOLD ENTERPRISES Box 18.2 Cautionary Advice * How much of the draft module is new and unproven? In its accuracy of measured enterprise income may be judged. basic design, the household enterprise module provided Even with these changes there is no guarantee that data here follows the same approach taken in many previous will improve.At a minimum, though, implementing the rec- LSMS surveys. The module contains parts that inquire ommended questionnaires will inform future work on the about the general operation of the enterprise, its rev- design of household enterDrise surveys. which is still a new enues and expenditures, its work force, and its assets. In field of research. many ways, however, the module attempts to collect infor- * Which parts of the module most need to be customized? mation in more precise detail than has been achieved Several parts particularly need to be customized to coun- before. This is true in particular with respect to sales, try-specific circumstances. In Part D, questions referring to expenditures, employment, and seasonality. apprenticeships may not apply, and social security systems * How well hos the module worked in the past? Past house- must be mentioned by name according to the country's hold enterprise modules have produced somewhat ques- governmental structure. In Part E, units of measurement tionable income data.The current revisions to this module for articles should be specified according to local custom, aim to improve the reliability of the enterprise variables, providing suitable codes for weight and content measures; This will be done by: dealing with seasonality and the con- also, in some countries the list of possible buyers may dif- fusion it creates for answering questions about averages fer from what Part E provides. In Part F, questions con- and 12-month totals; asking the trading enterprses a set cerning licensing problems and practices may be made of questions on sales and expenditures on raw materials more country-specific, and survey designers should that is more suitable to their context; more carefully include a specific example of a tax levied on small busi- accounting for the use of inputs and business assets; and nesses. In Part G, one question about documents of own- asking for several cross-check measures, by which the ership may need to use a specific local terminology. number of questions one household is asked in the ed in part G. Sixty percent of these assets are owned household enterprise module; this average accounts by the enterprise. for all households regardless of whether they have an It is important to note that questions E35 and enterprise. E36, which record employment and income by The proportions of respondents branching off at month, are recorded for every month (totaling 13 for each skip point (D20, D5 1, and so on) is calculated based E35 and 12 for E36) rather than as a single question on responses to similar questions in the Ecuador (1993), each.While this adds substantially to the total number Pakistan (1991), andVietnam (1992) LSMS surveys, to of questions, questions E35 and E36 are not onerous the extent that this is possible; a substantial number of to administer. questions in the standard model are new. However, the most important assumptions are the following: Annotations to the Draft Module * Of all households, 30 percent operate one house- hold enterprise, 8 percent operate two, and 2 per- This section explains the motivations behind the ques- cent operate three. If the size of a sample were tions so that researchers may better customize the 10,000 households, this would yield 5,200 nonagri- model questionnaire to fit the circumstances of the cultural enterprises. country they are studying. The explanations in this * Eighty-five percent of the household enterprises section will also assist interviewers in implementing are actively operating at the time of the interview, the survey iri the field. * On average, 1.5 family members work in each Each group of questions is described in turn. enterprise. Occasionally an alternative format is discussed. The * Fifteen percent of enterprises employ someone question numbers refer to the expanded version. from outside the household. * A typical enterprise reports using inputs from three Part A: Household Exposure to Entrepreneurship of the seven input categories listed in part F Half of A1-A2. These questions seek to establish whether these inputs are purchased. there is a pattern of entrepreneurship in the family. All * A typical enterprise reports using business assets households should answer these questions so that it from four of the eight business asset categories list- becomes clear which households are most likely to 125 WIM P M.VIJVERBERG AND DONALD C. MEAD operate a household enterprise. It might be useful to C3. This piece of information might show the level of include this part of the household enterprise module technology, amount of experience, or degree of suc- in the part of the LSMS questionnaire where general cess (longevity) of the enterprise. household background information is gathered. This would increase the response rate on the part of house- C4. Location of operation is one way to distinguish holds that do not operate household enterprises. the type of an enterprise. Is the enterprise likely to be a significant contributor to the economy, or is it a sub- Part B: Establishing the Existence of Nonagricultural sistence enterprise? Enterprises B2. For households that do not operate an enterprise, C5. If the enterprise operates from within the home, the this question asks for reasons why. (Households that do home is a business asset, although not typically reported operate an enterprise will be asked an almost identical as such in Part G. For this question to be meaningful, the question in part H: question H14.) This question value and size of the home should be measured else- allows insights into how households cope with obsta- where in the questionnaire. The value of the home as a cles to private entrepreneurship. business asset depends on the proportion of the home used for business, the amount of time that the rooms in B3-B4. At several points in the household enterprise the home are used by the business (see questions E19, module the interviewer will refer back to informa- E26, E29, and E34), and the value of the home. tion gathered in B3-B4-information concerning either the enterprise being surveyed or other enter- C6-C7. These variables help adjust the sampling prises in the household (see Cl, C3, F13, F23, G9, weight for an enterprise. G12, and G25). Because the interviewer must have ready access to the names and enterprise code num- C8. This question aims to reveal the amount of enter- bers of each enterprise in the household, it is rec- prise income flowing to the household. ommended to record the responses to B3-B4 on a fold-out piece similar to the household roster. The C9. This is a very important question; the answer is answer to C9 should also be recorded on this fold- used several times as a filter, directing the interviewer out piece. to different parts of the questionnaire later on. (See questions D2, D20, E8, Eli, and E18.) Because the B3. The interviewer (or the person coding the question is so often referred to, it is recommended that responses) must have access to the International the answer to C9 be recorded on a fold-out piece sim- Standard Industry Classification. ilar to the piece for the household roster, along with the answers to questions B3-B4. B4. This question identifies who should be the respondent for the enterprise module.The respondent Part D: Employment should be the member of the household most knowl- D2. Questions D3-D13 are addressed to currently edgeable about the enterprise and/or the person in operating enterprises and questions D14-D19 are charge of the enterprise. (In this chapter this person is addressed to enterprises not currently in operation. often referred to as the "entrepreneur.") The inter- Although these sets of questions are parallel, merging viewer must make every effort to schedule an appoint- them creates difficult skip patterns. ment with this person. D3-D4. The respondent is usually also one of the fam- Part C: General Information ily members working in the enterprise. The phrasing Cl-C2. In exceptional cases the interviewer may have of question D3 implies that the entrepreneur will to conduct the interview with a household member automatically be listed. If the entrepreneur does not other than the entrepreneur. If so, the interviewer work in the enterprise, the responses to D5 and D1O should at least report who the actual respondent is so will be "O."AIl household members should be listed in as to indicate the credibility of his or her responses D3-D4 before the interviewer proceeds with about the enterprise. D5-D13 for each person in turn. 126 CHAPTER 18 HOUSEHOLD ENTERPRISES The names are recorded here for use by the intrahousehold allocation of enterprise income will interviewer; there is no need to code them into the view them as a portion of enterprise income. computer. The IDs will be used by researchers to link the personal characteristics (for example, age, educa- D20. This question is a filter that starts the portion of tional attainment, and sex) of workers to enterprise the module dealing with nonhousehold labor. performance. Questions D21-D40 are addressed to currently oper- ating enterprises and questions D41-D50 are D7-D8. Although household members receiving these addressed to enterprises not currently in operation. payments will view them as earnings, analysts of intra- These sets of questions are parallel, but merging them household allocation of enterprise income will view creates difficult skip patterns. them as a portion of enterprise income. D22. This question is used as a check on the answers D9. The researcher must assume that during the pre- to questions D32 and D33. vious 12 months, people responding to this question have contributed the same number of hours per day D23. This question is used as a check on the more and received the same payment as they did during the detailed question D24. An important note: if the past two weeks. (It is necessary to ask about the total entrepreneur states in D23 that the enterprise did not number of weeks because it is not wise to assume that employ nonhousehold labor during the past 14 days, it all household members who ever worked for the would be awkward to ask question D24. Instead, the enterprise during the 12-month period worked for interviewer should enter "0" into D24 without asking the entire 12 months.) and follow D24's skip pattern to D33. D12-D13. Just as for questions D7-D8, household D24-D26. If there is no apprenticeship system in the members receiving these payments will view them as country, the second answer row should be dropped. If earnings, but analysts of intrahousehold allocation of the number of workers in a specified category in D24 enterprise income will view them as a portion of is 0, the interviewer should skip to D33, which asks enterprise income. about work effort over the last 12 months only. D13. After all the enterprise workers have been ques- D27-D28. If apprentices and unpaid nonhousehold tioned, the interviewer can skip to D20. (Questions workers never receive any compensation in the coun- D14-D19 refer to enterprises not in operation at the try of the study, these questions need not be asked (in time of the interview.) which case question D27 disappears altogether). D14-D15. As in question D3, the phrasing of question D29-D31. These questions are designed to yield infor- D14 implies that the entrepreneur will automatically mation about the personal characteristics of the non- be listed. If the entrepreneur does not work in the household labor force.They should help establish labor enterprise, the response to D16 will be "O."AIl house- demand patterns for various demographic groups.The hold members should be listed in D14-D15 before schooling cutoff of six years is arbitrary and should be the interviewer proceeds with D16-D19 for each per- adjusted to an appropriate level for the country of the son in turn. study. If there is no apprenticeship system in the coun- The names are recorded here for use by the try, D30 may be replaced with a suitable question interviewer; there is no need to code them into the related to the training of unskilled workers. computer. The IDs will be used by researchers to link the personal characteristics (for example, age, educa- D32. For enterprises that employed workers in the spec- tional attainment, and sex) of workers to enterprise ified category during the previous 2 weeks, this question performance. is the only measure of these workers' efforts during the previous 12 months. The assumption is that days per D18-D19. Although household members receiving week, hours per day, and payments per worker are simi- these payments will view them as earnings, analysts of lar for the 2-week and 12-month periods. If the enter- 127 WIM P M.VIJVERBERG AND DONALD C. MEAD prise employed workers in the specified category during considered more reliable than retrospective responses the previous 2 weeks, no further questions will be asked about income. about the previous 12 months; the interviewer should turn to the worker category in the next column. Part E Revenues and Operation Schedule E1-E3. These questions describe types of enterprises by D36-D37. If apprentices and unpaid nonhousehold their output. Together with E17 they provide informa- workers never receive any compensation in the coun- tion about the extent to which enterprises generate try of the study, these questions need not be asked (in demand and supply effects, and in which markets they do which case question D36 disappears altogether). so. Another option would be to specify in detail some (perhaps up to five) commodities produced by the enter- D38-D40. These questions are designed to yield infor- prise-as was done in the Ecuador 1993 survey. This mation about the personal characteristics of the non- option was not chosen because such information would household labor force.They should help establish labor be tremendously time-consuming to code and analyze. demand patterns for various demographic groups. The Note that these questions do not substitute for B3 schooling cutoff of six years is arbitrary and should be (on type of industry).While B3 allows the respondent adjusted to an appropriate level for the country of the to specify a single industry, a substantial number of study. enterprises are involved in more than one of the three economic sectors. The skip pattern associated with the D41-D50. This block of questions, about nonhouse- answer to E3 ensures that only enterprises involved in hold labor for enterprises not currently operating, is trade can respond to questions E4-E16. parallel to D21-D22 and D33-D40 taken as a block. The sets of questions are written out separately since E4-E10. This set of questions aims to compute the typ- skip patterns in a merged block of questions would be ical gross profit margin on resold items (that is, items confusing. that are not modified in any way before being resold). Since many traders sell more than just five items, the D51-D52. These questions address the effect of mini- responses to E5, E9, and ElO cannot be used to com- mum wage legislation on the wage paid in the small- pute total revenue from sales in trading. However, since scale private sector.They measure the degree to which profit margins may differ between items, the inter- the entrepreneur feels bound by the legal minimum viewer has to inquire about the five most important wage. Whether the enterprise pays minimum wage or items-computing the typical gross profit margin as a higher can be deduced from questions D26-D27, weighted average across these five items. The gross D36-D37, and D46-D47 (although these questions profit margin can be applied to expenditures on pur- include the values of in-kind benefits). chasing items for resale (questions E12-E13) to esti- mate the enterprise's revenue from sales. Later in the D53-D55. These questions collect information about questionnaire, questions E20 and E35-E38 will record the social security coverage of all workers and about total (cash) revenue from sales; these numbers should payments made by each enterprise to the social secu- correspond with the sales figures derived from rity system. If greater detail is desired, these questions E12-E13 and the gross profit margin.As such, E20 and may be merged into the household labor grid and the E35-E38 can be thought of as accuracy checks on the two nonhousehold labor grids.The precise phrasing of responses given to E4-E13, and vice versa. the questions should be adjusted according to the pre- A substantial number of enterprises are probably vailing circumstances in the country. These questions involved in both trading and production-and-sales establish the coverage of potentially important social activities. For these enterprises, E20 and E35-E38 are security legislation. a mixture of revenues-so E21 and E39 will serve as accuracy checks for E4-E13, and vice versa. D56-D57. These questions help measure how much It is important to note that the gross profit margin an enterprise has contributed to employment growth. only represents the difference between purchase and These are the only retrospective questions in the mod- sale prices of resold items. To compute enterprise ule; retrospective responses about employment are income, other cost components must still be deducted. 128 CHAPTER 18 HOUSEHOLD ENTERPRISES E4. Writing down the information requested here will credit. If this is the case, an analyst of credit markets help the next five questions go smoothly. It is not will want to know what percentage of the goods were intended to be coded. purchased on credit (E14), how the creditor was paid To ensure the best possible responses, the inter- back (E15), and what the terms of credit were (E16). viewer should list all the items first, before asking Concerning the terms of credit, one would really like questions E5-E10 for each item. The idea is that to know the (implicit) interest rate, but finding this out traders list five of their most frequently traded com- would require several more questions: were you modities. They may be reluctant to list these once they charged interest; what was the interest rate; if you had understand what questions the interviewer is going to purchased these goods with cash, would you have ask them about the commodities. been able to purchase them at a lower price; if so, how much lower? Without such questions the analyst has to E5-E6. These questions ascertain the trading margin impute the customary regional interest rates that have for the five items that the enterprise purchases for been uncovered in the credit module of the question- resale. The aim of these questions is to find out from naire (see Chapter 21 on credit). the entrepreneurs how much they spend to purchase these items and how much they think they can sell E17. The customer base helps describe the enterprise them for. and its growth potential. E7. Items for resale can be purchased in bulk (for E19. This question makes it possible to compute example, by the bag) and sold by the piece. Since each enterprise revenues on a daily basis. This can be com- entrepreneur may be working with different units (for pared to the income that waged or salaried workers example, different-size bags), it is necessary to ask him earn. It is possible to go further and ask how many or her the relationship between the unit of purchase hours per day the enterprise was open for business; the and the unit of sale. In exceptional cases, items for model questionnaire has skipped this question for rea- resale can be purchased in smaller units than they are sons of brevity. sold. In such cases the interviewer is instructed to adjust the unit recorded in question E6. E20. These are receipts from sales for cash or credit. E9. Based on this question and on questions E5-E7 it E21. The answer for pure trading enterprises will be is possible to compute sales revenue and expenditures 100 percent. The answer for pure manufacturing or on items of resale. service enterprises will be 0 percent.The target of this question is enterprises that mix trading with other El0. Based on this question and on questions E5-E7 activities. it is possible to compute sales revenue and expendi- tures on items of resale. E22-E23. Besides cash or credit sales, some 10 percent of enterprises also receive payments in the form of E11-E13. Questions about expenditures on items for goods or services. In addition, the products that the resale are relevant only for enterprises involved in entrepreneur has used to purchase inputs should be trading. As the questions at the beginning of Part E counted as in-kind sales revenue. The value of such deal with trading anyway, it is proper to ask about such products counts both as a cost and as a revenue item. expenditures here rather than in Part F. Note that traders in business assets such as dealers in cars, bicy- E26. Currently operating enterprises have not neces- cles, and furniture (see Part G) should report their sarily been in operation for all of the previous 12 expenditures on these commodities here, because the months. Measuring annual income requires this ques- commodities are used for trading rather than for oper- tion and information about income flows. See ating the trading enterprise. E37-E44 for a more detailed explanation. E14-E16. When an enterprise buys goods with intent E28-33. This should be used only for enterprises that to resell, a portion of the goods may be purchased on are not currently operating. 129 WIMP R M.VIJVERBERG AND DONALD C. MEAD E29. This question is parallel to question E26 for cur- E40. The answer for pure trading enterprises will be rently operating enterprises. Measuring annual income 100 percent. The answer for pure manufacturing or of non-operating enterprises requires this question service enterprises will be 0 percent.The target of this and information about income flows. See E37-E44 for question is enterprises that mix trading with other a more detailed explanation. activities. E30-E33. Taken together, this series of questions E41-E42. Taken together with the responses to ques- should yield insights into the dynamics of economic tions E22 and E23, these questions will measure annu- activity and employment fluctuations in the enterprise. al in-kind sales revenue. Question E42 refers to pay- ments during a month with "average" sales (as defined E34. This question is asked of both operating and non- in question E36). For the sake of brevity, the question- operating enterprises. It establishes their rate of eco- naire does not repeat this question for months with nomic activity over the past 12-month period. This "high" or "low" sales.The analyst must assume that in- makes it possible to convert annual or monthly enter- kind sales revenue either varies proportionally with prise income into a daily rate, which is then compara- cash sales revenue or remains relatively constant. ble to the income that waged or salaried workers earn. E43-E44. Taken together with questions E24 and E25, E35. This question establishes seasonality and trends in these questions will measure the annual value of home employment opportunities in an enterprise. The grid consumption. Like question E42, E44 refers to a is set up with the 12 calendar months listed over two month with "average" sales (as defined in question years. If the interview takes place in May, the inter- E36). viewer should fill in entries for January to May of the current year and May to December of the previous Part F: Input Use and Expenditures year. An alternative system would be one row of 13 Fl. This question begins a section on generic expen- cells, with the last cell referring to the current month, ditures.Trading companies may not need to report on the next-to-last cell referring to the previous month, all items here, because their major expenses may and so on until the first cell refers to 12 months earli- already have been captured in questions EPO-Ell. er. However, references to specific months seem easier Question Fl establishes whether an item is used. The to interpret. list of items does not include rental, maintenance, It is important to note that the question asks for taxes, and fees. Rental and maintenance are expenses information for the full previous 12-month period. related to tangible business assets, and taxes and fees are These data enable analysts to compute year-to-year addressed separately in F26-F34. None of these four change in employment, indicating the employment expense categories fits the set of questions posed in trend, and month-to-month change in employment, F2-F25 very well. For the same reason, some questions measuring seasonality around the trend. are blocked out for the insurance expense category. E36-E39. The sales pattern revealed by question E36 F2. This question establishes the purchase of an item, will probably resemble the pattern of employment which is different from its use. over the 12 months. Most entrepreneurs are probably better informed about employment variations than F3. Expenditures during the previous month are easi- about monthly fluctuations in sales revenue. On the est for the entrepreneur to remember. The previous other hand, many enterprises employ only one or two month is used rather than the previous two weeks (as household members, with no monthly variation. In for recent sales revenue) because expenditures fluctu- the case of these enterprises, income fluctuation will ate both with and ahead of surges in sales revenue. be the most useful piece of information about season- Measuring expenditures by month may smooth these ality (other than the enterprise shutting down out-of- fluctuations somewhat. season). The level of detail implicit in questions E36-E39 is intended to focus the entrepreneur and to F4-F6. These questions attempt to measure expendi- enhance the accuracy of his or her responses. tures on an annual basis, linking them to the level of 130 CHAPTER 18 HOUSEHOLD ENTERPRISES sales as established in questions E36-E39.This link may cost items, and "other" items. Existing LSMS surveys reveal both "lean" and "fat" months in the year-round have never before included questions about these cat- activity of the enterprise. Also, by linking expenditures egories, so it is not clear how frequently these cate- to the agricultural calendar, these questions yield addi- gories are relevant. What is known is that in many tional insights into the workings of the local economy. existing surveys enterprises have reported making rel- atively few purchases of the inputs they might logical- F7-F9. If some inputs are bought on credit, an analyst of ly be expected to use given the industry in which the credit markets will want to know what proportion of the enterprise operates. inputs are bought on credit (F7), how the creditor is paid back (F8), and the terms of credit (F9). Concerning the F18-F19. These questions establish the value of items terms of credit, one would really like to know the that were not purchased or acquired in exchange for (implicit) interest rate, but finding this out would require enterprise products. Even if this value is positive, the several more questions: were you charged interest; what direct cost to the entrepreneur was zero-although was the interest rate; if you had purchased these goods there may have been indirect costs in terms of labor with cash, would you have been able to purchase them and goodwill or direct costs to produce goods in at a lower price; if so, how much lower? Without such another enterprise of the household. The fact that questions, the analyst has to impute the customary these items cost nothing to this enterprise may regional interest rates that have been uncovered in the increase the profitability of the enterprise, but it credit module of the questionnaire (see Chapter 21 on remains to be seen whether this enterprise improves its credit). In this questionnaire questions F7-F9 have been efficiency by using them. (This is an example of the blocked out for electricity, water, insurance and other distinction between private profitability and social inputs, all of which are unlikely to have been purchased returns.) with credit. Local conditions may differ, of course, in which case modifications should be made. F20-F21. The household may use some of the items acquired by the enterprise. These questions aim for a F10-Fll. The entrepreneur may have acquired some full accounting of both the profitability of the enter- items in exchange for some of the enterprise's output. prise and the consumption of the household. The entrepreneur may not think of this action as pur- chasing the item, but it is an expense nonetheless. F22-F24. These questions ask about inter-enterprise flows of inputs, so that each enterprise's profitability F12-F13. The entrepreneur may obtain some items can be accurately measured. from another enterprise in the household. If so, the other enterprise should indicate this in its responses F25. This question verifies the responses of the entre- to F22-F23. Most households operate only one preneur regarding shared resources. enterprise, but in households where several enterpris- es are in operation, each enterprise's expenditures F26. This question begins a section on the registration must be taken into account in order to accurately of the enterprise with government authorities measure performance. (F26-F32). This may be a sensitive question for the entrepreneur, so the question is phrased to appear as if F14. The household may provide the enterprise with the interviewer is only interested in the expenses relat- some of its inputs. While in this questionnaire the ed to registration of the enterprise rather than in the value of household-provided inputs is not ascertained registration itself. separately from the value of "free" inputs in other cat- egories (F12 and F15-F17), it might be appropriate to F27. The researcher may want to cross-check the reg- insert a question in order to account fully for intra- istration expenses reported here with common regis- household flows. tration fees reported by governmental agencies. The entrepreneur's figures may be higher if he or she F15-F17. These questions refer to three remaining includes any necessary bribes in responding to this categories-"gifts" from outside the household, zero- question. 131 WIM P M.VIJVERBERG AND DONALD C. MEAD F28-F32. These questions are posed to entrepreneurs questions are motivated by the concern that in many who have not registered their enterprise. The ques- countries women are less likely to formally own the tions aim to find out how much entrepreneurs know assets of their enterprises. This may have implications about registration requirements and costs and whether when women entrepreneurs apply for credit as well as registering an enterprise has disadvantages from the implications for intrahousehold allocation patterns. household's point of view. In some cases it might make For more information see Chapter 21 on credit and sense to modify the response codes to F32 to reflect Chapter 24 on intrahousehold analysis. practices in the country of the study. It might also be appropriate to add questions about the amount of G6. Current market value is a reasonable way to value time required for registration-for registered enter- a business asset that has been purchased at one time in prises, unregistered enterprises, or both. the past. F33-F34. It is recommended that the questionnaire list G7. Ownership of an asset does not necessarily mean several examples of taxes that enterprises may have to that other enterprises do not use the asset. pay. This will depend on the tax code in the country. G8-G13. Assets may be used by another enterprise in Part G: Business Assets the household; if so, the other household enterprises Gl. The following set of questions establishes the total should also report this in G24-G25. However, it is value of the business assets in use. These assets must be quite possible that some entrepreneurs fail to mention used as part of the production process. An enterprise that assets that they borrow from other household enter- trades business assets should list the assets that it owns for prises, especially if they own one of the other enter- the purpose of trading under questions E4-E13. For prises. Responses of the entrepreneur who owns the example, a car dealer may keep six cars, five of which are asset help establish the value to the other household for sale (and should be listed under E4-E13) and one enterprises of having access to this asset. that he drives around for his own business (and should be listed here). Even if he is willing to sell this particular G14-G15. The household may borrow an asset (such car as well, one car is still a business asset because he as a vehicle). This is an intrahousehold transfer; it always uses one of his cars for his own business. reduces the opportunity for the asset to be used for the enterprise and increases household consumption. G2. The first source of asset use is ownership at the time of the survey. Questions G32-G39 deal with pur- G18-G19. The purpose of these questions is to ensure chases and sales of assets. Current ownership is not that the entrepreneur has plausibly accounted for the equivalent to use during the past 12 months. Question use of his business assets. G2 acts like a filter; if the answer is "No," the inter- viewer should skip to question G20. G21-G23. One way an enterprise can use an asset with- out owning it is to rent it. In all previous LSMS surveys, G3. If ownership of an item is shared with another rental questions were listed under expenses. This ques- enterprise, the entrepreneur of that other enterprise is tionnaire places rental expenses among business assets. likely to report this asset as well. Using question G3 To conserve space, "rental" is also interpreted to mean together with question G5, a data analyst can ensure borrowing from a neighbor or relative for free (in that business assets within the household are not dou- which case question G22 would be answered with "0"). ble-counted even if several enterprises report them. The data could be made a little more user-friendly by G24-G25. A second way an enterprise can use an asset adding a question after G3: with which other enter- without owning it is to borrow it from another enter- prise is ownership shared? prise in the household.The other enterprise will report the value of this asset and for how long it is lent out. G4-G5. For the large business assets categories, these questions establish whether a partner's ownership has G26-G28. A third way to use an asset without owning legal weight and who legally owns the assets. These it is to borrow it from the household itself. Here, the 132 CHAPTER 18 HOUSEHOLD ENTERPRISES value of the asset must be ascertained-unless this has it. Local conditions may differ, of course, in which case already been reported in the modules on housing modifications should be made. (Chapter 12) or consumption (Chapter 5). G39. Asking why assets were sold gives researchers a G29-G30. Maintenance expenses constitute some of glimpse into the dynamics of the enterprise. the costs of using tangible business assets (apart from their implicit rental cost). In previous LSMS surveys, Part H: General Business Conditions maintenance questions have always been listed under H1-H5. These questions deal with raw materials. expenses. By relating maintenance to specific assets, this questionnaire should prompt more accurate H1-H2. Questions H1-H2 measure the enterprise's responses from entrepreneurs. inventory of raw materials. Question Hi is phrased in the present tense because, as a stock, inventories ought G32-G33. If an enterprise has acquired assets during to be measured at one point in time. However, if the the previous 12 months, this implies that the value of enterprise is not in operation at the time of the inter- its business assets at the beginning of the year was view, the question is not entirely appropriate. To ask lower than it is at the time of the survey and that the for inventories one year previous to the survey pro- enterprise's ownership (and use) of assets varied dur- duces the same problem (along with causing potential ing the course of the year. To relate business assets to recall problenis) since some enterprises were not in enterprise income, it is necessary to measure business operation one year previously. Another way to phrase assets during the year rather than at the end of the year the question-"In the last month that the enterprise as is done in question G2. The questions about asset was in operation, what was the inventory of raw mate- acquisitions and sales attempt to establish the value of rials?"-is equally useless, because the enterprise was assets for a more meaningful period of time. For an winding down. Asking the question in the present even fuller accounting of asset use, a question could be tense establishes a definite time and makes recall easy. added about the date of acquisition. However, since It might then be appropriate to assume that enterpris- the asset categories are fairly aggregative, several acqui- es not currently in operation typically maintained an sitions may have occurred, making a question about inventory of a size similar to those of their competi- date of acquisition ambiguous. The model question- tors in the industry. naire assumes that any acquisitions occurred in the middle of the previous 12-month period. H3. This question is as much as check on the response to H2, about inventory, as it is a check on reported G34-G36. If the acquired business assets have been levels of sales (E20, E37-E39). Aberrations in either bought on credit, an analyst of credit markets will inventory or sales should show up in the responses to want to know how much of the purchase was financed question H3 and its parallel questions, H8 and Hi11. with credit (G34), how the creditor was paid back (G35), and what the terms of credit were (G36). H6-H8. These questions are about inventories of items Concerning the terms of credit, one would really like requiring further processing (or intermediate inputs). to know the (implicit) interest rate, but finding this out Question H8 is a check on enterprise sales. The com- would require several more questions: were you ments on questions Hl-H3 also apply here. charged interest; what was the interest rate; if you had purchased these goods with cash, would you have H9-H13. These questions refer to products ready for been able to purchase them at a lower price; if so, how sale-either products purchased by traders (as question much lower? Without such questions, the analyst has H12 implies) or products produced by the enterprise. to impute the customary regional interest rates that Question HI 1 is a check on sales revenue. The com- have been uncovered in the credit module of the ments on questions H1-H3 also apply here. questionnaire (see Chapter 21 on credit). In this ques- tionnaire, questions G34-G36 have been blocked out H14. It is important to ask about more than one prob- for furniture, tools, and other durable goods, all of lem for enterprise startup, because the primary problem which are unlikely to have been purchased with cred- may be an internal household problem and thus not 133 WIM P M.VIJVERBERG AND DONALD C. MEAD amenable to policy solutions. B2 is an almost identical and how credit obligations are fulfilled (H29-H30). question for households that do not operate enterprises. H29 and H30 are parallel to similar questions about the entrepreneur's own use of credit (El 5-E16, H15. This question illustrates how lack of start-up F8-F9, G35-G36). capital can be a barrier to entering an industry. If the Questions H26-H30 are interesting, as credit and total sum of money needed is large, few households product markets may be linked. There is little survey- will be able to afford starting a new business. The based information about the involvement of small- response to this question may need to be adjusted for scale entrepreneurs in the credit market. inflation, according to the age of the enterprise (C3). H31-H33. This set of questions asks about fluictuations H16. The source of money is always an issue for both of daily sales revenue.These questions are placed here, households and policymakers. Stimulating entrepre- away from section E of the module, in order to check neurship requires understanding how an individual the revenue responses in section E. finanices the start-up of his or her enterprise. Thus it is important to gather full information on all sources H34-H35. As xvas evident in Table 18.1, many enter- of finance. Hopefully, three sources will cover most prises seem to be losing money. This question exam- situations. ines whether the entrepreneur is aware of this or whether, instead, his or her responses to questions on H17. Does the entrepreneur receive any assistance sales and expenditures are possibly erroneous. from government or business organizations? It is diffi- cult to be more specific in these questions and yet H36. Here the entrepreneur is asked to report his maintain enough generality; to increase specificity the monthly expenses as a check on his or her responses questionnaire could be modified to incorporate local in Part F (which were used to derive monthly expens- assistance programs. The effectiveness of such pro- es). Responses here should include rental and mainte- grams could then be evaluated by relating the respons- nance expenditures and possibly purchases of business es here to the profitability of the enterprise. assets-all of which are covered in Part G. H18-H22. The list of macroeconomic variables to H37-H40. These questions are an independent check which the entrepreneur is exposed can be expanded to on enterprise income. Questions H37-H38 are asked fit the conditions of the country of the survey. in acknowledgement of the fact that the budgets of Openness of domestic markets and access to imports the enterprise and the household sometimes blend are a starting point. Depending on the industries in into each other. Questions H39-H40 let the entrepre- which they operate, entrepreneurs may or may not neur make an estimate of the (cash) profitability of his have to deal with markets or imports. or her enterprise. To derive enterprise income, it is still necessary to add consumption by the household of the H23-H25. These questions deal with the financial products of the enterprise (E25, E44). security of the enterprise. When the debt level is high and the weekly or monthly payments are large in corn- H41. This question gives an insight into how the parison to the cash flow, the enterprise is in trouble. entrepreneur uses his enterprise's income.While some These questions allow investigation of the relation operate enterprises to furnish their household with an between financial security on the one hand and seasonal income, others may operate enterprises with the goal trends, regional econotnic characteristics and macroeco- of making investments in schooling, land, or a new nornic conditions on the other. If similar information can business. Reinvesting in the present business is anoth- be extracted from the household credit module, these er form of saving. questions might be omitted (see Chapter 21 on credit). AdditionalAnnotations to the Standard Questionnaire H26-H30. The enterprise itself may also extend cred- D16-D26. Here the structure of the questionnaire dif- it. This set of questions measures how much credit is fers somewhat from that of the expanded version. extended to the enterprise's customers (H27-H28) Questions D16-D21 are to be answered by entrepre- 134 CHAPTER 18 HOUSEHOLD ENTERPRISES neurs who currently employ nonhousehold labor, and for resale, transport, electricity, water, fuel, rental, main- questions D22-D26 are to be answered by entrepre- tenance, taxes, registration fees, and insurance)?" neurs whose enterprise is currently not in operation With these changes,Table 18.5 would also change. or who have employed nonhousehold labor in the past Revenue in Part E of the table would total 4.70 ques- 12 months but not in the past 14 days. tions per household and 9.04 questions per enterprise, and Seasonal Revenue in part E would total 0.00 ques- GI. Even though the standard questionnaire does not tions per household and 0.00 questions per enterprise. attempt to distinguish use of tangible business assets from The overall number of questions would total 24.03 ownership of these assets, Gl is the best introductory questions per household and 46.22 questions per question to this section. It allows the interviewer to ask enterprise. But the tradeoff for this time saving would about all assets in use by the enterprise before launching be a loss of precision in the estimate of annual enter- into more detailed questions. Moreover, "use" covers prise income and a loss of information on seasonality. both owning and renting, details of which will be asked for specifically in questions G2-G4 and G5-G7. It also F2. Grouping all expenditure categories into a single ag- indicates whether the entrepreneur had access to assets gregate undoubtedly reduces the accuracy of the report- even if he or she ncither owned nor rented them. ed statistic. The only alternative is to collect the detailed kind of information that the standard version gathers- Additional Annotations to the Short Questionnaire although the standard version records inputs purchased D4, D9, D14. These questions record the number of rather than inputs used (as in the expanded version). family members working in an enterprise. While the standard and expanded versions of the household G1-G2. These questions aim to estimate the value of enterprise module also record the number of hours the enterprise as measured by its business assets. To worked by family members, the short version needs a shorten the questionnaire even further, these asset cat- link with the employment module for time allocated egories could be grouped into a single aggregate, but to work in nonagricultural self-employment activities. this is not advisable because the asset value of the This is a difficult link to make, as the C6te d'Ivoire enterprise represents household wealth, which ought (1985-1988), Ghana (1988-1989), andVietnam (1992) to be measured carefully in any LSMS questionnaire. surveys have shown. If the needed time allocation information is not collected in the employment mod- G3-G6. Deleting these questions would save 1.22 ule, these questions may as well be omitted. questions per household and 2.35 questions per typi- cal enterprise. The tradeoff for this time savings would E4-E7, E14-E17. These questions are retained in the be that less would be known about the dynamics of short version because nonmonetary transactions and the enterprise and the household. intrahousehold consumption constitute a significant part of the sales revenue of a small number of enterprises. Notes E10-E13. These questions make up 15 of the 60.22 The authors appreciate comments by participants of the LSMS questions asked, on average, of an enterprise. The workshops, in particular Paul Glevwe, Margaret Grosh, and Julie answers to these questions make it possible to com- Schaffner. pute an estimate of annual sales revenue and provide 1. The model questionnaire presented in this chapter does not information about the seasonal nature of the enter- provide sufficient detail to answer this question; the question can- prise. If it were necessary to shorten the questionnaire not be answered unless the list of products bought and sold is fur- even further, these questions could be replaced with ther disaggregated. the following: "During the past 12 months that your 2. Referred to in this chapter simply as a "household enter- business was in operation, how much money did it prise." This definition represents a break from tradition; in earlier receive from sales of its products, goods, or services?" LSMS questionnaires, a food-processing household business that Question F3 could be rephrased as follows: "During did not purchase any raw agricultural ingredients was not classified the past 12 months, how much have you spent in total as a household enterprise but rather as an extension of the house- on the purchase of inputs (labor, raw materials, items hold's farming activity. 135 WIM P M.VIJVERBERG AND DONALD C. MEAD 3. One question is what the smallest scale of operation is that that have been tried. The Vietnam and Pakistan questionnaires should be sampled. Should a household that butchered one animal, added the possibility of barter to question 1. Several questionnaires processed its skin and sold it for money be sampled? Should the added other family businesses as candidates for sharing the pur- survey include someone who once received some money for fix- chased inputs. The Ecuador questionnaire asked for the value of the ing a car or for transporting some commodities in his truck to a last purchase rather than the usual purchase in question 2. While neighboring toNvn? A reasonable criterion for inclusion in the sam- most questionnaires included expenditures on wages in their list of ple is xvhether the activity was purposefully intended to earn an input categories, the Ecuador, Perl (1990), and Peru (1991) ques- income rather than an incidental event of daily life. tionnaires included a question about wage payments to be asked 4. An enterprise survey can also be carried out on a door-to- immediately after the entrepreneur indicated that he paid for out- door basis, with interviewers asking at each household or place of side help.The obvious advantage of this strategy is that it eliminates business whether there is a nonfarm enterprise in operation at that the possibility of entrepreneurs reporting that they pay their svork- location, and, if so, adm-inisterinig the eniterprise questionniiaire. If the ers but failinig to report ansy wage paymiienits or reporting wage pay- sample is built on the basis of household enumeration hsts or a ments but reporting that they do not pay their workers. The complete enumeration of all activities in a random sample of local- Ecuador questionnaire listed expenditures on raw materials and ities, this type of enterprise sample will be the same as the LSMS articles for resale right after the series of revenue and home con- survey If the sample is drawn up by visiting places of business, it sumption questions instead of treating this as a regular business will include larger enterprises that are not captured by household- expense. This seems illogical and has the potential to confuse both based surveys. Using this approach carries the risk that enterprises the interviewer and respondent. The Pakistan survey distinguished with a variable location (such as taxis, fishing enterprises, and some inputs purchased weekly or more often from those purchased less vending activities) may be undersampled. frequently. It is not clear that anything substantial is gained with this 5. Estimated from LSMS surveys in Cote d'lvoire (1985-1988), information. The Pakistan questionnaire added two other ques- Ghana (1 987-1989), and Vietnam (1 992-1993). tions-one asking whether the item was purchased by cash or cred- 6. In some cases the enterprise's organizational structure may be it (and if by credit, xvhether the supplier or someone else extended such that one household owns it, perhaps by virtue of providing the credit), and the other asking wyhether the entrepreneur had ever financial start-up capital, while another operates it. The first house- encountered shortages of the item. These are interesting questions, hold would not describe itself as operating an enterprise, wvhile the although they did not yield much variation of answvers. Except for second may see itself as working for the first household rather than purchases of rawv materials and items for resale, virtually all transac- as operating the enterprise.This could lead to a nonrandom sample tions were in cash. The model questionnaire in this chapter draxvs of enterprises even though the sample of households was random- from these experiences. ly selected. This is presumed to be rare. 11. Virtually all LSMS questionnaires contain: "Q4: Does your 7. Not all tIoxvs are of living standards concepts. Some are of household use this ... ?" By itself, this question contains little infor- changes in stocks; for example, saving is a change in the asset posi- mation; the value of what is shared remains unknown. The tion of the household. researcher only knows for sure that the enterprise uses less of the 8. A few surveys have used a different set of questions.The early input than question 3 reports. Moreover, this question 4 refers only questionnaires for Cote d'lvoire, Peru (1986), and Ghana skipped to inputs purchased and shared with others.There is no mention of the 12-month revenue question if the entrepreneur gave his recent the possibility that the enterprise receives inputs (such as electrici- revenue. This meant that it was impossible to compare the revenue ty, wvater, or the use of tools) from the household or from other responses wvith one another. The Peru (1990 for Lima only and household enterprises. 1991) and Ecuador questionnaires asked only about revenue 12. Detailed tabulations show that in every data set there are received during the last month of operation, xvhich of course syas some enterprises that have expenditures in the losvest quintile and the current month if the enterprise was in operation at the time of revenues in the highest quintile, thus apparently generating huge the second visit. positive profits, and others for which the reverse is true. 9. The Pakistan questionnaire also separated out receipts from 13. The first question was in fact used in the GlianaVietnarri, subcontracting. Fexv entrepreneurs reported any such receipts, but and later Cote d'lvoire questionnaires; the second question for those xvho had some receipts, subcontracting was an important occurred in the Ghana questionnaire. Interviewers vyere not per- part of business operations. Note, though, that receipts from sub- mitted to accept negative answers to these questions. but in fact contracting should already be covered by cash sales receipts. only a few entrepreneurs answered "0." 10. These questions have been used in virtually all LSMS ques- 14. It is possible to put a positive rxvist on these figures. About tionnaires. This footnote documents additions and modifications 80 percent of the enterprises xvere either along the diagonal or in 136 CHAPTER 18 HOUSEHOLD ENTERPRISES the niext adjacenit quintile. This is riot to mininrize the fact that, in enterprises, but interviewers were instructed to fill out an extra terms of absolute values, the agreement between the two measures questionnaire form if a given household operated more than two is not good. enterprises, and to inform their supervisors that they had done so. 15. In a few questionnaires the interviewer was instructed to check the responses about the economic activities of household References members, as reported in the employment module, to determine whether the household appears to be operating a nonagricultural Liedhohm, Carl, and Donald C. Mead. 1995. "The Dynamic Role of enterprise, even if household members report that they do not. Micro and Small Enterprises in the Development Process." Presumably the interviewer will probe more vigorously if he sus- Development Alternatives, Inc., Growth and Equity through pects that an enterprise exists. Microenterprise Investments and Institutions (GEMINI) Project, 16. A remaining problem with using this approach is the time Action Research Program 1, Final Report, Bethesda, Md. gap between the first visit, xvhcn these questions are asked, and Mead, Donald C. 1994. "The Contribution of Small Enterprises to the second visit, when the rest of the enterprise information is Employment Growth in Southern and Eastern Africa." World collected. Developsnent 22 (12): 1881-94. 17. This strategy would have solved all matching problems were . 1995. "How the Legal, Regulatory, and Tax Framework it not for two other shortcomings: these questions were asked only Affects the Dynamics of Enterprise Growth." In P. English and if the enterprise employed more than one worker including the G. Henault, eds., Agents of Change: Studies on the Policy entrepreneur, and IDs and hours were ascertained for no more than Environmentfor Small Enterprises in Africa. Ottawa: International four family members. Because of the first shortcoming, the hours Develop Research Centre. that household members worked were still unknown for two-thirds Moock, Peter, Philip Musgrove, and Morton Stelcner. 1990. of the enterprises, and the second shortcoming meant that infor- Education and Earnings in Perun Informal Nonfarm Fansily mation was missing for enterprises employing five or more family Enterprises. Living Standards Measurement Study Working members. The Peru (Lima 1990; 1991) qtuestionnaires asked for the Paper 64.Washington D.C.:World Bank. total number of hours worked in the enterprise by all workers, but, Otero, Maria, and Elisabeth Rhyne, eds. 1994. The New Role of like the Ecuador questionnaire, they did so only if other members Microenterprise Finance: Building Healthy Financial Institutionsfor besides the entrepreneur worked in the enterprise. This shows how the Poor. West Hartford, Conn.: Kuoisariaii Press. carefully questionnaires must be designed in order to yield user- Schultz.Theodore W 1975. "TheValue of the Abihty to Deal with friendly data. Disequilibria."Journal of Econoniic Literature 13 (3): 827-46. 18. This is preferable to asking about the ownership of assets at Vijverberg,Wim P.M. 1992. .Mleasuring Inconiefromn Family Enterprises any time during the previous 12 months; ownership at any time uitli Household Surveys. Living Standards Measurement Study during the past year does not imply usage during the entire year, as Working Paper 84.Washington D.C.:World Bank. some assets are bought or sold during the year. . 1998. "Nonfarm Household Enterprises in Vietnam." In 19. In Pakistan, 13 percent of all assets are shared with the David Dollar, Paul Glewwe, and Jennie Litvack, eds., Household household or another enterprise; in Ecuador this figure is around Welfare and Vietnam's Transition. Washington, D.C.:World Bank. 40 percent. Young, Robert C. 1993. "Policy Biases, Small Enterprises, and 20. A similar strategy was followed by LSMS teams in Ecuador. Development." Small Enterprise Development: An International The Ecuador (1994) questionnaire allowed for responses about tvo Jouirnal 4 (1) 137 , ^ ~Agriculture J 9 Thomas Reardon and Paul Glewwe Promoting sustainable growth in agriculture can reduce rural poverty and increase employment and welfare in both rural and urban areas in developing countries, for five reasons. First, agricul- ture is a major source of household income and employment in most developing countries, both directly, through own-farm production and agricultural wage labor, and indirectly, through activi- ties that use farms' outputs or that provide products and services to farmers. Second, the poor benefit disproportionately from the welfare and employment gains brought about by agricultural growth because the majority of poor people in the developing world live in rural areas (World Bank 1995).Third, the types of crops and livestock produced by farmers and the farming tech- niques that they use affect the health, nutrition, and environmental conditions of families living both in rural areas and in nearby cities, through their ecological impact on farmland and on forests, wetlands, and rivers. For example, farmers who are unable to increase the productivity of their existing land may extend their cultivation into forests and other ecologically sensitive areas in order to maintain or increase their incomes. Fourth, the agricultural sector affects welfare and employment in urban areas because of its influence on food prices, wages, the input costs of the food and fiber processing industries, and the balance of payments. Finally, if serious problems develop in the agricultural sector (such as a drought that induces a crop failure), many rural dwellers will migrate to urban areas to seek work, which leads to overcrowding and increased unemployment in urban areas. Agriculture is usually defined in the national accounts In particular, the agriculture module presented in this of developing countries as the set of activities involved chapter includes only the activities of the farm that in the production of annual and perennial crops involve crop (annuals and perennials) and livestock (including trees for timber) and the production of live- production. It omits hunting, fishing, and gathering stock.This set of activities can be broadened to include activities as well as the processing of agricultural prod- hunting, fishing, and the gathering of wild flora and ucts.Those activities can be treated as nonfarm enter- fauna. In this chapter, the narrower definition of agri- prise activities and, therefore, should be included in culture is used for the purposes of making recommen- the household enterprise module of an LSMS or sim- dations on the design of the agriculture module in ilar multitopic survey (see Chapter 18 for further Living Standards Measurement Study (LSMS) surveys. details). 139 THOMAS REARDON AND PAUL GLEWWE In past LSMS surveys the agriculture module has Agricultural Policy Issues in Developing often been the longest module in terms of both pages Countries and interview time. The agricultural module has gen- erally had three objectives: measuring net income In many developing countries the agriculture sector is from the household's production of crops and live- changing rapidly, and in multiple ways.Yet policymak- stock; measuring the value of household agricultural ers and analysts often know very little about these assets such as land, animals, and equipment; and meas- changes because up-to-date, reliable data on agricul- uring the household's use of agricultural services such ture are scarce. This lack of information can lead poli- as extension programs, cooperatives, and veterinary cymakers to adopt inefficient or inequitable policies services (Ainsworth and van der Gaag 1988). Despite and may also cause them to miss some opportunities the length of these agricultural modules, the data col- to implement policies that can raise household wel- lected in them have been analyzed less often than the fare. data from almost all the other modules of the surveys Well-designed multitopic household surveys can containing them. Fewer than 10 percent of all of the help policymakers by providing them with accurate publications that analyze LSMS survey data have used data on agriculture and related activities, on nonagri- data from the agriculture module either directly or cultural activities, and on the characteristics of both indirectly Jolliffe 1995). One explanation for this is agricultural and nonagricultural households (and the that in almost all developing countries, researchers communities in which they live).These data can then have found it difficult to find out about the survey, to be analyzed using either "descriptive analysis" (statisti- get access to the data, and to find the time and funds cal analysis of correlation among variables, usually to study the data. Addressing these problems is beyond reported in graphs and tables) or "causal analysis" the scope of this book (see Blank and Grosh 1999). (econometric analysis to measure causal relationships), Another possible reason for the underuse of data from as explained further in the second section. previous agricultural modules is that the data collect- This section presents the most important policy ed were not useful to researchers. If this were indeed issues in the agricultural sector in developing coun- the case, it suggests that the agricultural module tries. The first subsection describes current patterns should be revised to ensure that it collect data that are and trends in agriculture in developing countries.The much more useful for policy analysis. The purpose of next subsection briefly reviews the agricultural out- this chapter is to advise survey designers on how to comes that are of greatest interest to policymakers.The accomplish such a revision. remaining subsections each discuss one of the four dif- The first section of this chapter lays out the most ferent kinds of agricultural policies, focusing on issues important agricultural policy issues in developing that are highest on the policy agenda. countries, including some links between agricultural issues and other topics on which data are usually col- Current Patterns and Trends in Agriculture in Developing lected in LSMS and other multitopic household sur- Countries veys. The second section discusses the data needed to Four broad trends have profoundly influenced agricul- analyze these issues, as well as measurement concerns ture in developing countries in recent decades and are related to those data. The third section introduces likely to prompt still more changes in the years ahead. three versions of a draft agriculture module. (The First, land in the developing world has become modules themselves are presented inVolume 3.) Each increasingly scarce as populations have grown. As a version is designed to gather information at a different result, agriculture in most developing countries has level of detail; the choice for a given survey will changed from being "extensive" (increasing produc- depend on the degree of emphasis that survey's tion by bringing more land under cultivation) to being designers wish to place on agriculture. All of these ver- "intensive" (increasing production by raising the pro- sions must be adapted to reflect the agricultural con- ductivity of the land already under cultivation).This is ditions and policy issues in the country where the sur- true even in Africa where, until recently, land scarcity vey is to be implemented. The final section of this was usually not an issue (Binswanger and Pingali chapter consists of annotated notes to specific ques- 1988). Policymakers need to know more about this tions or submodules of the draft module. inevitable move toward intensive cultivation, including 140 CHAPTER 19 AGRICULTURE whether it is leading to land degradation, water pollu- dustrial firms contract with farmers to supply crops) tion, or other environmental problems-all of which can affect the agriculture sector. can undermine long-term growth. They also need to In addition to these four trends, a key characteris- know whether the poor have access to modern tic of agriculture is that it has always been subject to inputs-such as chemical fertilizer-that enable them recurring "exogenous shocks" that affect farms and to raise the productivity of their current landholdings. farm families. Examples of such shocks are new crop Second, the agriculture sector is rapidly becoming diseases, periodic pest infestations, floods, droughts, commercialized in many countries and thus is increas- epidemics (of which AIDS is a recent example), and ingly linked to the urban and export sectors. At the civil war. Since agriculture is inherently a risky activi- same time, farmers in many countries are partially ty, it is essential for policymakers to understand how diversifying their activities by producing fewer staple farm households deal with these risks. crops and investing in dairy farming, livestock farm- ing, and the production of fruits and vegetables. Agricultural Outcomes that Interest Policymakers Policymakers need to know which farming house- As a frame of reference for the rest of this chapter, it is holds are making these changes and which households useful to identify the agricultural phenomena that are face barriers that prevent them from doing so. They of greatest interest to policymakers. In this paper they would also like to know which diversification activi- will be referred to as the "basic agricultural out- ties are most productive. comes." These outcomes are: Third, environmental degradation has become a * Production of crops, livestock, and related byproducts. major problem in most developing countries. * Use of inputs in the production processes, includ- Degradation problems include land degradation in the ing physical inputs, labor, and capital. form of soil erosion, the reduction of soil nutrient lev- * Technologies adopted and technology packages els, loss of tree and bush cover, and the salinization of used by agricultural households. soils from intensive irrigation. Another problem is the * Marketing activities that agricultural households runoff of farm chemicals into water sources, which undertake to sell their products. occurs mainly in areas where a "green revolution" has * Profits (net incomes) earned by households from taken place (in other words, where high-yielding their agricultural activities. seeds, chemical fertilizer, irrigation, and sometimes * Investments that households make in agriculture, pesticides are being used intensively). In some coun- such as the purchase of equipment and the tries a loss of biodiversity has become a problem, while improvement of land. in other areas (such as the West African Sahel) inade- * Nonincome welfare indicators of agricultural quate use of chemical fertilizers and manure have households, such as child nutrition, school enroll- exacerbated problems of soil degradation as farmers ment (of children), and household amenities. cultivate their farmiland more intensively in order to * Environmental phenomena that are affected by produce enough to survive. agricultural activities. Fourth, technological developments in recent Policymakers are interested in these basic agricul- years have greatly affected agriculture in many devel- tural outcomes for many reasons.The total production oping countries, and even more rapid change is on the of food and nonfood agricultural products has a large horizon.To the extent that new agricultural technolo- impact on the national economy and on the welfare of gies-"green revolution" technology, biotechnology the population as a whole, particularly in developing advances such as genetic modification of plants to countries where many households derive a large pro- enhance disease resistance, integrated pest manage- portion of their income from agriculture. The use of ment, and new types of agricultural equipment-are inputs is also of crucial importance, since the extent to adopted by farmers, they affect agricultural productiv- which a household uses inputs efficiently will affect its ity. It is important for policymakers to understand net income. Also, each input has an opportunity cost; what determines whether farmers adopt these new if it had not been used in agriculture, it could have technologies. Policymakers also need to know how been used elsewhere in the economy.The relationship the emergence of new institutions such as farmers' between total production and inputs depends greatly organizations and contract farming (in which agroin- on the technology used, which in turn depends on the 141 THOMAS REARDON AND PAUL GLEWWE availability of different technologies and their associat- work will be necessary to provide information useful ed inputs; in most cases policymakers would like to to policymakers.The remaining subsections review the encourage the adoption of new technologies so that current state of knowledge about how these four kinds inputs are used more efficiently. of policies affect agricultural outcomes, and emphasize Marketing activities are important for ensuring that areas where further empirical research is needed. food and nonfood products are brought to urban areas. They also affect agricultural exports, which can have a Macroeconomic Policies major impact on a country's balance of payments. Macroeconomic policies are very broad economic Agricultural profits contribute to households' incomes policies that are always implemented at the national and to their welfare as measured by nonincome indica- level and affect not only agriculture but also many tors (such as health and education outcomes), and cur- other sectors of the economy. The four macroeco- rent investments in agriculture are clearly important for nomic policies that can have strong effects on agricul- future production. Finally, the environmental conse- ture are exchange rate policies, trade policies, banking quences of agricultural activities are important policy and credit policies, and the overall size of the govern- issues in many developing countries. ment budget. Government policies affect these basic agricultur- al outcomes (as well as other more specific ones) by EXCHANGE RATE PoLIcIEs. Exchange rate policies influ- influencing their determinants. The most important ence, and in some countries completely determine, the determinants of agricultural outcomes are: value of foreign currencies in terms of domestic cur- * The prices that farmers face for both inputs and rency, directly affecting the domestic prices of all agri- products (which can be affected by taxes, subsidies, cultural inputs and products that are imported or and exchange rate policies). exported. To reduce trade deficits, governments often * Past investments that agricultural households have devalue their currency when it is thought to be over- made in their stocks of productive capital, including valued, since overvalued currencies discourage exports human capital. while stimulating imports. Such currency devaluations * The technology available to farming households. can increase the prices that farmers receive (in terms of * Farming households' access to credit. domestic currency) for their export crops and will also * Information and extension services available to tend to raise the prices of any crops produced by farm- farmers. ers that are also imported. Currency devaluations can * Large-scale investments in infrastructure, such as also have a negative effect on farm incomes if farmers transportation and irrigation networks. depend on imported inputs, since a devaluation will * The institutional environment in which households usually increase the domestic price of those inputs. operate (such as the system of land tenure, insur- These general effects of changes in the exchange rate on ance opportunities, and laws). the prices faced by farmers can be counteracted by sec- * The risks and uncertainty faced by farming house- toral policies. For example, the government can set the holds such as price variations, weather variability, domestic prices of specific agricultural inputs and prod- crop diseases, and harmful pests. ucts and not adjust these prices after a devaluation. Such * Any direct taxes, such as income taxes, for which sectoral policies are discussed further in the next sub- farm families are liable. section. A final issue regarding exchange rate policies is Four different kinds of policies-macroeconomic the extent to which exchange rates are allowed to fluc- policies, sectoral policies, policies that affect the insti- tuate over short periods of time. The government can tutional environment, and public investment pro- smooth out these fluctuations-for example, by setting grams-affect agricultural (and even nonagricultural) a fixed exchange rate in terms of U.S. dollars and main- households by altering one or more of these determi- tain the same rate for several years. This kind of policy nants of agricultural outcomes. However, the way in may reduce the risks borne by farmers through reduc- which these policies affect agricultural outcomes is ing fluctuations in the prices that they face. often ambiguous, and, even when it is clear, econom- ic theory reveals little about the magnitudes of the TRADE PoLIciEs. Trade policies usually take the form effects of specific policies. Therefore, some empirical of tariffs and various nontariff trade barriers. Examples 142 CHAPTER 19 AGRICULTURE of nontariff barriers on agricultural goods are grades government spending can mean reduced price sup- and standards related to food quality, food safety, and ports for outputs and lower price subsidies for inputs, the environmental effects of food production. as well as reductions in other kinds of government Reducing export taxes, nontariff barriers that discour- programs that benefit agricultural households, such as age exports, and tariffs imposed by importing coun- extension services and agricultural research stations. tries should make the production of export crops Similarly, policies that increase taxes can reduce farm- more profitable. In contrast, reducing import tariffs ers' income either directly (for example, through an and nontariff barriers that discourage imports have a income tax) or indirectly (through taxes that affect more ambiguous effect on agricultural households; prices). As with nationwide banking and credit poli- such reductions tend to increase competition from cies, the specific ways in which spending is cut or taxes imported agricultural products but can also reduce the are increased determines the actual effect on agricul- prices of imported agricultural inputs. Nontariffbarri- tural households. This again leads to sectoral policies. ers can also limit the technology available to farmers. For example, phytosanitary regulations-rules on SUMMARY. The discussion of macroeconomic policies importing plant matter-can limit the access of farm can be summarized in terms of the following specific households to imported seeds, and chemical standards policy questions: can limit these households' access to imports of certain * How do exchange rate policies affect the prices of chemicals, such as DDT. The standards that apply to agricultural inputs and products and the variability the goods exported by developing country producers of those prices? can also influence the use of inputs, many of which are e How do tariffs and nontariff trade barriers affect associated with technologies commonly used by farm- the availability of agricultural inputs and the prices ers in developing countries. For example, if Mexican farming households face for inputs and outputs? producers want to export to the U.S. market, they can- * How do national banking and credit policies affect not use as much of certain kinds of pesticide as they agricultural households' access to credit and the can use on goods that they produce for the Mexican terms on which credit is available? market. Finally, nontariff barriers in the household's * How do across-the-board reductions in spending own country, such as import quotas and domestic and increases in taxes affect direct taxes, prices for component regulations, can limit farmers' access to agricultural inputs and products, and the availabili- imported agricultural inputs. ty of programs that benefit farmers? There are no simple answers to these questions, BANKING AND CREDIT POLICIES. The availability of because agricultural systems vary widely across devel- credit and the terms on which credit is available can oping countries, and even within a given country the have a large impact on the activities of agricultural characteristics of farming households can vary enor- households. At the national level, policies that affect mously. For example, devaluations can increase the credit are set by finance ministries and central banks. income of producers of export crops, but may have lit- A nationwide tightening of credit can increase the tle effect on farming households that produce mainly interest rates that farmers face, while changes in bank- food crops for subsistence. Thus the answers to these ing regulations can make credit either more or less questions will vary from country to country and by accessible. Of course, the impact of national banking the different types of households within each country. and credit policies can be altered by sectoral policies In general, these effects of macro reforms on agri- such as special programs within the ministry of agri- culture have not been adequately studied. Much of the culture that provide banking services to rural areas. literature on the impact of macroeconomic policies focuses on the impact of exchange rate policies, par- SIZE OF THE GOVERNMENT BUDGET. In many develop- ticularly the effect of exchange rate devaluations. ing countries fiscal deficits are a serious problem. Another strand of the literature focuses on general Governments often seek to reduce these deficits both price liberalization-the removal of price controls and by reducing spending and by increasing taxes. In many other policies that influence prices such as subsidies cases the emphasis is on reducing public spending. and taxes on specific products. Evidence from rural From the perspective of farmers, broad reductions in areas in a variety ofAfrican and Latin American coun- 143 THOMAS REARDON AND PAUL GLEWWE tries suggests that a concentration of market power but this is mostly an issue of semantics; there is no and market entry barriers tends to produce greater need to rigidly classify each policy as one or the other. price instability when prices are liberalized, while There are many different kinds of policies in agri- devaluation has an ambiguous effect on farm prof- culture at the sectoral level.They can be grouped into itability (Reardon and others 1997). Overall, much policies that directly affect the prices of inputs and more research needs to be done on the impact of products (such as price subsidies, taxes, and price floors macroeconomic policies on rural households. and ceilings), programs that directly provide technolo- There is also a literature that focuses on the gy, information, and specialized services to farmers, macroeconomic policies that are typically included in and policies that affect the availability of credit to structural adjustment programs. The evidence in this farmers. literature tends to be quite mixed. For example, some authors have found that structural adjustment pro- PoLIcIES THAT AFFECT THE PRICE OF AGRICULTURAL grams have had positive effects (Sahn 1994), while INPUTS AND PRODUCTS. A wide variety of sectoral other authors have found that these policies have gen- policies can affect the prices for agricultural inputs and erally negative effects (Taylor 1993), and still others products. Taxes can raise (and price subsidies can have found a mixture of positive and negative effects lower) the prices of agricultural inputs. In some coun- (Commander 1989; Duncan and Howell 1992). The tries all marketing (and even production) of certain variety of results in the literature suggests that the inputs is controlled by marketing boards, which are effects of macroeconomic policies in a given country sometimes referred to as parastatal corporations. depend on the characteristics of that country, which in Agricultural products can also be taxed or subsidized, turn implies that LSMS and similar multitopic surveys and in some cases all marketing is controlled by a are useful sources of information for illuminating the national agency. Some governments decree price potential effects of these policies in specific countries. floors or ceilings, although it is not always possible to enforce such regulations.These policies can affect not Sectoral Policies only prices but also fluctuations in prices. For exam- Sectoral policies differ from macroeconomic policies ple, the government may be ready to purchase specif- in that they focus directly on a given sector, such as ic agricultural products at guaranteed minimum agriculture, and are often designed and implemented prices, which will reduce the price fluctuations faced by the appropriate ministry-in this case the ministry by farmers. Some policies that affect the prices of of agriculture. Sectoral policies for agriculture include inputs and outputs depend on aid received by the gov- taxes and price subsidies for specific agricultural inputs ernment. For example, the government may sell food and products, marketing boards that purchase agricul- or fertilizer received as aid in order to reduce food or tural outputs and sell agricultural inputs (in some cases fertilizer prices in some areas. Of course, macroeco- monopolizing these markets), regulations that govern nomic exchange rate and trade policies can also affect prices of both agricultural inputs and outputs, agricul- prices, so the overall effect will be determined, rough- tural extension services, programs that provide credit ly speaking, by the sum of the effects of macroeco- for farming households or promote new agricultural nomic and sectoral policies. technologies, and public investments in agricultural infrastructure and research. AGRICULTURAL EXTENSION SERVICES. Ministries of Sectoral policies in agriculture can be closely agriculture often provide a variety of agricultural related to macroeconomic policies. Price subsidies for extension services to farmers, such as basic agronomic certain agricultural products or inputs may be information, information on new types of technology, designed to counteract the impact of general tariffs or visits by extension agents to farms to investigate spe- an exchange rate devaluation. Alternatively, the gov- cific problems, advice on the use of pesticides and her- ernment's attempts to reduce macroeconomic budget bicides, and vaccinations and other services for farm deficits may reduce spending on price subsidies or on animals. Some ministries of agriculture are also the provision of agricultural services. At times it may involved in the production of new technology at agri- not be clear whether a given policy should be classi- cultural research stations.Agricultural extension agents fied as a macroeconomic policy or a sectoral policy, may periodically visit farming households whose 144 CHAPTER 19 AGRICULTURE members might otherwise never visit an extension * What impact do interventions that are intended to center. Many of these services are undoubtedly quite change the prices of agricultural products actually useful to farmers, but others may not be. Some of the have on prices, and what is the impact of any price advice provided may even have a negative impact on changes on basic agricultural outcomes? the welfare of agricultural households. Obviously, until * What impact do different agricultural extension policymakers understand the effects of these different services have on basic agricultural outcomes? services, they will not know which services to expand * What prices, if any, should be charged for agricul- and which ones to reduce or eliminate. tural extension services, and how are the benefits Another issue is how much to charge for agricul- and costs of those services distributed among farm tural extension services.While economic theory pro- households? vides clear reasons to subsidize some services, such as * How do credit policies and programs in the agri- the provision of information, other services may not culture sector affect the availability of credit to agri- need to be heavily subsidized and could even be made cultural households and the development of private more widely available if some element of cost recov- credit institutions? ery were introduced. A related issue is the distribution As with macroeconomic policies, there are no simple of the benefits and costs of agricultural extension serv- answers to these questions because of the enormous ices. Do they reach the poorest households? Who ulti- variation in both agricultural systems and sectoral mately pays the costs of providing these services? For policies in developing countries.The individual effects an interesting discussion of these issues and research of most sectoral policies (such as taxes, subsidies, and evidence to date see Purcell and Anderson (1996). price controls) on output or input prices are unam- biguous, and the effect of price changes on household POLICiES THAT AFFECT THE AVAILABILITY OF CREDIT. welfare and agricultural output is also well known. While macroeconomic policies clearly affect credit That is, increases in output prices raise output and markets, the ability of farmers to invest in working household welfare, while increases in input prices have capital is affected by a variety of policies implemented the opposite effect. However, the magnitude of these by ministries of agriculture at the sectoral level. In effects are usually not known, which implies that the many countries governments have directly offered size of the benefits are unknown, and the benefits may credit to agricultural households-with mixed success not be worth the costs. Moreover, when several poli- at best. For recent reviews of past experience see cies are implemented simultaneously-such as when a Besley (1994),Yaron (1994), and Zeller and Sharma structural adjustment program is implemented that (1998). In some countries access to credit has been removes many sectoral policies designed to influence increased by the development of rural bank programs prices-the overall impact on farm production and operated by nongovernmental organizations (NGOs) household welfare is uncertain. such as the Grameen Bank in Bangladesh, yet the suc- The literature on sectoral policies is large and can- cess of NGO programs has also been mixed (Morduch not be easily summarized. Nevertheless, it is clear that 1999; Rahman 1999). Economic theory shows that many questions remain unanswered. Some observers credit markets can suffer from a variety of market fail- argue that sectoral policies designed to alter market ures, including problems of moral hazard and adverse prices are inherently distortionary and inefficient, and selection in risky environments with incomplete thus should be removed (Schultz 1978).Yet the empir- information. Economists' understanding of credit mar- ical evidence has revealed several cases where the kets has increased substantially in recent years due to elimination of sector-level interventions did not lead both research and innovations in credit institutions. It to the expected outcomes. For example, in some is important for policymakers to know how credit countries the supply of fertilizer and seed from private policies and programs affect capital formation in agri- merchants increased much less than expected after the culture so they can design effective policies and create elimination of fertilizer and seed marketing boards, a policy environment that promotes the development which had depressed prices for these inputs (Rukuni of efficient credit institutions. 1996; Dembele and Savadogo 1996; Rusike and oth- This discussion of sectoral policies leads to the fol- ers 1997). Other studies have claimed that the govern- lowing specific policy questions: ment has an important role to play in developing mar- 145 THOMAS REARDON AND PAUL GLEWWE kets for agricultural outputs. For example, some econ- of law in rural areas. Examples of such policies include omists have argued that fertilizer markets in Africa are regulations specifying the acceptable range of rights plagued by a series of fundamental problems and idio- and responsibilities in contracts between agroindustri- syncrasies such as risk, seasonal demand, high transport al firms and farms, establishment and regulation of costs, underdeveloped financial markets, and cash- government-managed crop insurance and drought constrained farmers (Barrett and Carter 1999). Thus, insurance schemes, and the establishment of civil court while it is true that fertilizer subsidies and domestic systems for land disputes. fertilizer production schemes have suffered from fiscal The most important policy questions regarding unsustainability and problems of implementation in the institutional environment in developing countries Africa, it also appears that private markets in rural are: Africa may not operate in ways that some policy advi- * What impact do traditional forms of land tenure sors expected they would. It may be that governments have on basic agricultural outcomes, and what can need to invest in improving transportation infrastruc- government policies do to overcome inefficient ture before private markets can function well (Ahmed, outcomes or to change the system of tenure? Falcon, and Timmer 1989; Rusike and others 1997). If * How politically feasible is major land redistribution better agricultural data can be collected in LSMS and in countries where the distribution of land is high- similar multitopic household surveys, more light may ly unequal, and what impact will such redistribution be shed on this question. have on basic agricultural outcomes and on the dis- tribution of these outcomes across households? Policies That Affect the Institutional Environment * What policies, regulations, and enforcement mech- Institutional policies primarily concern changes in the anisms can governments implement to promote the "rules of the game," such as land tenure rules, con- rule of law, and how does the rule of law affect basic tracts, and so on. Systems of land tenure are particular- agricultural outcomes? ly important. In some countries, land rights in many * What programs can governments implement to rural areas are determined by traditional systems that provide insurance directly or to promote the provi- may discourage efficient use of the land. One example sion of insurance in the private sector, and how do of this is the designation of some land as community these insurance schemes affect basic agricultural grazing areas for livestock, which usually leads to over- outcomes? grazing of that land. As land constraints grow in many The focus of recent empirical research on the countries, there is a tendency to formalize land titling effects of institutional change on basic agricultural in response to increased competition for land. outcomes has been on changes in land institutions, In other countries the distribution of land is high- particularly land tenure policies and land redistribu- ly unequal, which may also encourage inefficient use of tion. In recent decades land redistribution from col- agricultural land. This is particularly the case in coun- lective farms to individual households has occurred in tries with "dual" agriculture sectors-where a small many socialist or formerly socialist countries, such as number of very large farms coexist with large numbers Eastern Europe, China, andVietnam. Land redistribu- of very small farms, as in Brazil, Central America, tion has been limited in other developing countries. Mexico, South Africa, and Zimbabwe. In these coun- Recent empirical evidence on the effects of land tries potential or actual land redistribution (land titling-providing more "formal," and thus more reform) has prompted heated political debate. secure, land tenure-is mixed. In some countries Policymakers need to knoxv how land reform programs researchers have found that more secure land owner- have affected or could affect the concentration of land- ship increases productive investments in land (see holdings, farmers' access to land, income distribution, Place and Hazell 1993 and Migot-Adholla, Hazell, and and the incidence ofpoverty.A particularly contentious Place 1990 for evidence from Rwanda and Ghana). issue in this context is the effect that land reform and But this was not the case for Kenya; Migot-Adholla, redistribution have had on farming productivity, capital Hazel], and Place (1990) found that the relationship to labor ratios, and the welfare of rural households. between tenure and land improvements was weak Government policies also have direct effects on because farmers already felt secure in their use rights contract enforcement and, more generally, on the rule under the traditional land use system. Overall, the 146 CHAPTER 19 AGRICULTURE impact of more formal titling appears to depend in additional person can use them at little or no cost to part on the kind of system that is being replaced and others and it is difficult to prevent people from using the kind of investment or farm practice examined. them. Others, such as education and health care, may (For example, long-term investments were more sen- have significant benefits in the form of externalities; sitive than short-term investments to land insecurity.) that is, they may provide benefits to members of soci- There is a fair amount of empirical evidence on ety beyond those that directly use the service. Some whether smaller farms are more productive, which is a large physical infrastructure projects, such as irrigation key issue concerning land redistribution policies. In and electric power grids, may have large economies of India, for example, Bardhan (1973) and Deolalikar scale, which is another reason for government involve- (1981) show that smaller farms have higher land pro- ment. A final argument in favor of government ductivity but lower labor productivity. They point to involvement is imperfect information; for example, the greater labor intensity of smallholder farms as the residents in remote rural areas may not be aware of the reason. Empirical studies tests in Africa ( Carter and benefits of education or modern medical treatments. Wiebe 1990 on Kenya; van Zyl, Binswanger, and Investments in transportation such as roads, rail- Thirtle 1995 on South Africa) also find an inverse rela- roads, water transportation, ports, and air transporta- tionship between farm size and land productivity. tion can have a dramatic impact on markets-and Another example is Barrett (1996), who shows an particularly on market prices-by linking local mar- inverse relationship for rice farmers in Madagascar. On kets more closely with regional, national, and interna- the other hand, larger farmers could in theory com- tional markets. Similarly, modern communications pensate for less family labor per hectare by using hired infrastructure (such as postal service, telephones, labor, nonlabor variable inputs, and capital to meet or radio, television, electronic mail, and even satellite surpass land productivity on small farms. Adesina and connections) also links markets more closely with Djato (1996) show this for large rice farms in Cote each other and can facilitate the flow of useful infor- d'Ivoire, and Rao and Chotigeat (1981) show it for mation to farming households, including information large farms in India. Smaller farms may also have lower on prices, new technologies, insurance opportunities, land productivity because their more intensive farm- and procedures for obtaining government assistance. ing fatigues and degrades the soil, yet a zone with bet- Finally, government investments in electric power ter soils might attract more farmers, giving rise to generation and large-scale irrigation projects can have smaller farms with better yields than in other zones. an enormous impact on households' welfare and agri- Almost no research has been done on the rule of cultural productivity. law and the agriculture sector in developing countries, Government investments in basic social services, and only a small amount has examined insurance mar- particularly schools and health facilities in rural areas, kets. Crop insurance is sometimes available for large can also affect agricultural productivity and household commercial farms, but administrative costs usually pre- welfare. There is a large literature that shows how bet- vent it from being offered to small family farms. For an ter health and higher education make agricultural introduction to crop insurance in developing coun- workers more productive (see Strauss and Thomas tries see Gudger (1990). 1995 for a recent literature review). As explained in Chapters 7 and 8, there are sound economic reasons Public Investrnents for governments to invest in these services. Of course, Public investment policies include investments in policymakers need to make decisions on the extent physical infrastructure-such as transportation, com- and nature of these investments based not only on munication systems, electric power grids, and large- their impact on agricultural outcomes but also on scale irrigation schemes-and investments in basic their impact on other outcomes. social services, particularly in schools and health clin- Thus the two specific policy questions regarding ics. From the viewpoint of economic theory there are public investments and agriculture are: many reasons why such investments should be * What impact do government investments in trans- financed (though not necessarily implemented) by the portation, communications, electric power genera- government. Many infrastructure investments, such as tion, and large-scale irrigation schemes have on roads and canals, are public goods in the sense that an basic agricultural outcomes? 147 THOMAS REARDON AND PAUL GLEWWE * What impact do government investments in outcomes of interest. But before considering in detail schools and health services (clinics, hospitals and the kinds of data needed to assess these impacts, it is public health services such as immunizations) have useful to consider how agricultural households behave on basic agricultural outcomes? and how government policies can affect their behavior. The bulk of recent empirical work on the impacts Agricultural economists and agronomists often of infrastructure development on agriculture points to think of agricultural activities in terms of a production positive effects on the rate of commercialization and process or production function. When various inputs productivity growth. For a recent review of the litera- are combined in certain ways using a certain technol- ture see Raisuddin Ahmed and Cynthia Donovan ogy, the result is the crops and animals ("outputs") that (1992). For investments in social infrastructure see agricultural households produce. These products can Strauss and Thomas (1995), who address, among other either be consumed by the household or sold to oth- things, the impact of nutrition and education on agri- ers.The overall value of these activities to each house- cultural productivity. Again, the literature usually finds hold can be measured as farm profits (net agricultural positive impacts, but in some studies the link is weak income), which include not only earnings from selling or even nonexistent. products but also the value of products that the house- hold consumes. Households can invest some or all of Analytical Approaches, Data Needs, and Data the income generated by agricultural activities (as well Collection Issues as income from other sources) in ways that will increase their agricultural production in the future. The policy questions presented in the previous section How do agricultural households decide what to cannot be resolved by appealing to economic theory. produce, what inputs to use, and related choices? They can be answered only by examining data using Economists often portray households and their mem- appropriate empirical research methods. However, col- bers as organizing their activities to maximize some lecting data is not easy, and quite often the data avail- kind of utility function.1 Their utility ("happiness") is able are insufficient for answering important policy higher when they consume more goods and services questions. LSMS and similar multitopic household and lower when they increase the amount of time they surveys can provide policymakers with detailed, accu- spend working. Given this situation, agricultural rate information that can be used to understand the households organize their crop- and livestock- impact of current trends and proposed policy changes producing activities in ways that increase farm pro- on the agriculture sector. How well they do so ductivity and economize on the amount of time that depends on the type of survey data collected and on household members spend working on these activi- the methods used to gather and analyze the data. ties. This may involve more than the organization of This section draws on the methods used by agri- agricultural activities; for some households it may cultural economists and other researchers to assess what make sense for one or more members to find employ- data are needed to answer the policy questions raised in ment in nonagricultural activities, since that may pro- the first section. The first subsection, drawing on eco- vide the household with more income than would be nomic theory, discusses the behavior of agricultural the case if those members xvorked in agriculture. households and links this to the agricultural policy Agricultural economists have developed mathe- issues. The second subsection continues the discussion matical models of the behavior of farm households by explaining what data are needed to provide answers that provide useful insights for doing empirical to the various policy questions.The third and final sub- research (see Singh, Squire and Strauss 1986). One les- section shows how the required data are collected in son from these models is that it is important to distin- practice, emphasizing the difficulties involved and rec- guish factors that are beyond the control of the house- ommending practical solutions whenever possible. hold from outcomes that can be determined, at least in part, by the household. Some examples will make this An Economic Approach to the Behavior of Agricultural distinction clear. In general, households and individu- Households als have no control over prices of agricultural inputs Ultimately, policymakers would like to know the and outputs, the technological packages available, or impacts of different policies on the basic agricultural access to credit. In contrast, households typically do 148 CHAPTER 19 AGRICULTURE have control over the crops that they plant, the * Market conditions for buying, selling, and renting amounts of inputs that they use, and the time they land. spend working in agriculture. Given the factors * Traditional land tenure arrangements. beyond their control, which economists often refer to * Institutions for enforcing contracts and settling land as exogenous variables, and family characteristics (such disputes. as amount of land owned, education levels of family . Opportunities to purchase insurance. members, and family size), farm households make Nearly all government policies affect households by decisions about the things they can control, which influencing one or more of the exogenous variables economists refer to as endogenous variables. listed above; very few government policies directly In general, all the basic agricultural outcomes pre- affect the (endogenous) basic agricultural outcomes.2 sented in the first section are endogenous: they are Consider the macroeconomic policies discussed in the determined, at least in part, by the decisions of rural first section. Exchange rate policies affect agricultural households. Rural households decide what crops they households solely through the prices that the house- will grow, what animals they will raise, and how much holds face for their inputs and outputs.Thus to under- of each input (including their own labor) to use. The stand the impact of exchange rate policies it is neces- amounts of crops and animals that households produce sary to know how a specific exchange rate policy are partially under their control; they can increase pro- affects prices and how these prices affect the different duction by bringing more land under cultivation and agricultural outcomes. using more inputs. However, other factors, such as Tariff and nontariff policies also affect farming weather, also affect production. households, primarily through the prices that the Rural households also decide what technologies households face. These policies can also restrict the to use, whether and where to market some or all of access of farm households to new technologies. And their crops, and what kinds of long-term investments trade barriers occasionally work in other ways. A to make in agriculture. These decisions affect house- quota system for an imported agricultural input, such hold profits from agriculture. In addition, rural house- as fertilizer, may be accompanied by a distribution holds decide how much time, if any, each member mechanism not governed by price, such as rationing. should work in nonagricultural activities, and rural In this situation the rationed fertilizer can be thought households make many decisions about nonwork of as a "service" provided by agricultural extension activities related to health, nutrition, education, and centers.Yet flexibility is needed; because rationing can other areas. Finally, the activities of rural households take so many different forms, other types of rationing can affect the environment, the last basic agricultural may require a different approach. outcome presented in the first section. For example, By definition, credit policies at the macroeco- intensive use of fertilizers and pesticides can have neg- nomic level affect the terms and availability of credit. ative environmental consequences. They may do so directly (for example, by establishing How do government policies affect these basic government-run credit institutions) or indirectly (for agricultural outcomes? In general, they do so by alter- example, through regulations on private banks and ing the exogenous variables that households face.The other private lending institutions). following variables are exogenous and play important Finally, general reductions in spending and roles in determining basic agricultural outcomes: increases in taxes can affect both agricultural and * Prices for agricultural products and agricultural nonagricultural prices by reducing subsidies or inputs. increasing taxes on specific items.Taxes can also affect * Weather, pest infestations, and crop diseases. rural households (and often urban households) more * The availability of different kinds of agricultural directly-for example, through a general income tax. technologies. Many sectoral policies also affect prices. In partic- * The availability of agricultural extension services ular, subsidies, commodity taxes, price controls, and and the prices charged for them. marketing board policies almost always affect rural * Physical and social infrastructure. households by altering the prices that these house- * Taxes. holds face. The pricing of agricultural extension serv- * The availability of credit and the terms of that credit. ices is another sectoral policy that works through 149 THOMAS REARDON AND PAUL GLEWWE prices. Sectoral credit programs affect both the avail- "back of the envelope" calculations. For example, they ability and terms of credit. The government can also may estimate that reducing a tariff on an imported increase the availability of extension services-and in agricultural input will reduce the domestic price by the long run the availability of technology-by estab- the same amount that the tariff was reduced. In many lishing agricultural research stations. And some tax cases this approach may be too simplistic. A better policies in the agriculture sector may also have direct approach would be to use data from sources other than effects (that is, effects that do not work through a household survey to answer this question-for prices); an example is a tax on agricultural land. example, to use macroeconomic time-series data on Policies that affect the institutional environment exchange rates and domestic prices. Neither of the and public investment policies have both direct and two approaches uses household survey data. indirect effects on the exogenous factors that agricul- In contrast, household survey data are crucial for tural households face when making decisions about answering the second question. Since this book focus- their various activities. Any policy that affects tradi- es on the design of household surveys, the rest of the tional forms of land tenure will have major implica- chapter will focus on what is needed to answer the tions on the land markets and tenure arrangements second question, which is relevant for a range of that agricultural households face.The same is true for important macroeconomic and sectoral policies. For land redistribution policies. Such policies can have example, in examining the effects of exchange rate indirect effects (for example, changing the price of policies, the discussion will focus on how changes in land) and direct effects (for example, by outlawing spe- the prices of traded inputs and outputs affect agricul- cific forms of land tenure or requiring larger landown- tural outcomes, rather than on how changes in the ers to sell some of their land). Policies that affect the exchange rate affect prices.To answer the first question rule of law, such as the establishment of a legal appara- researchers need to consult the appropriate literature, tus for enforcing private contracts, can also have direct which in general does not make use of household sur- and indirect effects.A direct effect would be that con- vey data (two examples are Krueger et al 1992 and tracts become enforceable, and an indirect effect Barrett 1999). would be a change in prices due to the increased con- The second case concerns government policies tract enforcement. Insurance policies affect the terms that directly affect households' basic agricultural out- and availability of insurance, and they may do so comes. A good example of such policies are agricul- directly or indirectly. Finally, direct investments in tural extension services provided by the ministry of either physical or social infrastructure alter the oppor- agriculture. In this case the analysis is simpler because tunities available to households either directly (for household survey data can be used to examine the example, through a new road that reduces the wear (direct) impact of such policies on the phenomena of and tear on vehicles owned by the household) or indi- interest. rectly through prices (for example, as the new road The next subsection explains two ways to use provides better access to distant markets). LSMS-type household survey data to analyze agricul- In discussing how these four kinds of policies tural policy issues and describes in detail how to col- (macroeconomic, sectoral, institutional, and public lect the data needed for both types of analysis. investment) affect the exogenous variables and how these variables in turn affect the basic agricultural out- Analytical Methods and Specific Data Needs comes, it is useful to distinguish between two cases. In Research methods for analyzing the relationship the first case the policy affects households indirectly by between the exogenous variables affected by policies influencing one or more intermediate variables, such and basic agricultural outcomes can be divided into as prices or the availability of technology. In such cases two types: descriptive analysis and causal analysis. there are two separate questions.The first is: How does the policy affect prices or the availability of technolo- DESCRIPTIVE ANALYsIs. The main objective of descrip- gy? The second is: How do changes in prices or in the tive analysis is to describe what is occurring without availability of technology affect households' basic agri- rigorously explaining why it is occurring. The meth- cultural outcomes? Data analysts sometimes answer ods used are simple calculations of various shares and the first question by making simple assumptions or levels, so the biggest challenge is not doing the analy- 150 CHAPTER 19 AGRICULTURE sis but obtaining sufficiently accurate data. Simple behavior of agricultural households. For example, data descriptive statistics can be used to answer many dif- on which households currently grow a particular crop ferent questions of interest to policymakers, such as: do not show precisely who might benefit from a * What crops are agricultural households growing, reduction in a tax on that crop (which would increase and how does this vary in different types of agri- the prices received by farmers), because households cultural households? that are not currently growing the crop may decide to * What are the net farm incomes (profits) of different do so to take advantage of the new policy. Moreover, types of agricultural households? some of the households that are currently growing the * What agricultural inputs, such as fertilizer, irriga- crop may decide to increase their production of it, tion, pesticides, and farm equipment, are used by while others may hold their production constant. Thus different types of farming households? statistics on how much each household is currently * How does access to and use of agricultural exten- producing will not fully reflect who will benefit-nor sion services vary among households? the extent of the benefit-from a tax reduction. An * Which rural households are making investments in even more difficult problem is that many policies agriculture, and what form do these investments affecting prices benefit consumers as well as produc- take? ers. For example, if the supply of a taxed crop is quite There are two reasons why simple descriptive statistics elastic while the demand is not, the market price will are useful for policymakers. First, the better informed fall, so most of the benefits of the tax cut will accrue policymakers are about the basic characteristics of the to consumers rather than to producers. agricultural sector, the better prepared they will be to Descriptive analyses of agricultural issues that are make day-to-day policy decisions. Second, descriptive of interest to policymakers require data on all of the statistics can be used to obtain estimates of who cur- basic agricultural outcomes listed above, plus data that rently benefits from specific government policies. For can be used to classify households into different socioe- example, data on which households obtain informa- conomic categories. Therefore, the agriculture module tion from (or are visited by) agricultural extension should collect the following kinds of information: agents show who benefits from agricultural extension * The amount of each crop produced and of each services. Similarly, data on who obtains loans from type of livestock produced (including any crop or government credit programs help show the distribu- animal byproducts). tion of benefits from those programs. * The amount of purchased inputs used (including Simple descriptive statistics can also show which fertilizer, herbicides, insecticides, fungicides, seeds households currently benefit from programs that affect and seedlings, and irrigation services) and the the prices of agricultural inputs and outputs. For household's use of credit and agricultural extension example, price subsidies for a particular crop benefit services. Information on the amount of labor used only the farmers who grow that crop, and the benefit (including both household members and hired any individual farmer receives is proportional to the labor) may yield a clearer picture of the agricultur- amount that he or she grows.Thus information about al sector but is not necessary for making rough esti- which households grow each crop, and how much mates of the impact of government policies. they grow, can be used to estimate the distribution of * The technologies used by each household, includ- the benefits provided by both current and proposed ing high-yielding varieties of particular crops or price subsidies (Deaton 1989). Similar calculations can new types of capital equipment. be done for price subsidies on agricultural inputs (such . The marketing activities of agricultural households, as fertilizer and insecticides) and for taxes on either including how much of their crop and livestock inputs or outputs. Other policies that affect the prices production was sold and to whom it was sold. of agricultural inputs and outputs-such as tariffs, . The net income (profits) that households earn from exchange rate devaluations, and price controls-can their agricultural activities. also be evaluated. * Households' investments in agriculture, both in While such descriptive analysis is useful, it is terms of their current stock (ownership of farm important to keep in mind that these estimates are machinery, irrigation equipment, farm tools, build- approximations because they do not account for the ings, and land), and recent additions (purchases of 151 THOMAS REARDON AND PAUL GLEWWE these items plus building construction and land CAUSAL ANALYSIS. While descriptive analysis can be quality improvements). quite useful for policy discussions, it has serious limi- No mention was made here of welfare indicators tations. In particular, it cannot be used to explain such as consumption expenditures, health status, and household behavior. This is why estimates that use school enrollment. Information on welfare should be descriptive methods to show who benefits from gov- collected in other modules of the household ques- ernment policies are only approximations. More gen- tionnaire such as the consumption, education and erally, descriptive information cannot be used to health modules. Similarly, data on how environmental examine the causes of various phenomena of interest, outcomes may be affected by agricultural activities such as the impact of additional inputs on farm pro- should be collected in the environment module. ductivity and the reasons why some households plant However, if the survey does not include an environ- certain crops, use certain inputs, or adopt certain tech- ment module, it may be necessary to collect this infor- nologies, while others do not. Understanding the mation in the agriculture module. See Chapter 14 on causal relationships between inputs and outputs and the environment for a detailed discussion of how to the determinants of household behavior requires collect such data. causal analysis, in which analysts use econometric and The definitions of most of the data listed above are statistical methods (most commonly, regression analy- quite clear. However, the calculation of net farm income sis) to explain why households make the choices that (profits) merits a brief discussion. This variable is the they do.' Ultimately this information can be used to value of agricultural outputs minus the value of the cor- estimate how existing government policies affect responding agricultural inputs.The values ofoutputs and households' activities and well-being, and to predict inputs used can be obtained directly or derived from the likely effects of any new policy options that may quantity and price information. As discussed further be under consideration. below, collecting data on quantities can lead to difficul- When analyzing agricultural issues, two kinds of ties if respondents are only able to provide answers in causal relationships can be estimated: structural rela- local units, but their responses need not be converted tionships and reduced form socioeconomic relation- into national or international units if the price data are ships.The most important example of a structural rela- also expressed in local units. The price information can tionship is the agricultural production function-the be obtained from either the household questionnaire or physical relationship between outputs (crops, animals, the community questionnaire. Two points should be and their byproducts) and inputs (including those con- kept in mind xvhen designing the survey. First, the set of trolled by households, such as labor, fertilizer, and land, outputs and inputs consists of both those that are sold for as well as random factors outside the control of house- (or purchased with) cash and those that are given (or holds, such as rainfall and pest infestations). Strictly received) in kind; market prices can be used to impute speaking, production functions do not depend in any the value of the in-kind inputs and outputs. Second, to way on the characteristics of households. Thus pro- calculate total net farm income it is not necessary to duction functions contain no information about know which inputs were used on each farm product, household behavior. although calculation of net farm income for each prod- The second kind of causal relationship, a reduced uct would require such information. form relationship, shows how exogenous variables In summary, for descriptive analysis, it is only nec- affect the different basic agricultural outcomes (all of essary to collect data on the phenomena of interest. In which are endogenous). Since these relationships all most cases there is little need to collect price data involve choices made by households, they depend on unless analysts wish to see how prices vary among dif- each household's characteristics, such as how many ferent geographical areas or socioeconomic groups. people belong to a household and the household's However, experience with past LSMS surveys has utility function. This kind of causal relationship does shown that data survey designers initially thought measure household behavior. While a production would not be useful for analysts have often later function is the same for all kinds of households, a proved to be of considerable analytical interest. Thus if reduced form relationship can vary among different possible the survey should aim to collect a broad range types of households because it depends in part on their of information. characteristics. 152 CHAPTER 19 AGRICULTURE Some examples may make the distinction or as a factor that modifies the contribution of between structural and reduced form relationships other inputs) affect the farmers' productivity, the clearer. A production function is a technological rela- crops that they grow, the inputs they use, the tech- tionship between inputs and outputs. It can be nologies they adopt, and other basic agricultural thought of as a "formula" that shows how much out- outcomes? put is produced by combining different sets of inputs. Econometric methods can be used to estimate One input may be fertilizer, or even the labor of structural relationships such as production functions, as household members, but the relationship between well as reduced form relationships. Estimates of pro- inputs and outputs is completely unaffected by duction functions can be used to answer the fifth and whether a household uses fertilizer or how it uses its eighth questions above, on the impacts of education own labor; the relationship itself is not affected by and farm size on farm productivity. Estimates of household behavior. In contrast, one example of a reduced form relationships can be used to provide reduced form relationship is the determinants of the answers to the other six questions listed above. One amount of labor that a household devotes to agricul- useful reduced form relationship is the "profit func- tural activities.This can also be thought of as a formu- tion." This shows how much profit a farm household la, but in this case the parameters of the formula may can make given certain input and output prices depend on household characteristics such as the rela- (which are clearly exogenous), household assets such tive weights on leisure and consumption in a given as farm size or capital holdings (which can be consid- household's utility function. ered exogenous in the short run), and other household Estimates of these two types of causal relationships characteristics.4 Profit functions are particularly useful can be used to answer policy questions such as: because they can be used to derive estimates of func- * How much, and how quickly, do changes in the tions that show how the supply of outputs and the prices of agricultural products and inputs affect the demand for inputs are affected by these same variables production of exported and nonexported crops, the (Sadoulet and de Janvry 1995). use of agricultural inputs, and other basic agricul- Two things must be borne in mind when consid- tural outcomes? ering causal analysis. First, more data are needed to do * How do different agricultural extension services causal analysis than to do descriptive analysis. affect farmers' output, use of inputs, incomes, and Unbiased estimation of reduced form relationships productivity? requires that the analyst have data on all of the vari- * How does the availability of credit affect the use of ables that affect the basic agricultural outcome of agricultural inputs, the adoption of new technolo- interest. For example, if an analyst is interested in find- gies, capital investments, and other basic agricultur- ing out how the price of fertilizer affects the amount al outcomes? of fertilizer used, data are needed not only on the price * How do traditional forms of land tenure and invest- of fertilizer and the quantity used but also on every- ments in physical infrastructure affect crop produc- thing else that determines the use of fertilizer, includ- tion, households' incomes from agricultural activi- ing the prices of other agricultural inputs (pesticides, ties, and other agricultural outcomes? herbicides, and hired labor), the prices of the crops on * What is the causal relationship between farm size which the fertilizer may be used, the availability of and farm productivity, and how does farm size credit, the education level of adult household mem- affect other basic agricultural outcomes? bers, the prices of other crops (and even of animals), * How do policies that promote the rule of law and rainfall and other weather conditions, the land, enforcement of contracts affect households' agri- machinery, and other types of capital owned by the cultural activities? household, and so forth. Basically, data are needed on * What impacts do different types of investments in almost all of the exogenous variables listed at the physical infrastructure have on the marketing of beginning of this section. Collecting such data is not crops, use of purchased inputs, adoption of new easy, and almost always requires a long and detailed technology, and other basic agricultural outcomes? agricultural module. * How does the educational level attained by farmers The second thing to realize about causal analysis (which can be thought of either as a distinct input is that many problems can lead to biased estimations of 153 THOMAS REARDON AND PAUL GLEWWE these relationships. If data are missing on some causal exogenous factors is always a challenge, which implies factors, the estimates are likely to suffer from omitted that a long agricultural module will be needed to variable bias (see Chapter 26 for further explanation of undertake rigorous causal analysis. this point). A classic example of omitted variable bias Table 19.1 summarizes relationships between the is the inability to observe the managerial talent of the policy issues raised in the first section and the data farmer (see Griliches 1957). Suppose analysts want to needs discussed in the second section. It also shows the estimate a production function for a certain crop and varying ability of the three different versions of the that they are particularly interested in how the use of agricultural module (which are introduced in the next fertilizer affects productivity. Suppose as well that section) to supply the data needed to answer each those farmers with more managerial talent are both question.5 As discussed above, several of the policy more productive, all else being equal, and more likely questions in the first section can be broken into two to use fertilizer. If a regression is estimated without parts: the impact of a specific policy on prices or tech- accounting for the impact of managerial talent, it will nology availability and the impact of prices or tech- overestimate the impact of fertilizer on output because nology availability on basic agricultural outcomes. In part of the impact measured is the impact of manage- Table 19.1 such questions are similarly divided. For rial talent, which is positively correlated with use of example, the table has no question on the impact of fertilizer. The "simple" solution to this bias is to exchange rate policies on basic agricultural outcomes, include all of the variables that are needed, but it is not but it does have a question on the impact of exchange always possible to collect data on some of the causal rates on prices and a question on the impact of prices factors. on basic agricultural outcomes. In addition, the table Estimates of causal relationships also suffer from includes several policy questions raised in in this sec- such other potential problems as measurement error in tion's discussion of policy questions that can be the variables (which affects both structural and answered by using descriptive analysis. reduced form relationships) and the endogeneity of the explanatory variables (which affects only structur- Some Difficulties and Some Potential Solutions al relationships such as production functions). These Collecting data on the variables discussed above is econometric issues are discussed in detail in Chapter often complicated by the complexity of the produc- 26; suffice it to say here that these problems are seri- tion process and by the great variety of agricultural ous ones that constantly plague empirical researchers. producers in developing countries. Farms vary in size However, many of the problems (including measure- from small garden plots in Russia to giant grain farms ment error and omitted variable bias) can be mini- in Argentina. Farms also vary in the extent to which mized if detailed data are collected and appropriate they are connected to markets, and in the degree to procedures are used to ensure data accuracy. The fol- which they are privately or collectively owned. lowing subsection provides advice on how to collect Moreover, within any given rural household one often such data. finds a bewildering array of common and individual In summary, the data required to undertake causal plots (the latter often controlled by different house- analysis include data on all of the basic agricultural hold members) with a variety of crops grown on each outcomes that are of interest and all of the exogenous plot. factors that determine these outcomes. These exoge- Because of these complexities, the agricultural nous factors include prices for agricultural products module of a multitopic survey must be carefully and agricultural inputs, various "shocks" (such as designed to reflect the prevailing circumstances of the weather, pest infestations, and crop diseases), the avail- agricultural sector in the country where the survey is ability of technology and of extension services (and being fielded, often within the constraint of a limited any prices associated with their use), physical and survey budget.This subsection examines several specif- social infrastructure, taxes, the availability and terms of ic issues involved in the collection of agricultural data credit and insurance, the characteristics of the local and provides practical advice on how to resolve them. land market, traditional land tenure arrangements, and It begins by broadly distinguishing between situations the country's institutional capacity for enforcing con- in which it is relatively easy to collect agricultural data tracts and settling disputes. Collecting data on these and situations in which it is relatively hard to do so. 1 54 CHAPTER 19 AGRICULTURE Table 19.1 Policy Issues, Methods of Analysis, and Household Survey Data Adequacy of data from Policy issue Data needed Short version Standard version Expanded version Policy issues that can be addressed using descriptive analysis of household survey data What crops and livestock are Quantities of crops and livestock grown (household Good Very good Very good being grown by different types questionna re) plus data from other modules for of agricultural households? classifying households. What are the net farm incomes Net income earned from agricultural activities Poor Good Very good (profits) of different types of (household quest onnaire), plus data from agricultural households? other modules for classifying households. What agricultural inputs are used Quantities of agricultural inputs used (household Fair Very good Very good by different types of farming questionnaire), plus data from other modules for households? classifying households. How does access to and use of Existence of extension services (community Very good Very good Very good agricultural extension services questionnaire) and use of those services vary among households? (household questionnaire), plus data from other modules for classifying households. Whicn rural households are Purchases of capital goods and land (household Poor Fair Good making investments in questionnaire), plus data from other modules agriculture, and what form do for classifying households. these investments take? What agricultural technologies Use of hybrid seeds, new types of capita Poor Good Good are being used by different kinds equipment and specific farming methods of agricultural households? (household questionnaire), plus data from other modules for classifying households. ... ................... ................................................................................................................................................................................................... How does access to and use Local sources of credit (community questionnaire) Poor Good Good of credit vary among and use of credit (household questionnaire), plus data households? from other modules for classifying households. How do marketing opportunities Distance to nearest local and periodic m arkets and Far V ry good Very good and activities differ across means of transportation to get to them (community different households? questionnaire), marketing activities (household questionnaire), plus data from other modules for classifying households. How are the benefits of price Amounts of outputs produced and amounts of inputzs Good Very good Very good subsidies to inputs and outputs used (household questionnaire), plus data from other distributed across different modules for c assifying households. Rough estimates households? of change in prices due to policies must be assumed or obtained from other kinds of data. ................................................................................................................................................................................................................................... Policy issues that can be addressed using causal analysis of household survey data How do changes in the prices of Reduced form estimates, which have the basic Fair Good Very good agricultural products and inputs agricultural outcome as the dependent variable.The affect production of crops, use most important explanatory variables are prices of of inputs, and other basic inputs and outputs (price questionna re and agricultural outcomes? household questionnaire): plot, farm, and household characterist cs (household questionnaire); and community characteristics such as physical infrastructure and access to extension services (community questionnaire). ............................................................................................................................................................................................................................. How do different agricultural Reduced form estimates, with the basic agr cultural Fair Good Very good extension services affect farmers' outcome as the dependent variable.The most output, use of inputs, incomes, important explanatory variables are: pr ces of inputs and productivity? and outputs (price questionnaire and household questionnaire); plot, farm, and household characteristics (household questionnaire); and community characteristics such as access to extension services (community questionnaire). ................................................................................................................................................................................................................................... (Table continues on next page.) 155 THOMAS REARDON AND PAUL GLEWWE Table 19.1 Policy Issues, Methods of Analysis, and Household Survey Data (continued) Adequacy of data from Policy ssue Data needed Short version Standard version Expanded version How does the availability of Reduced form estimates, with the bas c agricultural Fair Good Very good credit affect capital investments, outcome as the dependent variable.The most use of agricultural nputs, important explanatory variables are: prices of adoption of new technologies, inputs and outputs (price questionnaire and and other basic agricultural household questionnaire); plot, farm, and household outcomes? character stics (household questionnaire); and community characteristics such as access to, and terms of, locally available credit (community questionnaire). How do traditional forms of Reduced form estimates, with basic agricultural Fair Good Very good land tenure and investments in outcomes as the dependent variables.The most physical infrastructure affect crop important explanatory variables are: prices of production, households' incomes inputs and outputs (price questionnaire and from agriculturai activities, and household questionnaire); plot characteristics other agricultural outcomes? (including type of tenure) and household characteristics (household questionnaire); and community characteristics such as the nature of traditional tenure arrangements (community questionnaire). What is the causal relatio nship A production funct on, the estimate of which will Poor Fair Good between farm size and farm require data on all inputs, including plot size, and the productivity? output of interest (all from the household questionnaire)-and may require prices (from the household or community questionnaire) or other instrumental variables. How does farm size affect Reduced form estimates, wnich wil have the basic Pair Good Very good basic agricultural outcomes? agricultural outcome as the dependent variable.The most important explanatory variables are: prices of inputs and outputs (price questionnaire and household questionnaire); plot size, other plot characteristics, and household characteristics (household questionnaire): and such community characteristics as physical infrastructure and access to extension services (community questionnaire). -row do policies that promote Reduced form estimates, with the basic agricultural Poor Pair Pair the rule of law and enforcement outcomes as the dependent variables.The most of contracts affect households' important explanatory variables are: prices of inputs agricultural activities? and outputs (price questionnaire and household questionnaire); plot characteristics (including any relevant contract information) and other household characteristics (household questionnaire); and such community characteristics as the local legal system and its degree of enforcement (community questionnaire). What impacts do different Reduced form estimates, with the basic agricultura Fair Good Very good types of investments n physical outcome as the dependent variable.The most nfrastructure have on marketing important exp anatory variables are: prices of of crops, use of purchased inputs, inputs and outputs (price questionnaire and adoption of new technology and household questionnaire); plot characteristics and other basic agrcultural other household characteristics (household outcomes? questionnaire); and such community characteristics as physical infrastructure (community questionnaire). How do farmers' ieveis of A production function requiring diata on: all inputs, Poor Fair Good education and health status including education and health of household members, affect their productivity? and all outputs (household questionnaire) as well as prices (from household or community questionnaire) or other instrumental variables. ............................................................................................................................................................................................................................ 156 CHAPTER 19 AGRICULTURE Table 19.1 Policy Issues, Methods of Analysis, and Household Survey Data (continued) Adequacy of data from Policy issue Data needed Short version Standard version Expanded version How does the availability of Reduced form estimates, with the basic agricultural Poor Fair Good insurance affect capital outcome as the dependent variable.The most investments, the use of important explanatory variables are: prices of inputs agricultural inputs, the adoption and outputs (price questionnaire and household of new technologies, and other questionnaire); plot, farm, and household characterist cs basic agricultural outcomes? (household questionnaire); and such community characteristics as access to, and terms of, insurance opportunities (community questionnaire). How do farmers' levels of Reduced form estimates, which will have the basic Fair Good Very good education and health status agricultural outcome as the dependent variable.The affect other agricultural most important explanatory variables are: prices of outcomes? inputs and outputs (price quest onnaire and household quest onnaire); plot characteristics, health and education status of household members, and other household characteristics (household questionnaire); and such community characteristics as physical infrastructure and access to extension services (community questionnaire). ................................................................................................................................................................................................................................... Pol cy Comments Policy issues that cannot be addressed using household survey data Impact of exchange rate policies on prices of agricultural inputs and Requires times ser es data on exchange rates and local prices. outputs and on the variab lity of these prices Impact of tariff and nontariff trade barriers on prices of agricultural Requires times series data; studies of results from other countries might inputs and outputs also be useful. ................................................................................................................................................................................................................................... Impact of national and sectoral credit policies on the availability of Would require a survey of private providers of cred t, and perhaps even credit from private sector providers a series of surveys over several years. See Chapter 2 1 for further discussion. Impact of sectoral price subsidies and taxes on prices of agricultural Would require time series data on taxes, prices, and subsidies. In theory, inputs and outputs several household surveys over many years could provide the price data. Impact of government po.icies on traditional forms of land tenure Would require a special survey focusing on traditional forms of land tenure.A more sociological or anthropological approach might be useful. See Chapter 25 on collection of qualitative data. Politicai feasibility of land redistribution Could require a multidisciplinary approach. See Chapter 25. ..................................................................................................................................................*................................................................................ Impact of investments in social infrastructure (health and education) Dealt with in Chapters 7 (education) and 8 (health). on social outcomes Source: Author's summary The subsection then discusses the implementation of more disaggregated level (for example, by plot rather the survey, including issues that are important because than by entire farm and by season rather than by cal- they determine the extent to which measurement endar year) to minimize the chances that the data will errors creep into the data as the data are collected. suffer from serious measurement error. Countries (or areas within a country) where EASY-TO-SURVEY VERSUS HARD-TO-SURVEY farms tend to be easy to survey are mostly in non- AGRICULTURAL SITUATIONS. In some countries the highland, non-semiarid Asia, in non-highland Latin organization of the agriculture sector is fairly straight- America, and in the cash crop zones of Africa. With a forward, which simplifies the design of the agricultur- few exceptions, these countries are in less poor areas of al module. In contrast, the organization of the agricul- the developing world. Many of these countries also ture sector in other countries is more complex, which have a strong National Agricultural Research System can greatly complicate the design of the agricultural (NARS) and reliable national farm surveys. module. Box 19.1 presents characteristics of farms that Countries (or areas within a country) where farms tend to make them easy or hard to survey. In general, tend to be hard to survey are more commonly in those in "hard" situations it is necessary to gather data at a parts of Africa outside cash-crop areas, in highland Asia 157 THOMAS REARDON AND PAUL GLEVVWE respond to questions about all plots and crops or Box 19.1 Easy-to-Survey versus Hard-to-Survey should the individuals in charge of each plot or crop Situations respond separately to questions about the plots or Characteristics of an easy-to-survey situation crops for which they are responsible? * Farmers grow rice and wheat. A prominent feature of most hard-to-survey situ- * Farmers specialize in a few crops or types of livestock. ations is that farmers grow crops on a large number of * The farm economy is highly monetized and commer- plots. For example, in the 1996 Nepal LSMS survey cialized. the typical farm household had five or six plots.6 * Standard units (such as hectares and kilograms) are Similarly, in the 1985-86 Peru LSMS survey, 7 to 10 widely used. plots were common, and most households had 10 to * Each farm has only a few plots,.lt eecmo,admothueod a 0t The plotsfare hspaly concentr. 15 plots in the 1981-85 Burkina Faso survey by the * The plots are spatially concentrated. * There is a literate adult in the household. International Crops Research Institute for the Semi- * Agriculture is irrigated. Arid Tropics (ICRISAT).Typically such multiple plots * Where there are many livestock, they are kept in are managed by several different plot managers, many fenced-in areas. of whom manage more than one plot. The agriculture modules in most previous LSMS Characteristics ofo hard-to-survey situation surveys gathered information using the whole farm, * Farmers grow tubers, bananas, roots, and coarse grains. rather than the plot, as the unit of analysis, with the * Farmers typically produce many different products, most knowledgeable" person at the household level including both crops and livestock. * The farm economy is only partially monetized and as the sole respondent. In countries where easy-to-sur- commercialized. vey agricultural circumstances predominate, this * Nonstandard (local) units of measurement are widely approach may work reasonably well for collecting data used. that can be used to calculate basic agricultural out- * Each farm has many plots. comes (such as net farm income) for the whole house- * The plots are spatially dispersed. hold. However, this approach is unlikely to work well * There is no literate adult in the household. in countries characterized by hard-to-survey agricul- . Farmers rely solely an rainfall for irhgation. tural circumstances, where each agricultural house- * Many livestock are kept in open pastures. hold has many plots on which many different crops are Source: Authors' summary grown-with different inputs and practices used on different plots. There are several reasons why collecting data at and Latin America, and in parts of semiarid South Asia. the plot level, as opposed to the "whole farm" Because these tend to be the poorest areas of the devel- approach, is preferable in hard-to-survey situations. oping world, national farm surveys are less likely to First, using the plot as the survey unit reflects the nat- exist in those countries. Yet agriculture is often the ural flow of a conversation between an interviewer largest economic sector in such countries, and thus has and a respondent. Experience with farm management a disproportionately large impact on economic surveys has shown that farmers usually refer to each growth, nutrition, poverty, and the environment. plot as they describe the tasks undertaken and the crops produced; they refer to the whole farm only SURVEY UNIT AND CHOICE OF RESPONDENT. Once sur- when discussing the purchase of inputs (Matlon 1988). vey designers have established the extent to which the Forcing respondents and interviewers to collect data at country surveyed is an "easy" or "hard" case, they must the farm level may lead to significant errors in the choose the unit of analysis for the agriculture module. data. In most developing countries agricultural households Second, gathering data on each plot yields more work on several different plots of land, so the question observations and more variation in the data, both of regarding the unit of analysis is whether data should be which allow for more precise estimation of farm pro- collected for the farm as a whole or separately for each duction functions and other agricultural relationships. plot or field. A related issue is which household mem- Third, plot-level data are very useful for examining ber should be the respondent. Should the same person intrahousehold allocation issues (see Chapter 24 for a 158 CHAPTER 19 AGRICULTURE full discussion). A fourth reason to collect data at the ticides) are handled by a single household member plot level is that on any given farm, plots usually differ instead of by the plot managers. If the questionnaire is in terms of land quality, the degree of land degradation carefully field-tested, it should become clear where and erosion, and other characteristics such as the this is occurring; in such cases the questionnaire degree of slope or whether the plot is situated on a should be designed to ask a single respondent (the hill, a valley, or a plain. It is important to control for household member best informed on the subject) the these differences when estimating production func- questions on these topics.This was the approach taken tions, and it is more difficult to control for these dif- in the 1996 Nepal LSMS survey. ferences when the data are not disaggregated by plot. The implication of this discussion is that when A fifth and final reason to collect plot-level data is that surveys are implemented in countries with hard-to- this approach keeps all options open for future data survey agricultural sectors, they should collect data at analysis. Researchers can aggregate the plot data to the the plot level. In fact, collecting plot-level data is also level of the whole farm, whereas they would not be reasonable for easy-to-survey situations, since there are able to disaggregate whole-farm observations to the few disadvantages to collecting such data. All LSMS level of the plot. surveys that collect detailed information on outputs There are also good reasons for interviewing the and on use of labor and nonlabor inputs should do so manager of each plot rather than choosing only one for each plot, and the manager of each plot should be respondent per household. First, the operator of each the respondent for all questions concerning that plot. plot is most likely to know specific details about the This is crucial if the data are to be used in production size and quality of the plot, and about how much time function analysis. It is less essential if only rough each household member has spent working on various observations of farm income are needed, although tasks on that particular plot. Second, spreading the even here survey designers may want to collect data at burden of completing the agricultural module over the plot level in order to minimize measurement error. several different respondents avoids placing too much In contrast, in most cases the acquisition of inputs can of that burden on a single respondent. be recorded at the household level by the person best The advantages of using the plot as the unit of informed about those purchases. observation must be weighed against two possible dis- In contrast to crops, livestock are mobile.They can advantages. The first is that gathering more detailed be kept either in a specific location on the farm, such data may lengthen the time needed to complete the as in the corral or compound, or on communally interview.Yet this disadvantage may be more apparent owned pastureland. Households may move livestock than real. Respondents in farm surveys in developing from one place to another, perhaps several times, over countries often provide answers plot-by-plot even the course of a year. In addition, many of the inputs when the agricultural module is designed using a pertaining to animal husbandry are not necessarily "whole farm" approach-forcing the interviewer to linked to the plot or pasture where the animals reside. add up the responses on the spot to produce the For example, veterinary services, labor inputs, and "total" numbers required by the questionnaire.When purchased feed are not necessarily provided on the this is the case, there may be little difference in the land where livestock are usually kept. Thus there is lit- interview time, and reducing the measurement error tle reason to collect animal husbandry data at a plot generated by ad hoc aggregation on the part of the level; instead, questions about livestock husbandry interviewer would probably be worth the modest should be asked at the household level. amount of extra time needed to complete a plot-by- Nevertheless, there are cases in which the quality plot agricultural module. Moreover, the interview of livestock data would probably be higher if more time could be reduced by collecting detailed plot-level than one respondent were interviewed, particularly data only for a subsample of plots while collecting a when some household members are responsible for much smaller amount of information for the remain- herds that are separate from those of the head of ing plots, as was done in the 1995 LSMS survey in household. This is often the case in agropastoral areas, Northeast China.7 where livestock husbandry is an integral part of the The second possible disadvantage can occur when farming system. Indeed, livestock are a sign of wealth some farm activities (for example, the purchase of pes- and status, and an important form of savings. Al of this 159 THOMAS REARDON AND PAUL GLEWWE implies that the ownership of livestock can be a sensi- middle of an agricultural season, which can produce tive issue. When livestock ownership is a sensitive data that are not very useful. For example, suppose issue, serious consideration should be given to using there is only one cropping season per year and a more than one respondent to answer questions about household is interviewed a few weeks before the har- the livestock and herds of the household. Still, it may vest of that season.The output reported will be for the be difficult to elicit accurate answers if respondents previous cropping season, and nearly all of the inputs cannot be interviewed privately, since the respondents reported will be for the current cropping season. may not want other household members to know the The implication of this discussion is that the recall details of their livestock holdings. period for recording agricultural outputs and inputs should be the cropping season, not the preceding 12 RECALL PERIOD. Most previous LSMS surveys have months. More specifically, the agricultural module used a recall period of the previous 12 months in the should be administered after the end of the cropping agriculture module because this coincided with recall season; it makes little sense for the interviewer to visit periods used in other modules of the survey. Farm sur- a farm in the middle of the season and count the use veys typically use a one-vear recall period, but in con- of inputs in that season because they will not be able trast to LSMS surveys, this period cannot be a random to link these inputs to the outputs, which are not yet 12-month interval. Instead, it must correspond to known. Although this is simple enough in principle, it either the most recent "harvest year" (the 12 months may complicate the field organization of an LSMS- from one harvest until just before the next year's har- type survey (or any multitopic survey that visits house- vest) or the most recent "production year" (the 12 holds only once)-particularly by affecting the timing months from planting until right before planting of the interviewer's visit to sampled households. As begins one year later). These two ways of defining an explained in Chapter 3, in many LSMS surveys the agricultural year can incorporate one or more "crop- interviews are spread out over 12 months, which ping seasons" (also known as "agricultural seasons"), makes it almost inevitable that households are inter- which are the times from the beginning of planting viewed in the middle of a cropping season. The best until the end of the harvest. In some areas, such as the way to resolve this problem is to have interviewers ask West African semiarid tropics, there is only one agri- about the most recent 12-month period that includes cultural season, wvhich reflects the fact that there is only one or more completed cropping seasons. The follow- one rainy season per year. In other areas, such as the ing paragraphs explain how this would work for both East African highlands, there are two agricultural sea- countries with one cropping season per year and sons, since there are two rainy seasons per year. Of countries with two cropping seasons per year. course, even in areas with only a single rainy season per Consider semiarid West Africa, an area with just year, there can be two cropping seasons if irrigation is one cropping season a year. In semiarid West Africa the used in the dry season or if certain crops can be grown cropping season runs from June to October. and harvested without irrigation in the dry season. Interviewing after the end of a cropping season can be The recall periods used for agricultural activities done relatively easily. Any interview that takes place in previous LSMS surveys had several disadvantages. between November and May and uses a one-year recall First, respondents were asked to provide the total period wvill cover that one season. Interviews can be amount of outputs and inputs on their farms in the 12 done after May as long as the respondent remembers months immediately preceding the date of interview. what happened the previous May. Thus in semiarid Such an approach can lead to aggregation bias in West Africa the recall period runs from May of the pre- countries where there are two or more cropping sea- ceding year until April of the current year. More gen- sons per year (a problem discussed later in this sec- erally the recall period is the most recent 12-month tion). Second, there is evidence that the use of one period that contains a complete cropping season. long recall period, such as one year, can have a detri- Things become only slightly more complicated in mental effect on the quality of the household survey places where there are two cropping seasons, such as data (see Kelly and others 1993). Third, and perhaps the East African highlands or the Indian semiarid trop- most important, the 12-month period immediately ics. Suppose that one cropping season runs from April preceding the interview often starts and ends in the to August and the other cropping season is from 160 CHAPTER 19 AGRICULTURE November to January. If the interviewer were to visit first interview would collect data only on the first sea- the household during, say, September or October, the son and the second interview would collect data only recall period would refer to the current year's on the second season. Of course, the advantages of April-August season and the previous year's these options must be weighed against the advantages November-January season, and the module would of having the interviews spread out over a long peri- contain separate questions for each season. If the inter- od of time. (Recall from Chapter 3 that the three viewer's visit occurred sometime in February or advantages were controlling for seasonality patterns, March, the interviewer could administer the module reducing equipment costs, and reducing the for the just-completed November-January season and number-and hopefully increasing the quality-of the for the April-August season of the previous year. lf an interviewers trained.) In surveys for which agriculture interview were to fall in the middle of either season, is a lower priority, the advantages of deviating from the the reference period used would be the 12 months common LSMS interviewing scheme may be out- that contained the two most recently completed crop- weighed by the disadvantages. ping seasons. For example, if the interview were to In summary, 12-month recall periods must be take place in June, the recall period would run from used that include cropping seasons in their entirety as March of the previous year to March of the current opposed to recall periods that begin or end in the year. If the survey interviews were done continuously middle of a cropping season. For many households this over a 12-month period (as has been done in many means that the 12-month recall for agricultural activ- past LSMS surveys, for reasons given in Chapter 1), the ities will not be the 12 months immediately preceding data on the April-August season would be in one cal- the interview but will exclude the previous 2 or 3 endar year for some households and in another calen- months. One disadvantage of this is that if analysts try dar year for other households; the same would be true to calculate total household income they will be faced for the November-January season.While this is prob- with data on different sources of income that do not ably not a serious disadvantage, analysts need to keep cover the exact same time period. For example, agri- it in mind when using the data. cultural income may refer to the 12-month period In countries with two or more cropping seasons beginning 15 months before, and ending 3 months per year, data should be collected separately for each before, the date of the interview. In contrast, the data cropping season, as it is common for there to be sub- on income from wage labor may refer to the 12 stantial differences in the extent and nature (for exam- months immediately preceding the interview. ple, the crop mix or the technology) of agricultural However, this disadvantage is probably not a major activities in different seasons within one year, even problem, and it is generally preferable to forcing the when irrigation is used in the dry season. Even for agricultural module to fit the previous 12 months plants with long-growing cycles that are harvested lit- even if the interview takes place in the middle of a tle-by-little throughout the year (such as roots, tubers, cropping season (as was done in most previous LSMS bananas, and plantains), there are differences in the surveys). Of course, if agriculture is the main focus of intensity of harvesting depending on the season and the survey, the common practice of interviewing on the point in the growth cycle of the plant. Thus households over a 12-month period could be dropped respondents must be asked to provide separate answers in favor of either interviewing all households in a one- for each cropping season. or two-month period between cropping seasons or If agricultural issues are the top priority of the interviewing each household twice, once after the first survey, survey designers should seriously consider season and again after the second season. More exper- deviating from the common LSMS practice of inter- imentation in future LSMS-type surveys should pro- viewing households only once and spreading those vide useful experience on how best to collect agricul- interviews out evenly over a 12-month period. One tural data within the context of a multitopic option is to perform all interviews in a short period of household survey. time between two cropping seasons. An even better, though more expensive, option is to visit each house- SAMPLE SizE. Because of the inevitable budget con- hold twice-once soon after the first season and a sec- straints involved in any household survey, there is a ond time soon after the second season. In this case the tradeoff between using a large sample to keep sampling 161 THOMAS REARDON AND PAUL GLEWWE errors low and using a small sample to fully implement use of panel data often invoke questionable assump- all of the procedures designed to reduce nonsampling tions, as explained in Chapter 23.Thus panel data will (measurement) errors. For a given sample size, nonsam- probably prove useful, but the magnitude of the bene- pling errors xvill generally be smaller in easy-to-survey fit is difficult to gauge at first glance. situations than in hard-to-survey situations. As If panel data are collected, serious consideration explained in Chapter 1, LSMS surveys usually have rel- should be given to matching individual plots of land atively small samples (2,000-5,000 households) due to across the different surveys.The last section of Chapter their emphasis on reducing nonsampling errors. The 23 presents detailed recommendations for linking size of the whole sample for a given survey depends household members across a series of surveys that not only on the agriculture module but also on all of cover the same households. In principle, these meth- the other modules in the survey. Survey designers ods for matching household members can be adapted should bear in mind that the size of the sample for the to plots of land, but there is little experience in doing agriculture module will be smaller than that of the so for surveys separated by several years. Thus serious overall sample because many households do not have thinking and field testing need to be done to success- any agricultural activities. Also, when there is more fully link plots of land in panel surveys, using the dis- than one agroclimatic zone in the country or region cussion in the last section of Chapter 23 as a starting being surveyed, the sample for each zone will be even point. In some cases data analysts may also want to link smaller. Consider, for example, a survey with a total individual bullocks or other large animals, or individ- sample of 3,000 households. About half of that number ual pieces of capital equipment. The same general can be expected to engage in agricultural activities.5 methods may be used, but again there is very little Thus in a country with 3 distinct agroclimatic zones experience in doing this. there will be, on average, about 500 agricultural house- holds in each zone. These relatively small samples have UNITS OF MEASUREMENT. It is inherently difficult for direct implications for the design of the questionnaire. analysts to use data that are not standardized in mean- For example, if sharecropping (or orchard farming or ingful terms. When surveys are fielded in developing some other farming practice) were common only in countries, respondents frequently provide answers one of the zones and if only 10 percent of households using nonstandard units and terms. For example, in the in that zone engaged in sharecropping, it xvould not be LSMS surveys done in Ghana, fewer than 5 percent of worth asking a large number of detailed questions on the responses used standardized units (in the metric sharecropping because they would apply to just a few system or the English system).There were also a large households. In other words, a detailed analysis of share- number of nonresponses to questions concerning the cropping would not be possible due to the very small conversion of local units into kilograms and other sample (about 50 households). metric units Uolliffe 1995). One possible solution is available for local units PANEL DATA. Causal analysis of agricultural activities that are containers, such as a hollowed-out gourd of a can benefit from the collection of panel data. For certain size.The best way to determine the weight of example, panel data can be used to estimate produc- say, rice, in such a container is to fill it full of rice and tion functions that remove certain biases caused by then weigh the rice (scales are common pieces of unobserved "fixed effects" (see Chapter 23 for a equipment for collecting price data, as explained in detailed discussion).They can also be used to examine Chapter 13, and thus the survey team should have at the impact of certain kinds of government programs least one). The weight of one "gourd" of rice can be on agricultural outcomes and on the welfare of agri- recorded in the community questionnaire or the cultural households. A third use of panel data is to price questionnaire. This need not be done for every study the impact of household-specific shocks such as type of product that may be measured using such a crop disease. This issue is particularly important given container. Instead, at the national level a table can be the inherent risk of agricultural activities. Overall, prepared that shows the weights of various products there are several potential benefits of collecting panel in terms of a standard unit of volume, such as a liter. data. At the same time, survey designers should be This table can be used to determine the weight of aware that many of the estimation methods that make one "gourd" of other products without having to 162 CHAPTER 19 AGRICULTURE weigh each product described in terms of the local occasionally even no observations-for some items in unit of the community. some communities. Another disadvantage is the fre- An alternative way to resolve this problem is as quent unit conversion problems noted above. follows. For each transaction by a household, note the On the other hand, there are also problems with quantity (in local units) and total cost of the transac- prices collected in a community-level price question- tion (or, for bartered goods or consumption from own naire such as the one presented in Chapter 13. First, production, the estimated cost). In the community or these prices are retail (consumer) prices at the time of price questionnaire, ask knowledgeable people about the interview, and to calculate household income ana- the price per kilogram of that product; use that price lysts need data on producer prices at the time the and the total cost of the transaction to calculate the crops were harvested. Consumer prices at the time of weight, in kilograms, of all transactions by each house- interview can be misleading because prices vary over hold.9 If no one in the community can give a price per the year. In particular, prices may drop significantly at kilogram, weigh out one kilogram of the product harvest time, in which case the prices farmers receive using a scale and ask community members the value for crops sold immediately after the harvest may be of that amount. Unlike the method described in the lower than consumer prices prevailing at the time of previous paragraph, this method can also be used for the survey.A second problem with data from the price units of measure that are not containers, such as questionnaire is that some agricultural products may "heaps" and "bunches." not be sold in local markets throughout the year. For example, cocoa, coffee, rubber, and other export crops PRICES OF AGRICULTURAL PRODUCTS. In most previous may not be sold in the community other than at har- LSMS surveys, price questionnaires were administered vest times. In addition, subsistence crops, such as fod- in each community from which households were sam- der for animal feed, may not be sold in the communi- pled. In many surveys data on the prices of actual ty. The best way to resolve these problems is to design transactions have also been collected in the agricultur- the price questionnaire to collect data both on current al module of the household questionnaire. When col- consumer prices and on the wholesale prices that pre- lecting prices for agricultural products using either vailed during the one or two most recent harvest sea- method, two potential problems should be addressed. sons.The information on harvest prices should not be First, prices are needed not only for heavily marketed collected by visiting local markets, but instead may be crops but also for subsistence crops (including crops done as part of the interview of community leaders used solely for livestock feed) because subsistence (see Chapter 13). crops constitute-very roughly-two-thirds of crop In some countries there may also be a third alter- output in most rural areas in Africa and one-third to native: district-level price information on producer one-half of crop output in rural areas in Asia and Latin prices available from government agencies or from a America. Second, in some countries, many communi- public "market information service."This is most like- ties reported too few transactions in the household ly in countries with easy-to-survey agricultural condi- questionnaires to generate reliable community price tions. However, such official information is best used estimates. This second problem is exacerbated in situ- as a backup for the other two sources of price data ations where differences in household prices reflect rather than as a substitute for one or both of them. differences across households in the quality of the crop. ESTIMATING THE SIZE OF PLOTS. It can often be diffi- These two problems lead to the more general cult to obtain an accurate measurement of the size of issue of whether agricultural prices should be collect- plots.The most obvious option is to ask plot managers ed in a community-level price questionnaire, in the to estimate the dimensions of their plot or plots, but in household questionnaire, or both. The safest approach many countries they often find it difficult to do so, is to collect both types of agricultural price data, particularly when plots are irregularly shaped. Such a because each type has advantages and disadvantages. problem was encountered in the ICRISAT Burkina Community-level price data should be collected Faso survey (Matlon 1988). In that survey the inter- because it is often the case that data collected at the viewers had to measure the plots themselves, using a household level include few observations- compass and a measuring tape. This clearly yielded 163 THOMAS REARDON AND PAUL GLEWWE more accurate data than asking the plot managers for MEASURING SOIL QUALITY. Information on soil quali- estimates; when the interviewers' measurements were ty can be extremely useful, but data on land or soil compared to farmers' estimates of the size of their quality have not been collected in most previous plots, large differences were found.1 This problem was LSMS surveys Jolliffe 1995). The 1992-93 and also evident in a survey carried out in southern Haiti; 1997-98Vietnam LSMS surveys and the 1995 China researchers checked the accuracy of four informants' LSMS survey included questions about the quality of estimates of the size of 21 plots and found that the land but only in the form of a simple ranking (for margin of error in the respondents' estimates ranged example, from 1 to 5).While this kind of question can from 0 to 400 percent." be useful, there are ways to ascertain soil quality with Resolving this problem is not easy because meas- greater precision. One option would be to have soil uring plots is very time-consuming for interviewers, scientists from the country's NARS examine the plots especially when the number of plots to be measured is to obtain these data. This could be very expensive. A large.The standard method to measure plots is to use less precise but more practical approach would be to a tape measure and a compass. More recently, devise a series of local or folk classifications for soil Geographical Positioning System technology has quality and land configuration and then pretest them appeared that uses satellite signals to pinpoint the loca- in local languages. This has been done effectively in tion of any object on earth to within 100 feet. Yet India by Dvorak (1988), in Nigeria by Dvorak (1993), there is little experience with using this technology to and in Burkina Faso by Matlon (1988) and Prudencio measure plot size in developing country situations, and (1983). This approach is offered as one of the code such a method might be very inaccurate for small options (local or folk soil type) in the recommended plots. More experience in the use of Geographical module discussed in the next section of this chapter, Positioning System technology for this purpose would because it is less costly than soil testing but probably be extremely valuable. Another option would be for more accurate and useful than simple ranking. One the interviewer to ask all of the other questions con- disadvantage of folk classifications is that they may not cerning the plot and then send out a separate team, be very consistent across communities; however, this possibly lent by the country's National Agricultural may also be true for "official" land quality classification Research System (NARS), to measure the plots. This schemes such as those found in China andVietnam. two-step procedure has been adopted successfully in a variety of hard-to-survey situations.12 This would be LABOR. In the agriculture modules of almost all previ- justified only in places where a pilot test has shown (by ous LSMS surveys, each household member has been comparing respondents' answers with actual plot asked to give summary information on the work done measurements) that respondents do not know the size on the family farm during the previous week and in of their plots, so that relying on respondents' estimates the previous 12 months. In most cases this approach would generate very large measurement errors. If the has yielded inadequate data, for several reasons. First, pilot test shows that respondents' estimates are reason- the "past week" data are not very useful because the ably accurate, questions about plot size can be asked in week used has a 50-60 percent chance of falling out- the household questionnaire, and these direct meas- side a crop production season in areas with one rainy urement methods will not be needed. In situations in season and roughly a 20-30 percent chance of doing which the pilot test shows that the respondents cannot so in areas with more than one rainy season. Second, accurately estimate the size of their plots, it will be the type and amount of labor needed on the farm necessary to conduct an independent measurement of varies greatly throughout a given cropping season, each plot using one of the methods discussed above. because there are peaks and troughs associated with This is most likely to be the case in situations where tasks such as preparing plots (for planting), weeding, the plot shape is irregular or changes by season and and harvesting; the same holds true for livestock pro- where the operator does not measure in standard area duction. During slow periods household members units or even in commonly used local units. Of course, may need to devote only part of the day to farming, independent measurements will raise the cost of the while the rest of their time is devoted to leisure, survey, though the magnitude of this added cost will housework (including collecting firewood and making vary widely from one survey to another. repairs to the house), and off-farm employment.Third, 164 CHAPTER 19 AGRICULTURE labor input per hectare generally varies among the RAINFALL. Data on rainfall have almost never been plots on a farm depending on the crop grown, the collected in previous LSMS surveys. Since rainfall technology used, the soil type, and the operator's often differs substantially from year to year, among sea- access to labor. Fourth, analysts need to know the sons within a given year, and across communities, ana- amount of labor used per crop to estimate a produc- lysts would ideally like to have data for each commu- tion function for that crop. nity for each season. In practice, very few communities A much better approach to collecting data on have such data.Yet on a more positive note, in some labor in agriculture is for the interviewer to ask ques- countries rainfall data for each season can often be tions on a task-by-task and plot-by-plot basis.There are obtained from rainfall charts kept by district branches two ways for this to be done. One is to ask plot man- of the country's NARS or International Agricultural agers how much labor time (measured in person days) Research System (IARS).This could be useful because was spent on different tasks (plot preparation, planting, respondents at both community and household levels weeding and harvesting) for each plot. The other is for are unlikely to be able to recall rainfall levels with any each household member to be asked how much time degree of precision. If the survey has a strong focus on he or she spent doing each of these tasks for different agriculture, a generous budget, and relatively few rural plots of land.The first approach was used in the 1995 communities (less than 100), survey designers may LSMS survey in Northeast China and is used in the want to consider whether it is worth the cost to hire standard version of the agricultural module inVolume people to measure rain in each community covered by 3. The second approach was used successfully in the the survey. However, this would have to be planned ICRISAT Burkina Faso survey (see Matlon 1988) and well ahead of time-at least 12 months before the first is used in the expanded version of the draft agricultur- interviews were to begin. al module.While one might object that this approach is more time-consuming than the approach used in Recommended Questionnaires for the past LSMS surveys, this is not necessarily the case.The Agriculture Module disaggregated method may be quicker and easier than more aggregated methods in which the household This section introduces a short version, a standard ver- head struggles to estimate how much time is spent on sion, and an expanded version of the agriculture mod- given crops for the household as a whole. The two ule.The three versions are provided inVolume 3 of this approaches can be compared using a pilot test. book. On the other hand, it may be best to enumerate Each version of the agriculture module was certain tasks (for example, the maintenance of irriga- designed to address the policy questions raised in the tion infrastructure or marketing) at the level of the first section and the data needs presented in the sec- farm rather than the plot. It is generally best to distin- ond section. All three versions are divided into 6 sub- guish three types of labor: family labor, hired labor modules labeled Part A through Part F Part A collects (either tenants or permanent or casual laborers), and information on agricultural land. Part B asks the exchange labor. In addition, one may want to distin- household about its inventory of agricultural equip- guish laborers by whether they are men, women, or ment, such as tractors, threshers, and pumps. Part C children. gathers information on the amounts of each crop har- It is better to put questions about farm labor in vested, and what was done with these crops. Part D the agricultural module than in the employment collects data on inputs used in agricultural activities, module (see Chapter 9), for two reasons. First, it is best such as fertilizer and pesticides. Part E is designed to to ask all questions about tasks and use of inputs (such gather information about livestock, and Part F asks as animal traction or the application of chemicals or about the use of agricultural extension services. The manure) together, as this helps the plot manager short and the standard versions of the draft modules remember details of both. Second, respondents usually are presented in full. To avoid needless repetition, only find it easier to remember how farm labor was those parts of the expanded module that differ from deployed on individual plots rather than across the the versions in the standard module are presented. whole farm, and only the agriculture module deals The purpose of the short version is to collect data with farm plots. on agricultural assets owned by the household (land, 165 THOMAS REARDON AND PAUL GLEWWE farm equipment, and large livestock) and to record Box 19.2 CautionaryAdvice summary information on crops grown and inputs pur- chased. Data on assets are useful to obtain an approxi- *How' much of'the draft module is new and unproven? In its basic design, the agricultural module in Volume 3 follows mate measure of each household's wealth and the the approach taken in many previous LSMS surveys, in form that wealth takes. Data on crops grown can be that it contains submodules covering land, capital, output used to classify agricultural households into different and marketing, input use, livestock, and services. types, such as producers of export crops and producers However, it also contains several substantial innovations. of subsistence crops. Such crop data are also useful for First, the recommendation that the recall period be the calculating a rough measure of the incidence of any previous agricultural season (if there is only one season taxes or price subsidies for specific crops. Similarly, data per year) or the previous two agricultural seasons (if on purchased inputs can he used to estimate the ici- there are two seasons per year) differs from past prac- dence of taxes or price subsidies on agricultural inputs. tice in LSMS surveys, although it is standard in farm man- agement surveys. This is not a risky innovation and The short module also collects summary information indeed should make the survey more accurate and eas- on use of agricultural extension services, to see which ier to implement. A second innovation is collecting data households benefit from these services. This module is on outputs and inputs by plot.This has been done only considerably shorter than the agriculture module used in a few recent LSMS surveys, such as those in China, in previous LSMS surveys, and is intended for use in Nepal, and Vietnam. This approach is also standard in surveys for which agriculture issues are only of minor farm management surveys, which makes it a low-risk interest.This module does not collect the information innovation.Third, collecting data on land quality is inno- vative for LSMS surveys, but the methods suggested here required to calculate household income from agricul- are well-tried in farm management surveys. A fourth tural activities. innovation is that the respondents are the plot managers The standard version collects a large amount of rather than the single household member best informed information on agricultural activities, and can be used about agricultural activities. Fifth, the draft module col- to calculate household income from these activities. lects data on labor use by task by season, and by plot. In addition, this module can be used to estimate pro- The fourth and fifth innovations also involve little risk duction functions using the plot as the unit of obser- because they are common in farm management surveys. vation. It can also be used to estimate cost functions How well has the module worked in the post? The agri- and cultural module used in past LSMS surveys generated profe Moduleis s e t lner data that have been underused. This is not surprising than those used in past LSMS surveys, hut the time because key variables were frequently missing or per- required to administer it may be no longer than the ceived by potential users as likely to be very inaccurate. time required in those surveys because the interview- Thus it is fair to say that the agricultural module has not er and the respondent probably had to discuss the worked well in the past. additional detail in this version just to complete those Which ports of the module most need to be customized? past versions. Moreover, interview time can be The land and input use parts (Parts A and D) require by reduced by collecting detailed plot data for only a far the most pretesting and customization. Part C on crop output and disposition requires the second-most p customization. The sections of the draft questionnaires LSMS survey conducted in northeast China in 1995. pertaining to capital (Part B), livestock (Part E), and use This version of the agricultural module should be of agricultural extension services (Part F) should require used when one of the main objectives of the survey is relatively little customization. However these differences the analysis of agriculture issues. It should also be used in customization are all on a relative scale. The average when the decision has been made to collect total amount of customization required for the agricultural household income (see Chapter 17 for a full discus- module is quite large compared to the customization required in most other modules of LSMS surveys, so ftiscoc) require in mos othermodulesof LSMSsurvey, The expanded version collects all the inforimation because farming systems and agricultural and land policy the exanded version collects alle information issues differ greatly from country to countryTherefore, in the standard version and adds detailed information extensive pretests should be done, and the team doing on land transactions in the past five years. In addition, the pretesting should include specialists in conducting it collects more detailed information on labor inputs, agricultural field research in develop ng countries. both of household members and of hired laborers.This version of the agricultural module should be used 166 CHAPTER 19 AGRICULTURE when the main objective of the survey is to study agri- who should respond to certain questions, which sec- cultural issues. tions to include, and which questions to include in a All three versions of the agriculture module pre- given section. sented in this book are merely starting points for Before examining these three versions in detail, developing a module to fit any particular country; the one should review the general rules about question- great variety of agricultural systems and issues across naire formatting in Chapter 3. These rules are partic- developing countries implies that survey teams must ularly important for the agricultural module because adapt the agricultural module to their circumstances many codes, including unit codes, land area codes, and and to the issues that they want to explore in depth. In crop codes, are used to fill it out. Applying these sim- some countries substantial changes will be needed. For ple rules will greatly reduce errors in filling out the example, in transition economies where the land mar- questionnaires and should also reduce interview time. ket is privatizing, the survey team may want to collect Two formatting rules are especially important. information on land transactions during the past few First, the codes should be consistent across different years in the standard version, and perhaps even in the parts of the agricultural module (and indeed across the short version. Another example is countries where entire household questionnaire). For example, if one urban households have small farms or gardens such as developed crop codes with rice as 1, wheat as 2, and "dachas" in Russia; the survey team may want to maize as 3, these codes would need to be the same on develop a new submodule that focuses on these types every page of the module that asks about rice, wheat, of activities. Oliver (1997) provides an LSMS-type or maize. Second, the codes needed to fill out any par- questionnaire for the countries of the former Soviet ticular page of the questionnaire should appear some- Union, including a detailed agricultural module. Yet where on that page of the questionnaire or, if there is her design of the agricultural module is based on pre- not enough room, on a facing page or a laminated vious LSMS surveys and thus does not take into sheet that provides all the codes. account of some of the suggestions of this chapter, such as the collection of data on each plot of land. Short Version A final point regarding the different versions of As explained in the third section, this version of the the agricultural module is that they should be thought agricultural module is used when agriculture is only of as three points along a continuum of possible levels of minor interest and agricultural data are collected of detail. One could create a hybrid version that lies primarily for analyzing nonagricultural issues. The halfway between the short and standard versions or data obtained are not as accurate as those collected in halfway between the standard and expanded versions. standard and expanded modules because the questions The key to success is to develop very specific objec- are directed to a single household member (the mem- tives for the module, and to design the module with ber best informed about agricultural activities). A final those objectives in mind. difference between the short version and the two more extended versions is that the recall period of the Annotations to the Recommended short version is the past 12 months instead of the past Questionnaires two cropping seasons, since there is no intention of matching outputs with inputs to estimate a production This section provides detailed notes on the three ver- function or any other causal relationship. sions (short, standard, and expanded) of the agriculture The following notes provide specific information module presented in Volume 3. The notes serve three about the design of the short version of the agricul- purposes: to explain the recommended module, to tural module. point out where the survey team might introduce modifications to suit specific circumstances, and to flag PART A. potential difficulties related to survey questions, sug- A.2. The main purpose of this question is to get a gesting how interviewers can minimize these prob- name or short description of the plot for reference lems when interviewing households. In some cases during the interview. Only the code numbers for the modifications are suggested, including changes in plots, immediately to the left of the answers provided, response codes for questions, ways of posing questions, are needed for data analysis; the data entry operator 167 THOMAS REARDON AND PAUL GLEWWE need not enter the names of the plots in the electron- tion economies of Eastern Europe and the former ic files. The one exception to this rule is the case in Soviet Union, as well as for the socialist countries of which panel data on plots are collected; in this case the East and Southeast Asia. name of the plot would be useful in the electronic files for matching plots when reinterviewing the same A.10. In regions of certain countries land transactions households in a future survey. may be rare, in which case respondents would have a hard time answering this question. In such cases one A.3.The codes for land area often vary by country; the possible response could be "DON'T KNOW." appropriate codes should be obtained by consulting However, if at all possible interviewers should try to the ministry of agriculture or the local National obtain an estimate-however rough-of the value of Agricultural Research System (NARS), if there is one. the land. A.4. The different types of land will also vary by coun- PART B. The list of items for which these questions are try; again the appropriate codes should be obtained asked must be customized for each country. from the ministry of agriculture or the local NARS. Further comments on this question are provided B.3-B5. If joint ownership of these items is rare below in the notes for the same question in the stan- (which can be determined during the field test) these dard agricultural module. questions may be dropped. In this case the instruction in uppercase letters in Question 2 should be dropped A.5-A.6. If the land has been rented out for all of the and the instruction in Question 6 should be modified. past 12 months, there is probably little reason to ask what crops are grown on it. The respondent may not PART C. The answers to these questions will be even know what crops are grown on the land. approximate because they are not being asked for each plot. The list of crops must be customized for each A.6. In most countries there is a large number of dif- country; again the appropriate codes should be ferent crops; each crop can receive a different code obtained from the ministry of agriculture or the local number (as is done in Part C).The code numbers used NARS. do not fit in the space available in Question 6.Yet it is very convenient for the interviewer (and thus reduces C.2. The main purpose of this question is to obtain a errors in filling out the questionnaire) to have crop rough idea of the quantity of each crop produced by code numbers easily accessible. One way to do this is the household. This information can also be used to to have the codes printed on the facing page of the classify agricultural households by the kinds of crops questionnaire. Another possibility is to have a laminat- they grow-for example, distinguishing producers of ed sheet with crop codes (and other codes) that the export crops, such as coffee, cocoa, or rubber, from interviewer can set next to himself (or herself) during producers of domestically consumed food crops. each interview. The codes used should include a code for "FALLOW" for cases where no crop is grown on a C.3. The main purpose of this question is to obtain an field in a given season. approximate estimate of who benefits from price sub- sidies. Technically speaking, farmers only benefit from A.7. In countries where there are several different price subsidies if they sell their crops. The information kinds of irrigation systems, and differences between gathered by this question is also useful because it gives them have important implications for the productivi- a rough idea of the impact of price changes on farm- ty and value of the land, one could add a question that ing households. Households that sell few of their crops asks for the type of irrigation on the plot. will be more insulated from price changes than house- holds that sell most of their crops. A.8-A.9. These codes regarding how the land was acquired and the type of ownership rights must be PART D. customized to the circumstances in each country. D. 1. Fertilizers, kinds of manure, pesticides, herbicides, Customization is particularly important for the transi- and fungicides should all be referred to by their 168 CHAPTER 19 AGRICULTURE explicit names, either brand names or generic names. inputs used in the production process (Part D), and This will vary significantly by country. The appropri- finishing with similar questions for livestock (Part E) ate names should be obtained from the ministry of and a general discussion of access to agricultural serv- agriculture or the local NARS. ices (Part F). Activities involving the transformation of agricultural products into processed foods or other D.3. The codes for the source of purchased inputs agricultural goods are recorded in detail only in the must be modified for each country. household enterprise module (although the fact that some of the household products were used for this PART E. These purpose of these questions is to obtain purpose should be recorded in Part C of the agricul- a rough estimate of the stock of animals. Only large tural module), since such activities generally do not livestock are included. The types of livestock will vary depend on whether the raw materials were produced by country; the questionnaire must be modified by the household or purchased from some other accordingly. source. Questions regarding assets (equipment), input PART E In some countries there may be different kinds transactions, livestock, and agricultural services (Parts of agricultural extension organizations or agents. If so, B, C2, D3, E, and F) are asked of a single household the questions in Part F should be modified to distin- member-the person who is best informed. Detailed guish between the different types. questions regarding plots (characteristics, products, and inputs) are asked of each plot manager. As discussed in F.1, F.2, F.7, £8 £13,AND F.14. The difference between the second section, this should reduce recall error with what constitutes a visit by a household member to an only a small increase in interview time. agent and a visit by an agent to a household is usually The following notes provide specific information clear. However, in some countries the distinction may about the design of the standard version of the agri- be less clear, such as when an agent comes to a meet- cultural module: ing held by farmers near their homes. This distinction will depend on the nature of agriculture extension PART A. This part is divided into three different sets of services in the country; survey designers should seek questions. The first, Al, collects information on plots the advice of the ministry of agriculture and the local of land owned and farmed by the household. The sec- NARS in designing these questions. ond, A2, collects information on plots of land rented from other households, and the third, A3, gathers data F.3 AND £9. As with Question 6 of Part A, the crop on plots rented out by the household.The appropriate and animal codes should be visible to the interviewer, questions to ask on the ownership and renting of plots either on the opposing page or on a laminated code of land can vary greatly over countries, so a large sheet. amount of customization is needed. F.15. After this question, the interviewer may want to PART Al. ask what kinds of crops or animals were discussed dur- A1.2. This question obtains a name or brief descrip- ing the visits, as is done in Questions 3 and 9 for vis- tion of the plot, for reference during the interview. its to extension agents. Asking for the name of each plot should work well in farming systems with relatively few plots per house- Standard Version hold, such as Vietnam (with an average of 5)-but it The standard version of the agriculture module pro- becomes increasingly difficult in systems with more vides data that can be used to calculate net farm plots, such as China and Peru (which average 9) or income and to estimate both production functions and Burkina Faso (with an average of 10-15). Another reduced form relationships.The order of the submod- method may be needed if farmers do not have names ules is designed to follow the contours of a typical or simple brief descriptions for their plots. One possi- conversation with a farmer-first establishing the bility is to combine information on the plot manager, stock of land and equipment (Parts A and B), then ask- the location of the plot, and the primary crop grown ing about crops produced and marketed (Part C) and into a plot numbering system.13 For example, the 169 THOMAS REARDON AND PAUL GLEWWE interviewer tells the plot manager, "Let's talk about to approximate the plot size, even in local units.This is household plot 4 now, your cotton plot 100 yards from often the case in farming systems where mainly sub- the residence." It may help to have a blank page in the sistence crops are produced and there is little reason to questionnaire for the interviewer to draw a rough map have accurate measures of one's plots. In some cases that can be referred to during the interview.A second households know the area of their major plots and issue is that in some countries farmers may exclude cash crop plots, but not of minor or subsistence crop fallow or pasture plots, or plots with a failed harvest. In plots. If the pretest shows that this is the case, the sur- such cases interviewers must be trained to prompt vey team should consider adding an explicit plot each respondent (using the code list in Question 5) measurement component to the survey. This will about plots that he or she would tend to omit. A final increase survey cost, but may be the only way to get a point is that only the code number for the plot reasonably accurate measure of plot size in a rural (immediately to the left of the answers to Question 2) economy in which the majority of rural incomes are is needed for data analysis; the data entry operator from subsistence farming. One option for measuring need not enter the name of the plot in the electronic plot size would be to hire a separate team, possibly lent files. However, if panel data on plots are to be collect- by the NARS, to take the measurements. In most cases ed (see Chapter 23), including the name of the plot in the cost would be relatively low. Finally, note that the electronic files would be useful for matching plots accurate plot size data are usually not needed for fal- when reinterviewing the same households in a future low, pasture, or wood lot plots. survey. A1.5. The different types of land will also vary by Al.3. The general approach of Part Al is to ask the country, and again the appropriate codes should be best-informed household member to list the plots obtained from the ministry of agriculture or the local owned and farmed by the household and to indicate NARS. which household member manages each plot. All fur- ther questions about the characteristics of each plot are A1.6. In most countries there will be a large number addressed to the plot managers. Because plot manage- of different crops, each of which can receive a differ- ment arrangements vary widely over farming systems, ent code number (as is done in Part C). All of the this approach may have to be altered to fit the coun- code numbers cannot fit in the space available in try studied. In a "centralized" system (usually found on Question 6. Yet it is convenient for the interviewer smaller farms with fewer plots and centralized opera- (and thus reduces errors in filling out the question- tions such as irrigation), the household head manages naire) to have crop codes easily accessible. One way to operations on all plots. In this case question 3 can be do this is to have the codes printed on the facing page omitted because the household head or the household of the questionnaire. Another approach is to prepare a member "most knowledgeable about agriculture" can laminated sheet with crop codes (and other codes) answer all the questions for each plot. that the interviewer can set next to himself (or her- self) during each interview. One crop code should be A1.4. The codes for land area often vary by country; "FALLOW" indicating that no crop was grown on a the appropriate codes should be obtained by consult- field in a given season. In cases in which two or more ing the ministry of agriculture or the local National crops are cultivated on the same plot in the same sea- Agricultural Research System (NARS), if there is one. son, detailed farm management surveys often attempt It is important to pretest this area question carefully. to determine how much of the plot is assigned to The survey team should note two points in pretesting, each crop.This could be done here by asking an addi- adapting, and posing the question. First, in many tional question about the percentage of the plot countries farmers can only recall plot sizes in local devoted to each crop. If the two crops are inter- units. The pretest will easily reveal this, and local agri- cropped (planted together), a separate question asking cultural researchers usually have the conversion coeffi- about this could also be added. A final point about cients from local units to hectares. Second, in situa- Question 6 is that the interviewer should be instruct- tions in which plots are numerous, small, and ed to ask about what crops were planted, not what irregularly shaped, it may be quite difficult for farmers crops were harvested, and the interviewer should 170 CHAPTER 19 AGRICULTURE probe the respondent about this; otherwise farmers crop mix (cover), agronomic techniques (plow or may omit crops that failed. not, mulch or not, terrace/bund or not), and time since last fallow, one can compute an index of soil A1.7. In countries where there are several different erosion and degradation that is very useful in pro- kinds of irrigation systems, and differences between duction functions. (See Byiringiro and Reardon them have important implications for the productivi- 1996 for an illustration.) ty and value of the land, one could add a question that asks for the type of irrigation on the plot. Questions PART A2. Questions 2-5 are identical to Questions could even be asked about how the irrigation is man- 2-5 in Part Al, so the comments above apply here. aged. Similarly, Questions 7-8 are the same as Questions 6-7 in Part Al, and Questions 9-12 are the same as A1.8-A1.9. These codes regarding how the land was 11-14 in Part Al. acquired and the type of ownership rights must be customized to the circumstances in each country. A2.1l. A related question, useful for some types of These questions will be inadequate for farming sys- research, is the length of time that the household has tems in a socialist or "transition" economy. For two been farming this land. examples in East Asian transition economies, see the LSMS survey questionnaires used in northeast China A2.13. These codes need to be adapted to each coun- in 1995 and inVietnam in 1992-93 and 1997-98. For try. In addition, more detail could be added, such as countries of the former Soviet Union see Oliver which relative and how far away the landlord lives. (1997). A2.14. These codes also need to be modified for each A1.10. In regions of certain countries land transactions country. may be rare, in which case respondents would have a hard time answering this question. In such cases one A2.17. In some countries it may be common for live- possible response could be "DON'T KNOW." stock to be used as in-kind payments for renting. In However, if at all possible interviewers should try to such cases the "CROP CODE" column must include obtain an estimate-however rough-of the value of code numbers for livestock as well. Codes for livestock the land. One possibility in such difficult cases is to products may also be needed. For example, a common drop the question entirely and ask about land prices in arrangement in the Sahel has one pastoral family the community questionnaire. obtaining user rights to a pasture plot in the dry sea- son in return for milk provided to the landholder. A1.11-A1.14. These questions collect general infor- mation about the characteristics and quality of each PART A3. Questions 2-3 are identical to Questions plot. Other characteristics could be added depending 2-3 in Part Al, so the comments above apply here. on the nature of agriculture in the country and the Similarly, Questions 6-7 are the same as Questions issues of greatest interest to policymakers.An alterna- 4-5 in Part Al, and Questions 8-15 are the same as tive to these questions on quality is directly measur- 7-14 in Part Al. ing soil quality or other plot characteristics, but this is much more costly. Additional questions to consid- A3.4-A3.5. If the answer to Question 4 shows that the er adding are the distance to the plot from the land was rented out for only one season, presumably it household's dwelling, local or "folk" soil classification was farmed by the household members in the other codes, and, for tree crops, the fraction of the plot area season and thus was listed in Part Al. In this case there with trees that are still too young to bear fruit.. is no need to ask Questions 6-15; instead, one should Usually there is information available from agricul- simply note the plot code from Part Al and the inter- tural researchers on local farmers' "folk" soil classifi- viewer can proceed to Question 16. However, if the cation systems and how these correspond to scientif- answer in Question 4 is that the land was rented out ic classification of soils. 14 Combined with in both seasons, it presumably was not listed in Part Al information about simple categorization of slope, and the interviewer should proceed to Question 6. 171 THOMAS REARDON AND PAUL GLEWWE A3.16. These codes need to be adapted to each coun- manager for each plot is that in the relatively decen- try. In addition, more detail could be added, such as tralized farming systems commonly found in "hard- which relative the plot is rented from and how far to-survey" situations there is no single household away the landlord lives. member who knows the details of each plot. In some "easy-to-survey" situations a single person may be able A3.17. These codes need to be modified for each to respond. Part Cl could easily be modified to target country. only one respondent (by dropping the column for the name of the plot manager in Question I), but it is still A3.20. See the comment above on Part A2 Question useful to collect output data by plot and by season. In 17. contrast, there is little reason to collect disposal infor- mation by plot or by season; indeed, if the harvest from PART B. many plots is stored in a single place and then divided PART B1. The types of farm equipment must be mod- up for different uses it may be difficult, if not impossi- ified to fit the specific country. Some items, such as ble, to do so. On the other hand, in some countries rice winnowers, may pertain to specific crops. In these each plot manager may tightly control disposal of the cases the type of crop should appear in the name of products of his or her plot, in which case the questions the item. on disposal must be addressed to the plot manager for each plot, which would mean moving all of the ques- B1.3-B1.5.Joint ownership is most likely for tractors, tions in Part C2 into Part CI.This could significantly mills, and irrigation equipment. If joint ownership is increase the interview time if the same crop were rare for any item, these questions can be "blacked out" grown on many different plots. in the lines with these items. Ifjoint ownership is rare As discussed in the second section, the questions for all items, which can be determined during the field in Part C must explicitly distinguish between the dif- test, these questions can be dropped. In this case the ferent cropping seasons (as opposed to asking about instruction in uppercase letters in question 2 should be "the past 12 months"). dropped and the instruction in question 6 should be nmodified. PART Cl. There is not enough room on this page to provide all the crop codes for the interviewer. The best B1.6. It may be difficult for the respondent to state the approach is to list the codes on the facing page. This is price of each item as there may be no market for the not only more convenient for the interviewer but item, the item may be home-produced, or the item should also reduce error. An alternative is a plastic lam- may have been bought long ago and the household is inated sheet that lists crop codes and other codes as not familiar with current prices. In these cases, the sur- well. vey team may need to add a question to the commu- nity questionnaire regarding prices, but this should be C1.l. This list of plots will be used not only for Part avoided if possible because these would only be aver- C1, but also for Parts DI and D2.This being the case, age prices. A final possibility is that an item may be it may be useful to have this question as a fold-out flap obsolete and thus not sellable for any price. In this case in the questionnaire, similar to the list of names of the price should be recorded as zero. household members in the household roster.The sim- plest way to list the plots is to list the plots in Part Al PART B2. The list of hand tools needs to be modified in order, then list the plots in Part A2. In countries to fit the circumstances in each country. where households commonly own large numbers of plots, two pages may be required (which would also PART C. The general approach is to ask each plot man- imply two pages for Dl and D2). ager to recall, for each cropping season, the production of crops on each plot, then to ask a single household Cl.2. In most countries room for four crops per plot member (the one most knowledgeable about the over two seasons should be sufficient, but in some household's farming activities) about the disposition of countries room for five, six, or even more may be crops. The reason for recall of production by the plot needed. One might also ask about whether any of 172 CHAPTER 19 AGRICULTURE these crops were intercropped with each other, but in reasons for sales or to find out more about the location most cases it is more convenient to ask this in Part A. or periodicity of the sales. Another issue is that crops with long-growing cycles, such as roots, tubers, bananas, and some fruit trees, are C2.6-C2.11. Crops may be disposed of in other ways, harvested little by little throughout the year. It may be such as presenting some as gifts to relatives or neigh- hard for respondents to recall these harvests, and bors or saving some for seed. If these are common pretests will suggest the best way to resolve this prob- ways of disposing of crops, questions along those lines lem. One approach is to ask about the typical off-take should be added here. Disposal in the form of gifts per week or month during the season; for example: may be hard to recall because it often consists of many "How many cassava root pieces did you dig out of small transfers; losses are the hardest to recall because your plot per week in this season?" Finally, note that this requires retmeimibering what was harvested, sub- nonstandard units may be common in "hard-to-sur- tracting all other uses, and looking into the storage vey" situations. Methods for resolving this problem are bins or sheds to see what is left. Only in the case of discussed in the second section of this chapter. theft or loss of large amounts is it easy to recall losses. Pretests will reveal how difficult these questions are in PART C2. a specific situation and whether the questions need to C2.1. The list of crops must be adapted to each be modified. country. In some cases the list could be 60-70 items. It is good survey practice to explicitly ask for each C2.9. This includes only crops that were processed crop, since farmers may forget "minor" crops if they then sold to others. This is considered a household are simply asked to provide a list of the crops they business, and these activities are covered in detail in the cultivate. household business module (see Chapter 18). Crops that were processed then consumed by household C2.2-C2.11. Nonstandard units may be common in members should be included in Question 10. "hard-to-survey" situations. As discussed in detail in the second section, there are several methods for C2.11. In some farming systems the harvested crop is resolving this problem. stored in nonthreshed form, and in others it is stored in threshed form. The latter should be the standard, C2.4. In theory there is no need to ask for "total sales," and if the unit is cited in nonthreshed form, that form as that can be deduced from the unit price and num- should be noted. Coefficients that convert from ber of units sold. But it may be common for a respon- threshed to nonthreshed units should be provided by dent to recall easily the total cash; if this is the case, a the survey team in the documentation for the survey, column for "total sales" can be added. Another reason based on consultation with the ministry of agriculture for a total sales column is that in some situations crops and any NARS. The question can be simplified in are bartered, in which case the value of total sales can farming systems in which the respondents normally be in the form of in-kind payments. Finally, in some recall in one or the other form. situations the seller does not get full payment at once, even within the recall period. The total payment in PART D. such cases may be divided into how much has been PART Dl. This section collects plot-by-plot informa- received so far and how much is still owed. tion on the labor time of household members on dif- ferent tasks and in different seasons. An alternative, C2.5. This question establishes to whom a sale was more detailed way to collect such data is provided in made. The codes need to be adapted to each country. the expanded version of the agriculture module. In Other possible codes arc a government cereal market- contrast to the expanded module, individual house- ing agency or an export firm. Additional questions can hold members are not identified in the standard ver- be added for detail on output marketing-for exam- sion; rough estimates of the time spent working in ple, whether the sale was in response to, or part of, a agriculture can be obtained from the employment public campaign to increase commercialization of cer- module (see Chapter 9). Putting detailed agricultural tain crops. Questions can also be added to find out the labor questions in the employment module is verv dif- 173 THOMAS REARDON AND PAUL GLEWWE ficult because it would be hard to design the employ- Another issue is whether each of these questions ment module in a way to get plot-by-plot data. Also, should be asked separately for men, women, and chil- since much of the labor may be that of the plot man- dren. That would increase the length of the question- ager, it is easier for him or her to recall all the agricul- naire and presumably the interview time, but the ture information at the same time, as opposed to split- increase in interview time may not be very large. This ting it between the employment and agricultural can be evaluated in the pilot test of the questionnaire. modules (which may not be administered on the same day). Finally, as discussed in the second section, it is PART D1.6 AND DI.AL Harvesting labor may be quite best to collect all agricultural data in terms of the 12- difficult to recall for crops, such as cassava, that are har- month period that contains one or two complete agri- vested little-by-little over the season or between sea- cultural seasons. This would also be hard to do in the sons. Pretests should reveal ways of modifying the employment module, which in general collects data question to recall such labor, but it may be necessary for the past 12 months and the past 7 days. to come up with a rate and apply that to the output. In some situations, which can be discerned in For example, if it takes 1 hour to dig up a basket of pretests, it may be best to enumerate certain tasks, such potatoes, and 100 baskets of potatoes were harvested as marketing or the maintenance of irrigation infra- from the garden, about 100 hours of labor were need- structure, at the farm level. Information on such activ- ed. Dividing this by 8 hours per day yields about 12 ities could be asked of thc household member most days of labor for potato harvesting. informed about agriculture, and could be put in a set of household-level questions at the end of DI. This PART D2. Certain labor tasks may be done collective- could include time spent by household members car- ly (with other households) as "exchange labor." One ing for livestock. common variant of this is the communal work party that goes from farm to farm and performs a single D1.. As mentioned above, this question need not be task. In Part Dl there could be a separate question on filled out if Question 1 of Part Cl is written on a fold- such donations of labor on other people's farms. out flap, as is done with the names of the household Receipt of such labor should be treated as receipt of members in the household roster. "hired" labor and noted in D2. (A question could be added to distinguish such labor from "ordinary" hired D1.3 AND D1.8. In some farming systems (such those labor.) as in West Africa), animal traction rental is done as labor hire, with the laborer bringing his own plow and D2.1. As mentioned above, this question need not be team. In India, one usually rents bullocks but uses one's filled out if Question I of Part Cl is written oni a fold- own labor and one's own equipment. In some coun- out flap, as is done with the names of the household tries one can rent traction equipment and supply one's members in the household roster. own bullocks or horses, or rent a tractor and supply the diesel fuel. In general, if the arrangement is one D2.3-D2.4. An alternative for hired labor is to ask where household members rent equipment or draft about the labor days spent in each of the different tasks animals but supply their own labor, the labor time listed in question 4. This would lengthen the ques- should be noted in these questions. If the system is one tionnaire and thus this option has been left to the in which labor is hired and the laborer brings his oNvn expanded version. equipment or uses the equipment of the household, that labor time should be noted in Part D2. In some D2.13 AND D2.24. The types of fertilizer must be countries one may want to ask who supplied the adapted to the country of the survey. This list of codes equipment. (The cost of renting such equipment is should be developed in consultation with the ministry recorded in Part D3.) of agriculture and any NARS in the country. D1.3-D1.6 AND D1.8-Dl.11. These are the main tasks D2.16, D2.18, D2.27, AND D2.29. The types of manure in most countries, but in certain situations other tasks used must be adapted to the country of the survey. could be listed, or these tasks could be aggregated. This list of codes should be developed in consultation 174 CHAPrER 19 AGRICULTURE with the ministry of agriculture and any NARS in the D3.3 AND D3.6. These codes must be adapted to the country. circumstances of the country. D2.20 AND D2.31.Very specific codes for the different PART E. The information on anirmals is gathered over pesticides, herbicides, and fungicides must be provid- the t2-month period that contains the last two crop- ed. As with other agricultural inputs, the list of codes ping seasons (as opposed to the 12 months immedi- should be developed in consultation with the ministry ately preceding the day of the interview), to be con- of agriculture and any NARS in the country. sistent with the crop data in the agricultural module. Part E is designed to focus on farm households that PART D3. Prices for the different farm inputs were not hold livestock, rather than pastoral households that obtained in Part D2. The reason for this was to avoid engage in transhumance or other mobile systems. A needless repetition, since the same input could be survey of the latter type of household would require applied to many crops, and it is only necessary to ask much more detail on livestock activities. for the price once.The prices are obtained more con- cisely in Part D3. If prices vary from season to season, E.2. The types of animals listed will vary by country. one may want to make this distinction in Part D3.This In some countries it may be useful to distinguish could be done by dividing Question 2 into two ques- between mature animals and young animals. This can tions, one for each season. be done either by having separate lines for adult and young animals, or by distinguishing between the two D3.1, LINES 1-11. Lines 1-ll in this grid should be types in selected questions. replaced with explicit names of fertilizers, types of manure, pesticides, herbicides, and fungicides that are E.3. Livestock holdings questions can be very sensi- commonly used in the country. tive subjects, except in farming systems in which there are few livestock. In some farming systems, D3.1, LINES 12-15. Provision is made for recording livestock are an important store of wealth and self- different wages paid for different tasks. It is usual for insurance, a status symbol, and a means of financial there to be such variation, and such variation should freedom for family members (in relation to other be fairly easy to recall. However, there may also be dif- members of the family). In Senegal, for example, it is ferences in wage rates depending on whether men, difficult to solicit information from wives concern- women, or children are hired. In this case more lines ing the animals they own because they do not want can be added that distinguish between wage rates for the household head to know their wealth, which men, women, and children. Finally, hired labor is usu- would reduce their autonomy (Kelly and others ally paid an hourly, piece, or daily wage, perhaps with 1993). Where there is livestock taxation, households an in-kiind payment such as a meal. Rather than enter- may also be reluctant to provide accurate informa- ing into details on the latter payments, the respondent tion. In an LSMS-type survey it is best not to spend is asked simply to make a rough guess of daily wage too much effort trying to ascertain livestock holdings including in-kind payments such as meals. More very accurately. In most farming systems the data will detailed information on hired labor is obtained in part probably be reasonably accurate, and greater accura- D2 of the expanded module. If piece rate payments are cy would probably require much longer contact with common, this could be accommodated in the standard the households. In farming systems in which live- module by adding "PER HECTARE," "PER BAS- stock holdings are more complex (mainly in hard-to- KET," or similar codes into the unit code box. survey situations), it would be better to interview each household member, but this can absorb large D3.2-D3.3. One or more questions could be asked amounts of time and create tension. The person best about credit arrangements at this point, such as informed at the household level can provide a rough whether credit was obtained to purchase some or all of estimate, and where he or she does not have a good the input, and the terms on which credit was idea, it is unlikely that the interviewer can use anoth- obtained. For further discussion on collection of cred- er method to get a better estimate in a single visit to it information see Chapter 21. the household. 175 THOMAS REARDON AND PAUL GLEWWE E.6, E.11, E.14-E.17, AND E.20. Barter is often more Expanded Version common in livestock transactions than in transactions The expanded version of the agriculture module takes in other agricultural products; provision can be made the standard module as its starting point, adding Parts for this by asking about in-kind payments in a separate A4 and A5, which obtain information on land transac- question in systems in which such payments prevail. tions during the past 5 years, and replacing Part D of the standard module with a more detailed Part D. The E.15. More detailed questions can be added concern- comments below apply only to Parts A4 and A5 and ing animal health expenditures and services in coun- the expanded version of Part D. For details on the tries in which animal disease is an important policy parts of the expanded module that come from the issue, such as in the trypano zones of Africa. standard module without any modification, see the comments above. E.18-E.20. This question should exclude any animal byproducts that are used as inputs for food processing PARTS A4 AND A5. The questions in Parts A4 and A5 are that is part of a household business. Such activities are merely a point of entry for survey teams that wish to covered in the questionnaire module on household add a series of questions on the impacts of land enterprises (see Chapter 18). If such food processing reforms, decollectivizations, or other land market poli- activities are important, the use of animal byproducts cies. The general approach is to ask the best-informed for that purpose could be noted here by adding ques- household member to describe all land transactions. tions similar to 18 and 19. However, in some farming systems plot managers have a large degree of autonomy in buying and selling their .19. The types of fresh byproducts will vary by coun- plots, so they may not need to obtain the permission of try. The codes should be developed in consultation the head of household or even to inform the head. with the ministry of agriculture and the local NARS. Such situations should be ascertained in the pretest; in countries where this is the case, provision must be PART E In some countries there may be different kinds made to administer these questions to all plot managers of agricultural extension organizations or agents. If so, in the household. A final general comment is that ask- the questions in Part F should be modified to distin- ing questions about the past 5 years is rather arbitrary. guish between the different types. The length of the recall period should depend on the specific conditions and policy issues in each country. F.1-£2, F.7-F.8, AND F.13-F.14. The difference between a visit by a household member to an agent PART A4. In countries where farmers are expanding and a visit by an agent to a household is usually clear. production by bringirng new land under cultivation, However, in some countries the distinction may be one may want a separate section on such activities. The less clear, such as when an agent comes to a meeting interviewer could ask the best-informed household held by farmers near their homes. Here the distinction member to describe land clearing activities to find out will depend on the nature of agriculture extension how much land the household cleared in the past five services in the country; survey designers should seek years, the use rights the household now has on that the advice of the ministry of agriculture and the local land, and the use to which the land is being put. The NARS in designing these questions. arnount cleared may be hard to establish if the land is fragmented, partially cleared, or irregularly shaped. F.3 AND £9. As with Question 6 of Part Al, the crop This is important to ascertain in a pretest. It may also codes and the animal codes should be visible to the be hard for households to value this land in areas interviewer, either on the opposing page or on a lam- where one cannot rent or buy land (because there is inated code sheet. no land market), or if the household in question has little experience in the land market. Another issue is F.15. After this question the interviewer may want to that the household might need permission from the ask what kinds of crops or animals were discussed dur- village chief or other public authorities to clear land, ing these visits, as is done in Questions 3 and 9 for vis- and may consider that the equivalent of title. Finally, its to extension agents. questions concerning land clearing and land title may 176 CHAPTER 19 AGRICULTURE be quite sensitive because clearing may be officially A5.7. In some places, selling land does not necessarily banned, so the interviewer should be instructed to mean losing the right to produce on it. For example, assure the respondent that all answers provided will be in Burkina Faso, land can be transferred to another completely confidential. household but the original owner can still collect sheanuts from his trees on the land. Provision can be A4.2. Past land transactions involving family members made in the module in countries where this is the who now live outside the household may not be vol- case. unteered by the respondent unless the interviewer is instructed to be sensitive to this possibility and prompt A5.11. In some regions of the world it may be inter- for it.This comment also applies to Question 2 of Part esting to track "urban-based buyers" as a separate cat- A5. egory, in order to track absentee ownership. A4.3-A4.4. Because most of the plots acquired within PART D. The expanded version of Part D is intended the past five years will still be in the household's pos- to replace the Part D presented in the standard version session, information on these plots will already be avail- of the agricultural module. It collects information on able from Part Al or Part A3. When this is the case, it labor and nonlabor inputs in much more detail than is noted in Question 3 and then the plot code from does the standard version of Part D.The main purpose Part Al or Part A3 is recorded in Question 4. This of collecting such detailed data is to estimate produc- allows the interviewer to skip Questions 5, 6, and 7. tion functions and other causal relationships in much more detail. The data are also useful for detailed A4.5-A4.7. These questions are essentially the same as descriptive analysis of the agricultural activities of rural Questions 4, 5, and 8 in Part Al of the standard mod- households. ule. See the comments above on those questions. PART D t. The expanded version of Part D 1 is very dif- A4.7. It may be difficult to distinguish between long- ferent from the standard version. The main difference term lease and sale or gift. Also, in some cultures there is that each household member is asked about work may be sensitivity or shame attached to selling land done on plots farmed by the household. The informa- (just as anthropological research has shown that there tion collected is disaggregated by plot, season, and type is shame attached to selling grain within one's own of task. Each household member is expected to village in some regions), so the interviewer should be respond for himself or herself. This amount of detail instructed to be sensitive to the propensity of a will increase interview time, but such an increase respondent to describe a sale as a "gift." Finally, in areas should be expected since the expanded module with a mix of private and collective lands, allowing the assumes that agricultural policy issues are the top pri- reincorporation of land into the collective may be ority of the survey. considered neither a sale nor a permanent divestiture; it might be seen as "long-term lending" land to the Dl.l. Because the unit of observation for each line is cooperative or collective. Questions specific to such household member, as opposed to plot of land, the land systems should be added. These potential prob- fold-out list of plots in Part Cl cannot be used to lems should be explored in the pretest. replace question 1, unlike question 1 of the standard Some additional questions to consider adding are module for Part Dl. Instead, the fold-out list of house- what kind of ownership rights the household has hold members that is part of the household roster (see (similar to Question 9 in Part Al), whether the land Chapter 6) should be used. was purchased on credit and, if it was, the terms of that credit. See Chapter 21 for further details on collecting D1.2-D1.16 AND D1.18-D1.32. For each season, credit information. detailed data are gathered for the three plots on which the household member spent the most time. In addi- PART A5. These questions are mirror images of the tion, more cursory data are collected for three other questions in Part A4; thus the comments in Part A4 plots in each season (Questions 34 and 36). In some generally apply to the same questions in Part A5. countries there may be a need to increase the number 177 THOmAS REARDON AND PAUL GLEWWE of plots for which detailed data are collected, and per- D2.3-D2.10 AND D2.12-D2.19. The four main tasks list- haps even the number of plots for which cursory data ed in these sets of questions (preparing and sowing the are collected.The pilot test of the questionnaire should field, applying inputs, weeding and pruning, and harvest- reveal whether this is the case. ing) could be expanded into a more detailed list of tasks, but this would increase the time required to complete D1.2, D1.7, D1.12, D1.18,D1. 23, D1.28, D1.34, AND the questionnaire.The level of detail of the tasks depends D1.36. Plot ID codes are provided in Parts Al and A2. on the specific policy questions to be addressed and on the agriculture system prevailing in the country. D1.3-D1.6, D1.8-D1.11, D1.13-D1.16, D1.19-D1.22, D1.24-D1.27, AND D1.29-D1.32. These four D2.3, D2.5, D2.7, D2.9, D2.12, D2.14, D2.16, AND activities-preparing and sowing the field, applying D2.18. For some or all of these tasks, labor days could inputs, weeding and pruning, and harvesting-should be disaggregated into adult male, adult female, and suffice for most crops in most countries. Some data child labor days. This would lengthen the question- analysts may want a more detailed list of activities; this naire but might be xvorthwhile under some circum- can easily be provided but it will increase the inter- stances. If this were done it wvould be useful to disag- view time. Other activities may be needed for certain gregate payments for hired labor in the same way. types of crops; for example, smudge pot operations may be needed for tree crops. Labor used to maintain D2.4, D2.6, D2.8, D2.10, D2.13, D2.15, D2.17, AND or repair agricultural equipment is not explicitly men- D2.19. In countries where in-kind payments to hired tioned. In most cases respondents will include this farm laborers are common, survey designers could labor with the task for which the equipment is used- split each of these questions into two questions, one that is, maintenance of plowing equipment will be for cash payments and one for in-kind payments. included in plot preparation and repair of irrigation However, before expanding the number of questions, equipment will be included in time spent irrigating the purpose of the survey and the results of the pretest the plot. should be reviewed to see whether this is worthwhile. D1.6, D1.11, D1.16, D1.22, D1.27, AND D1.32. For PART D3. The only difference between Part D3 of the cassava and other root crops that are harvested little- expanded questionnaire and Part D3 of the standard by-little over the year, respondents may have difficulty questionnaire is that the expanded version has fewer answering questions on days spent harvesting. For sug- items for questions 1-3. Because payments to hired gestions on how to handle this, see the note on labor arc recorded in Part D2 of the expanded ques- Questions 6 and 11 of Part Dl of the standard version tionnaire, there is no need to ask about prices paid for of the agricultural module. different kinds of hiired labor in Part D3. D1.37-D1.39. These questions on hours per week Notes spent on animal husbandry activities work best for households where raising animals is not the main The authors are very grateful to Bonnie Banks and Andrea activity. In agricultural systems that where raising live- Ramirez for helping to create the questionnaire. They are also stock can be a full-time activity, much more data are thankful for comments from Harold Alderman, Christopher needed on labor inputs. Barrett, Jeanine Braithwaite, Michael Carter, Carlo del Ninno, Margaret Grosh, Courtney Harold, Juan Munioz. Scott Rozelle, PART D2. The main difference between the expanded Kinnon Scott, and the participants at two LSMS seminars. Finally, and the standard versions of Part D2 is that the expand- Reardon thanks Peter Matlon for years of training on farm surveys. ed version adds much more detail about hired labor. 1. While it is analytically convenient to think of the household as a single entity that has one utility function, it is more realistic to D2.1. Unlike Question I of Part Dl of the expanded assume that each household member has an individual utivty func- version, this question can be removed if Question 1 of tion.As explained in Chapter 25, this distinction can have impor- Part Cl is placed on a flap that is visible when the tant policy implications.Yet the discussion in this subsection is quite questions in Part D2 are being asked. general and thus applies in either case. 178 CHAPTER 19 AGRICULTURE 2. The main exception to this general statement is that in some location ("by the paved road on the south bank of the rivcr"), by socialist countries the government may stipulate that certain crops manager ("my first wife"), and often by principal crop ("a cotton should be grown. For example, in some parts of China households field"). are required to grow rice in order to meet quotas for rice produc- 14. A practical approach would be to devise a soil classification tion. However, this practice is becoming rare even in socialist and land configuration scheme and to pretest it in local languages, economies. There may also be cases where the government direct- asking the plot operator for the soil type and land configuration. ly rations scarce agricultural inputs, but this practice, too, is becom- This has been done effectively by Dvorak (t988) in India, Dvorak ing rare. (1993) in Nigeria, and Matlon (1988) and Prudencio (1983) in 3. Another use of causal analysis is to estimate parameters that Burkina Faso. researchers can use for applying optimization models. These mod- els can be used to derive what farmers should do to maximize prof- References its, or to optimize some other objective. 4. Whether farm size and capital holdings are really exogenous Adesina, A. A., and K. K. Djato. 1996. "Farm Size, Relative is a matter of debate. The answer often depends on the specific Efficiency and Agrarian Policy in Cote d'lvoire: Profit details of the data and of the agricultural system in the country Function Analysis of Rice Farms:' Agricultural Economnics 14 (2): where the data were collected. 93-102. 5. The ability of the three different versions of the agricultural Ahmed, Raisuddin, and Cynthia Donovan. 1992. "Issues of module to answver these questions will depend on whether the agri- Infrastructural Development: A Synthesis of the Literature:' cultural sector of the country in question is hard to survey or easy IFPRI Occasional Paper. International Food Policy Research to survey These two kinds of situations are discussed below. In gen- Institute, Washington, D.C. eral, the ratings in Table 19.1 are an average of the two situations. Ahmed, J., Walter Falcon, and Peter Timmer. 1989. "Fertilizer 6. Many regions of Nepal are hard-to-survey situations, but Policy for the 1990s" Harvard Institute for International some (such as the Terai region) have characteristics that make them Development Discussion Paper 293. Cambridge. Mass. easy to survey. In such cases the household questionnaire needs to Ainsworth, Martha, and Jacques van der Gaag. 1988. Guidelines for be designed to handle hard-to-survey situations. Adapting the LSMYS Living Standards Questionnaires to Local 7. Two other examples in xvhich detailed data were not gath- Conditions. Living Standards Measurement Study Working ered for "minor" plots are the 1988-90 multiround farm survey in Paper 34. Washington, D.C.:World Bank. Senegal by the International Food Policy Research Institute and Bardhan, Pranab. 1973. "Size, Productivity, and Returns to Scale:An the Senegal Agricultural Research Institute (Kelly and others 1993) Analysis of Farm-Level Data in Indian Agriculture." Journal of and the 1981-85 ICRISAT survey in Burkina Faso (Matlon 1988). Political Economy 81 (6): 1370-86. 8- Since couintries wvith hard-to-survey areas tend to he more Barrett, Christopher. 1 996. "On Price Risk and the Inverse Farm rural, this proportion will be higher in those countries. Size-Productivity Relationship." Journal of Development 9. Survey designers should resist the temptation to have inter- Economnics 51 (2): 193-215. viewers make these conversions in the field and then write down . 1999. "The Effects of Real Exchange Rate Depreciation quantities in standard units in the questionnaire. Data analysts can ois Stochastic Producer Prices in Low-Incoiise Agriculture." do the conversions much more quickly and wvith far fewer errors. .4gricultural Economics 20 (3): 215-30. 10. In the Burkina Faso survey, Matlon asked the plot managers Barrett, Christopher, and Michael Carter. 1999. whether the plot was the same size as it had been the year before. "Microeconomically Coherent Agricultural Pohcy Reform to Most said yes. For each plot said to be the same, he measured the Africa." In JoAnn Paulson, ed., .4frican Economies in Transition, plot and compared the data with measurements taken the year Volume 2: The Reform Experience. London: Macmillan. before, often finding a 100 to 200 percent difference. Besley, Timothy. 1994. "How Do Market Failures Justify 11. The authors thank Kinnon Scott for this information. Interventions in Rural Credit Markets?" World Bank Research 12. This method has been used in Burkina Faso (see Matlon Observer 9 (1): 27-47. 1988), Niger (see Hopkins and Reardon 1989), Senegal (see Fall Binswanger, Hans, and P. Pingali. 1988. 'Technological Priorities and others 1989), and Rw.anda. for Farming in Sub-Saharan Africa." World Bank Research 13. In the Burkina Faso ICRISAT survey (1981-85), the house- Observer 3 (1): 81-98. hold head named the common field or fields that he or she man- Blank, Lorraine, and Margaret Grosh. 1999. "Building Social Policy aged at the household level, then named the individual plots man- Analysis Capacity in Conjunction with Household Surveys." aged by household members, referring to them by approximate World Bank Research Observer 14 (2): 209-27. 179 THOMAs REARDON AND PAUL GLEWVVE Bviringiro, E, and Thomas Reardon. 1996. "Farm Productivity in Griliches, Zvi. 1957. "Specification Bias in Estimates of Production Rwanda: Effects of Farm Size, Erosion, and Soil Conservation Functions." Jourfnal of Farm Economics 39 (1): 8-20. Investments." Agricultural Economics 15 (2): 127-36. Grosh, Margaret, and Paul Glevvwe. 1995. A Guide to Living Carter, Michael, and K. D. Wiebe. 1990. "Access to Capital and Its Standards Mleasurement Study Surveys and Their Data Sets. Living Impact on Agrarian Structure and Productivity in Kenya." Standards Measurement Scudy Working Paper 120. AmericanJournal ofAgricultural Economics 72 (5): 1146-50. Washington, D.C.:World Bank. Clav, D., F Byiringiro, J. Kangasniemi, Thomas Reardon, B. Gudger, Michael. 1990. "Crop Insurance: Failure of the Public Sector Sibomana, L. Uwamariya. and D. Tardif-Douglin. 1995. and the Rise of the Private Sector." In D. Holden, P Hazell, and Promoting Food Security in Ru'anda Through Sustainable Agricultural A. Pritchard, eds., Risk in Agriculture: Proceedings of thle Tenth Productivity: Meeting the Challeniges of Population Pressure, land Agriculture Sector Synrposunm. Washington, D.C.: World Bank. Degradation, and Poverty. Michigan State University Jolliffe, Dean. 1995. "Reviewv of the Agricultural Activities Module International Development Paper 17. East Lansing, Mich. from the Living Standards Measurement Study (LSMS) Commander, Simon, ed. 1989. Structural Adjustment and Agriculture. Survey" World Bank, Poverty and Human Resources Division, London: Overseas Development Institute. Policy Research Department,Washington, D.C. Deaton, Angus. 1989. "Rice Prices and Income Distribution in Hopkins, J., and Thomas Reardon. 1989. 'IFPRI Survey Thailand: A Non-parametric Ar.alysis." Economic Journal 99 Methodology: Agricultural Price Policy Reform Impacts and (395): 1-37. Food Aid Targeting in Niger." Project Document 2. DembeLe, N. N., and K. Savadogo. 1996. "The Need to Link Soil International Food Policy Research Institute, Washington, Fertility Management to Input and Output Market D.C. Development: Key Issues." In S. Debrah and W Koster, eds., Kelly, V, Thomas Reardon, A. A. Fall, B. Diagana, L. McNeilly Linking Soil Fertility Management teAgricultural Iniput and Output 1993. "Final Report of IFPRI/ISRA Project on Market Development. Lome: IFDC-Africa. Consumption and Supply Impacts of Agricultural Price Deolalikar, Anil. 1981. "The Inverse Relationship Between Policies in Senegal." International Food Policy Research Productivity and Farm Size: A Test Using Regional Data from Institute and Institut Senegalais de Recherches Agricoles, India." American Journal ofAgricultural Economics 63 (2): 275-79. Washington, D.C. Donovan,WG. 1996. Agriculture and Econtomic Reform in Suib-Sahlaran Krueger, Anne, Matnrice Schiff, and Alberto Valdes, eds. 1992. Thte A4frica. Working Paper 18. World Bank, Africa Technical Political Economic ofA gricultural Pricinig Policy Baltimore, Md.: Department, Environmentally Sustainable Development Johns Hopkins University Press. Division, Washington, D.C. Matlon, Peter. 1988. 7he ICRISAT Burkeina Faso Farm-Level Studies: Duncan, A., and J. Howvell, eds. 1992. Structural Adjustmfient and the Survey Mlethods anid Data Files. International Crops Research zlfrican? Farmer. London: Overseas Development Institute. Institute for the Semi-Arid Tropics, Andhra Pradesh. Dvorak. K.A. 1988. Indigenous Soil Clcssffication in Serni-Arid Tropical Migot-Adholla, Shem, Peter Hazell, and F Place. 1990. "Land India. Economics Group Progress Report 84. Andhra Pradesh: Rights and AgricLdtural Productivity in Gharia, Keriya anid International Crops Research Institute for the Semi-AridTropics. Rwanda: A Synthesis of Findings." World Bank, Agriculture 1993. "Characterizing System Dynamics for Soil and Rural Development Department,Washington, D.C. Mcnagement Research in the Humid Tropics: Studies in Morduch, Jonathan. 1999. "The Microfinance Promise."Journal of Southeastern Nigeria:" In K.A. Dvorak, ed., Social Science Economic Literature 37 (4): 1569-614. Research fir Agricultural Technology Development: Spatial and Oliver, Raylynn. 1997. AMcdel Living Standards Measurement Study Temnporal Dimnensions. Wallingford: CAB Int'l. Survey Questionnairefor the Couintries of the Forrmer Sovirt Union. Fall, A.A., V Kelly, and Thomas. Reardon. 1989. Household-Level Living Standards Measurement Study Working Paper 130. Survey Mlethods Used in the IFPRI/ISRA Study of Consumption Washington, D.C.: World Bank. and Supply Imipacts of Agricultural Price Policies in Senegal. Place, F, and Peter Hazel. 1993. "Productivity FifectsofIndigenous Document lI.Washington, D.C. and Dakar: International Food Land Tenure Systems in Sub-Saharan Africa." American Journal Policy Research Institute and Institut Senegalais de of Agricultural Economnics 75 (February): 10-19. Recherches Agricoles Project. Prudencio, YC. 1983. "A Village Study of Soil Fertility Ferreira, M.L. and C.C. Griffin. 1995."Tanzania Human Resource Management and Food Crop Production in Upper Volta- Development Survey: Final Report."World Bank, Population Technical and Economic Analysis." Ph.D. diss., University of and Human Resources, Eastern Africa Department, Arizona, Department of Agricultural and Resource Washington, D.C. Economics, Tucson, Ariz. 180 CHAPTER 19 AGRICULTURE Purcell, Dennis, and Jock Anderson. 1996. "Achievements and Sahn, David, ed. 1994. Adjusting to Policy Failure in African Economiies. Problems in Developing National Agricultural Research Ithica, N.Y: Cornell University Press. Systems?' Report 15828.World Bank, Operations Evaluation Savadogo, K., Thomas Reardon, and K. Pietola. 1995. Department,Washington, D.C. "Mechanization and Agricultural Supply Response in the Rahman, Aminur. 1999. "Micro-Credit Initiatives for Equitable and Sahel: A Farm-level Profit Function Analysis. "Journal of African Sustainable Development: Who Pays?" World Developrment 27 Economies 4 (3): 336-77. (1): 67-82. Schultz,T.W 1978. Distortions of.4gricultural Incentives. Bloomington, Rao,V, andT. Chotigeat. 1981."The Inverse Relationship Between Ind.: Indiana University Press. Size of Land andAgricultural Productivity." AmericanJournal of Singh, Inderjit, Lyn Squire, and John Strauss. 1986. Agricultural Agricultural Economics 63 (3): 571-74. Household Models: Extensions, Applications and Policy. Baltimore, Reardon,Thomas,T.Thiombiano, and Christopher Delgado. 1988. Md.:Johns Hopkins University Press. La Substitution des Cereales Locales par les Cereales Inport&es: La Strauss, John, and Duncan Thomas. 1995. "Human Resources: Consommation Alimentaire des Menages a Ouagadougou, Burkina Empirical Modeling of Household and Family Decisions." In Faso. Serie des Resultats de Recherche 002. Research Report J. Behrman and T. N. Srinivasan, eds., Handbook of Development for International Food Policy Research Institute-Centre Ecootmics Volume 3. Amsterdam: North-Holand. d'Etudes et Recherches Economiques et Sociales Project. Taylor, Lance. 1988. Varieties of Stabilization Experience: Towards Burkina Faso: CEDRES, Universite de Ouagadougou. Sensible Mlacroeconomics in the Third World. Oxford: Clarendon Reardon, Thomas., V Kelly, E. Crawford, B. Diagana, J. Dione, K. Press. Savadogo, and D. Boughton. 1997. "Promoting Sustainable In- , ed. 1993. The Rocky Road to Reform: Adjustment, Income tensification and Productivity Growth in Sahel Agriculture After Distribution, and Grow'th in the Developing World. Cambridge, Macroeconomic Policy Reform." Food Policy 22 (4): 317-28. Mass.: MIT Press. Rukuni, M. 1996. "A Framework for Crafting Demand-driven van Zyl, J., Hans Binswanger, and C. Thirtle. 1995. "The National Agricultural Research Institutes in Southern Africa." Relationship Between Farm Size and Efficiency in South Department of Agricultural Economics Staff Paper 96-76, African Agriculture." Policy Research Working Paper 1548. Michigan State University, East Lansing, Mich. World Bank, Agriculture and Natural Resources Department, Rusike, J., Thomlas Reardon, J. Howard, anid V Kelly. 1997. Waslhington, D.C, "Developing Cereal-based Demand for Fertilizer Among Small- World Bank. 1995. World Development Report. New York: Oxford holders in Southern Africa: Lessons Learned and Implications for University Press. Other African Regions." Pohcy Synthesis 30, Food Security II Yaron, Jacob. 1994. "What Makes Rural Finance Institutions Project, Michigan State University, East Lansing, Mich. Successful?" World Bank Research Observer 9 (1): 49-70. Sadoulet, Elizabeth, and Alain de Janvry. 1995. Quantitative Zeller, Manfred, and Manohar Sharma. 1998. "Rural Finance and Development Policy Analysis. Baltimore, Md.: Johns Hopkins Poverty Alleviation." IFPRI Policy Report. International Food University Press. Policy Research Institute,Washington, D.C. 181 q ^ ~Savings 2 O Anjini Kochar The savings module is an essential part of a multitopic household survey like the LSMS surveys.This module gathers data on the value of the household's stock of financial assets. Such data are necessary to accurately estimate household wealth, a variable that is required for research on almost all aspects of household behavior. And the savings module can collect information on both the types of financial assets held by households and recent transactions in such assets during the period of the survey- information that is directly relevant for analyzing household savings, particularly financial savings. Although policymakers in most countries are interest- in detail in other chapters of this volume. Therefore, in ed in a wide range of issues relating to household sav- order to design a survey that can aid research on sav- ing, the savings modules in most multipurpose house- ings, survey designers should read this chapter in con- hold surveys (including many LSMS surveys) typically junction with the other relevant chapters of this vol- collect information only on financial assets and liabil- ume, particularly Chapters 5, 17, and 21 on ities. However, this need not limit the research on sav- consumption, income, and credit. ings that can be done using the data, for two reasons. The first section of this chapter lists the major First, the data set generally includes information on policy issues associated with household savings. The the household's nonfinancial assets in other modules of second outlines the methodologies that can be used to the survey. This is appropriate as questions about, for analyze these issues and the specific data needed for example, the capital used in farm and nonfarm enter- such research-focusing particularly on how best to prises should be asked in the same part of the ques- measure savings.The third section discusses the design tionnaire that gathers information on how these assets of the savings module and the elements required in are used. Second, while analysts need data on assets to other modules to gather data for savings research, and address some policy issues, other issues can be presents two prototype savings modules: a standard addressed using data on income and consumption. version and a short version. The fourth section pro- Given that research on savings requires data from vides explanatory notes on the two versions of the other modules of the survey, it is important to keep module presented in the third section. such data needs in mind when designing these mod- ules. Many of the data problems inherent in undertak- Policy Issues in Household Savings ing research on savings arise from difficulties in col- lecting data on income, consumption, and transactions Governments are interested in the rate of savings (in in nonfinancial assets, difficulties which are discussed other words, the fraction of a household's income that 183 ANJINI KOCHAR the household saves) and the types of assets held by rate of growth of inputs and the productivity of inputs. households for several reasons. Perhaps the most The rate of growth of inputs reflects both the econo- important reason is the relationship between the level my's savings rate and the productivity of inputs. Thus and form of household savings and the rate of growth the savings rate and the productivity of capital deter- of a nation's per capita output. And because savings mine the rate at which output grows. It is this rela- choices also affect individual incomes, concern about tionship between output growth and savings that is the the level and distribution of household income pro- main reason policymakers are interested in household vides another motive for policy interest in saving. savings. Moreover, in recent years policy interest in savings has Historically, economists have emphasized the rela- been heightened by the recognition that people's wel- tionship between the rate of savings and income fare depends not only on their level of consumption in growth rather than the relationship between output any given period but also on the stability of their con- growth and the productivity of savings. Lewis (1954) sumption over time. Since saving provides a means to maintained that the "central problem in the theory of transfer income across periods and hence smooth con- economic development is to understand the process sumption over time, policy concern about the welfare by which a community which was previously saving of households requires knowledge of whether house- ... 4 or 5 percent of its national income or less converts holds have access to the assets that allow such itself into an economy where voluntary saving is run- intertemporal income transfers, the costs of these ning at about 12 to 15 percent of national income or assets, and differences in such costs across households. more." The central importance of savings in the The interest of policymakers in predicting the groxvth process was also emphasized in the works of level of household savings and in influencing the level Harrod (1939) and Domar (1946), as well as in Solow's and distribution of household (and hence national) (1956) neoclassical growth model. income implies that savings studies must examine not However, recent empirical evidence on the deter- only the level and form of savings but also the deter- minants of economic growth in a number of minants of savings. For example, if policymakers want economies has emphasized the importance of factor to predict how anticipated changes in national demo- productivity growth (World Bank 1991). Growth in graphic structure will affect national savings, they need factor productivities is believed to explain as much as data that enable a study of which demographic vari- 50 percent of the output growth in the United States ables influence savings and how. Similarly, if policy- between 1960 and 1985. While the contribution of makers want to use interest rates to bring about desired factor productivity growth to output growth has been changes in the level and form of household savings, much smaller in Asian, African, and Latin American they need data that enable an analysis of the sensitivity economies (particularly relative to the contribution of of savings and portfolio choices to interest rates. capital growth), as much as 50 percent of the differ- The rest of this section outlines the reasons poli- ence in growth rates across economies is attributable cymakers are interested in the rate, form and determi- to dffferences in productivity growth. Moreover, a slow- nants of savings. While a number of factors determine down in the productivity of inputs is responsible for savings, policymakers often find it easiest to influence much of the fall in growth rates experienced by most savings by intervening in the financial sector.Thus this of the world's economies after 1973. section ends with a discussion of policy issues con- While much of this evidence relates to the pro- cerning the financial sector. ductivity of all inputs, not just capital, data from a number of developing economies similarly suggest The Level and Form of Household Savings that there is more reason for concern about the pro- Policy interest in the level and form of household sav- ductivity of capital than about the rate of savings. ings exists because of the effects of savings on nation- Official estimates show that in recent years the popu- al income as well as their effects on both current and lations of poor countries (including many economies fiuture household incomes and consumption. of Sub-Saharan Africa) have routinely managed to save at a rate equal to approximately 20 percent of their EFFECTS OF SAVINGS ON NATIONAL OUTPUT GROWTH. country's GDP. However, the rate of return on these Growth of output in an economy is determined by the savings appears to be very low-about 8 percent a 184 CHAPTER 20 SAVINGS year-meaning that the high savings rate has con- from future income fluctuations. However, like the tributed little to economic growth (World Bank effect of saving on household income, the usefulness of 1989). saving for smoothing consumption depends on which It is widely believed that the low return on assets assets the household has chosen to invest in. This is in developing economies partly reflects the fragment- because assets may differ in terms of their liquidity. ed nature of capital markets and, hence, the inability of Thus, to assess whether households have the means to households to hold the assets that yield the highest smooth their consumption over time, policymakers rates of return.This explains policy interest in financial need to know not just the level of a household's sav- intermediation-in the spread of financial institutions ings but also what kinds of assets it owns. and the willingness of households to entrust these institutions with their savings. Policy issues relevant to The Determinants of Savings and of Portfolio Choices the spread of financial intermediation will be discussed Several theories explain the determinants of house- in greater detail later in this section. hold savings and portfolio choices. Because these the- ories are predicated on different motives for house- EFFECTS ON INDIVIDUAL INCOMES. While the aggregate holds to save, they sometimes yield contrary savings level and the productivity of savings affect the predictions for how any given set of variables, includ- growth rate of national output, the level and especial- ing policy instruments, will affect household and ly the forms in which households save also affect national savings. This fact underscores the importance household incomes, particularly in countries where of research on the determinants of savings and portfo- agricultural or nonfarm enterprises constitute a major lio choices. source of household income. (This is the case in most developing economies; see Chapter 18 on the house- DETERMINANTS OF SAVINGS. Much of the current hold enterprise module and Chapter 19 on the agri- research on savings stems from the "life-cycle model" culture module.) Income from agricultural or non- (Modigliani and Brumberg 1954). This model assumes farm enterprises reflects, in part, the household's that people's tastes or preferences are relatively stable ownership of physical capital or "productive" assets throughout their life cycles, while their income is sta- such as the machinery and tools used in such enter- ble only during their working years, falling to zero at prises. Investment in such assets represents an act of retirement. Correspondingly, an individual's savings saving, thereby linking savings and portfolio choices to will peak in his or her prime earning years and fall as household income. the savings are drawn down to finance consumption Because of this link between a household's income during retirement years. Since this model assumes that and its stocks of productive assets, developing country the young save while the old consume in excess of governments have for a long tinme encouraged house- their income, it predicts that any changes in the holds, particularly poor households, to invest in such incomes or numbers of young people relative to old assets. Indeed, projects aimed at encouraging households people will affect national savings.Thus it predicts that to invest in productive assets have been at the heart of economies in which there are more prime-age earners many poverty reduction programs, such as the Integrated than elderly people will show positive aggregate sav- Rural Development Programs implemented in the ings, and that any increase in the relative size of the eld- 1970s and 1980s in a number of Asian economies. erly population will reduce national savings. It also pre- dicts that higher productivity growth-which increases THE IMPORTANCE OF SAVINGS AND PORTFOLIO the incomes of the young-will increase savings. CHOICES FOR SMOOTHING CONSUMPTION. Recent An alternative model of savings, the "precaution- research has shown that the welfare of households ary savings model," theorizes that even though life depends not only on the level of their consumption in cycle motives may exist, the primary motive for saving any given period but also on their ability to protect is not to guard against reductions in income in later this consumption from income fluctuations-in other life, but rather to protect consumption from annual or words, to "smooth" their consumption across periods seasonal uncertainty in incomes (Deaton 1989). This within a year or over several years. Saving provides one model implies that assets must constantly be run down means by which they can protect their consumption to protect consumption from income fluctuations. 185 ANJINI KOCHAR Thus, according to this model. household savings will members, particularly members who work, interven- as often be negative as positive and will average zero tions that improve health and sanitary conditions may over the years. If precautionary savings are important, increase household investment in productive assets. national savings are likely to be low in economies Assessing the link between health inputs and savings wvhere a significant proportion of the population is choices requires data from the health module of the employed in occupations with fluctuating short-run multitopic household survey (see Chapter 8). incomes, such as agriculture. Growth in occupations with more stable incomes may cause precautionary Financial Intermediation motives for savings to be replaced by life cycle con- In this chapter, the term "financial sector" refers to cerns-increasing national savings. The precautionary "formal" financial institutions-institutions either reg- savings model also implies that policy interventions ulated or owned by the government. Such institutions affecting the volatility or uncertainty of incomes and include banks, life insurance companies, and housing expenditures (for example, programs aimed at stabiliz- finance institutions, as well as any businesses, post ing prices or output or programs that provide unem- offices, or other government agencies that accept ployment or disability insurance) will significantly household savings (either in the form of shares or increase aggregate savings. deposits) but do not offer loans. "Informal" institu- tions, institutions not regulated by the government, DETERMINANTS OF PORTFOLIO CHOICE. In order to include moneylenders as well as relatives and friends of encourage investments in specific assets, policymakers household members who provide loans to households. need information on what determines the choices that Informal institutions also include group savings and households make about their savings portfolios. For credit associations, such as the susu men in Ghana and example, governments need to know what factors Gambia, the tontines in Senegal, and the hui in China. underlie the demand for financial assets before they Some of these associations only maintain deposits for can implement policy intended to increase deposits in their members, while others also provide their mem- financial institutions. It could he that a low demand for bers with loans. financial assets primarily reflects difficulties in with- The formal financial sector in many developing drawing money from accounts in financial institutions. economies is small, both in absolute terms and relative If so, policies that reduce such difficulties-say, by to the informal sector. (Chapter 21 on credit provides introducing passbook savings schemcs-may have a far statistics on the relative sizes of the formal and infor- greater impact on the number and amount of deposit mal sectors in a number of countries.) One common accounts than policies that increase interest rates measure of the financial depth of an economy is the offered by the accounts. percentage of the country's gross domestic product Understanding the determinants of total savings held either in currency or in bank deposits (including also helps policymakers understand households' port- currency, demand, time, and savings deposits). In 1993 folio choices. One implication of the precautionary this percentage was less than 15 percent in some savings model is that households (particularly ones that African economies, including Sierra Leone, Uganda, lack access to sources of credit for consumption as Guinea Bissau, and Ghana. In South Asian economies opposed to production) underinvest in illiquid it ranged from 33 percent in Bangladesh to 44 percent assets-such as productive capital-because they need in India (World Bank 1995). to maintain their stocks of currency, food grains, and There are several good reasons why the govern- other liquid assets in order to meet consumption ments of developing economies are keen to develop needs (Morduch 1994). In this case lack of investment the formal financial sector. The primary reason is to in productive assets primarily reflects a household's make the savings of households available to investors uncertainty about its income; thus any policy inter- so as to enable households to realize the welfare gains vention that reduces this uncertainty would increase from "trading" funds with other households over time the household's willingness to invest in productive (see Besley 1995 and Chapter 21 ofthis book).While assets. This in turn requires policymakers to under- the informal sector can serve this purpose, its geo- stand the sources of income variability. If such vari- graphic scope is relatively limited because informal ability partly results from the ill health of household sector transactions primarily occur between borrowers 186 CHAPTER 20 SAVINGS and lenders who are well acquainted with each other. In addition to being concerned about the devel- Economies with relatively large informal credit sectors opment and profitability of the financial sector, poli- thus tend to have fragmented capital markets, in which cymakers are interested in assessing whether formal investors only have access to the funds of savers they financial institutions directly benefit households either are linked to through informal institutions. In practice by providing opportunities to invest in financial assets this has meant that costs of credit and, equivalently, or by providing credit. Later sections of this chapter returns to savings, may vary tremendously across address the data requirements and research method- households. Indeed, equity concerns about the access ologies for evaluating the impact of formal financial of all households, particularly poor ones, to low-cost institutions on savings. Chapter 21 discusses how the sources of credit underlie much of government inter- availability of formal credit can benefit households. est in developing the formal financial sector, even though available evidence on formal financial institu- Data Requirements and Research tions in a number of developing countries suggests Methodologies that formal institutions do not always perform better than informal ones in this regard. This section discusses the data required to study the Policymakers in most developing countries are level, forms, and determinants of savings and other interested in knowing the level and spread of the for- issues relating to financial intermediation.This section mal financial sector across regions and among house- also discusses some of the methodologies commonly holds within a given region, both in absolute terms used in empirical savings research. and in relation to the informal sector. Concerns about The primary requirement for any research on sav- the level of financial intermediation are closely linked ings, whether it be motivated by interest in the level to concerns about the profitability of formal financial and forms of savings or in the determinants of savings, institutions, which depends on both the volume of is some measure of savings. This section starts by dis- these institutions' business and the costs of doing busi- cussing two alternative ways to measure savings: by ness relative to the returns. Net costs are determined subtracting consumption from household income and by such factors as the interest rate on deposits, the by observing changes in stocks of individual assets. interest rate at which loans are made, the ability of Since savings can be measured using income and institutions to collect on their loans, and other costs consumption data alone, is it necessary to include a of servicing deposit accounts and loans, including savings module in a multitopic household survey? In administrative and transactions costs. Many of these most cases, yes. Policymakers are interested not only factors are influenced by bank procedures and poli- in the amount of savings but also in household port- cies, which affect both borrowers' repayment incen- folio composition and issues related to the financial tives and incentives to maintain deposits. The factors sector-both of which require data on assets. Data affecting net costs are also influenced by the govern- on financial assets are best collected in the savings ment's monetary and fiscal policies, which directly or module. indirectly determine inflation and interest rates, and In this section the discussion on different ways to by education and infrastructural development poli- measure savings is followed by a discussion on types of cies, which affect transaction costs and the costs of assets for which data should be collected. Next are dis- training bank personnel (Gurgand, Pederson, and cussions of the data needed to inform financial sector Yaron 1994). Other critical inputs into the profitabil- research and of the benefits for such research of using ity of financial institutions include the willingness of panel data relative to using a single cross-section of households to maintain bank deposits and repay loans. data. The section concludes by considering whether it While household surveys may not provide all the is necessary to disaggregate the household data in the information needed to comprehensively analyze the savings module to the individual level. effects of government and bank policies on financial institution profitability, they do provide the means to Measuring Household Savings analyze the importance of several factors that may Empirical studies have measured savings by subtracting explain households' willingness to use formal financial consumption from household income and also by institutions. observing changes in household assets. Several studies 187 ANJINI KOCHAR have presented results from using both measures whereas the reference period for consumption data is (Paxson 1992;Wolpin 1982). generally a month or a week. The necessary extrapo- This subsection first discusses each method of lation of annual consumption from monthly data may measuring savings. Then the two methods are com- yield misleading estimates. This is because even if pared using data from the LSMS surveys in Pakistan households smooth their consumption against fluctu- and Ghana, to assess their relative merits and thus ations in income, consumption may vary from month determine what data are needed to measure savings. to month as a result of, for example, price fluctuations. Savings may be overestimated for households that are SAVINGS AS INCOME MINUS CONSUMPTION. Most interviewed during months in which their consump- empirical studies on household savings measure sav- tion was relatively low on account of such price fluc- ings as the difference between a household's income tuations.Thus it is crucial to follow the recommenda- and consumption (Deaton 1992a, 1992b).This meas- tion of Chapter 5 that data on consumption ure is subject to all the problems that arise when meas- expenditures should be collected by asking about a uring income and consumption, including the diffi- "usual month." culty of obtaining accurate measures of the income of A final problem relates to the separation of labor the self-employed, measures of the consumption of income and asset income. Models of savings are usual- home-produced goods, and measures of the value of ly used to test the relationship between consumption inputs that are only imperfectly marketable. Chapters (or savings) and exogenous changes in income. Such 17 and 5 of this book detail the problems in obtaining tests require that labor inconme be separated fronm asset accurate measures of income and consumption in income, since the model is testing a theory about asset household surveys. Survey designers should read these income. For households with a family farm or with a chapters carefully to understand the biases inherent in nonfarm household enterprise, it is generally not pos- measuring savings as the difference between income sible to separate labor income from asset income, since and consumption and how, to some extent, these bias- the profits realized from the business represent a return es can be overcome by improving available measures. both to the family's labor and to any fixed assets used When using income and consumption data to to produce this income. Thus, even when households measure savings, it is also necessary to be aware of do report payments to members engaged in family some issues that are not as important when the intent enterprises, it is not possible to ascertain whether this is simply to measure income or consumption. For is a return to labor or to capital. Researchers have example, when calculating income for the purpose of commonly dealt with this problem by assuming that measuring savings it is necessary to allow for the farm profits primarily reflect a return to the family's depreciation of capital inputs and the appreciation of labor. Measurement error in income and potential various stocks. In addition, the consumption measure biases caused by endogeneity of labor choices are then used should include only the value of the services pro- addressed using instrumental variable techniques (see vided by the household's current stock of consumer the statistical appendix in Chapter 26). Nevertheless, durables, while the actual investment in those durables inaccuracies remain. The assumption that farm profits should be included in the savings measure.This neces- primarily represent a return to family labor is more sitates dividing consumption goods into durables and likely to be true in countries such as Cote d'Ivoire, nondurables. While such a division is clear-cut for where labor is scarce relative to land (Deaton 1992b), goods like vehicles and heavy appliances (both than in land-scarce countries such as the South Asian durables), it is less obvious for clothing, kitchen uten- economies. sils, and even jewelry. After it has been (arbitrarily) decided which goods are durables and which are non- SAVINGS FROm DATA ON ASSET TRANSACTIONS. If data durables, a value must be imputed to the services pro- are available on a household's asset transactions or on vided by each durable good. the stock of assets at the beginning and end of a refer- A further problem arises if different reference ence period, savings can be measured as the net value periods are used for collection of income and con- of transactions in all assets or as the change in a house- sumption data, as is commonly the case.The reference hold's stock of assets over the reference period. period for income data may be a year or a season, However, it is notoriously difficult to collect informa- 188 CHAPTER 20 SAVINGS tion on all the various assets in which households that health expenses reflect consumption more than invest their savings. Researchers tend to be skeptical investment, particularly in poor agrarian economies about measuring savings based either on households' where out-of-pocket expenses for preventive treat- asset transactions or on stocks of assets at different ment are often insignificant. Nevertheless, health points in time, because data on certain asset transac- expenditures do have effects that extend beyond the tions may be missing or inaccurate. current period, and this certainly merits including Particularly difficult to collect are data on transac- health expenses in measures of savings. tions in assets such as foodgrains, fodder, building A final cautionary note relates to the means by materials, seeds, and other inputs. Possibly because of which households acquire or dispose of a particular the difficulties involved in estimating quantities and asset.The value of asset transactions should not include value of such stocks, a number of highly reputable data the (imputed) value of gifts received by the household sets, including that collected by the International Crop or given by the household. If the gift of a consumer Research Institute of the Semi-Arid Tropics durable is recorded as the purchase of an asset, the (ICRISAT) in India, contain no measures of such measure of savings derived from the asset data will not stocks. If survey designers do wish to collect these equal that calculated as the difference between income data, it is probably best to collect them at the point in and consumption. One way to ensure that the value of the agriculture module when the interviewer is asking such gifts is not included in data on asset transactions questions about the household's agricultural output is to ask explicit questions about whether any assets and its disposal. To ensure the reliability of these data were acquired as gifts (in the savings module or in it is probably best to collect them by crop. other modules recording data on asset transactions). It is also difficult to obtain accurate information This procedure was used in the Pakistan LSMS. on a household's credit transactions and stock of cur- rency.1 The direction of the bias in credit transactions COMPARING SAVINGS MEASURES TO ASSESS THE may also be hard to predict. Deaton (1992a) reported RELIAILuY OF THE DATA. Data from the LSMS sur- that there were many more creditor than debtor veys in Pakistan and Ghana illustrate the difficulties households in C6te d'Ivoire and suggested that this inherent in each of the two ways of measuring savings. may be because respondents were more willing to A lack of data on important assets is a problem when report their assets than their liabilities. The opposite measuring savings using data on asset transactions. And situation appears to prevail in Pakistan, where 1,667 the difference between income and consumption does households reported receiving loans from informal not always provide a reasonable estimate of savings, sources, primarily relatives and friends, and only 252 often because of weaknesses in the design of the households reported making loans. An apparent reluc- income and consumption modules. tance to report loans made to others has also been For the purpose ofthis exercise, data on income and noted in other South Asian economies, such as India consumption are taken from the aggregate files of the (Kochar 1997). Pakistan LSMS data set. In addition to the usual prob- If savings are to be measured using data on asset lems of measurement error, these data are subject to all transactions, a survey designer has to make judgment the assumptions that were made in arriving at these calls about how to treat certain expenditures, primari- aggregates; no additional cleaning of the data has been ly expenditures on education and health-related items. done. Data on asset transactions were obtained from sev- Typically, consumption and savings research has treat- eral different modules of the survey, primarily the ed such expenditures as items of current consumption. income modules2 and the savings module.3To the extent However, research on the economics of education has possible, the value of gifts that households received was emphasized that such expenditures represent invest- excluded from measures of income and consumption ments in future income. As Gersovitz (1988) argues, (because, as noted above, such gifts do not represent an even if education is desired only as a consumption act of saving by the household); the Pakistan LSMS good, the benefits of education are spread out over a specifically asked respondents whether any reported lifetime, so these expenditures should be regarded as a acquisition or sale of assets was in the form of gifts. consumer durable. There is less agreement on how One difficulty in estimating savings in the Pakistan health expenses should be treated. A case can be made data arises from insufficient data on transactions in jew- 189 ANJINI KOCHAR elry. While data on jewelry purchases were recorded in possibility that measuring savings as the difference the inventory of durables module, this module did not between income and consumption is just as error- provide information on the sale ofjewelry. As a result, ridden as measuring savings using asset data. Survey in this exercise jewelry purchases were considered con- experts generally believe that existing ways of measur- suniption expenditures rather than savings.4 ing income tend to underestimlate the inconrc of thie Table 20.1 shows substantial differences in the self-employed. Thus, if self-employment is more wide- Pakistan LSMS between the estimates of savings spread in rural areas, rural incomes (and hence savings) derived from asset data and the estimates derived from may be underreported relative to urban inconmes. the difference between income and consumption data. Table 20.1 also provides data across income class- Using asset transactions to measure savings yielded a es, which are created by dividing households by their lower value than the value obtained by computing median level of income. It can be seen that in the savings as the difference between income and con- Pakistan survey the difference between the measure of sumption. However, all of the assets held by house- savings derived from asset data and the measure of sav- holds may not have been enumerated in the data. In ings derived as the difference between income and order to assess this possibility and to identify which consumption was particularly evident among the rich- asset transactions are most likely to have been misre- est households. Taking high-income rural and urban ported, Table 20.2 provides details on the number of households together, there is a difference of Rs. 41,130 households reporting savings and the mean level of between income and consumption. This is probably savings by type of asset.These data show a low level of due both to the undernumeration of assets (causing savings derived from loan transactions-possibly error in asset-derived savings data) and to consump- reflecting difficulties collecting data on loan transac- tion by the wealthy (causing error in measures of sav- tions. On average, households in the Pakistan survey ings as income minus consumption). sample reported net borrowings of Rs. 19,805 (Table Official statistics from the Government of 20.2), a number that is particularly suspect given the Pakistan report a domestic savings rate of 1 1 .8 percent relatively small size of the formial sector in Pakistan. in both 1990-91 and 1991-92 (National Bank of The low level of savings measured using data on trans- Pakistan 1992).5 While use of the transactions data actions may also reflect the lack of data on stocks of reveals a negative household savings rate, measuring foodgrains and fodder. savings as the difference between income and con- Although stocks of foodgrains are more likely to sumption yields a household savings rate of 16.5 per- contribute significantly to savings in rural areas than in cent. Another source of data on household savings in urban areas, the absolute value of the difference Pakistan, the Household Income and Expenditure between income and consumption is greater, on aver- Survey, showed a household savings rate of 4.6 percent age, in urban areas than in rural areas. This raises the for the year 1987-88. Table 20.1 Alternative Measures of Mean Savings, Pakistan LSMS Survey (rupees) Savings from asset dataa Income minus consumption Mean income Mean -onsumption0 (I) (2) (3) (4) All households -3,901 8,652 52,246 43,783 poorest 50% -4,621 -21,266 13,915 36,724 richest 50% -3,088 38,463 90,577 51,743 ......................... ................................................................................................................................................................................................ Urban households -1,162 19,266 65,014 45,959 poorest 50% -3,244 -15,304 20,365 37,972 richest 50% -5,200 53,962 109,783 55.036 r.ural houseoids -3.642 - 913 39,534 4;,6- i5 poorest 50% -5,099 -24,687 9,738 35,243 chest 50% -1,998 20,778 69,463 48,791 Note: Income and consumptlor are from aggregate World Bank files. Households are divided into rich and poor by treir median level of ncome. a. Savings n this columr are calculated from asset data from different modules, as detailed nTab e 20.2. b. Consumpt on is total consumption m nus the follow ng tems: education (VEXP5240), durab es (VEXP5300), ewelry (VEXP33 2), other household effeots VEX14240), ktchen equipment (VEXP4320), fumiture and fi-tings (VEXP4330), and other durable housing expenses (VEXP4390). lewelry purchases (PV33 7) are inc uced in consumption and excluded from savings. Source Authors ca culations based on Pakisian LSMS survey. 190 CHAPTER 20 SAVINGS Table 20.2 Household Savings by AssetType, Pakistan LSMS Survey All households _ Urban Rural Mean savings Mean savings Mean savings File Number in rupees Number in rupees Number in rupees Asset number reporting (standard deviation) reporting (standard deviation) reporting (standard deviation) Agricultural land fO9a4 39 -9.626 4 16,400 35 -12,600 (77,169) (42,940) (80,027) ~~~~~~~~~~~~~~~~~~.......... " '".... ... , ,,,,,........,, ,,,,,,,,, ......... , ,,,,,,,,,,,,,,,(716,,, ...............I........ , ," ,-,"I"Ill,.94)..... " 'l............. .80 Agricultural equipment fO9dm 36 50,897 4 130,250 32 40,978 (69,900) (93,682) (61,262) .......................................................... ................ ,.......................... ................ ............. .............................................. ..... ........... 1......... . .... Livestock fO9f2 1,063 -322 200 1.683 863 -787 (8,885) (10,552) (8,390) Nonfarm assets flOcl 362 8,130 216 6,817 146 10,072 (63,442) (42,174) (85,889) Business improvement f Oc2 75 9,436 5 i 11,717 24 4,587 (36,585) (44,019) (7,615) Cash fl5di 4,559 -810 2,212 -1,349 2,347 -302 (38,334) (54,787) (S,065) ................. ................................ .................... ........................... ...................... ............................ ... .........................'........................... 8 .. ..... Residential land fl 5dm 4,030 998 1,938 2,012 2,092 58 (37,220) (51,862) (13,260) Investment land fl.5dm 565 0.0236 228 0.0008 337 0.039 (0.5219) (0.609 1) (0.4539) Shares fl5d3 261 1,787 217 598 44 7,651 (35,101) (30,054) (53,578) Deposits fi 5d4 1,159 2,856 703 3,620 456 1,676 (4 1,257) (46,59 1) (3 1,3 1 6) BSisi/savings committee fl 5d5 867 -315 670 -704 197 1,004 (12,490) (12,265) (13,174) Durable goods Aggregate 4,799 1,047 2,400 1,508 2,399 585 files (8,233) (10,870) (4,122) Education Aggregate 4,799 2909 2,400 4,527 2,399 1,290 fles ( 12,663) (17,425) (3,443) H-ome improvement Aggregate 4,799 942 2,400 1,415 2,399 469 files (10,355) (13,754) (4,984) Credit fl 5b3 2,542 -19,805 1,262 -26,711 1,280 -12,997 f 15c2 (132,689) (184,185) (37,927) Source: Author's calculations based on Pakistan LSMS survey In contrast to the data from Pakistan, the data 1987-88. Not surprisingly, calculating household sav- from the Ghana LSMS yielded higher mean savings by ings at the median (Table 20.3) considerably reduced using data on asset transactions than by subtracting the discrepancy between the two measures. However, consumption from income (Table 20.3). The data on the difference was still large. Data on asset transactions asset transactions yielded positive average savings in yielded a median level of savings close to zero in both both years of the survey (3,510 cedis in 1987-88 and years of the survey, while the income minus consump- 10,168 cedis in 1988-89), whereas the income and tion measure yielded median savings of -89,237 cedis consumption data yielded average savings far less than in 1987-88 and -100,191 cedis in 1988-89. zero (-100,490 cedis in 1987-88 and -114,851 cedis Both measures of savings are suspect. Measuring in 1988-89).6 savings as the change in the stock of households' assets The considerable discrepancy between these two was bound to provide misleading estimates in this case measures of savings in Ghana partly reflects the fact because the Ghana LSMS data set included no data that the mean net asset transactions of the sample on transactions in financial assets by the household households was skewed by exceptionally high pur- although it did include data on the value of the chases of assets by a small number of households. A household's stock of these assets. The bias introduced mere 1 percent of the survey households accounted by this lack of data is probably substantial, given that for 44 percent of the total recorded purchases of con- financial assets comprised a significant percentage of sumer durables by all of the sample households in household wealth. In 1987-88, 97 percent of sample 191 ANJINI KOCHAR Table 20.3 Comparison of Savings Measures in the Ghana LSMS Survey 1987-88 1988 89 Mean savings in cedis Mean savings in cedis (standard deviation) Median savings in ced s (standard deviation) Median savings in cedis Household income' 251,576.83 170,671.4 254,71 1.12 180,833.3 (293,219.69) (256, 143.66) Household consumption 352,066.84 292,269.6 369,561.87 309,802.4 (260,702,49) (270,024.36) ................................................................................................................................................................................................................................... Savings as change in assets0 3,509.69 -600.00 10,167.72 600.0 (99,627.49) (82,846.83) ..... .................................................................................................................................................................................................................... Savings as income 100,490.01 -89,237. -I 14,850.74 -100,191.0 minus consumption0 (260,108.76) (251,159.24) Savings as income -83,822.38 -74,622.8 92,325.03 -92,103.8 minus consumption' (284.676.95) (243,969.92) a. Co ombe, McKay, and Round; 993. b. Calculated as the va ue of sva able data on net transactions in land and build ngs vestock, farm equipment, business assets, consumer durables, and cred t transac- t ons Deta Is of these transact ons are in Tab e 20.5. c. Ca culated by the World Bank. Source Author's calculations based on Ghana LSMS survey households reported owning financial assets, the mean these transactions appear to be primarily short-term, value of such assets being 15,540.80 cedis. Much of with as much as 94 percent of outstanding debt con- this wealth xvas probably held as cash. Only a minor- tracted within the reference year. While this may ity of households reported holding financial assets in reflect a relatively low demand for long-term credit, it such forms as bank deposits, deposits in other finan- also undoubtedly reflects the limited availability of cial institutions, or stocks and bonds.7 In contrast, 84 credit for financing long-term capital investment and percent of households reported owning "other" forms investments in consumer durables. Households also of financial savings, a category that includes stocks of have the option of selling some of their other assets to cash. finance purchase of consumer durables. However, this If households finance their investments in other is unlikely to be a viable option for many households assets primarily through cash transactions, a lack of as the data indicate that most of these other assets con- data on such transactions means that any measure of sist of farm and nonfarm capital, which are relatively savings based on data on transactions is likely to be illiquid. Table 20.4 reveals that, with the exception of overestimated. Evidence on types of assets purchased livestock, households report relatively few sales of their and sold by sample households supports this conclu- other assets. sion. The disaggregated data on asset transactions in It is likely that households pay a significant share Table 20.4 reveal that the positive savings estimates of the cost of purchasing consumer durables and derived from the data on asset transactions reflect a net business assets from their accumulated stocks of cash. purchase of consumer durables and, to a lesser extent, If so, an absence of data on cash transactions in a of business assets. The disaggregated data on the con- given data set may result in an inflated estimate of the sumer durables purchased by households reveal that increase in net savings when instead it should show a the high mean level of such purchases primarily shift in the composition of households' asset portfo- reflects the purchase of high-cost indivisible items lios. It is not surprising that without data on cash such as cars and television sets.8 It is unlikely that transactions, measures of savings based on asset trans- households finance such large purchases from their actions show positive mean savings for the Ghanaian current income alone. sample. One possibility open to households is to finance Other sources of error in the transactions data for the purchase of consumer durables through loans. Data Ghana are a lack of information on the household's from the Ghana LSMS survey on credit transactions stock of foodgrains and fodder and on the loans made (Table 20.4) reveal a fairly active credit market; 39 per- by the household during the reference period. The cent of households reported borrowing and 42 per- data set does, however, provide information on the cent reported loaning to others in 1987-88. However, amount borrowed by each household during the ref- 192 CHAPrER 20 SAVINGS Table 20.4 AssetTransactions, Ghana LSMS Survey, 1987-88 and 1988-89 1987-88 1988-89 Purchases Sales Purchases Sales Land and buildingsa Mean (cedis) 138.01 196.29 177.33 444.67 Standard deviation (2,040.64) (2,508.14) (3,879.92) (1 1,268.90) ................................................................................................................................................................................................................................... Frequency (42) (25) (44) (21) ......... ......... ......... ......... ......... ......... ......... ......... ......... ...................................................... .................................... ......... ......... ......... Livestockb Mean (cedis) 875.57 3,387.89 1,393.78 3,867.01 Standard deviation (7,734.67) ( 15,818.36) (9,505.17) (14,908.44) Frequency (398) (2)(496) (774) ............................................................................................................... ............................................................................................................... Farm equipmentc Mean (cedis) 452.60 288.25 330.94 43.27 ................................................................................................................................................................................................................................... Standard deviation (1I2,261.11)i ( 10,9,,47.,61i) (6.......................,126.50) ,,, ,,(1,303.27,), Frequency (32) (3) (31) (3) ................................... ................................................................I......................... .............I............................... .~).......................................... 3 Business assetsd Mean (cedis) 4,270.42 216.97 3,420.06 464.98 Standard deviation (62,121.81) (4,134.84) (35,763.10) (16.752.39) .............c"y......................................................")...................................................................... ............................................................ Fr,uenc,y ........(...................57.1).,,5......I..,,, (.5) (657) (12) Durablese Mean (cedis) 9,093.38 1,651.14 13,688.04 1,501.45 ....... r-d................... "' 'n............................ -,1-8........... ............................................................. ......................... ................................. . .... Standard deviation (85,1 18.29) (34,690.47) (73,441.02) (25,646.65) ................................................................................................................................................................................................................................... Frequency (365) (48) (589) (6 Credit transactions' Mean (cedis) 6,173.16 9,073.60 7,852.23 8,631.21 Standard deviation (24.099.16) (80,052.52) (31,352.02) (50,297.17) ..................................................... 9- .......................................................................3-)-1111,11,11111--"I............. ............................................................. Frequency (900) (843) (1 100) (996) a. Data on land purchases are from fi e 9a of the Ghana data set. Data on land sales are from files 9a and 1 4b. b. Data are from file 9f c. Data are from file 9k1 d. Data are from file I Od. e. Data on purchases of durables are from file I c and represent the value of durab es reported as being bought during 1987-88. Data on sales of durables are from file 1 4b and include income from sales of veh cles as well as from rental of other durables. f Loans made by the household are recorded under purchases; a household's borrowing or the debt that the househo d contracted during the reference year are recorded under sa es. Source: Author's calculations based on Ghana LSMS survey erence period as well as on each household's total out- to which consumption exceeds income and the corre- standing loans and debt. Because 94 percent of these spondingly large negative savings estimated at both the loans were contracted during the reference year, the mean and the median also call into question the accu- stock data on total outstanding loans payable to the racy of the consumption and income data.9 Chapter household were used as an approximation for the loans 17 on income suggests that the problem in the income made by the household during the reference year. minus consumption measure may result from an Using the stock variable would, however, overestimate undernumeration of income. Chapter 17 also suggests households' savings. several ways in which the collection of data on income Thus in the Ghanaian data set much of the dis- can be improved-in turn increasing the accuracy of crepancy between the two measures of savings may be savings estimates. It will probably be easier to put these explained by the absence of data on transactions in improvements into effect than to attempt to collect financial assets, on foodgrains and fodder, and on loans data on households' cash holdings, stocks of food- made during the reference year. However, the extent grains, or credit transactions. 193 ANJINI KOCHAR Collecting Data on Stocks of and Transactions in Specific ductive assets to estimate the profitability and riski- Assets ness of household portfolios. Even if household savings are measured as the differ- ence between income and consumption, data on Research on Financial Intermediation stocks of and transactions in specific assets can signifi- Because of the great importance of financial sector cantly enhance the value of the survey for research on development for overall economic development, poli- savings. Data on stocks of different types of household cymakers are perhaps most interested in financial assets are of interest to policymakers because they pro- assets.This subsection outlines methodologies and data vide information on the productivity of assets and, requirements for researching the policy issues con- hence, on the contribution of savings to both house- cerning financial institutions that were identified in hold and national income. Data on transactions in spe- the first section of this chapter. The relevant policy cific assets are useful because they provide insights into issues include the level and distribution of financial issues of specific concern to policymakers, such as the intermediation, the effects of government and bank determinants of savings. For example, Rosenzweig and policies on financial intermediation, and the impact of Wolpin (1993) used regressions of sales and purchases the development of the financial sector on house- of bullocks on measures of income variability to holds. explore whether households use productive assets to Financial institutions have two distinct, though smooth consumption. Udry (1995) used data on trans- related, objectives: to maintain deposits and make prof- actions in livestock and stocks of grains and other itable loans.Their success in achieving these objectives goods to assess the responsiveness of household savings determines the financial sector's profitability and to income shocks. hence its growth. Thus researchers need to evaluate a Data on stocks of assets are also necessary to esti- country's government and bank policies in terms of mate household wealth. Experience has shown that how these policies affect both of the functions of the accuracy of estimates of household wealth can be financial institutions. Similarly, research on the impact improved if households are asked about the value of of financial institutions on households should consid- different types of assets rather than being asked to pro- er not only how socioeconomic outcomes are affect- vide an estimate of their total wealth. Since estimates ed by access to credit but also their effect on house- of household wealth are required for almost all aspects hold savings. of socioeconomic research on households, this rein- This chapter focuses on general issues that arise in forces the importance of collecting data on assets, even the context of financial sector research, issues that if income and consumption data are available in the apply as well to research on factors determining the survey. willingness of households to hold financial assets as to Much of the research on savings that has utilized the credit functions of financial institutions. However, data on assets has been conducted at a fairly high evaluating the credit functions of financial institutions degree of aggregation, analyzing, for example, the also raises a number of specific data and methodolog- determinants of transactions in livestock, liquid assets ical issues including how best to collect interest rate (such as financial assets, stocks of grains and other information and what determines households' access goods, and currency), or "productive assets," which to bank funds. Since these issues require data from the are defined as all fixed assets used in the production credit module, they are addressed in Chapter 21 on of either farm or nonfarm income (Udry 1995; credit. Therefore, this chapter's discussion of the data Kochar 1998; Alderman 1996). However, as discussed requirements and methodologies for research on in Chapters 18 and 19 on the household enterprise financial institutions should be read in conjunction and the agriculture modules, collecting accurate with the discussion of these issues in Chapter 21. information on any broad category of assets general- ly requires collecting data on narrowly defined THE DISTRIBUTION OF FINANCIAL INSTITUTIONS. groups of assets within the broad category. Having Researchers examining the spread and distribution of data at this level of detail may also facilitate savings financial institutions generally use the disaggregated research. For example, Rosenzweig and Binswanger data on stocks of financial assets collected in the sav- (1993) used details of stocks of different types of pro- ings module of household surveys.This is a major jus- 194 CHAPrER 20 SAVINGS tification for including a savings module in household interest rates on household savings have used time- surveys. On the basis of disaggregated data on stocks of series data, either for individual countries or for a financial assets, it is possible to discover which house- number of countries pooled together (VanWijnbergen holds are most likely to have accounts in financial 1982; Giovannini 1983; and Fry 1988). institutions and, hence, whether households differ by, This lack of variability in key variables also limits for example, socioeconomic status in their access to the usefulness of a single cross-section of data for ana- such institutions (Kochar 1997). lyzing the effects of various bank policies-such as The extent to which researchers can conduct such specific lending procedures and organizational innova- an analysis using data from a multitopic household tions including group lending-on various aspects of survey will vary from country to country depending financial intermediation.10 Most of the research in this on the level of development of the formal financial area (Yaron 1992; Gurgand, Pederson, andYaron 1994; sector. Analysis of data from the Ghana LSMS survey Hossain 1988) has been based oIn case studies of spe- in Deaton (1992a) revealed that only 7 percent of cific financial institutions. For example,Yaron (1992) loans to the sample households were made by formal examined four rural financial institutions in Asial 1 and sector institutions (such as private banks, government reviewed the factors underlying their success or failure banks, and cooperatives). Only 79 of 2,397 rural in a number of areas including financial self- households in the Pakistan LSMS-3 percent of the sustainability and outreach. rural sample-reported receiving loans from formal As a corollary, cross-sectional data can be used to financial institutions.This small sample limits what can analyze the effectiveness of any policy instrument that be learned about the formal sector from such multi- varies across a sample of households. For example, such topic household surveys. Thus research on the finan- data can be used to assess whether access to formal cial sector in such economies may require "stratified" financial institutions, as measured by a household's dis- surveys that identify borrowers from financial institu- tance from the nearest such institution, affects the will- tions and ensure that sufficient numbers of such ingness of the household to hold deposits, as well as households are included in the sample. the attractiveness of the formal financial sector as a source of loans relative to the informal sector. Such GOVERNMENT AND BANK PoLIcIEs. Household data information was used by Behrman, Foster, and can be used to analyze the effectiveness of any partic- Rosenzweig (1997) to assess whether the availability ular government or bank policy if the data meet two of a bank within 5 kilometers affected the savings of criteria. First, the sample needs to include a sufficient rural households in Pakistan. Household survey data number of households that are affected by the policy can also be used to analyze the extent to which inter- in question. Second, identifying the role of any partic- est rates affect the demand for deposits or credit in ular policy instrument-such as the interest rate or the economies where there is sufficient regional variation rate of return on deposits-in achieving a stated poli- in such rates. cy objective requires this instrument to display signif- icant variation across the sample of households. THE EFFECTS OF FINANCIAL INSTITUTIONS ON Data from one random cross-sectional survey will HOUSEHOLDS. Financial institutions can have an generally not be enough to evaluate government and impact on household fiduciary outcomes such as sav- bank policies relating to the financial sector, both ings and consumption and also on other aspects of because of the limited size of this sector in many well-being such as the health and education of chil- developing economies and because government poli- dren. However, it is very difficult to ascertain the cies in this area generally involve changes in variables, extent to which a household's behavior reflects its such as interest rates, that do not vary significantly transactions with the financial institution in question. across the sample.This is particularly true if the data set In particular, a researcher must address two issues. provides information on only a single cross-section of First, households that do report transactions with households, but it is also true in short panels of data financial institutions may differ in their socioeconom- that survey households over a period of two to three ic characteristics from households that do not report years.This point is also made in Chapter 23 on panel these transactions; if households with higher income data. Not surprisingly, most studies on the effects of are also more likely to borrow from such institutions, 195 ANJINI KOCHAR a positive correlation between consumption levels and in economies where the formal financial sector is loans from financial institutions may merely reflect an poorly developed. underlying correlation between consumption and While inferring causality may not always be pos- income. Second, the developers of a given financial sible, it is possible to use household surveys to assess institution or credit program may have purposely cho- the correlation between the level of development of sen to locate in a particular location because of the the financial sector and various outcomes of interest to socioeconomic characteristics of local residents or of policymakers.Thus the study by Behrman, Foster, and the agroeconomic characteristics of the region Rosenzweig (1997) was able to show how households (Rosenzweig and Wolpin 1986; Pitt, Rosenzweig, and with relatively easy access to financial institutions had Gibbons 1993). Therefore, any observed differences in a higher level of savings than households without such savings between households with access to financial access, even though this study could not explain the institutions and households without such access may factors that caused that difference in savings. merely reflect the unobserved socioeconomic or agroeconomic characteristics that motivated the place- Assessing the Determinants of Households' Savings and ment of the institution in its current location. Portfolio Choices Addressing these selection problems requires, at a Most theories of savings are based on standard models minimum, data on sufficient numbers of borrowers of intertemporal choice. In these models, households and nonborroxvers or, more generally, participants and are assumed to choose their level of consumption in nonparticipants in any particular program. Also, in any given period to maximize the present discounted order to deal with the endogeneity of program place- value of utility over the life cycle, subject to a budget ment, the nonborrowers need to be drawn from sam- constraint that equates the present value of the sum of ples of households both in areas with financial insti- the household's consumption in each period with the tutions and in areas without such institutions. For present value of its lifetime income-and also subject these reasons, studies that have analyzed the effects of to any other constraints, such as credit constraints, that financial institutions on households have generally may affect its decisionmaking over time. Given this been based on data sets that were specifically designed common framework, theories of savings differ prima- for such an analysis. For example, Pitt and Khandker rily in the importance they ascribe to the various (1997) analyzed the impact of a number of credit pro- determinants of savings, such as the variability of grams in Bangladesh, including the Grameen Bank, short-term versus long-term income, variability in on household outcomes such as consumption, labor households' preferences, uncertainty about house- supply, and the health and education of children. To holds' incomes and expenditures, and liquidity con- do so, they used a stratified random sample of house- straints. Thus, in order to distinguish among these var- holds both from villages with credit programs ("pro- ious theories, researchers need-in addition to gram" villages) and from villages without credit pro- measures of savings-data on income, on the demo- grams. The households within program and graphic variables that determine preferences and dis- nonprogram villages were further distinguished count rates, and on measures of the uncertainty in according to whether they met the eligibility criteri- income and of the liquidity constraints to which a on for participation in the credit programs. Finally, household is subject. With these data, researchers can within the program villages, households that met the use both regression analyses and simulation techniques eligibility criterion were divided into participant and (simulating savings on the basis of consumption and nonparticipant households, with 12 participants being income data and hypothesized values of other factors randomly selected for every five nonparticipants. that determine savings, such as interest rates) to under- Using this technique ensured that there were enough stand savings. Alternatively, they can use simple tech- data on a sufficient number of participants as well as niques, such as plots of income and consumption by on a "control" group against whom the outcomes for age, to assess whether savings display the "hump" shape the participants could be evaluated. In contrast, the predicted by life cycle models (Mirer 1979; Danzinger "random" survey techniques usually used in the col- and others 1983; Deaton 1992c). lection of multitopic household survey data rarely Empirical research on household savings therefore provide a sufficient sample of borrowers, particularly requires income and consumption data, as well as 196 CHAPTER 20 SAVINGS nmeasures of demographic variables and other factors is dcfined as the difference between current income that affect savings. Data on household consumption and consumption and hence must be positively corre- are required not only because such data allow lated with current income, such tests of how savings researchers to estimate savings as the difference change in response to anticipated changes in income between income and consumption but also because cannot be based on regressions of current savings on many theories of savings can be tested using data on current income. If a household's current income is consumption. For example, the hypothesis that house- used in a savings regression, it needs to be instrument- holds use savings to protect their consumption from ed by values of the household's income in previous income shocks can be tested through regressions that years to test the responsiveness of its savings to the reveal the relationship between changes in income and component of current income that was anticipated on either consumption or savings. In the past, researchers the basis of last year's value. have preferred to use consumption data to test theo- It is even more desirable to use panel data when ries of savings; this is primarily because error in con- analyzing how households' savings change in response sumption measurements is likely to be less than error to long-term changes in their income.The most con- in savings measurements, which include the measure- vincing studies of the importance of retirement sav- ment error in both consumption and income. ings, such as the Longitudinal Retirement History While LSMS surveys provide data on household Surveys in the United States, have used long panels of consumption and all the necessary demographic vari- data. The Longitudinal Retirement History Surveys ables, they do not always collect the data needed to followed over 11,000 people of retirement age for 10 estimate total income (see Chapter 17 on total years (Hurd 1987; Bernheim 1987). income). Therefore, survey designers should be aware While several studies have tested the life cycle that if savings research is an important justification for model using a single cross-section of data (Darby the survey, total income data need to be collected. 1979; Deaton and Paxson 1992), it is difficult to infer Researchers investigating the effects of liquidity con- life cycle motives from one round or even from a straints or income uncertainty on savings need either panel of data that provides information on households some measure of these variables or data allowing them over only two or three years. In order to use such data, to estimate these variables-in addition to data on the researcher has to assume that the preferences, income, consumption, and demographic variables. And prices, and constraints that influence a household's life researching the determinants of portfolio composition cycle experiences and, hence, its behavior will remain requires data on stocks of or transactions in disaggre- identical from one cohort to another. In other words, gated groups of assets. it has to be assumed that the behavior of a currently 60-year-old man is a reliable measure of how a cur- THE VALuE OF PANEL DATA. To infcr savings motives rently 40-year-old man will behave 20 years from accurately, it is generally necessary to have a panel of now. Using only one cross-section of data can also data, primarily because models of intertemporal cause sample selection problems. Good health and the choice imply that a household's savings reflect its probability of living a long life are generally positively expectations of future income and consumption. To correlated with wealth, so that the rich are over-rep- test these models, researchers need data that span a resented among the (surviving) elderly. Unless the number of years. Studies of how households change researcher controls for this bias, he or she may con- their savings in response to anticipated changes in clude that there is little decumulation of wealth with either annual or seasonal income have generally been age, even though such decumulation may in fact based on tests of the relationship between either sav- occur. ings or consumption and the change in income across Panel data are also necessary for estimating the periods (Hall 1978; Deaton 1992b; Flavin 1981, 1993; variability in individual incomes and, hence, the Alderman 1996; Kochar 1998). Deaton (1992b) importance of precautionary savings to hedge against regressed the change in income between two periods this variability. The scarcity of research on the impor- on the previous year's values of income and savings in tance of the precautionary motive in developing order to test the hypothesis that households save in economies is probably due to a lack of the long panels anticipation of changes in their income. Since savings of data required for such estimates. In contrast, consid- 197 ANJINI KOCHAR erable research on precautionary savings has been done Other researchers have used data on the number in developed countries such as the United States-in of days of illness reported by working household part because long panels of data have enabled members as a measure of income shocks (Cochrane researchers to estimate the variability in individual 1991). Reported days ofillness, however, may not rep- incomes (MaCurdy 1982; Hall and Mishkin 1982). resent a shock to the individual, not only because ill- ness is often predictable but also because there is often ENHANCING THE VALUE OF CROSS-SECTIONAL DATA systematic measurement error in self-reported meas- FOR RESEARCHING THE DETERMINANTS OF SAVINGS. ures of health. For example, there is considerable evi- While panel data can be quite valuable, collecting long dence that the number of self-reported days of illness panels of data is costly and thus not always feasible. For is correlated with household characteristics such as this reason it is likely that for the foreseeable future, income and education (see Chapter 8 on the health research on savings, particularly in developing mnodule). Nevertheless, the availability of such data economies, will have to be based on single cross- does give researchers some insights into the factors sections or short panels of data. And in spite of the that determine households' savings and portfolio deficiencies of cross-sectional data, these data can choices. reveal important information about households' The availability of data on the earnings of indi- motives for saving. vidual household members and the sources of these Since one of the merits of panel data is that they incomes can also facilitate research on particular sav- enable researchers to estimate anticipated changes in ings motives. Having data on the earnings of individ- household income, the usefulness of cross-sectional ual household members enables researchers to ascer- data can be increased if they provide measures of tain whether parents and their coresident adult expected household income. In a single cross-section children combine their incomes so that the consump- of data, Flavin (1993) used information on what tion of each individual depends not on his or her own households expected their income to be in the current income but on the combined incomes of all house- year to estimate savings responses to anticipated hold members (Hayashi 1995). Such evidence informs changes in income. Guiso, Jappelli, and Terlizzese research on life cycle savings; there may be little need (1996) used data from a single cross-sectional survey on to save for old age if the income of the young provides households' subjective perceptions of income risk and for the consumption requirements of the elderly. This on whether they had been denied credit in the past to research methodology is feasible in economies where analyze how households' choices of assets changed in individual sources of income are significant. However, response to income risk and credit constraints. it is difficult to implement in economiies where indi- Data on whether a household has experienced vidual earnings are rare, and where households instead unexpected changes in consumption expenditures or earn their incomes from family enterprises through income can also be useful in assessing whether house- the joint labor of their members. In such cases, esti- holds use savings to protect their consumption from mating the individual claims on jointly produced fam- such shocks and which of their assets they use for this ily income in order to test the hypothesis of pooled purpose. Researchers interested in this issue have family income is fraught with difficulties (see Chapter occasionally designed and administered a cross- 17 on total income). sectional survey to collect such information. Udry Having income data by source also helps (1995) designed a cross-sectional survey for rural researchers understand savings. Researchers can use Nigerian households that provided information on data on pensions or insurance payments to explore whether the households experienced unexpected whether the households receiving such income are less changes in income over the reference period of the likely to save and, hence, whether savings are a survey; he used this information to assess the respon- response to the need for annuity income or insurance. siveness of savings, by assets, to such shocks. The avail- These data also enable researchers to investigate poli- ability of disaggregated data on the types of assets held cy issues xvithout having to delve into households' by households contributed significantly to the value of motives for saving. For example, the question of this research-highlighting the benefits of collecting whether publicly funded programs such as social secu- such data. rity "crowd out" private savings can be approached 198 CHAPTER 20 SAVINGS either by examining whether savings reflect individu- children to parents and the need (or lack of need) for als' perception of their old age and insurance needs or life cycle savings, because the entire sample yielded by examining whether the receipt of social security only 99 instances of transfers from children to parents. payments causes a reduction in private transfers. Such The usefulness of cross-sectional data for savings an analysis has been conducted by Cox and Jimenez research can also be enhanced if they are supplement- (1992) using a cross-section of data from the Peruvian ed by aggregate time-series data on variables affecting LSMS. The researchers found that elderly people's household income. Paxson (1992) andWolpin (1982) receipt of social security benefits significantly reduced have used available data on rainfall statistics over a the amount of transfers they received from other number of years to predict changes in agricultural households. incomes and, hence, to separate the transitory compo- The usefulness of cross-sectional data for assessing nents of income from its permanent components- the impact of government policies depends on how without the use of panel data. many households report being affected by the policies Box 20.1 classifies savings issues according to what in question. For example, it would not be possible to kinds of data are needed to analyze them. use the Pakistan LSMS to study the effect of pensions on savings because of the very small number of house- The Relevant Unit for Analysis holds in the sample who reported receiving pensions The previous subsection suggests that individual-level (64 out of 4,799). Nor would it be possible to use these data on incomes and on interhousehold and intra- data to analyze the relationship between transfers from household transfers may help researchers understand Box 20.1 Policy Issues and multitopic Household Survey Dat Issues that can be analyzed with cross-sectional household sur- Issues that can be analyzed with I 0 years of panel data vey data * The demand for life cycle savings and how this demand * The ratio of household savings to household income. affects the composition of households' savings portfo ios. * The forms in which households save and how households' * Better estimates of both precautionary savings and savings physical assets are divided for use among self-employment for short-run consumption smoothing (due to data on enterprises, real assets, and financial assets. income and consumption profiles over the life cycle). * Variation in the rate and form of savings among house- * The effects of aggregate income shocks on savings during holds with differences in socioeconomic characteristics the survey period. such as wealth, demographic characteristics, occupation, * Analysis of the effects on household savings of policy vari- and region of residence. ables that vary over the duration of the survey, including * The importance of financial assets in household portfolios. macroeconomic factors such as inflation, interest rate * The importance of formal financial institutions relative to changes, and fiscal and monetary policy. informal ones. * A similar analysis of the impact of government policies on * The difference in the importance of financial assets across the viabilty of financial institutions and, hence, on the households of different socioeconomic status. extent and spread of financial intermediation. * How total savings and savings in particular assets change in response to income shocks and expectations of Issues for which household survey dota are not sufficient and for income, if a measure of income shocks or expected which additional data or special sample designs are needed income is available in the data. * Studying a special policy such as group lending would * Insights into households' motives for saving if data on indi- require a special survey collecting data from a sufficient vidual incomes and income by source are available. number of households affected by the policy and from a * The effect of the availability of financial institutions on control group of households not affected by the policy. financial savings. * Studying group lending would additionally require collect- ing suffic ent data on borrowers and nonborrowers, Issues that can be analyzed with two or three years of panel data including both the nonborrowers who have access to the * Whether savings are used to smooth short-run fluctua- program n question and the ones who do not. tions in income and which assets are used for this pur- pose. Issues for which household survey data are of little use * The importance of precautionary savings (even without * Analyses of the profitability of financial institutions by type of direct measures of income shocks). institution and policyThese usually require bank-level data. 199 ANJINI KOCHAR household savings. Several recent studies have argued lecting individual-level data on these measures, particu- that data on incomes and consumptions of "private" larly in the cases of Joint (or household public) goods goods, or goods for which individual consumption lev- shared among household members and income from els can easily be identified and measured, should be col- joint family activities. Similarly, assigning individual lected at the level of individuals within the household ownership to household assets-particularly consumer rather than at the level of the household. (This is dis- durables and productive assets used in the production of cussed in detail in Chapter 24 on intrahousehold issues.) income from joint activities-may be an impossible task. The authors of these studies argue that the theory of the Therefore, it is not necessary to attempt to assign unified household, which underlies much of neoclassi- individual ownership to all household assets. However, cal economics, is invalid, and that individuals within a it is worthwhile to record the ownership of assets, such household are likely to have distinct and unique prefer- as jewelry, that can easily be assigned to any given indi- ences. Empirical research generally supports this vidual in a household, and to collect data on the hypothesis. For example, available evidence suggests that wealth inherited by particular individuals in the income earned by women is spent differently from household.These data are useful inputs into tests of the income earned by men, with more of women's addi- validity of the neoclassical "unified" household model tional income spent on food and education (Thomas relative to the "individualistic" models, since they pro- 1990 and 1993; Schultz 1990; Quisumbing 1994). vide a set of variables that may be correlated with Such gender differences are likely to affect house- individual incomes. The variables can then serve as hold saving decisions, as they imply that the propensity to instruments to correct for problems caused by meas- save may vary among individuals within a household. If urement error and endogeneity in incomes. Chapter governments want to increase the savings rate they 24 discusses these and other variables that can facilitate should target individuals within a household who are research on intrahousehold issues. most likely to be responsive to savings initiatives.While a significant number of microfinance institutions, such as Two Versions of the Savings Module the Grameen Bank in Bangladesh, do target individuals wvithin the household-notablv women-this policy has This section introduces two prototype savings usually been adopted for reasons other than its potential modules-one standard length version and one short effect on savings. Evidence on the effects of such pro- version-and discusses their design. (The modules are grams on household savings is slowly accumulating (Pitt presented in Volume 3.) The primary purpose of the and Khandker 1997); however, much more research is savings module is to provide information on a house- needed in this area before firrm conclusioins can be drawn. hold's financial assets, including currency, and on its Even if programs that target individuals within a other assets such as land and buildings held for invest- household are found to increase household savings, it ment purposes. Details of this information are not vill be difficult to infer whether this is because individ- generally available in other modules.While it is possi- uals differ in their savings propensities.Why? One reason ble to provide data on stocks and flows of all house- is that testing differences in the savings propensities of hold assets in the savings module, available evidence individuals is much harder than testing the validity of the suggests that data on most assets are best collected in neoclassical model. General tests of the validity of the modules other than the savings module.Thus the stan- neoclassical model simply require identifying the dard savings module is relatively short. income earned by some individuals within the In some cases, lack of time and resources may household-an easy task to accomplish if there are wage necessitate fielding an even shorter survey-one that earners in the household. In contrast, testing differences collects just the basic information necessary to assess in the savings propensities of individuals requires assign- household welfare.This survey need not collect infor- ing the totality of household savings (or consumption) to mation on household asset transactions but must individual members, either by fully identifying the indi- include questions on the value of stocks of assets- vidual income and consumption of all of the household's including financial assets-since these data are needed members or by assigning individual ownership to all to obtain a measure of total household wealth. While assets. Chapter 17 on income and Chapter 5 on con- much of the discussion in this section pertains to the sumption both discuss the difficulties involved in col- standard-length savings module, Table 20.5 summa- 200 CHAPTER 20 SAVINGS rizes the relative merits of the short and the standard tions in assets, given the general hesitancy of house- savings modules for addressing policy issues. holds to disclose details regarding their wealth. It is necessary to keep this in mind in designing the ques- Collecting Data on Household Assets in the Savings tionnaire in general and the savings module in partic- Module ular. For example, as discussed in Chapter 3, it is gen- A recurring theme in this section is the difficulty of erally desirable to place the savings module near the collecting data on households' stocks of and transac- end of the questionnaire. This serves two purposes. Table 20.5 Summary of Policy Issues and How Multitopic Household Surveys Can Be Used to AddressThem Usefulness of questionnaire for Other modules from Issue analysis which data are needed Would need to add Short Questionnaire Rate and form of savings and differences across households Good Income and income-related modules;' Nothing Consumption from core; Socioeconomic indicators from core Lvel an srad of fnncia instittions Good Socioecooi indicatrs from core Nothing Determinants of savings Poor Income and income-related modules0- Generally requires panel data for details such as measures of income expectations and shocks Determinants of portfolio choice Poor Income and income-related modules.- Details of financial transactions for details such as measures of income from the standard questionnaire; expectations and shocks; Generally requires panel data Asset-related modulesb .............................................................. .............................. .................................... ................................................................................................................................ Effects of government ano bank policies Poor Community survey-for data on Details of transactions in financial on financial institutions distance to banks assets from standard questionnare; Generally requires special survey . ................................................................................................................................... ............................................................................................... Impact of financial institutions on Poor Details of transactions in financial households assets from standard questionnaire; Generally requires special survey .................................................................................................................................................................................. I................................................. Standard Questionnaire Rate and form of savings and Good Income and income-related modules;n Nothing dfferences across households Consumption from core; Socioeconomic indicators from core ........ ........ ........................ *............... *......................................................................................................................................................................... Level and spread of financial institutions Good Socioeconomic indicators from core Nothing Determinants of savings Poor Income and income-related modules0- Generally requires panel data for details such as measures of ncome expectations and shocks ................. *.................................................................................................................................................................................................................. Determinants of portfolio choice Fair Income and income-related modules0- Generally requires panel data for details such as measures of income expectations and shocks; Asset-related modules' Effect s of government and bank poicies Fair Commnity survey-for data on distance Generally requires special survey on financial institutions to banks .............................................. *..................................................................................................................................................................................... Impact of financial institutions Fairc Generally requires special survey on households to ensure data on sufficient numbers of borrowers . ......................................................... *...................................................................................................... *..................................................... *............. a. Estimates of annual income additionally require information on agricutural income (farming and ivestock module), labor or wage income (abor module), nonagricul- tura or enterprse income (nonfarm enterprise module), livestock income (farming and livestock module), and other income including pensions, remittances, rental and interest income (other income modu e). b Studying issues melating to portfolio composition requ res disaggregated data on the stocks of and transactions in all assets. As above, this i-format on needs to be col- ected in the modules that gather information on the different assets held by the househo d. In genera these modu es will be the housing module (for data on res den- t.a. wealth), the farming and livestock modules (for data on agricultural assets, livestock, and Inventories of foodgrains and fodder), the nonfarm enterprise module (non- farm assets and inventories), and the -onsumption module (for data on consumer durables). c. Using the standard questionna re oue can assess the mpact Of features of finaricial institutions thdt vary across a sample. Source. Authors summary. 201 ANJINI KOCHAR First, it enables the interviewer to develop a rapport questions of any kind about the financial assets of the with the respondent, hence, increasing the likelihood sample households because the survey designers that the respondent will be willing to answer questions assumed white households would not want to provide about the household's financial wealth. Second, it accurate details of their financial assets for fear that ensures that important data will already have been col- these assets would be confiscated after the imminent lected from the other modules if, upon being asked change to black majority rule.TheVietnam and Ghana about the household's financial wealth, the respondent LSMS surveys yielded only estimates of the total cur- should decide to terminate the interview. Similarly, rent value of all of the household's financial assets, within the savings module, it is best to place questions without providing any details about financial transac- about particularly sensitive assets, such as currency, tions the household may have engaged in within the toward the end of the module. previous 12 months. Conversely, the Peruvian LSMS The difficulty in collecting data on financial assets survey yielded data on the household's transactions but suggests that it may be desirable to alloxv the respon- not on the current value of its financial assets. dent to put the value of an asset that he or she owns The difficulties LSMS surveys have encountered within a broad range of values rather than citing a spe- in trying to collect accurate data on financial assets is cific figure. This approach may make the respondent an indication that it may be better to measure savings more likely to part with this information. Depending as the difference between income and consumption on the willingness of the respondent to continue with rather than to measure it on the basis of data on asset the questions, the initial value range can be successive- transactions. However, the evidence so far is only sug- ly narrowed until the value of the asset in question can gestive, and the question needs to be studied further be placed within a fairly narrow range (Juster and before any conclusive recommendation can be made. Smith 1997). However, the feasibility of this approach Until solid evidence on the reliability of one measure should be tested in a pilot survey before it is incorpo- over another becomes available, it may be desirable to rated into the questionnaire-as should the appropri- collect data on asset transactions in addition to data on ate range of values to offer the respondent as options income and consumption, so as to enable the for the value of the household's financial assets. researcher to use both measures of savings.This under- Therefore, the prototype modules presented in this scores the need to improve techniques for collecting chapter do not include such questions. income and consumption data, as well as to ensure that reliable data on all asset transactions are recorded. DATA ON HOUSEHOLDS' STOCKS AND FLOWS OF F;INANCIAL ASSETS. The standard savings module DxrA ON OTHER ASSETS IN 1IHE SAVINGS MODULE. should gather information on the household's stock of Policymakers also need information on other house- and transactions in such financial assets as stocks, hold assets, including: stock and flow of physical assets shares, bonds, other securities, and deposits in financial used in the production of farm or nonfarm income; institutions. These data not only facilitate research on real assets such as land and housing wealth other than savings but are also essential inputs into the estimation that used in the farm or nonfarm enterprise; land or of household wealth. As noted earlier in this chapter, buildings held purely for investment purposes; and an accurate measure of a household's total financial stores of foodgrains, fodder, building materials, and assets is more likely to be obtained if respondents are other inventories. As with financial assets, not all past asked to provide disaggregated information on the LSMS surveys gathered information on both the value of their financial assets such as bank deposits, stocks and flows of such assets.Those in which this was currency, and savings in informal savings institutions done differ in terms of where in the questionnaire this than if they are asked to give the sum value of all their information was collected. In the Vietnam LSMS sur- financial assets. vey, information on the stock of real assets (buildings Hoowever, households are frequently hesitant to and houses) was collected in the savings module, while provide the interviewer with details of their financial information on the sale of these assets was collected in wealth. As a result, these data are not available in many the miscellaneous income module. In the Pakistan survey data sets, including many LSMS surveys. For LSMS survey, data on the stocks of and transactions in example, the South African LSMS survey asked no real assets were gathered in the savings module, where- 202 CHAPTER 20 SAVINGS as in the South African survey these data were collect- is that there is a greater probability that some assets will ed in a separate module. inadvertently be ignored, adding to any measurement One question that survey designers must address is error in estimates of household savings based on asset the module in which to locate questions on nonfinan- transactions data. One asset that is frequently missed is cial assets so as to maximize the accuracy of responses. buildings and land held for investment purposes.13 The answer to this question varies from survey to sur- While the miscellaneous module frequently yields data vey depending on which modules are included in each on rental income from such property, it gathers no questionnaire. If the survey contains a farm and a non- information on the property's value or on transactions farm enterprise module, as do about two-thirds of in such buildings and land.The savings module may be LSMS surveys, data on the stocks and flows of land, the best place to gather this data. Separating questions buildings, and other assets used in these enterprises are on investment property from questions on residential best collectcd as part of these modules, along with property may also reduce measurement errors in the questions on the operations of the enterprise in ques- resulting data in both categories. tion. In the absence of these modules, however, data on the relevant assets should be included in the savings DATA ON STOCKS AND TRANSACTIONS OF CONSUMER module. DURABLES IN THE SAVINGS MODULE. The savings mod- Data on other real assets, such as residential land ules in several existing LSMS data sets contain infor- and housing, are frequently collected in both the sav- mation on households' stocks of and transactions in ings module and the housing module. This was the consumer durables. As with the value of property case in the Pakistan survey, where the housing module owned by the household, this information is also col- gathered data only on residential property, and the sav- lected elsewhere in the survey, most commonly in a ings module collected data on the sum of residential separate module that collects data on expenditures for and investment property.'2 This difference in method nonfood items and consumer durables and additional- led to a difference in the value of land and buildings ly provides an inventory of durable goods. Again, as in reported in these two modules. In the housing mod- the case of the value of residential and nonresidential ule 3,900 households reported owning residential property, comparing the data in this module with the land, with a mean value of Rs. 155,000, whereas in the data in the savings module suggests that the data in the savings module 3,988 households reported owning savings module are unreliable. For example, the residential land or land rented out for residential pur- Vietnam survey asked households about the current poses, with a mean value of Rs.179,000. The number value of their durable assets (such as motorbikes or of households that reported owning residential land washing machines) in the savings module. Only 22 other than land they occupied was just 88. This small households reported owning such assets, with a mean number suggests that some households did not report value per household of 8.3 million dongs.This figure their ownership of residential land, though there is no can be compared to the total value of all household means of verifying this with the data at hand. durables from the "inventory of durable goods" mod- Discrepancies of this kind suggest that respon- ule (Section 12, Part C of theVietnam questionnaire), dents may be averse to providing accurate information in which 4,663 households reported owning con- on their assets when it is clear that the intention of the sumer durables, with a mean value per household of interviewer is to collect data on the household's 2,9 million dongs. Indeed, disaggregated information wealth. It appears that respondents give far more accu- on the ownership and value of diffcrent types of assets rate information if the relevant questions are asked in reveals that the numbers of households owning just other modules in a less sensitive context, such as when washing machines and motorbikes exceed the num- the interviewer is asking about housing characteristics. bers reported in the savings module. While only 15 Whatever the reason for these discrepancies, it seems households possessed a washing machine, as many as that there may be little advantage to collecting data on 512 owned a motorbike, with a mean value of 8.8 mil- individual assets in the savings module if they can be lion dongs.Thus the data in the savings module appear collected in other modules of the survey. to bear little relation to ownership either of total assets One drawback to gathering information on asset or of the individual assets explicitly mentioned in this transactions in modules other than the savings module module. 203 ANJINI KOCHAR Because it appears that collecting data on the modules a few additional questions on any variables value of or transactions in consumer durables in the that affect savings. For example, information on savings module results in considerable measurement whether the household's total current income exceed- error, these data should, to the extent possible, be col- ed or fell short of the expected amount could be con- lected in a separate module along with details on other veniently located in the income module, and would items of expenditure. facilitate research on the responsiveness of savings to anticipated changes in income. In a similar vein, Modifing Other Modules to Facilitate Research on Savings Chapter 21 on credit suggests including in the credit Given that collecting data on real and physical assets in module questions concerning whether or not house- modules other than the savings module generally holds applied for credit over the reference period and yields more accurate data, it is important to ensure that the results of this application. The response to such these modules are designed to yield the data needed questions may provide information on the importance for doing research on savings. of liquidity constraints, information that can be used At a minimum, data are needed on the stock of to assess the effect of such constraints on savings. and transactions in all assets. Most past LSMS surveys It is important, however, to keep the overall size of have collected such data for households' agricultural the relevant module in mind when including such and business assets in the farm and nonfarm modules. questions. For relatively long modules such as the agri- However, a number of surveys have not gathered this culture module, further increases in length may com- information-compromising their value for savings- promise the quality of the data. In such cases it may be related research. For example, the Vietnam LSMS sur- best to limit the questions asked to ones that are essen- vey gathered no data on the purchase of buildings and tial for obtaining accurate measures of the stocks of lands by households for nonfarm enterprises, while and transactions in the relevant assets.Thus it may be the South Africa survey did not gather information desirable to omit questions relating to reasons for the either on the current value of household livestock or sale or purchase of any particular asset, or the timing on the value of livestock purchased by farm house- of such transactions, despite their usefulness for savings holds during the year in question. research. Survey designers need to think carefully about how The loss of data that results from collecting asset best to gathcr data on consumer durables. In most pre- information in modules other than the savings mod- vious LSMS surveys the inventory of durables module ule is counterbalanced by the fact that collecting sav- gathered data on the current value and purchases of ings-related information in these other modules yields durables. Data on sales have generally been gathered in more accurate data on the value of asset stocks and the miscellaneous income module, as in the Vietnam, transactions. China, and Ghana surveys. In all of these instances, how- ever, respondents were asked how much income their DATA ON INVENTORIES OF FOODGRAINS, FODDER, AND households received from the sale of all durable goods, OTHER MATERIAL. As noted earlier, experience has and this degree of aggregation may have increased the shown that it is notoriously difficult to gather data on measurement error in this figure. Research on savings households' stocks of foodgrains, fodder, and other also requires distinguishing between income from the materials.Yet it is widely believed that stocks of such sale of durables and income earned from their rental. liquid assets account for a significant share of house- (This distinction is recommended in Chapter 11 on hold savings in any given period. Finding ways to transers and other nonlabor income.) Similarly, as dis- study the usage of this portion of household savings is cussed in Chapter 5 on consumption, it is important to likely to yield insights into what influences household separate purchases from gifts and bequests received by decisions about savings (Chaudhuri and Paxson 1994). the household. As noted earlier, the best way to do this Some surveys, such as the Vietnam LSMS survey, may be to include an explicit question about whether an have gathered data on stocks of foodgrains (in this case, asset was received as a gift or was purchased, along with paddy and rice) in the savings module. The fact that in questions about transactions in such goods. the Vietnam survey a very low number of households The value of the data set for savings research can (337) reported owning such assets raises doubts about also sometimes be augmented by including in other the validity of the data. Other surveys provide no infor- 204 CHAPTER 20 SAVINGS mation at all on such stocks. For example, in the data which have been used successfully in past LSMS set from the Pakistan LSMS survey it is only possible to surveys. estimate the nonmarketed surplus from foodgrains pro- duction during the reference year. No data are available Explanatory Notes on the Standard and Short on any stocks carried over from previous years. Versions of the Questionnaire Practical experience suggests that such data can only be reliably collected at the point in the agricul- The Standard Questionnaire ture module when the interviewer is asking specific For each asset the survey covers, the respondent should questions about the household's crop output and its be the household member most knowledgeable about disposal (see Chapter 19 on agriculture). The China the asset. Who this person is will vary depending on LSMS survey, one of the few to follow this practice, which assets are being discussed. While the male head was designed so that data on inventories (by crop) at of household may be most knowledgeable about sav- the time of the harvest and at the time of the inter- ings in investment properties and financial assets, a view were collected in the agriculture module. To female household member may know more about sav- facilitate research on savings, a multitopic household ings held through informal savings committees. survey should include, at minimum, an extensive agri- culture module that gathers such crop-specific details. Box 20.2 Cautionary Advice While many past LSMS surveys have included such an agriculture module, questions on household stocks of * How much of the draft module is new and unproven? The and transactions in foodgrains and fodder need to be savings module presented in this chapter is similar to those used in many recent LSMS surveys. incorporated in future surveys. Since the reference * How well has the module worked in the post? While the period f taaisntrsavings modules used in past LSMS surveys have gen- previous 12 months, it would be desirable to use the erally produced reasnable data, households are typi- same reference period to record the change in inven- cally wary about providing information on their wealth tories and transactions in foodgrains and fodder. and financial savings. It is therefore necessary to follow the recommendations in the chapter regarding the SAVINGS-RELATED DATA IN THE COMMUNITY placement of the savings module toward the end of the QUESTIONNAIRE. Inserting questions on the availabili- survey, when sufficient trust has been built between the interviewer and the respondent. Some surveys have quesandncosts fcfinancalso serviceresethe comunsvit collected data on specific items (such as consumer durables or the value of residential property) in two Data on formal financial institutions may be best suit- different modules of the survey, and the wealth esti- ed for this questionnaire, because the terms of both mates derived from the different modules have some- borrowing and lending from such institutions are rel- times been significantly different from each other Such atively uniform across all households in the communi- differences appear to reflect confusion regarding ty. In economies where such terms vary across regions whether the data collected in the savings module are and hence across households, data on such variables stocks of assets at a particular point in time or changes can significantly contributc to research on houscholds' in stocks of assets dicing the reference period. These errors can be minimized with well-trained interviewers who understand the questions they must ask and who Chapter 4 on conIIIIunity and price data details can communicate these questions well to respondents. sonme of the information that can profitably be col- * Which parts of the mDdule most need to be customized? lected in the community questionnaire. This includes In countnes where the leve of financial savings is low, the kinds of financial institutions available to house- questions in the stancard questionnaire regarding details holds (such as government banks, private banks, and of different types of financial assets will not be neces- cooperatives), the kinds of savings instruments gener- sary; ir such cases only the short questionnaire may be ally used in the community, the distance of relevant feasible. In other economies, the disaggregated list of savings instruments (such as bonds, government certifi- insituion fr m he omm ni cete, andtrcates, savings accounts, and informal savings associations) age interest rates on both loans and deposits, if any. in the standard version of the questionnaire must be tai- Box 20.2 indicates which elements of the draft lored to reflect the availability of each instrument module presented here are new and unproven and .. 205 ANJINI KOCHAR PARI A. In the questions on land and property held for represent both new investments and any changes in investment purposes should be included in the savings the value of the asset over the course of the year. module only if such information is not included else- where in the survey. Thus data on income from rental PART D. The information in Part D is important in of agricultural land is perhaps best collected in the countries where local savings groups or rotating savings agriculture module. No questions on the value of associations are an important means of increasing sav- owner-occupied residential property or other land and ings. Such associations include "bisi" accounts in property used in farm and nonfarm enterprises are Pakistan, "susu" accounts in Ghana, and "tontine" included in this section, on the assumption that such accounts in C6te d'Ivoire. questions are in other modules. It is important to distinguish investments in resi- The Short Questionnaire dential land from investments in agricultural land in For each asset the survey covers, the respondent should countries where the two kinds of land are conceptu- be the household member most knowledgeable about ally distinct in terms of how they are taxed and the this asset.Who this person is will vary depending on markets in which they can be transacted. If data on which assets are being discussed.While the male head land is collected in the agriculture module, such a dis- of household may be most knowledgeable about sav- tinction is necessary to ensure that the information in ings in investment properties and financial assets, a the savings module does not overstate land ownership female household member may know more about sav- by double-counting the value of such land. ings held through informal savings committees. Questions 9 and 10 of Part A separate out receipt of gifts since not doing so will mean that the measures PART A. In this part the questions on land and prop- of savings yielded by income and consumption data erty held for investment purposes should be included will not tally with the measures of savings yielded by in the savings module only if such information is not data on asset transactions. Also, if gifts are received as included elsewhere in the survey.Thus data on income payment for previous services provided by the house- from rental of agricultural land is perhaps best collect- hold, they are more akin to credit transactions than to ed in the agriculture module. No questions on the outright gifts. value of owner-occupied residential property or other land and property used in farm and nonfarm enter- PARTS B AND C. In these two parts the lists of finan- prises are included in this section, on the assumption cial instruments are only suggestive. The actual items that such questions are in other modules. that should be included will depend on the nature of It is important to distinguish investments in resi- financial markets in the economy and the financial dential land from investments in agricultural land in instruments available to households. countries where the two kinds of land are conccptu- Question 3 of Part B collects data on interest ally distinct in terms of how they are taxed and the incomiie in the samne section wvhere data is collected on markets where they can be transacted. the household's financial assets.This will reduce meas- urement error more than will asking respondents PART C. Since the primary goal of including the value about the sum of all of the household's interest, divi- of financial assets in the short questionnaire is to dend, and profit income in some other module such as obtain a measure of household wealth, there is no the miscellaneous income module. need to collect data on transactions in financial assets. Since some of the assets in Part B may be trans- Even with such a narrow objective in mind, the broad acted several times in any given year, households may level of disaggregation in this questionnaire is useful not be able to provide accurate answers if asked about for minimizing measurement error. each separate addition/withdrawal of the particular asset in question. In such cases it is easier to ask the Notes respondent to give the value of that asset a year before the survey interview, as in question 4. The change in The author is grateful for commsents by Margaret Grosh, Paul this value during the reference year (in other words, Glewwe, Julie Schaffner, and other participants in the LSMS the difference in responses to questions 4 and 2) will authors'workshop. 206 CHAPTER 20 SAVINGS 1. One example of the difficulties in obtaining data on credit 12. The Pakistan LSMS does provide separate details of the transactions is the host of problems that can arise in trying to obtain agricultural land owned by nonfarm households. accurate details about the interest rates charged on informal loans. 13. In a few previous LSMS surveys, such as the South African (See Chapter 21 on credit.) survey, information on nonfarming land and other immovable 2. The relevant files from which the data are drawn are listed in property has been collected m a separate module. This separate Table 20.1. module provides information on only rental income from the land 3. Expenditures on education and consumer durables were and current value of the land (or property)-and not on purchases included in savings and excluded from consumption aggregates. or sales throughout the year. Consumer durables included household effects, kitchen equip- ment, furniture and fittings, and other durable housing expenses. References 4. No correction was made for the value of any jewelry received as gifts, because there were no data on this. Alderman, H. 1996. "Savings and Economic Shocks in Rural 5. Data on household savings are not available. Pakistan." Journal of Development Economncs 51 (2): 34345. 6. The estimates of aggregate income and consumption used in Behrman,Jere R., Andrew Foster, and Mark R. Rosenzweig. 1997. this exercise were generated by Coulombe, McKay, and Round "Dynamic Savings Decisions in Agricultural Environments (1993) and used by McKay in Chapter 5 of this book. Smce expen- with Incomplete Markets." Journal of Businiess and Economic ditures on health and education are included in the estimates of Statistics 15 (2): 282-92. aggregate consumption, these expeiises are excluded fromi the calcu- Beruheiins, B. Douglas. 1987. "Dissaving After Retirement: Testing lations of savings based on asset transactions. An alternative estimate the Pure Life Cycle Hypothesis." In Zvi Bodie, John B. of consumption aggregates calculated by the World Bank yielded Shoven, and David A. Wise, eds., Issues in Pension Econoniies approximately the same results. Problems encountered in estimating Chicago, Ill.: University of Chicago Press. income and consumption are detailed in Chapters 5 and 17 along Besley, Timothy. 1995. "Savings, Credit, and Insurance'' In Jere with suggestions for improving the collection of income data. Behrman and T.N. Srinivasan, eds., Hendbolok of Developnient 7. Of these financial assets, the most common was bank Economics. Vol. 34. Amsterdam: Elsevier Science, B. deposits, wvhich were held by 36 percent of the sample households. Chaudhuri, Shubham, and Christina Paxson. 1994. "Consumption Only a very small number of households reported owning financial Smoothing and Income Seasonality in Rural India. Discussion assets in other institutions. Paper 173. Princeton University, Woodrow Wilson School of 8. The purchase of cars, which was reported by just 1 percent of Public and International Affairs, Research Program in the sample population, accounted for 44 percent of the total expen- Development Studies, Princeton, N.J. ditures on consumer durables by surveyed households in 1987-88. Cochrane,John. 1991. "A Simple Test of Consumption Insurance:" 9. The consumption aggregates used to calculate savings as the Journal of Political Economy 99 (5): 957-76. difference between income and consumption include a number of Coulombe, Harold, Andrew D. McKay, and Jeffery 1. Round. 1993. items that may more appropriately be treated as investments. "The Estimation of Components of Household Incomes and Excluding education and health expenditures from the consump- Expenditures from the First Two Rounds of the Ghana Living tion aggregates calculated by theWorld Bank increases savings frons Standards Surveys 1987/88 and 1988/89?" University of -83,822.38 cedis to -68,732.57 cedis. Further excluding expendi- Warwick, Department of Economics,Warwvick. tures on lionse iisaiiLtesiance, kitchers equipinieiit, atid furniture Cox, Donald, anid Eimunailuel Jisiieriez. 1992. "Social Security aisd increases savings to -61,760.77 cedis. However, since the savings Private Transfers in Developing Countries:The Case of Peru." estimate is still highly negative, questions remain about the reliabil- World Bank Economic Revieu' 6 (1): 155-69. ity of the income and consumption data. Danzinger, Sheldon, Jacques van der Gaag, Eugene Smolensky, and 10. If the data set includes sufficient information on both bor- Michael Taussig. 1983. "The Life-cycle Hypothesis and the rower and nonborrower households, cross-sectional data can be Consumption Behavior of the Elderly." Journal of Post- effectively used to analyze such issues as the factors that determine Keynesian Economnics 5 (2): 208-27. households' access to bank credit and the effects of such access on Darby, Michael R. 1979. Effects of Social Security on Incomxe and thle various aspects of household behavior. See Chapter 22 on credit. Capital Stock. Washington D.C.: American Enterprise Institute. 11. The Badan Kredit Kecamatari (BKK) and the Bank Rakyat Deaton, Angus. 1989. "Saving in Developing Countries: Theory Indonesia Unit Desa (BUD) in Indonesia, the Bank for Agriculture and Review." In Stanley Fischer and Dennis de Tray, eds,, World and Agricultural Cooperatives (BAAC) in Thailand, and the Bank Atnual Conference on Development Economiics 1990. Grameen Bank in Bangladesh. Washington, D.C.:World Bank. 207 ANJINI KOCHAR 1992a. "Household Saving in LDCs: Credit Markets, Hurd, Michael D. 1987. "Savings of the Elderly and Desired Insurance, andWelfare." Scandinavian Journal of Economics 94 (2): Bequests." American Economic Review 77 (2): 298-312. 253-73. Juster, Tom, and Jim Smith. 1997. "Improving the Quality of .1992b. "Saving and Income Smoothing in Cote d'Ivoire." Economic Data: Lessons fronm the HRS and AHEAD." Discussion Paper 156: 3-27. Princeton University, Woodroxv Journal of the American Statistical Association 92 (440): Wilson School of Public and International Affairs. Research 1268-78. Program in Development Studies. Kochar, Anjini. 1997. "An Empirical Investigation of Rationing 1992c. UInderstandiug Consumption. Oxford: Clarendon Constraints in Rural Credit Markets in India." Journal of Press. Development Economics 53 (August): 339-71. Deaton, Angus, and Christina H. Paxson. 1992. "Patterns of Aging Kochar,Anjini. 1998. "Household Savings and Portfolio Choices in in Thailand and Cote d'lvoire." In David. A. Wise, ed., Topics in Developing Countries: The Impact of Ill-Health, Income the Economics of Aging Chicago, Ill.: University of Chicago Uncertainty, and Credit Constraints." Stanford University, Press. Department of Economics, Palo Alto, Cal. Domar, E. 1946. "Capital Expansion, Rate of Groxvth and Lewis,W Arthur. 1954. "Economic Development with Unlimited Unemployment." Economietrica 14 (2): 137-47. Supplies of Labor' ."lfanchester School of Economic and Social Flavin, Marjorie. 1981. "The Adjustment of Consumption to Studies 22 (May): 139-91. Changing Expectations about Future Income." Journal of MaCurdy,T. E. 1982. "The Use of Time Series Processes to Model Political Economy 89 (5): 974-1009. the Error Structure of Earnings in a Longitudinal Data - 1993. "The Joint Consumption/Asset Demand Decision: Analysis." Journal of Econonietrics 18 (1): 83-114. A Case Study in Robust Estimation." University of California, Mirer,ThadW 1979. "TheWealth-Age Relation Among the Aged." Department of Economics, San Diego. American Economic Revieuw 69 (3): 435-43. Fry, Maxwvell. 1988. Money Interest, and Banking in Econon,ic Modigliani, Franco, and Richard Brumberg. 1954. "Utility Analysis Development. Baltimore, Md.: Johns Hopkins University Press. and the Consumption Function: An Interpretation of Cross- Gersovitz, Mark. 1988. "Saving and Development." In H. Chenery section Data." In K.K. Kurihara, ed., Post-Keynesian Economics. and TN. Srinivasan, eds., Handbook of Development Economics. New Brunswick, NJ.: Rutgers University Press. Vol. 1. New York: North Holland. Morduch, Jonathan. 1994. "Poverty and Vulnerability" American Giovannini, Alberto. 1983. "The Interest Elasticity of Savings in Economic Review, Papers and Proceedings 84 (May): 221-25. Developing Countries: The Existing Evidence." World National Bank of Pakistan, Economic Research Wing. 1992. Developmttent 11 (7): 601-07. "Monthly Economic Letter." 19 (7). Guiso, Luigi,TullioJappelli, and DanieleTerlizzese. 1996. "Income Paxson, Christina H. 1992. "Using WeatherVariability to Estimate Risk, Borrowing Constraints, and Portfolio Choice." TDe the Response of Savings to Transitory Income in Thailand." American Economnic Review 86 (1): 158-72. American Econoniic Review 82 (1): 15-33. Gurgand, Marc, Glenn Pederson, and Jacob Yaron. 1994. Outreach Pitt, Mark M., and Shahidur R. Khandker. 1998. "The Impact of and Sustainability of Six Rural Finance Institutions in Sub-Saharan Group-Based Credit Programs on Poor Households in .Ifrica. Washington D.C.:World Bank. Bangladesh: Does the Gender of Participants Matter?" Journal Hall, Robert E. 1978. "Stochastic Implications of the Life Cycle- of Political Economy 106 (October): 958-96. Permanent Income Hypothesis: Theory and Evidence."Journal Pitt, Mark M., Mark R. Rosenzweig, and Donna M. Gibbons. qf Political Economy 86 (6): 971-87. 1993. "The Determinants and Consequences of the Placement Hall, Robert E., and ES. Mishkin. 1982. "The Sensitivity of of Government Programs in Indonesia." World Bank Economic Consumption toTransitory Income: Estimates from Panel Data Review 7 (Septenmber): 319-48. on Households." Econonmetrica 50 (2): 461-81. Quisumbing, Agnes R. 1994. "Intergenerational Transfers in Harrod. R. F 1939. "An Essay in Dynamic Theory" Economic Journal Philippine Rice Villages." Journal of Development Economics 43 49 (1):14-33. (2):167-95. Havashi, Fumio. 1995. "Is the Japanese Extended Family Rosenzveig, Mark R., and Hans Binswanger. 1993. "Wealth, Altruistically Linked? A Test Based on Engel Curves."Journal of Weather Risk, and the Composition and Profitability of Political Economy 103 (3): 661-74. Agricultural Investments." Economic Journal 103 January) Hossain, Mahabub. 1988. Creditfor Alleviation of Rural Poverty: The 56-78. Grameen Bank itn Bangladesh. Research Report 65.Washington Rosenznveig, Mark R., and Kenneth I. Wolpin. 1986. "Evaluating D.C.: International Food Policy Research Institute. the Effects of Optimally Distributed Public Programs: Child 208 CHAPTER 20 SAVINGS Health anid Fanily Planninig Interventions." Amnerican Economic Udry, Christopher. 1995. "Risk and Saving in Northern Nigeria." Review 76 (3): 470-82. American Economic Reviewu 85 (5): 1287-300. 1993. "Credit Market Constraints, Consumption Van Wijnbergen, Sweder. 1982. "Stagflationary Effects of Monetary Smoothing and the Accumulation of Durable Production Stabilization Policies:A Quantitative Analysis of South Korea." Assets in Low-income Countries: Investments in Bullocks in Journal of Developmenit Economitcs 10 (2): 133-69. India." Journial of Political Economty 101 (2): 223-44. World Bank. 1989. Sub-Saliaran .4frica: From Crisis to Sustainable Schultz,T. Paul. 1990. "Testing the Neo-Classical Model of Famnily Growth. Washington, D.C. Labor Supply and Fertility."Journal of Humian Resources 25 (4): . 1991. World Developtnent Report 1991: The Challenge of 599-634. Developtnent. NexwYork: Oxford University Press. Solow, R. M. 1956. "A Contribution to the Theory of Economic . 1995. lVobrld Development Report 1995: Workero in an Growth." QuarterlyJournal of Economics 70 (1): 65-94. Integrating World. Ne-wYork: Oxford University Press. Thomas, Duncan. 1990."Intra-Household Resource Allocation:An Wolpin, Kenneth I. 1982. "A New Test of the Permanent Income Inferential Approach:" Journal of Human Resources 25 (4): Hypothesis: The Impact of Weather on the Income and 635464. Consumption of Farm Households in India:" Internationial . 1993. "The Distribution of Income and Expenditure Economiic Review 23 (3): 583-94. Within the Household:' Annales d'Econtomie et de Statistique 29 Yaron, Jacob. 1992. Successful Rural Finance Itistitutionis. WVorld Banlk January/March): 109-36. Discussion Paper 150.Washington, D.C.:World Bank. 209 q ~~Credit - 1 ~~Kinnon Scott The importance of credit for household welfare and the imperfect nature of credit markets have inspired governments to intervene in credit markets in a variety of ways. In particular, govern- ments have adopted policies designed to increase access to and use of credit among specific groups and to promote the provision of credit for specific purposes. In principle, the returns to credit use are high. Credit and productivity and the role of credit in improving is a correlate of economic growth and, at the house- household welfare, and identifies the data needed to hold level, credit can be a tool to improve and protect implement these methodologies. The second section welfare. However, the fact that financial markets in also includes an assessment of how well previous many developing countries suffer from information LSMS surveys collected the data needed to analyze imperfections, monopoly power, and segmentation credit issues. The third section contains a summary of limits the, extent to which credit can promote eco- the key issues for data collection and reconimenda- nomic growth and help households insure themselves tions on how to collect adequate data from house- against risk. hold-level, community-level, and national sources.The Designing successful policies to increase access to fourth and final section provides detailed discussion of credit and to develop well-functioning credit markets the draft module provided in this chapter. requires an understanding of the effects of credit, household behavior regarding credit, and how credit Credit and Policy markets function. Household-level data can provide policymakers with information on how households There are two key reasons why governments intervene currently use credit as well as on the interrelations so often in the credit market. First, credit can provide between credit, household characteristics, and house- significant benefits. Financial intermediation, as meas- hold welfare. ured by the growth and development of the financial The purpose of this chapter is to provide guidance sector, in general is positively correlated with eco- on how best to collect policy-relevant data on credit nomic development. (The issue of financial interme- from households and communities. The first section diation is covered in Chapter 20 on savings.) More presents an overview of the key issues related to cred- specifically, credit has the potential to improve the it, issues that can be analyzed using data from house- welfare of households. Second, credit markets often hold surveys like the LSMS surveys. The second sec- operate imperfectly in ways that prevent many house- tion reviews the various methodologies that have been holds from using credit-thus limiting the extent to used to measure the impact of credit on households which credit can improve household welfare.The first 211 KINNON SCOan half of this section discusses ways in which credit can time againist the expectation of future increases in their benefit households.The second half outlines the most income. Without the opportunity to borrow, the common credit market failures. household would not be able to take advantage of investment opportunities. The need for agricultural What is Credit? households to make seasonal investments is a good Credit is the trade of money, goods, or services at the example of the cyclical nature of the need for credit.' present time for a payment in the future. Credit can be Another example of this is when nonagricultural busi- provided in many different forms and under a wide nesses invest to increase their output during peak sell- variety of arrangements.The standard loan, whereby the ing periods. A final example can be seen in older per- lender provides the borrower with money in exchange sons lending to younger ones. for a commitment from the borrower to repay the lender in cash, may not be the most common source of TECHNOLOGICAL OPPORTUNITIES. As technological credit in developing countries. Nor is formal sector or opportunities are not equally distributed among indi- bank lending necessarily a key source of financing for viduals or households, those with fewer or poorer households in those countries. While lenders may be opportunities can benefit from investing in (or lending individuals or institutions whose main function is the to) those with greater or more productive technolog- provision of financial services, they may also be traders, ical opportunities. Both parties can gain from such an employers, landlords, or relatives of the borrower who arrangement. lend money only in particular circumstances. Another important, albeit indirect, benefit is that Credit itself can be provided in the form of stan- having access to credit serves as an insurance mecha- dard, formal loans or by a variety of informal means. nism for households, increasing their ability to bear Some common forms of informal credit are: trade risks (Eswaran and Kotwal 1990). Households that credit, whereby credit is offered in-kind and no cash know they can borrow to maintain their consumption transaction occurs; tied credit, whereby the borrower if an investment fails will be able to make riskier but is tied to the lender through other arrangements than potentially more profitable investments.This also leads credit (such as tenant-landowner or worker-employer households to adopt technology more quickly than relationships); rotating credit associations in which they would have done in the absence of credit- members pool their savings and at certain times have potentially further increasing their productivity and, in the right to borrow from the pool of funds; the pawn- turn, their income. ing of goods; and simple agreements-often with no collateral or interest payments-between individuals INCREASING RETURNS TO SCALE. For new technologies who know each other. and activities, higher levels of investment may lead to increasing rates of return. An extreme example of this Credit and Household Welfare is when the rate of return to an activity is zero until Using credit can have a significant positive impact on some initial level of investment is made. If the initial the welfare of households, particularly by increasing investment is more than the household can afford from their productivity and decreasing the costs of smooth- its income or savings, the household will benefit from ing their consumption. The main ways in which being able to acquire credit to finance that investment. households stand to gain from credit use, as Besley (1995) argues, are similar to the ways in which the TASTES. Individuals difter in terms of the relative economy profits from intertemporal trade: specifically, weight that they put on present and future consump- the benefits arising from timing differences, technolog- tion. If an individual who prefers to defer his or her ical opportunities, increasing returns to scale, and tastes. consumption to a later date lends to someone who needs to finance his or her immediate consumption, TIMING DIFFERENCES. The gap between the time of they both stand to gain. One key factor affecting tastes investment and the time at which returns are realized may be where people are in their life cycles. is the primary way in which credit can benefit house- holds.When households have access to credit, they can CONSUMPTION SMOOTHING. Credit decreases the costs borrow money to make investments at the present to households of maintaining certain consumption 212 CHAPTER 21 CREDIT levels.2When insurance markets and credit markets are ADvERSE SELECTION. Stiglitz and Weiss (1981) argue absent, households self-insure. This can lead them to that lenders' returns are not monotonic in interest hold assets that are either nonproductive or have low rates. This implies that there is an optimum interest productivity, but that are highly liquid. In periods of rate that lenders will charge. If the rate is set above this income shortfalls, they can sell these assets to produce level, lenders will receive decreased returns. This is the income that they need to maintain their con- because lenders know the mean risk associated with a sumption. Such self-insurance is costly for households pool of borrowers but do not know the risk associat- in a number of ways. First, liquidating assets held for ed with each individual borrower. As the less risky precautionary purposes may yield more money than a borrowers in the pool have a lower rate of return than household really needs at that point. Second, if an asset the more risky borrowers, increasing the interest rate has some productive value (as, for example, does live- will push the less risky borrowers out of the market, stock), liquidating it reduces household production. leaving behind a pool of borrowers with a higher Finally, in areas with high levels of covariate risk (such overall level of risk who tend to have high default as agricultural zones affected by the same weather or rates.Thus the increased interest rate will leave lenders pest problems), the price of these assets may be worse off than if they charged lower interest rates.The depressed if many households with similar income net result is that lenders maintain interest rates below shortfalls sell off their assets at the same time. the level of a market-clearing rate and lend less than Access to credit provides households with insur- they could if they had full and perfect information. ance by decreasing their need to hold assets for pre- cautionary reasons (Deaton 1991). This frees house- ADVERSE INCENTIVES. The interplay of imperfect hold resources for investments in more productive information and moral hazard creates adverse incen- ways to generate income. tives in the credit market. A borrower's actions affect In the absence of credit of any sort, households' both his or her returns and the returns for the lender. productivity is limited by their ability to self-finance However, while a borrower has control over his or her consumption and investments. This can often lead actions, his or her lender has only limited control over households to underinvest or make inefficient invest- these actions, and borrowers and lenders have different ments. In addition, the absence of credit increases the preferences. This is essentially the problem of moral costs to a household of maintaining its consumption hazard; the differences in preferences and degree of levels. By taking advantage of credit, households and control lead to conflict between the two agents. Given individuals can improve their welfare by increasing this situation, lenders may choose not to increase their productivity and lowering the costs of smooth- interest rates-and thus avoid creating adverse incen- ing their consumption. tives that they cannot control. For example, if lenders increase interest rates, borrowers may be more likely to Market Imperfections: Imperfect Information substitute riskicr projects, hoping that these projects The design of most credit policy and government will yield higher returns that will enable the borrow- intervention in credit markets is driven by perceived ers to cover the higher interest rates. However, because imperfections in existing credit rnarkets. Policymakers many of these riskier projects are likely to fail, the see imperfections (or barriers to trade) in credit mar- expected return to lenders is reduced. Lenders volun- kets as obstacles that prevent the full benefits of credit tarily keep interest rates low to avoid or minimize this from being realized. problem. The asymmetry of information between lenders This assumes that deciding which project will be and borrowers leads to a phenomenon called equilib- funded involves a moral hazard problem caused by the rium credit rationing. This situation prevails when borrower's hidden actions and the asymmetry of infor- lenders voluntarily set the price of credit (the interest mation about the project. It may also be possible to rate) below the market-clearing price, leading to an have ex post asymmetry of information (S. Williamson unmet demand for credit. This underinvestment is a 1986, 1987), in cases where the borrower knows how rcsult of both advcrse selection and advcrse incentivc successful his or her project has been but the lender effects (Jaffee and Russell 1976; Keeton 1979; Stiglitz has less information and cannot find out more with- and Weiss 1981; Stiglitz 1987). out incurring significant costs.This problem of moral 213 KINNON Scoan hazard with hidden information creates further incen- important role in how credit markets function. In tives for lenders to keep interest rates at a level below places where legal systems are weak and it is difficult the market-clearing rate. The lender only incurs the to get contracts enforced, many borrowers will only costs of verifying the returns to a project if the project honor a loan contract when the costs of defaulting fails (or the borrower states that it does). As raising exceed the benefits of default. Over the long term, a interest rates can result in more projects failing lender can increase the costs to the borrower of (because lenders fund riskier projects), this increases defaulting by making repayment a condition of giving the costs to the lending institution of verifying project the borrower access to credit in the future. Eaton and outcomes. Thus the gain from increasing the interest Gersovitz (1981) studied how borrowers would be rate is offset by the increased costs of verification. more likely to honor their contract if they thought that their reputation for being creditworthy would be TYPES OF RATIONING. As indicated in the above dis- damaged by defaulting.3 Diamond (1989) argued that cussion, the most common result of equilibrium cred- there are other benefits to borrowers of not defaulting it rationing is that the quantity of credit is rationed: in addition to continued access to credit. Because not all people who want credit receive it and the price lenders see borrowers who have not defaulted as hav- of credit is set below the market clearing rate. Another ing a lower probability of defaulting in the future, they type of rationing exists when all potential borrowers are able to lower interest rates for this group.This has receive loans smaller than the ones they need. In the the effect of increasing investment in low-risk proj- past, explanations of this phenomenon were based on ects. It also means that, as the interest rate decreases the assumption that there was a fixed return to invest- over time, the cost of default to the borrower rises ments (Hodgman 1960). This meant that there was a because the borrower stands to lose access to ever maximum loan size that could be covered by the cheaper supplies of credit. return on the investment and that lenders would The threat of denying access to credit in the therefore deny requests for bigger loans. Subsequently, future also affects the behavior of borrowers, who will some analysts argued that lenders' unwillingness to be inclined to embark on low-risk rather than high- bankrupt borrowers-due to decreased earnings in the risk projects (Stiglitz and Weiss 1983). This threat can future (Ryder 1962) and the high costs of bankruptcy only be effective if the borrower and the lender have (Miller 1962)-led to size rationing. In general, size an ongoing relationship. The lender stops lending to rationing occurs when the investment of loan funds is the borrower after the borrower defaults, signaling to subject to decreasing returns to scale (Jaffee 1972). other credit providers that this borrower is not a good risk. Market Imperfections: Contract Design and Enforcement Lenders are also affected by threats to their repu- The problems associated with moral hazard can be tation. While renegotiating loan contracts can often affected by the ability of a lender to write and enforce prevent borrowers from defaulting, in certain cases contracts. If it were possible to write contracts that lenders may force bankruptcy on a borrower instead could completely cover all eventualities and that could of allowing him or her to renegotiate the terms of the be enforced without cost, there would be few prob- contract. The ability of a borrower to renegotiate his lems of moral hazard. However, it is extremely difficult or her contract has been observed to affect the bor- to create a contract that describes each possible state of rower's future behavior, leading him or her to invest in the world in enough specificity for outside verification high-risk projects (Eaton and Gersovitz 1981); the cost to take place. The incompleteness of contracts has of default is lower to the borrower when negotiation been shown to have deleterious effects on the eco- can take place. nomic relationship between borrowers and lenders (Williamson 1985; Klein, Crawford, and Alchian Government Intervention 1978). Even where it is possible to revise and renego- The fact that credit has the potential to improve wel- tiate an original contract, suboptimnal levels of lending fare has led to governmeints' interest in increasing and investment can still result (Hart and Moore 1988). access to credit and use of credit, both overall and Of course, if a contract is to have any value, it among specific groups. Governments justify interven- must be properly enforced. The legal system plays an ing in credit markets on the basis that markets have 214 CHAPTER 21 CREDIT failed or are inefficient. It is inmportant to clarify what ket indirectly are ones concerning land ownership, the qualifies as credit market failure. A failure to achieve legal environment, and taxation. Land titling and land Pareto efficiency is the usual standard for defining redistribution programs can lower the costs of both market failure, but this may not be appropriate in the screening and enforcement by increasing the use of case of credit markets. Instead, as Besley (1994) argued, collateral for loans (Hoff and Stiglitz 1990; Besley it may be more relevant to apply the standard of a con- 1994). Improving the regulatory framework and strained Pareto efficiency that takes into account the ensuring that existing laws are enforced can also special circumstances of credit markets brought about improve the way credit markets operate. (See Besley by the repayment problem and by the asymmetrical 1995 for an overview of this issue.) Taxation policies nature of information. such as taxing interest income (De Meza and Webb Governments also justify intervening in credit 1988) can have a significant impact on credit markets markets on equity or redistributional grounds-for by affecting the levels of funds available. example, if some areas or groups are underserved by the market and only the government is willing to step ROBUSTNESS OF THE MODELS. Typically, when govern- in and provide these areas or groups with credit. It has ments intervene in credit markets, they do so on the been argued that people living in areas with poor assumption that credit is underutilized. Due to infor- agroclimatic conditions are underserved by the market mation asymmetries betveen borrowers and lenders, (Binswanger and Rosenzweig 1986; Binswanger, lenders maintain interest rates below the market-clear- Khandker, and Rosenzweig 1993), and some govern- ing rate even when there is surplus demand for cred- ments have used this idea of material risk to justify it. This is the implication of the Stiglitz and Weiss their intervention in the credit market.When assessing (1981) model. However, the robustness of this model policy it is important to distinguish between market has been called into question. In the Stiglitz-Weiss inefficiencies and the redistributional goals of credit model, projects are taken from distributions with a policy; the market may be operating efficiently but not common mean but different variances. Riskier proj- in a way that achieves the government's redistribu- ects are those from distributions with larger variances. tional goals. The expected return for all projects is the same. As de Although governments have an array of policy Meza and Webb (1987) demonstrate, if it is assumed tools with which to influence credit markets, govern- instead that the returns to successful projects are the ment intervention is not necessarily an appropriate same but that the probability of their success differs solution. The evidence about whether the level of (and hence the expected returns also differ), then too credit in an economy is too low is often contradicto- much credit is being supplied and interest rates are too ry. And past government interventions in the credit low. In addition, even without changing assumptions, market have not always been successful; some have introducing other elements such as screening costs even created further problems. into the equation would contradict any conclusions obtained using the Stiglitz-Weiss model (de Meza and CREDIT POLICIES. Government interventions in credit Webb 1988). markets can be either direct or indirect. The most These theoretical contradictions make it difficult common direct interventions are policies that subsi- to design suitable policies. If overinvestment exists, dize interest rates, lower interest rate ceilings, enforce introducing a policy of taxing investment income may lending quotas, and provide guarantees for borrowers be appropriate. On the other hand, if there is underin- without collateral. Examples of direct policies include vestment, introducing this policy will have a negative the massive and widespread programs of subsidized impact in that it will discourage households from tak- credit for agricultural development (and recently also ing advantage of credit opportunities. for microenterprise development) that exist in many developing countries. PREVIOUS EXPERIENCE. The need for policymakers to Indirect policies can also correct for inefficiencies be cautious when designing and implementing credit in the credit market arising from imperfect informa- schemes is suggested not only by theoretical contra- tion and the subsequent costs of screening and dictions about which policies are most appropriate but enforcement. Some policies that affect the credit mar- also by the practical experiences of governments. 215 KINNON SCOTT When governments have adopted policies of pure efficiency). Either way the net effect is to under- financial repression (McKinnon 1973; Shaw 1973), mine the credit market (see Besley 1994). this has led to disequilibrium credit rationing. As dis- Finally, the large number of credit providers in the cussed above, governments sometimes impose interest market can complicate the design of credit policy as rate ceilings lower than the market-clearing rate, to well as its implementation. As Aryectey and others protect consumers against usury and to make credit (1997) have demonstrated, changes in financial policy affordable and accessible. However, these policies often may or may not have the effects that policymakers result in a decrease in credit access. The interest rate expect. Weak linkages between sectors of the market, ceiling forces financial institutions to pay lower rates niche markets where barriers to entry by other lenders on savings-limiting the stock of funds that the insti- exist, and other manifestations of the fragmentation of tution has available from which to lend (Wolken and the credit market in many countries can all affect the Navratil 1981). outcome of financial policy. In addition, both theoretical work (Stiglitz and Weiss 1981; Hoff and Stiglitz 1993) and empirical PolicyAnalysis and Household Data work (Aryeetey and others 1997; Siamwalla 1990) have shown that the outcomes of direct policies aimed Analysis of household-level data on credit use can help at increasing credit for specific borrowers or activities policymakers determine which-if any-credit have often been the exact opposite of the desired out- polices are appropriate by providing them with infor- comes. Target groups' access to credit has not mation on existing structures and behavior as well as increased, nor have interest rates fallen. And subsidized on the impact of previous policies. Household-level interest rates and directed lending have not only led to data can be used to answer three types of questions: disequilibrium credit rationing (Vogel and Adanms * What is the structure of the current credit market? 1986;Adams 1984) and a decrease in loanable funds, For example, what are the sources of credit, what they have also reduced access to credit for small or are the costs of credit, and for what purposes do first-time borrowers who are supposed to be the ben- households use credit? eficiaries of these government policies. Because inter- * What evidence is there that imperfections exist in est rate ceilings lower the profits that lenders can be the credit market? expected to make, the lenders aim to maximize their * What is the impact of credit on productivity? profits by lending to borrowers carrying the least risk-who tend to be the larger and wealthier bor- Limitations of Household Data rowers. This bias holds true even in the absence of Before looking at these topics in detail, it is important objectively measured higher default rates (Guttentag to note some of the limitations of household data col- and Herring 1984). lected through surveys in general and through LSMS- Moreover, to the extent that lenders incur fixed type surveys in particular. These limitations relate to administrative costs for processing loan transactions, a both the quantity and the accuracy of data that can be larger loan will generate greater profit (Anderson and collected. Khambata 1985). And the artificially low interest rate First, in the interview approach, especially the ceiling makes it less feasible for lenders to pay to gath- structured interview and the type of interview with er information about borrowers-causing lenders to closed-ended questions, only a limited amount of trust ration the provision of credit to first-time borrowers and rapport can be established between interviewer (Thakor and Calloway 1983). and respondent.To the extent that credit use (like any Another area in which government intervention activity associated with money) is a sensitive subject, has negatively affected how credit markets function is this relatively impersonal interview approach can enforcement. Government programs that subsidize sometimes limit the amount and quality of informa- and direct credit have often had high default rates, to tion provided by the respondent. which governments have responded by not enforcing Second, legal and cultural factors can also influ- the repayment of loans, by forgiving debts, or both. ence how much accurate data respondents are willing The reasons why governments do this can be political to provide to a survey interviewer. If they are involved or due to redistributional goals (that can conflict with in loans whose interest rates are higher than the legal 216 CHAPTER 21 CREDIT maximum or if they are embarrassed at having had to level not only omit potentially large amounts of cred- borrow money, they will underreport their credit use.4 it use but also provide little insight into the distribu- Third, because of the wide variety of loans in the tion of credit, the benefits of credit use to borrowers, market and of the ways these loans are priced, serious unmet demand, and, in many cases, the ways in which difficulties arise in trying to measure credit and its credit is used. costs by means of a household survey. This is the case As a result, household-level data are the primary not only because it is inherently difficult to properly source of information for determining the effective- investigate all aspects of credit use but also because ness of credit policies. However, to ensure that the there is a limit to the number of questions it is possi- overview of the credit market provided by household ble to ask in an interview and to the length of that data is accurate, a survey must cover every kind of interview. In addition, analyzing the effects of credit credit use, whether by a given household or in a spe- use often rcquires longitudinal data, which many cific community or region. The two key aspccts of the household surveys do not collect. Panel data certainly credit market that affect data collection methods are can be, and have been, collected quite successfully, but the range of providers and the variety of arrangements there are some costs associated with designing and under which credit is offered. Both aspects are dis- implementing panel surveys, not the least of which is cussed below, using examples from actual data sets to the need to have sufficient funds for more than one illustrate key points. Unless othervise indicated, the round of a survey. (The costs may not be as high as examples are from past LSMS surveys. they first appear; see Chapter 23 for a thorough dis- cussion of the costs and benefits of collecting panel SOURCES OF CREDIT. In any given country there is data.) often a wide range of lenders and credit sources. At Fourth and finally, because LSMS-type surveys one end of the spectrum are formal institutions, typi- typically have small national samples, the resulting data cally national or commercial banks. These institutions often do not allow analysts to study rare events. For provide a broad range of financial services, have estab- example, if only a small percentage of the population lished and often complicated rules governing access to borrows from a specific lender or is covered by a gov- their services, and are subject to government regula- ernment program, the survey data set will contain too tion. At the other end of the spectrum are informal few of these cases to enable analysts to draw accurate lenders-lenders not covered by the regulations of a conclusions. country's central monetary authority (Adams and Fitchett 1992). These lenders are usually individuals Describing Credit Markets who provide one financial service-credit-and who Household-level data is crucial for providing a detailed supply this service based on their personal knowledge picture of credit markets. No other source can provide of the borrower. Moneylenders, the best-known type information on the entire credit market and on the of informal lenders, have been the subject of debate characteristics of credit recipients. In any given coun- for years (see Ladman 1981; Rao 1980; Adams and try, the national statistics on savings and credit and Nehman 1979; Bottomley 1963; U Tun Wai 1957). more disaggregated figures from financial institutions However, moneylenders are only one source of infor- provide information on the overall magnitude of sav- mal credit. Relatives, shopkeepers, and rotating credit ings and credit in the formal sector of the economy. associations are all part of this informal credit sector. However, this is only one (and, in many countries, not And in between the extremes of formal institutions even the largest) credit activity in the economy. and informal credit sources are semiformal institu- Estimates for nine Asian countries show that the non- tions, which include credit unions and certain types of formal sector accounts for anywhere from one-third nongovernmental organizations (NGOs). to three-quarters of all credit that is disbursed Only formal sector and some semiformal sector (Montiel 1993). It has been estimated that the amount operations appear in national accounts and other of nonformal credit disbursed in Honduras is as much national credit databases. Thus household survey data as the amount of formal credit that is disbursed sets are the only source of information on informal (Larson 1990)-although this is probably an overesti- sector lending in developing countries. This is true mate (Scott 1992).The aggregate data at the national even in countries where the distinction between the 217 KINNON SCOTT formal and the informal sector is not very clear. In data show, trade credit is by far the most important Honduras, for example, even though moneylenders are type of credit (in terms of numbers of loans). Credit allowed to register themselves as lenders and seek provided by shopkeepers and hire purchase arrange- restitution from loan defaulters, only 10 to 33 percent ments constitutes the majority of all lending in South actually do register (Scott 1992). Africa.This is also the pattern of lending in Pakistan. An illustration from the 1994 South Africa LSMS Supplier credit is quite widespread in Pakistan; for illustrates the importance of household surveys for cap- some types of items, more households receive credit turing credit use in an economy. As can be seen in Table from suppliers than receive standard loans. As can be 21.1, only 9.3 percent of all loans reported by house- seen in Table 21.2, while approximately 20 percent of holds in the 12 months prior to the survey were dis- all agricultural households received general loans for bursed by formal sector institutions, with another 5.8 agriculture, 24 percent of agricultural households pur- percent disbursed by institutions in the semiformal sec- chased fertilizer on credit. (This pattern has also been tor. If these data alone had been used to estimate the found elsewhere; see Loria and Cuevas 1984.) share of households receiving credit, only 6.9 percent of Nonagricultural household enterprises use loans-and all households would appear to have received credit, supplier credit-somewhat less than do agricultural when the actual figure, taking into account credit dis- ones, but both types of- credit still exist. bursed from all sources, is 45.8 percent. Informal sector Not only will total credit use be underestimated if lending evcn accounts for a significant share-37.9 information on all types of credit is not collected, but percent-of the total money borrowed. If analysts had conclusions about the kinds of households using cred- drawn conclusions about credit markets and policy out- it will be biased. Households tend to specialize in the comes in South Africa without using household data, type of credit they obtain. For example, it is unlikely these conclusions would have been very inaccurate. that a household will receive both supplier credit and a cash loan. This can be seen in Pakistan, where only TYPEs OF CREDIT. Most credit markets offer many dif- one-fifth of all agricultural households that received ferent types of credit, ranging from formal loans some type of credit for agriculture obtained both a backed by collateral with a set repayment period and monetary loan and supplier credit (Table 21.3). For interest rate to loans from friends and family members nonagricultural enterprises this was an even smaller who may or may not attempt to cover their costs or fraction. make a profit. In between these extremes are loan In addition, given imperfect information and the arrangements such as supplier credit and tied lending effects of reputation in credit markets, it is likely that (see below). For analysts to get an accurate picture of borrowers will tend to specialize in a particular lender: the credit market, it is important for a survey to cap- once a borrower has established a credit relationship ture all of these types of credit. with a lender, he or she will continue to use that Cash loans are probably the least common kind of source rather than any others.While longitudinal data loan in most developing countries.As the South Africa would be needed to demonstrate this, a single cross- Table 21.1 Loans by Source: South Africa, 1993 Source of loan Number of loans Percentage of all loans Percentage of money borrowed Formaj' 552 9.3 47.4 Sem iorm ai 343 5.8 7.4 Informal 4,755 80.1 37.9 Moneylenders, fnends, relatives 955 16.1 Shopkeepers 2,002 33.7 H re purchase 798 30.3 ' o... n cIas ''ie d..................................................................2 8 4 ...........................................................B ........................................................... 7...2................................................. .............................................. N o nclass ifi ed 2 84 4 .8 7. 2 Totai 5,934 i100.0 100.0 ............................................. ................................................................................................ I...................................................................................... Percentage of all hojseholds receiving credit 45.8 a. rcludes banks, buildong societ es, and government agenc es. b. Irc udes NGOs, employers, cred t un ons, and Durial soc eties. c. Includes moneylenders, landlords, family and fr-ends shopkeepers, and hire purchase arrangements (pu.chases tha, are paid for over t me). Source: Project for Statistics on Living Standards and Development, 1994. 218 CHAPTER 21 CREDIT A Table 21.2 Division of Loans and Trade Credit among Credit-Receiving Agricultural and Business Households, Pakistan, 1991 Households withagricultural enterprses Households with other businesses Type of credit Percentage of all credit received Type of credit Percentage of all credit received Agricultural loan, monetary. 19.5 Business loan, monetary' 17.1 .................... ........................................................ ............................................................................................. ............................... ........................................ Loan in pact 12 months3 6.8 Loan past 12 months3 7.5 .......... i................. ........................................... ...................... ........... ...........I....................I................................................................................ Seed en credit 9.7 Frequent purchases, supplier credit 9.8 Seed pactiaiy on crede 13.2 Frequent purchases, other credit 1.2 ............ ............ ................ ........... ........... ........................................ ....... ...............................................I............................................................. Fertilizer on credit 24.0 Infrequent purchases, supplier credit 5.3 Herbic des oe pesticides en credit 13.8 Infrequent purchases, other credit 1.4 ........ I............................................................................................................... *............................................ *............................................................... Buy durable goods on credit 3.9 Any credit for agricuiture 42.0 Any cr edit for business 32.6 ................................ ..................................................................................................................... - .............................................. .................. Snare of all households receiving 15.6 Share of all households receving 13.0 some agricultural credit some enterprise credit Note: Categories are not exclusive since households can have more than one type of cred t. a.7wo cifferent subgroups of the household were asked about overall oan use; both are included here. Source: Authors calculations from Pakistan ntegrated Household Survey, Table 21.3 Specialization in Types of Credit, Pakistan, 1991 Agricultural households Households with nonagricultural enterprises Percentage receiving credit 42.0 21.3 ....... ................ :...... .................................................................. ................................. ....................... ............................... ......................... Percentoge of these receiving Monetary loan only 36.5 49.4 .................................................................... ............................................................................................................................................................... Supplier creditonly 43.7 34.6 Monetary loan and supplier credit 19.7 16.0 Source: Author's calculations from Pakistan ntegrated Household Survey data. Table 21.4 Number of Sources of Credit, Pakistan and South Africa Number of sources Percentage of households in Pakistan Percentage of households in South Africa None 46.6 54.2 5 ~ ...................................................................................................... 36...5.................................................................................. 32...6................................ One 36.5 32.6 ............................................................................................... *.................................................................................................................................. One or more 53.4 45.8 ........ ......................................................................................................... *.......................................................................................................... Of households receiving credit, percentage that receive it from only one source 68.3 71.2 Source: Pakistar Integrated Household Survey 199: Sout- Africa Project for Statistics on Living Standards and Development 594. section of data can give some indications about cessful and what the policy's redistributive effect has whether this is in fact the case. Credit use patterns in been. Depending on what policy is being evaluated, both South Africa and Pakistan provide some initial the most relevant information about who receives support for this hypothesis (Table 21.4). Of all house- credit may be at the household level (what types of holds receiving credit, 68.3 percent in Pakistan and households receive credit), at the sector level (what 71.2 percent in South Africa received credit from only percentage of agricultural activities are financed by one source. credit), or at the individual level (which people borrow). RECIPIENTS OF CREDIT.When examining credit mar- Any data needed to analyze credit use should be kets it is crucial to identify who has benefited from collected at the same level of disaggregation at which government credit policies. This task is simplest credit data are collected. If a policy affects whether or when a policy is targeted to specific areas or groups, not certain households receive housing loans, to study or when a policy is designed, either directly or indi- this policy it is enough for analysts to have data on rectly, to affect a specific type of loan (for example, households' income and assets (to analyze their collat- a loan for housing, agriculture, or education). If the eral). If, however, a policy aims to increase women's goal of a policy has been to direct credit to specific access to credit, analysts will also need data on indi- individuals, finding out who has received the credit vidual women, on their incomes and assets, and on indicates whether or not the policy has been suc- some of their other characteristics. 219 KINNON Scon CAN LSMS-TYPE SURVEYS PROVIDE ACCUlRATE in the future that would make this easier. Analysis of MEASURES OF TOTAL CREDIT USE? As was illustrated the data from the few surveys that have addressed this above in the cases of informal sector credit in South issue in depth5 has shown that it is vital to include Africa and supplier credit in Pakistan, not covering all explicit questions about the sources and types of cred- of the sources and types of credit in a multitopic it and about the purposes to which credit is put. Only household survey can lead to serious mismeasurement when these questions are included will surveys yield of credit use-and the subsequent design and imple- enough data to give an accurate picture of total cred- mentation of costly, ineffective credit policies. Thus it it use. Once an accurate measure of credit use and debt is essential for surveys to ask questions about every is available, it becomes possible to explore other key conceivable source and variety of credit to ensure that issues regarding the credit market. It also becomes pos- the full extent of credit use is accurately measured. sible to determine whether there is a need for the gov- How well have previous LSMS surveys succeeded ernment to change its existing credit policies, make in covering all credit sources? Not very well. While additional interventions in the credit market, or both. basic information on borrowing has been collected in It is also very important to gather data on the many past LSMS surveys (Table 21.5), few surveys characteristics of borrowers and of the credit markets have included detailed questions about credit sources in which they operate. Previous LSMS surveys have or even general questions about using supplier credit been fairly good at collecting data on which house- for productive purposes. Questions on the use of sup- holds receive credit, because they have collected cred- plier credit have most frequently been found in it data at the household level along with data on the inquiries about agricultural enterprises, but even in characteristics of households. Knowing which house- these cases very few questions were included. Only the holds have access to credit programs, what determines Kyrgyz questionnaire included a general question access to credit use, and wvhat prevents households about the use of supplier credit. from using credit are first steps toward determining Another critical omission in most previous LSMS the impact of government programs. In past LSMS questionnaires was the purchase of food on credit, surveys it has been more difficult to assess the distri- which is an important dimension of the analysis of not bution of credit in a particular sector. An example is only credit but also consumption. Only the Pakistan household enterprises. Because data on household questionnaire and, to some extent, the South Africa enterprises were not collected on all household enter- questionnaire included direct questions about this. prises in sample households, it was not possible to Given the analytical importance of being able to determine what percentage of family-owned enter- determine total credit use, it is clear that some prises use credit.This meant that it was not possible to improvements could be made to LSMS questionnaires calculate the probability of any given household enter- Table 21.5 Types of Credit Information Obtained by Selected LSMS Surveys Loans Trade credit Non- All Specific By source agricultural Other Country Mortgage loans loans Implict Explicit Agriculture enterprises Food consumption Service Ecuador 1994 0 *^ O ona s98i8 0 - S 0 0. ................................... ................ .................................... 6................................................................................................................................ Ivory Coasti198S 0 0 S * 0 ......st'"n... 9-1..........................0....................0...................0...................0........ ..............................0....................0...................0................................................... Kyrgyz Rep. 996 0 .. u'-h..........i.c......19-4 .................. (,)..................0.......................................0....................0......................................................... ().................. ............................... Pert. i985 0 0 0 ...................................... ;................4................6................ ......................................0.......................................................................................... Pakistan 199! 0 0 6 0 * *0 South Africa 1994 0 S 0 0 0 0 Vietnam 1992/30***V * Indiaste that the questionna l-e conta,ned thorough quest ons on th s topic. O Indicates that the questionnaire partially covered this top c. a. Orily agricuitura oan nforrniatou Was collected. Nete. Th s tasbe ony shows whether each questonnare included questons asking fthe household had obtained credit of a specfic type.The tabe does not show whether the design of the questionnaire would y eld the data necessary to calculate nhe size of the loan,the total cost of credi, or other loan terms. Source: Relevant LSM`S quest onra res. 220 CHAPTER 21 CREDIT prise using credit and, therefore, it was also impossible This section examines several potential market ineffi- to extrapolate to the universe of household enterpris- ciencies that result from imperfect information and es. It is recommended for future surveys to collect data enforcement problems in credit markets, including on all household enterprises. Doing this will clearly credit rationing, inarket segmizentation, and monopoly increase the usefulness of the household-level data. power. One drawback of past LSMS surveys is that they did not collect data on credit use at the individual EVIDENCE Or CREDIT-CONSTRAINED HOUSEHOLDS. level. This is a serious flaw in these data sets in terms Due to the problems of adverse selection and moral of the analysis of credit issues, one that should be rec- hazard, interest rates function not just as price mecha- tified in future surveys.6 The need for collecting infor- nisms but also as screening mechanisms. Much of the mation on credit at the individual level is twofold. economic thcory of credit markets indicates that this First, banks, credit unions, and moneylenders lend to will result in the non-price rationing of credit and to individuals rather than to households. It is the individ- less credit use than wvould be the case in the absence ual who receives the loan, and it is his or her com- of information imperfections. But, as has been dis- mand over collateral that affects the lender's decision cussed previously, the opposite could occur. A variety about whether or not to grant the loan. Second, not of empirical efforts have attempted to determine the collecting credit information at the individual level level of rationing in credit markets. may lower data quality substantially as one household The first and simplest method is to ask households member will not know the details of all other mem- about their ability to obtain credit. This method was bers and their activities.This is the logic behind much used by Barham, Boucher, and Carter (1996) in of the design of LSMS surveys. For such a sensitive Guatemala and by Feder and others (1989) in China. subject like credit use, asking individuals directly can Basically, households are asked whether they have only help the quality of the data collected. Thus indi- applied for credit; if so, whether their application was vidual-level data are clearly necessary for assessing the rejected; and if it was approved, whether they obtained impact of credit-related policies. the full amount that they requested. Feder and others In addition, to determine available collateral it wiDl (1989) found that defining credit-constrained house- be necessary to collect information about individual holds this way was compatible with data on the over- ownership of key assets such as land, housing, and major all liquidity of the households. However, using this durable goods. (Past LSMS surveys determined house- method of determining whether households are hold ownership, not individual ownership.) credit-constrained also requires collecting information Determining individual ownership should require the on households that did not apply for credit. This is addition of only a few questions to find out the identi- because, as Baydas, Meyers, andAguilera-Alfred (1992) fication code of the owner of each good. These can be pointed out, some of the households that do not apply added to existing sections on household asset ownership. for credit do not do so because they believe that they Future surveys should also gather data on govern- will be refused. (This point is very similar to the argu- ment credit programs and eligibility prior to designing ment for including discouraged workers in unemploy- the questionnaire. Questions should be added to the ment statistics.) questionnaire at the individual level aimed at identify- A second way to determine whether there is cred- ing both who is eligible for benefits from such pro- it rationing in the market is to estimate the shadow grams and who receives these benefits. price of credit, as was done by Sial and Carter (1996) in Pakistan and by Carter andWiebe (1990) in Kenya. Evidence of Market Imperfections These analysts examined differences in output and net The key argument that governments use to justify income among farmers based on their observable and intervening in the credit market is market failure. unobservable assets and skills and their use of credit. At Before a government decides to intervene in the mar- the observed average loan size, the marginal returns to ket, it should ensure that it has clear evidence of mar- borrowers were many times higher than the interest ket failure. rate charged on loans, and loan sizes were below the Adverse selection, moral hazard, and enforcement level that would allow the farm household to maxi- problems all affect credit markets in a variety of ways. mize incomne. 221 KINNON SCOan Third, many of the tests of liquidity constraints few surveys have also included questions about whether have becn donc in the context of analysis of consump- houscholds have ever applied for credit and, if so, on the tion. The theor,y of the optimization of consumption results of such application and the reasons given for over time says that household consumption is con- refusal-even though such questions are perfectly feasi- strained only by its lifetime budget constraint. If true, ble (and could easily be added to future surveys). this means that households' propensity to increase con- It would be more difficult to collect the necessary sumption is less if they receive a temporary increase in data for estimating the shadow price of capital, since income than if they receive a permanent increase in this requires being able to calculate rates of returns to income. However, empirical studies have shown that different activities. While it is possible to envision consumption is more sensitive to temporary income doing this in a study of a particular region or in a sam- changes than this theory would suggest, perhaps pie of a particular group such as coffee farmers or because households suffer from liquidity constraints. dairy farmers, it is unlikely that sufficient information To test this hypothesis, several analysts have could be collected at a national level to carry out this attempted to measure the impact of liquidity con- type of analysis. This is especially true given the small straints on household consumption.These studies have samples associated with LSMS surveys. Although the yielded some evidence of such constraints, but this sample used by Sial and Carter for their study in evidence is inconsistent. (See Chapter 5 oIn consunip- Pakistan (1996) was small (around 125 farniers), all tion and Chapter 20 on savings. See also Deaton 1992, sampled farmers were engaged in growing the same chapters 5 and 6.) Hall and Mishkin (1982), using the principal crop. It is not clear that such concentrations permanent income hypothesis, did find some evidence of activities will prevail in a small national sample. (See of such constraints. Zeldes (1989) explicitly used Euler Chapters 18 and 19 in this volume, on household equations (as first done by Hall 1978) to test the enterprises and agriculture, for a discussion of the fea- hypothesis that households maximize their lifetime sibility of calculating rates of return.) utility subject to liquidity constraints. Zeldes specified Whether tests of excess sensitivity using LSMS the model and split the sample households in advance surveys are feasible is a matter of some debate. First, into two groups, one of which (with low or negative these tests rely on panel data. Most past LSMS surveys wealth) he specified as being credit-constrained and have not collected panel data, although panel data may the other not. However, others have argued that if be more common in the future. (See Chapter 23 for a households are liquidity-constrained, they may change discussion of the advantages and disadvantages of col- their consumption without violating the Euler equa- lecting panel data.) A second issue is that the variables tions, thus weakening the usefulness of such tests needed for these analyses are often the variables most (Deaton 1992). difficult to measure adequately or accurately. Unlike the tests of liquidity constraints discussed Measurements of wealth and income may be tainted above, Hayashi (1985a, 1985b) and Japelli and Pagano by error. (See Chapters 17 and 20 for details of the dif- (1995) used cross-sectional data. This type of analysis ficulties involved in measuring income and measuring relies on splitting the sample households, based on the savings.) And these problems are compounded in level of their savings, into those that are credit- economies where a large proportion of the population constrained and those that are not. The analysts then works in the informal sector-adding to the difficul- estimated the level of consumption desired by the ties of measuring income. It is not surprising that the liquidity-constrained households and measured the bulk of empirical studies in this field have been done gap between desired and actual levels of consumption. on industrial economies. Hayashi as well as Japelli and Pagano found evidence that younger households-those with heads under the EVIDENCE OF SEGMENTED CREDIT MARrETS. One age of 30-were credit-constrained. common problem in developing countries is the frag- mentation of the credit market. In a fragmented cred- ADEQUACY OF LSMS DATA FOR STUDYING CREDIT it market, lenders provide loans only to certain sub- RATIONING. Most previous LSMS surveys have includ- groups of the population or to certain regions. The ed questions on what credit has already been obtained result is that there is often little overlap between the by households and on what money they owe. Only a clients of the various formal, informal, and semiformal 222 CHAPTER 21 CREDIT financial service providers. This outcome can arise for market that could indicate segmentation is easily gen- one of two reasons. Lenders may specialize in certain erated using variables readily available from LSMS sur- market niches-which is efficient when prices in dif- veys. If data are collected on the sources of credit used fereDt markets (formial, infornmal, and semiformal) by individuals in the sample as well as the purposes for reflect differences in the costs of funds and levels of which these individuals borrowed, it becomes possible risk in these different markets. Or credit markets may to deduce the degree to which borrowers are confined be fragmented because of market segmentation, in to one segment or another of the credit market. The which price differentials exist because funds do not fact that these data are needed proves again the flow between areas or groups. This is inefficient immense value of collecting exhaustive data on all because it limits the total amount invested in the sources and types of credit. economy, leads to suboptimal use of investment The other information required for determining resources, and, in rural economies with high levels of the segmentation in the market-information on covariate risk, increases the economic vulnerability of default rates, loan administration costs, and the costs of both borrowers and lenders. funds to lenders-is more difficult to collect with an Determining the difference between specializa- LSMS survey. While some informal lending may be tion and segmentation requires data from both bor- captured in the household survey, such surveys proba- rowers and the lenders. It is necessary to identify all bly cannot adequately cover issues such as lenders' cost potential lenders and to collect detailed information of funds and delinquency rates, which require infor- on both their costs and their risks. It is not usually dif- mation specifically provided by lenders. (Asking ques- ficult to identify formal lenders, but informal lenders tions about informal lending may also be risky. In the may be significantly more difficult to identify. Data for 1990 Thailand study by Siamwalla and others such a study of credit in Thailand were collected by having questions were not asked as it was felt that informants interviewers live for up to two months in villages in would not wish to answer. Still, many other LSMS Thailanid not only to be able to identify the key surveys-for example, the ones in Cote d'Ivoire, lenders but also to build up enough trust so that the Vietnam, Ghana-have asked such questions with no lenders would be willing to provide them with infor- reported difficulties.) In past LSMS surveys, questions mation on their costs, procedures, and default and on household lending were typically included more to enforcement problems (Siamwalla and others 1990).A identify a household's assets than to study its lending further way that informal lenders have been identified or borrowing activities. In addition, household-level is through interviews with households or borrowers, as data cannot provide a probability sample of lenders- were done in a study of credit markets in Ghana, limiting the usefulness of this information even if all of Malawi, Nigeria, and Tanzania (Aryeetey and others the details of household members' lending activities 1997). could be collected. Once lenders arc identified, it is possible to deter- Therefore, what is required to carry out this mine whether the market is segmented by comparing analysis is a survey of credit facilities, like the surveys risk-adjusted interest rates between or among sectors. of health and education facilities in such countries as As in Aryeetey and others (1997), this requires data on Ghana and Jamaica. A credit facilities survey has not explicit interest rates, rates of delinquency and default, yet been performed as part of an LSMS survey, but this costs of administering loans, and costs of funds to each type of data collection exercise has been done else- lender. As it was not possible in this study to gather where (see Aryeetey and others 1997; Udry 1990; panel data, two-year retrospective questions were Siamwalla and others 1990). Attempting to revise the added to the questionnaire to attempt to track any methodologies of these surveys to fit the LSMS may changes in these variables. Of course, this may have led not be entirely straightforward. One key concern will to bias if any informal lenders dropped out of the mar- be to establish what "the community" means in the ket during those two years and were thus excluded context of credit. (See Chapter 13 for a discussion of from the sample. this issue.) In many areas of the world, traders who live outside of the physical community are an important ADEQUACY OF LSMS DATA FOR STUDYING MARKET source of informal credit. However, in countries SEGMENTATION. The basic description of the credit where credit markets are a focus of government atten- 223 KINNON ScoTa tion, attemptinig to carry out such a facility survey may the case of teniaint farmers), such contracts appear to be warranted. provide the lender with an inordinate degree of mar- ket power. However, this is not necessarily the case. EVIDENCE OF MONOPOLY POWER. One facet of seg- When lenders use such contracts as screening devices, mentation can be the existence of monopoly power they are a source of monopoly power as only the within the market. Because lenders have only partial lender has the information on the borrower's potential and imperfect information about potential borrowers, for default (Bardhan 1989; Hoff and Stiglitz 1993). But they are forced to invest in screening mechanisms such when lenders use tied contracts to counteract prob- as establishing other economic links with borrowers lems of moral hazard and imperfect information, they and identifying collateral. This investment in screening may be doing so in the context of either perfect com- creates a "relationship-specific capital" (Hoff and petition or a monopoly (Braverman and Stiglitz 1982). Stiglitz 1990) that essentially ties borrowers to specific Determining whether any lenders have a monop- lenders and allows lenders to set their interest rates oly in the credit market requires data on the costs to higher than their marginal costs. The lack of barriers lenders of providing credit and on the fees that lenders to entry of lenders in credit markets does, however, charge. Aleem (1990) studied lender costs in order to limit the extent to which any one lender can monop- demonstrate market power in a rural area of Pakistan. olize the market. And in the informal sector, existing By collecting detailed information on the lenders' monopoly power may not be inefficient. As argued by screening and enforcement costs, Aleem calculated the Basu (1989), if the monopoly provider is a perfectly marginal and average costs of lending in that area. He discriminating monopolist, there is no market ineffi- found that the interest rates charged by informal ciency. The argument for the government to intervene lenders were close to average costs and above marginal in such a case is based on the fear that the lender is costs. The fact that lenders can only do an incomplete exploiting the borrower, rather than on the criterion screening of potential borrowers and the fact that bor- of market inefficiency. However, there is evidence of rowers have little information about other loan sources inefficient monopoly power in informal credit markets results in lenders specializing in specific types of loans in some developing countries and, to the extent that and in the existence of too many lenders in the mar- such power reinforces the segmentation of credit mar- ket.This causes interest rates to increase as lenders have kets, this creates additional inefficiency. to spread the costs of screening and enforcement over The substantial gap between formal and informal a smaller than optimal pool of borrowers. sector interest rates is often given as evidence of Aryeetey's study of four African countries (1997) lenders' monopoly power. However, the gap in inter- also focused on lenders' costs, comparing these to the est rates alone is not sufficient to support such an interest rates that they charged. He found evidence of argument. In the first place, it is not the nominal inter- significant market power among informal sector est rate that should be the focus of attention but rather lenders. Even though the informal lenders had lower the total cost of credit to the borrower. Adams and transaction costs, lower default rates, and lower costs of Nehman (1979) showed that the gap between formal funds than their formal sector counterparts, the inter- and informal sectors disappears if the total cost of est rates they charged were substantially higher than credit (calculated as the sum of the nominal interest those of formal lenders. rate and all of the borrower's transaction costs includ- ing loan charges, payments to third parties, and the ADEQUACY OF LSMS SURVEYS FOR MEASURING costs of his or her time and travel7) is used to compare MARKET POWER. It is probably beyond the scope of the formal and informal sectors. Data on formal sector LSMS and other multitopic surveys to gather the loans in Bangladesh, Brazil, and Colombia have shoxvn information necessary to calculate the total costs to that, for small borrowers, the nominal interest rate rep- borrowers of obtaining and repaying loans and the resents only a minimal share of the total cost of total costs to lenders of providing loans. First, doing credit-9 percent, for example, in Bangladesh. both of these calculations requires having accurate Tied contracts have also been cited as examples of information on interest rates. Second, estimating total monopoly power. By rendering borrowers dependent cost to a borrower requires having detailed informa- on lenders for other aspects of their welfare (such as in tion on all of the activities that the borrower under- 224 CHAPTER 21 CREDIT takes to obtain and pay back the loan. Third, estimat- In addition, lenders do not always charge explicit ing cost to a lender requires extensive information on interest rates. To compensate for imperfect informa- all of the activities associated with lending, including tion, they often use tied or interlinkcd contracts. This default and delinquency rates. further complicates the calculation of interest rates. If Determining the interest rates borrowers are pay- an interlinked contract does not specify an interest ing may be quite difficult. While information may be rate, analysts need detailed information on the com- readily available on formal sector interest rates, it is plete loan arrangement-including the in-kind com- harder to obtain information on informal sector rates. ponent, its value under the loan agreement, and an One of the characteristics of informal lenders is that alternative valuation-in order to be able to estimate they do not advertise (Aleem 1990). In theory it should an interest rate. If the survey is taking place prior to be possible to identify interest rates simply from house- the repayment of the loan, it may not be possible to hold interview data concerning loans received. But in calculate the real interest rate if the loan repayment practice simple questions to households about the involves, for example, some percentage of the borrow- interest rate they pay on a loan do not provide analysts er's harvest. with accurate information, for three reasons. First, Finally, religious bans on explicit interest may lead households may know that they pay an interest rate but lenders to use tied or interlinked contracts or other may not know what it is. Second, households may not repayment mechanisms besides charging interest. Islam be aware that they pay an interest rate because of dif- prohibits interest charges; the 1991 LSMS data set for ferences in terminology among lenders.Third, because Pakistan, an Islamic country, showed that 92.2 percent of religious laws against the charging of interest (as in of all the loans recorded in this country had no set Islam) or because the lender is attempting to improve interest rate. In the 1988/89 survey data for Ghana, screening and enforcement mechanisms, there may be which also has a significant Muslim population, a sim- no explicit interest rate, as an interest rate negatively ilar pattern was found.10 Lenders in these countries affects the expected return to the lender. (See the ear- presumably adopted alternative mechanisms to obtain lier discussion of adverse selection.) a return on their investments. Survey questionnaires Evidence from both LSMS and non-LSMS must be designed to ensure that data are collected on household surveys indicates that all of these factors these alternative mechanisms. confound analysis of the cost of credit and the exis- Determining interest rates is a difficult task-a tence of monopolies. A survey conducted by the task that past LSMS questionnaires have not effective- author in rural areas of Honduras showed that 40 per- ly accomplished. Most past LSMS questionnaires have cent of all borrowers were unable to tell the inter- asked about interest rates using questions such as: viewer what interest rate they paid on their loan (Scott "What is the annual rate of interest?"These questions 1992).Approximately 37 percent of people in Pakistan were often asked of the household respondent even if were unable to remember their interest rate.8 In sur- this person was not the borrower. veys that asked people not whether they paid an inter- One simple improvement would be to ask bor- est rate but simply what rate they paid, a large propor- rowers directly instead of asking household respon- tion of the answers were "zero." While these answers dents. Howevcr, this alone will not solve the problem; may have been true as UJdry (1990) found in highly borrowers were asked directly in the Honduran study segmented informal markets in Nigeria-the large cited above in which 40 percent of borrowers were number of zero answers may also have reflected unable to tell the interviewer the interest rate they respondents' lack of knowledge about the fees that paid on their loan (Scott 1992). they were being charged. In Peru in 1994 almost two- It is more important to focus on the wording of thirds of all borrowers stated that the interest rate for the question. This requires prior knowledge of both their loans was zero percent.While a number of loans the types of loans provided in a country or region and from family and friends might have had a zero explic- the terms used to describe these loans. Interest rates it interest rate, it is unlikely that two-thirds of all loans may vary among providers but be described in a uni- were interest-free.9 A more likely explanation is that form way. In such cases it becomes especially impor- the terms in which many informal sector loans are tant to find out the monetary (as well as linguistic) couched never mention the word "interest." "terms" of loans provided. Among moneylenders in 225 KINNON SCOaT Honduras, for example, a common form of informal The Impact of Credit credit is that for every 100 units borrowed, the bor- The willingness of governments to intervene in cred- rower must pay 110 when the (short-term) loan is it markets is based on the perceived benefits of such due.1' An interest rate is being charged here, but since intervention. As discussed in the previous section, gov- the borrower does not classify or understand it in this ernments often justify intervention by citing market way, asking about an interest rate will not yield accu- failures and inefficiencies that need to be rectified so rate data. Hoxvever, when borrowers are asked how that households can benefit from credit. This section much they pay per 100 units or per month, they prove examines how household-level data can provide ana- as knoxvledgeable about the subject as would be lysts with information on the effects of credit. expected. Properly wording questions about interest EVIDENCE OF THE IMPACT OF CREDIT ON requires a fair amount of investigation and experi- PRODUCTIVITY. An important reason why govern- mentation. Semiformal lenders can provide useful ments introduce policies that aim to increase the avail- information about informal credit arrangements, since ability and use of credit is to increase the efficiency they tend to view the informal sector as their compe- and productivity of investment. Making investment tition and are often quite knowledgeable about its more efficient and productive improves the economic activities. Siamwalla and others (1990) found that vil- welfare of households that depend on that investment lage headimien in Thailand were a good source of at activity-and of the economny in general. However, least partial information on informal lenders. even where government policies have increased the Just as past standard LSMS questionnaires have amount of available credit, it is not always apparent not collected accurate interest rate information, nei- that productivity has increased, or that it has increased ther have they enabled calculation of the total costs to a level high enough to justify the intervention.This of credit to borrowers. Little information has been seeming lack of success may be due to the fact that it collected on the transaction costs of credit. No data is difficult to measure the impact on productivity of has been gathered on the distances that borrowers increasing the amount of credit available. Or it maybe travel to get to lenders, and only a few of the addi- due to the fact that, for a variety of reasons, there has tional fees are ever mentioned. While several ques- been little or no impact on productivity. tionnaires have addressed the issue of agricultural Various studies have shown that credit increases extension agents, they have ignored the fact that bor- crop output-thus, if land is constant, increasing pro- rowers often pay fees to such agents for providing ductivity. Feder and others (1989) found that credit- verification needed for obtaining a loan. The ques- constrained households had lower output than non- tions that would need to be included in a household constrained households. In data from India, Khandker survey to calculate total borrower costs are not com- and Binswanger (1989) found that a 10 percent plex. Such questions have been incorporated into the increase in the amount of credit used led to a 2 per- draft module in this chapter. cent increase in output. Further work in India has Estimating the total costs to the lender of screen- shown the impact of credit on productivity to be of a ing and enforcement activities as wvell as of funds and similar magnitude (Binswanger, Khandker and related activities is beyond the scope of an LSMS sur- Rosenzweig 1993; Binswanger and Khandker 1995). vey. The data needed for such estimations can only be This increase was found to come primarily through collected in detailed lender surveys, which may face increased fertilizer use. sampling problems and the difficulty of getting access There are a variety of difficulties involved in meas- to the whole range of lenders (see Aleem 1990 for an uring the impact of credit on productivity. One key issue idea of the scope of this undertaking). Lenders have concerns the simultaneity of credit demand and supply. little incentive to provide interviewers with complete Actual credit use is the result of botli demiand and sup- informnation, particularly on the kind and extent of ply of credit. In order to measure the impact of credit on cost data that would be required for this analysis.Thus, productivity, analysts need an exogenous determinant of in the context of an LSMS survey, it is probably unre- the supply of credit. Khandker and Binswanger (1989) alistic to attempt to collect data on the total costs of used the number of rural commercial bank branches in lending. any given area to generate such a variable. A further 226 CHAPTER 21 CREDIT paper by Khandker and Binswanger (1989) employed a ductivity, a policy aiming to increase credit will not two-stage procedure wherein an equation that included necessarily increase productivity as expected. The lack the number of offices or branches of formal sector of complementary activities and the lack of input and lenders was used to estimate the volume of supplied output markets limit the extent to which credit can credit, and this estimated volume of credit was then used increase agricultural productivity (Feder and others to calculate the impact of the credit on input and out- 1990, 1991; Choe 1996). Third, the impact of puts for borrowers. These papers only addressed formal improved access to credit in one sector may actually sector credit, as there is no source of equivalent infor- affect a second, different sector. This will be the case if mation on all informal credit providers. the true credit constraint is in the second sector (Feder A single cross-section of data is of limited use in and others 1989). Also, to the extent that credit is fun- capturing the impact of credit on productivity. Instead, gible, the impact of credit may well be on an activity information is needed on the initial use of credit by a other than the one for which funds were nominally given borrower, the borrower's initial level of produc- borrowed. tivity, and the resulting productivity. In an effort to get around the need for multiyear or multiperiod data col- ADDITIONAL IMPACTS OF CREDIT. Increased credit use lection, Feder and others (1989, 1990, 1991) have used has been shown to raise the level of inputs used by cross-sectional data that incorporated retrospective farmers, boosting fixed investment in equipment such questions. However, survey designers should be cau- as tractors and irrigation pumps (Khandker and tious about including retrospective questions, as Binswanger 1989; Feder and others 1989; Binswanger respondents' memory errors can cause significant bias and others 1993; Binswanger and Khandker 1995). In in the resulting data set. (See Scott and Amenuvegbe addition, credit use affects labor markets in rural areas 1990 for an overview of the types of problems arising (Khandker and Binswanger 1989; Binswanger and from reliance on respondents' memory.) Khandker 1995). On the one hand, capital is substitut- In addition, measuring productivity requires hav- ed for labor and agricultural labor declines slightly. On ing detailed information about the economic activity the other hand, nonagricultural labor increases, often in question (whether it is a nonagricultural household substantially. The net result is a slight increase in agri- enterprise or an agricultural activity) both before and cultural wages. after the use of credit. For nonagricultural activities The use of credit also affects other household this implies collecting data on such factors as business behavior and the intrahousehold allocation of inputs, production processes, outputs, and general eco- resources and decisionmaking. In their study of the nomic trends for regions. For agriculture, information effect of credit on such household level outcomes in is needed on factors such as agricultural inputs and Bangladesh, Pitt and Khandker (1996) focused on the outputs, climatic conditions, and crops. (See Chapters impact of credit provided through group lending 18 and 19 in this volume for details of how to collect schemes on schooling, labor, household expenditures, such information.) the distribution of assets in the household, malnutri- A final point that needs to be considered when tion, and contraceptive use and fertility. The study measuring the impact of credit on productivity is that explicitly controlled for the endogeneity of participa- because of the nature of credit markets, policies tion in credit programs by using a quasi-experimental increasing credit access may have little or no effect on design that included both households that had the productivity in the activity for which funds were bor- choice of participating in the program and households rowed. First, for an increase in credit to have an impact living in other areas that did not. Additionally, the on productivity, access to credit must have been limit- authors estimated the effects of credit separately for ed to begin with. As shown in Feder and others male and female borrowers. The study showed (1991), many households do not borrow because they increased credit in a household to positively affect have no need for credit-not because they are credit children's years of schooling, the labor supplied by the constrained. Some support for this interpretation can household, and household expenditures. For contra- be found inWiens (1998) on Panama. ceptive use and fertility, the gender of the credit recip- Second, in cases where credit is neither the only ient affected the results. If the credit was received by a nor the most significant constraint for increased pro- man, fertility increased and contraceptive use declined; 227 KINNON SCOan the opposite held true for women recipients. Women household to list all of the loans that the household may receiving credit also increased their asset holdings. have received. This has a serious drawback for measur- Comparisons of these findings with the results ing the impact of credit on productivity. As Pitt and that would have been obtained if neither gender nor Khandker (1996) showed in their Bangladesh study, this the endogeneity of credit program participation had impact varies depending on whether men or women been taken into account demonstrate the importance received the credit. Thus, as discussed previously, credit of these refinements. A final point that will need to be data needs to be collected at the individual level and not addressed, however, is to separate the effects of the simply aggregated over all household members. credit itself from the technical assistance tied to the credit. The survey instrument used by Pitt and LSMS Surveys and Credit Summary Khandker incorporated questions concerning techni- A summary of the policy concerns discussed in this cal assistance activities. section can be found in Table 21.6. The table lists key credit issues for which household data can be used and ADEQUACY OF LSMS-TYnE DATA FOR ASSESSING THE the data required to carry out analyses of these issues. IMPACT OF CREDIT ON PRODUCTIVITY. To assess the This chapter introduces two draft LSMS credit mod- impact of credit on productivity, analysts need two types ules: a standard version and an abridged version. (The of data. First, analysts would need to have data from two questionnaire pages are presented in Volume 3.) The or more points in time-preferably panel data.To date, last two columns of the table show the adequacy of most LSMS surveys have not been designed or imple- each of the two draft modules in collecting the infor- mented to gather panel data, making it impossible to mation needed for each type of analysis. assess the impact of credit on productivity. As noted above (and discussed more thoroughly in Chapter 23), Data Collection it is feasible to collect panel data as part of the survey design. If measuring the impact of credit is important, Two points become clear from the previous sections. panel data should be collected.While it may be possible First, it is not a simple matter to collect adequate data to substitute longer recall periods for panel data by on credit use and on the impact of credit and govern- adding retrospective questions, analysts will need to ment credit policies. Second, past LSMS surveys have carefully assess the quality of such data because some not dealt with many of the complexities involved in respondents may incorrectly remember some details, gathering data on credit use. This section outlines the thus introducing significant bias into the data. key practical issues involved in collecting data on cred- Second, analysts need detailed information on the it use and presents tvo draft questionnaire modules-a productivity of a wide variety of economic activities. standard version and a shortened version-designed to Except in areas where only one crop is grown or in collect credit data. This section also provides recom- cases where analysts are examining only one specific mendations on how other modules of the questionnaire type of enterprise, the effort and expense required to will need to be adapted to ensure that sufficient data is collect such data may be too burdensome for a multi- collected. In areas where LSMS modules are insuffi- topic survey like the LSMS. It is not clear whether cient, the section lists other sources of relevant data. household enterprise and agriculture modules of past LSMS surveys have collected sufficient data for analy- Key Issues ses of the impact of credit on productivity. In deciding In any data collection exercise, the format of a ques- whether to gather information for such analyses, sur- tionnaire and its modules is shaped by the current pol- vey designers need to take into account the costs (par- icy priorities of the country studied. However, as has ticularly in terms of interview length) of attempting to been illustrated in the previous sections, several con- gather the information needed. See Chapters 18 and siderations will affect almost all surveys that aim to 19 for further discussion of this topic. gather information about credit. Such considerations A final consideration concerns the way LSMS sur- are summarized in this subsection. veys obtain data on credit. As noted in a previous sec- tion, LSMS surveys have typically collected credit data EXHAuSTIVE COVERAGE OF TYPES OF CREDIT. To ensure at the household level by asking one person in the that all types of credit are covered in the survey, design- 228 CHAPTER 21 CREDIT Table 21.6 Summary of Policy Issues and the Adequacy of the Proposed Modules to Address Them Data requirements Questions in Specific issue(s) from credit Other modules from Short module Standard module Standard Credit Policy issue for analysis module which data are required adequate? adequate? Module Total credit use Source, purpose. Listing of all Agriculture; Yes (if data are Yes (if data are Part A: 3-4, 7-8, by households amount of debt household credit Household enterprise; also collected in also collected 11-12, 15-16, use by source Consumption (data on other modules) in other modules) 19-20, 22-23 purchases of goods and food on credit) ....... ....... ....... ....... ....... ....... ...... ....... ....... ....... ........................................................................... ....... ....... ....... ....... ....... ....... ...... Incidence of Credit Credit use Characteristics of Yes Yes Part B: 3,31, 55 government programs by program households related to programs program, for example, housing module in the case of mortgage program, agricultural module in case of agrcultural lending program, and so forth Credit Ability of Refusals of loan Cnaracteristics of Yes Yes Part C: 1-6 rationed households tc applications, not households such as households obtain credit requesting loan welfare level due to belief that (consumption), collateral will not receive (housing module, durable goods); Shadow prico Loan size and Agriculture; No Possibly, depending Part B of credit cost Household enterprise on data collected (data on output, net in other modules incomes, other relevant characteristics of farmers and enterprise owners) .................................................................. Loan ... size................................................................... N... ............................. Probably..................... Part... .................... Cost of Loan size, No Probably Part B credit transaction and interest costs Market Sources and Lenders'costs (defau t Limit ed unless a Limited un e ss a Part B:. 1-3, segmentation uses of credit rates, loan ddministrdtion facility survey of facility survey of 29-31, 53-55 costs, cost of funds) from lenders is carried lenders is carried facility survey out out Impact of Cost of credit Agricultural: No Yes, if panel data Part B credit on Household enterprise are collected productivity (for measures of productivity in investment activity) Source: Authors summary, ers should include a series of explicit and probing ques- have rarely appeared in LSMS questionnaires include tions in the questionnaire. If questions are not included pawnshops and rotating savings and credit associations. on all types of credit, this will lead to a serious underes- timation of the use of credit in the econonsy as a whole, ACCESS TO CREDir. Access is not simply a function of and yield a distorted picture of credit users.Two impor- the number of lenders in a given area. Religious, eth- tant types of credit that have often been omitted in past nic, and economic factors can either facilitate or con- LSMS surveys are trade credit for productive activities strain access to different sources of credit. It may be and consumption credit for food and household goods. necessary to obtain information on credit access by examining both formal and informal rules for obtain- ExHAuSTrvE COVERAGE OF ALL SOURCES OF CREDIT. ing credit. Borrowers may be an important source of Survey designers should make a serious effort to include information about informal rules. explicit questions about all possible sources of loans. Many lenders in the informal sector have been missed COLLATERAL. Collateral requirements are an important by past LSMS surveys. Important sources of finance that determinant of access to credit, as well as of the cost of 229 KINNON SCOTT credit. Collateral requirements may take the form of it is necessary to go beyond the survey instruments to compensatory balances or real property. Care should collect additional information on any aspects of the be taken to determine both the ownership and value legal system that may assist or hinder the enforcement of collateral assets held by individuals. of contracts. COST OF CREDIT. Accurately measuring interest rates General Design Issues is critically important but surprisingly difficult. Given No one questionnaire or module can be used in all the wide disparity in views on interest rates and how countries-or even in the same country at different they are defined and understood, survey designers times. Which policy questions are high priorities will need to investigate this matter thoroughly before vary as policies are created or discarded.The resources attempting to design questions to obtain information available for carrying out a specific survey will also on the cost of credit. Questions on the cost of credit change. Thus the draft questions and modules in this should be put to direct informants-the household section must be modified to fit policy interests and members actually responsible for loans-rather than to credit market characteristics in the country of the sur- proxies. vey, as well as the resources available to conduct the The nominal interest rate is only one facet of the survey. total cost of credit. Data on other transaction costs- Questionnaire design is complicated by the vari- additional fees, legal work, and costs incurred in timne ety of ways in which credit is provided. For data col- and transport to obtain the credit-are also needed. lection purposes, a distinction is made between receiv- ing monetary loans and purchasing goods on credit. PANEL DATA. Panel data are often needed to analyze Monetary loans can be defined as loans given and the impact of credit on production and consumption received in monetary terms regardless of whether the and to identify liquidity-constrained households. The borrower receives cash or goods. For example, a loan need for panel data may be an important consideration for agricultural inputs made by an agricultural coop- when designing a survey. See Chapter 23 for an erative to a farmer is the same regardless of whether overview of the costs and benefits of panel data. the borrower can use the money at his or her discre- tion or whether he or she must use it to buy inputs INDIVIDUAL-LEVEL DATA. Households do not borrow from the cooperative that gave the loan. In the draft money, individuals do. Data must be collected at the questionnaire, information on monetary loans is col- individual level both on the use of credit and on assets lected in the credit module; data on purchasing goods (collateral) and income. This is important for deter- on credit are collected by adding questions to other minling who actually obtains credit and what the bar- modules ofthe questionnaire.Although the distinction riers to credit use are. Collecting individual-level data between these two types of credit is sornewhat artifi- enables analysts to study such issues as legal barriers to cial, interviewers must be trained to recognize this dis- credit use and the intrahousehold distribution of tinction and to classify credit activities correctly. resources and power. A distinct credit module is needed for several rea- sons. First, data must be collected at the individual GovERNMiENT POLICIES. If a survey aims to determine level to ensure that the data are highly accurate and the iiimpact of a specific policy or program, informa- comprehensive and to allow analysts to link a borrow- tion will be needed from sources other than the sur- er's characteristics with his or her credit use. Second, vey instruments themselves. Government sources will not all credit is linked to a specific productive activity. need to be used to collect detailed information on the If questions on credit were asked only at the level of exact benefits of a program, the value of any subsidies the household, agricultural activity, or nonagricultural it provides, its geographical coverage, and its eligibility business, it would be extremely difficult to ensure that criteria. Information will also be needed on programs all data on credit were collected and that no double and policies that affect credit use indirectly. counting of credit occurred. The credit module alone cannot capture all of the LEGAL FRAMEwoRK. For credit markets to operate types of credit used by households; questions also need effectively, contracts need to be enforceable.Therefore, to be added to other modules of the questionnaire. 230 CHAPTER 21 CREDIT Placing certain credit questions in other modules is ing loans, about five or six questions need to be added useful for several reasons. First, for some analyses it is to the housing module. These are shown in the important to associate credit with specific activities. Questionnaire Example 1. The first question, a stan- Second, putting questions on agricultural credit (say, dard one usually found in LSMS housing modules, is supplier's credit for fertilizers) in the agriculture mod- contained in the draft module in Chapter 12. The ule improves the flow of an interview, preventing it Chapter 12 module additionally includes Questions 7 from skipping back and forth among topics. Finally, and 8, which have also become more common in adding credit questions to other modules makes it pos- recent surveys. sible to place questions concerning other aspects of credit, such as collateral, in their logical places in the DURABLE GoODs. The only question that needs to be questionnaire. added to the durable goods part of the consumption A final, practical benefit of putting questions on module concerns who owns specific items. (Durable credit both in a credit module and in other modules is goods are discussed in Chapter 5 on consumption.) that this enables the survey team to cross-check cred- Because, like housing, many durable goods can be it data. Given that the use of credit can be a sensitive used as collateral for other loans, it is important to matter for some people, they may understate the determine who in a household actually owns the extent to which they use it-resulting eventually in goods in question. There is no reason to apply the the underestimation of total credit use in a given ownership question to all durable goods as many country. It is essential that survey designers take steps durable goods are not appropriate for collateral. to avoid such underestimation, especially since house- Which goods can be used as collateral and which are hold survey data are necessary for any estimates of most frequently used will vary from country to coun- total credit use in a country (and not just of formal try; some preliminary investigation on this topic will and semiformal credit). be needed in each country. For those goods that can be used as collateral, a question should be added to Adding Questions to Other Modules this part asking for the identification code of the per- Questions about credit need to be added to the mod- son who owns the item (Questionnaire Example 2). ules on housing (Chapter 12), savings (Chapter 20), Where feasible, the durable goods for which owner- consumption (Chapter 5), agriculture (Chapter 19), ship questions will be asked should be listed together and nonagricultural household enterprises (Chapter on the questionnaire. This will help ensure that these 18). questions are not accidentally omitted during the interview. HOUSING. Two types of questions need to be added to If analysts wish to know whether the durable the housing module. First, researchers must ask which goods were obtained through credit, one question household member owns the housing unit. This is should be added asking how each good was obtained. necessary because one of the keys to being able to However, this question can lead to double counting if borrow is owning marketable assets that can serve as the item was purchased with loan funds. Thus in the collateral, and the most important of these assets is resulting data set it must be possible for analysts to be housing. able to distinguish the type of credit used to purchase Second, researchers must ask questions to collect the good.And analysts will have to be careful to ensure data on mortgages and the terms and balances of that this distinction is respected. mortgage loans.These questions are important because obtaining housing is a common reason for borrowing SAVINGS. Savings can take a variety of forms, some of and incompletely paid mortgage loans are often the which can be used as collateral. Chapter 20 provides a largest of people's debts. detailed discussion of the various forms of savings, The housing module that has been used in most including liquid assets and nonliquid assets, both of LSMS surveys already contains much of the informa- which can be used as collateral for borrowing. Items of tion needed on credit (see Chapter 12). To capture value can be pledged or pawned, and money deposits information on the legal status of the housing, the (compensatory balances) are a standard requirement in ownership of the asset, and the existence of outstand- many credit unions and banks. 231 KINNON SCO1T Thus it is important to collect savings data in gen- tions about deposits. Chapter 20 discusses the difficul- eral but also specifically at the individual level. ties associated with asking for such information. The However, it can be difficult to get respondents to standard version of the savings module presented by answer questions about their savings, especially ques- that chapter provides examples of how to collect sav- Questionnaire Ekample 1: Credit Questions to Be Added to the Housing Module How did your household obtain this dwelling? PRIVATIZED I PURCHASED FROM A PRIVATE PERSON ............ 2 NEWLY BUILT ... .. 3 COOPERATIVE ARRANGEMENT 4 SWAPPED 5 (> 10) INHERITED 6 (> 10) OTHER 7 (> 10) 2 Did anyone in this household borrow money to purchase this housing unit? YES .........I NO. 2 >>> 7 3 From what person or institution was money borrowed to purchase this housing unit? GOVERNMENT HOUSING BANK ................. 1I.l.l OTHER PUBLIC HOUSING PROGRAM ................... 2 PRIVATE BANK 3 BUILDING SOCIETY 4 CREDIT UNION OR OTHER COOPERATIVE 5 EMPLOYER OR LANDLORD 6 RELATIVE OR FRIEND .... 7 OTHER INSTITUTION [SPECIFY ].......... 8 OTHER INDIVIDUAL [SPECIFY ] ............ 9 4 In what year was this money borrowed? Year 5 What was the total amount borrowed? Currency E L 232 CHAPTER 21 CREDIT 6 How much is still owed? Currency l I 7 Does any member of the household have a title or any document that shows ownership of this dwelling? YES .1 II NO . 2 (>10) 8 What type of title? FULL LEGAL TITLE, REGISTERED ............... 1 LEGAL TILE, NOT REGISTERED ........... 2 PURCHASE RECEIPT ...... 3 OTHER, SPECIFY .. 4 9 Which person holds the title or document to this dwelling? [INTERVIEWER: WRITE ID CODE OF THIS W PERSON IN THE BOX, FROM HOUSEHOLD ROSTER] Note: Question numbers are to show sequencing and skip patterns. Ouestions 1-2: The first question serves to determine which households might have borrowed for their present housing. Those who inherited their housing or swapped for it have clearly not borrowed and thus are sent on to the ownership questions. Question 2 is a further filter question; if no borrowing for housing occurred, the respondent is sent directly to questions on ownership. If borrowing for housing did occur, the respondent is sent to Question 3, the first in a series of questions about the characteristics of the loan. Ouestions 3-6: These questions are designed to collect basic information on the mortgage loans. If there are specific credit programs for housing, these need to included as options in Question 3; the other categories listed must be adapted to the local housing loan market. Depending on the importance of housing finance and policies, it may be worthwhile to add more questions here to obtain details about the terms of credit. While some of this information may be collected in the credit module, the relative rarity of mortgage borrowing means that very few households will have borrowed for a mortgage in the past 12 months and, hence, little information on this type of credit will be collected. Ouestions 7-9: These questions allow ownership of housing to be connected to the individual providing data that will be added to determine the assets (collateral) an individual owns. The categories in Question 8 will need to be very carefully adapted to the local legal situation, taking into account types of formal, legal ownership as well as the variety of informal claims that confer different values on the housing studied. ings information. Although the questions are designed module is extremely long and takes up a significant to be administered only at the household level, they amount of the total interview time-and may become can easily be adapted for use at the individual level. tedious for the respondent. Thus it is not feasible to The short version of the savings module is inadequate add new questions for each food item on the list. for determining ownership of assets and savings. Two recommendations are made here that should allow for a rough estimate of such credit without CONSUMPTION. Many purchases of food and common excessively lengthening the questionnaire. First, a household products are made on credit. Questions series of questions should be added inquiring about must be added to the consumption module to ensure purchases on credit at an aggregate level rather than that all such consumption credit is captured. While for individual products. The dilemma is whether it analysts would ideally like to know the exact quanti- makes more sense to group products by category (for ties of food and other items purchased on credit and example, fresh fruits and vegetables, meats, breads, the sources of credit for each item, there are practical cleaning products, and so on) or by point of purchase difficulties involved in gathering such information. (for example, supermarket, open air market, and so Even without such credit questions the consumption on). These different aggregations can have the same 233 KINNON SCOTT Questionnaire Example 2: Credit Questions to Be Added to the Duralle Goods Part of the Consumption Module I. Does any member of 2. Who owns this 3. How 4. How was this 5. How 6.How 7. If you the household own a: item? many [ITEM] obtained? much much wanted to [INTERVIEWER: years ago was was this sell this YES .1 WRITE IN did you OWN CASH .....1 paid for [ITEM] [ITEM] NO. 2>(NEXT THE ID CODE acquire CASH FROM this worth today, ITEM) OF this LOAN .2 [ITEM]? when how THE OWNER] [ITEM]? ON you much INSTALLMENT .. .3 received would OTHER (>7) it? you CREDIT .. 4 receive? GIFT ..... ... 5 (>6) PAYMENT FOR SERVICES .6 (>6) Refrigerator Television Stereo Tape recorder _ Sewing machine ________ Motorcycle X _ _ _ _ Car = T Truck T Note: Ouestions 1. 3. 5. 6. 7: These questions will vary depending on considerations other than credit. See Chapter 5 on consumption. Ouestion 2: This is the additional question for detennining ownership. Ouestion 4: This is designed to determine if the item was purchased using some credit mechanism. The terminology will have to be adapted, but the categories need to be maintained. results for specialty stores such as bakeries and phar- whether the information they give is likely to be accu- macies. However, in the real world, many stores do not rate and detailed. Thus, it will be necessary to do specialize in only one product, and survey designers extensive pre-testing of this module. It may be worth- must make some choices. The proposed questions while to test the two alternatives in a pilot exercise to below collect data by category of good rather than by ensure that the most effective design is chosen. point of purchase.The advantage ofthis method is that The second suggestion for collecting data on con- since food purchase data is collected product by prod- sumption credit without unduly lengthening the uct, analysts can estimate the value of credit for cate- questionnaire is to collect data only for the 12-month gories of goods without asking detailed additional reference period rather than for both the the 12- questions. If credit information is collected in some month and 30-day reference periods. This suggestion aggregate of these products, the value of the purchas- assumes that access to consumption credit does not es is already known and only the share of such pur- vary much throughout the year, although in areas with chases made on credit is needed to determine the severe weather fluctuations or high covariate risk or in value of the credit. If the questions were phrased in areas where poverty is severe, this may not be the case. terms of points of purchase, it would be necessary to If survey designers expect seasonality to affect access to gather further information on how much money was this kind of credit, it is more appropriate to apply the spent at each of these points. (If data on consumption shorter reference period to these questions (assuming were collected by point of purchase, this argument a 12-month interview schedule). Again, prior knowl- would be reversed. See Chapter 5 on consumption.) edge of the area where the questionnaire is to be The most important criterion for deciding administered is needed, as is pre-testing. whether to collect data by category of good or by Questionnaire Example 3 shows how to collect point of purchase is the ease with which respondents these data for food purchases. Similar series of questions can supply the necessary information; this determines should be created for other non-food purchases-from 234 CHAPTER 21 CREDIT common expenditures on items such as detergents to knowledge of the issue in the location where the expenditures on larger items such as clothing and small questionnaire is to be administered. Two different appliances. (Similar questions should be added for ways of obtaining this information are included here; durable goods; see preceding discussion on durable the pilot test should help determine which is prefer- goods.) able or if both are needed. A final consideration is the cost of acquiring credit for consumption. Are there price differentials HOUSEHOLD ENTERPRISES AND AGRICULTURAL between cash purchases and credit purchases? If so, ACTIVITIES. Four key issues concerning credit need to this has serious ramifications for valuing total con- be addressed in both the module on agriculture and sumption (see Chapter 5). While it will not be feasi- the module on nonagricultural household enterprises. ble to collect detailed information on such price dif- Survey designers will have to be careful to ensure that ferentials, some rough calculation may be possible these two modules are compatible with the credit with the answers to some very basic questions. module.This is especially true if they decide to use the Determining the exact phrasing and format of these short versions of either module as this will limit the questions will, however, require some preliminary possibilities for credit analysis. Questionnaire Example 3: Credit Questions to Be Added to the Food Purchases Part of the Consumption Module I . Do you or any members of your household purchase food on credit in other words, buy food and pay for it at a later date? YES................... NO .2 (>22) 2. On average, during the past 12 months, what percentage of all fresh fruits and vegetables purchased did you obtain on credit? Percent ll [IF ZERO, >> 71 3. When you purchase fresh fruits and vegetables on credit, do you pay a higher price? YES .1.......w.. NO .2 (>6) 4. On average, in percentage terms, what is this price difference? Percent 5. What was the difference in price between the cash price for [ITEM] and what you paid for [ITEM] on credit the last time you purchased [ITEM]? Cash Credit 6. Where do you obtain the majority of the fresh fruit and vegetables that you purchase on credit? PRIVATE INDIVIDUAL ..................I SMALL STORE (PRIVATE) ...........2 SUPERMARKET (PRIVATE) .3 GOVERNMENT STORE .................4 7. On average, during the past 12 months, what percentage of all fish, meat and dairy products purchased did you obtain on Percent [ ] credit? L W [IF ZERO, > 12] 8. When you purchase fish, meat and dairy on credit, do you pay a higher price? YES .................... NO . 2 (>11) 235 KINNON SCOTT Questionnaire Example 3 (continued) 9. On average, in percentage terms, what is this price difference? Percent 10. What was the difference in price between the cash price for [ITEM] and what you paid for [ITEM] on credit the last time you purchased [ITEM]? Cash Credit 11. Where do you obtain the majority of the meat and dairy products that you purchase on credit? PRIVATE INDIVIDUAL ..................1 SMALL STORE (PRIVATE) ...........2 SUPERMARKET (PRIVATE) ..........3 GOVERNMENT STORE ................4 12. On average, during the past 12 months, what percentage of all grains and cereal (unprocessed) purchased did you obtain on credit? Percent [IF ZERO, >171 13. When you purchase grains and cereals on credit, do you pay a higher price? YES ..... 1 w. NO . 2 (>16) 14. On average, in percentage terms, what is this price difference? Percent W 15. What was the difference in price between the cash price for [ITEM] and what you paid for [ITEM] on credit the last time you purchased [ITEM]? Cash Credit 16. Where do you obtain the majority of the grains and cereals you purchase on credit? PRIVATE INDIVIDUAL ..................1 SMALL STORE (PRIVATE) ...........2 SUPERMARKET (PRIVATE) ...... 3 GOVERNMENT STORE ................4 17. On average, during the past 12 months what percentage of all canned and processed food purchased did you obtain on credit? [IF ZERO, > 22] 236 CHAPTER 21 CREDIT Questionnaire Example 3 (continued) 18. When you purchase canned and processed food on credit, do you pay a higher price? I YES ..1...... I NO . 2 (>21) 19. On average, in percentage terms, what is this price difference? Percent 20 What was the difference in price between the cash price for [ITEM] and what you paid for [ITEM] on credit the last time you purchased [ITEM]? Cash Credit 21. Where do you obtain the majority of the canned or processed goods you purchase on credit? PRIVATE INDIVIDUAL ...................1 SMALL STORE (PRIVATE) ............2 SUPERMARKET (PRIVATE) ..........3 GOVERNMENT STORE. 4 Note: Ouestion 1: This is the filter question for food purchases on credit. While this question is needed to ensure that households not using credit to buy food are not annoyed by the following questions (which would be irrelevant), the question must be carefully worded so that all informants understand what is being asked, and no households that actually do purchase on credit are excluded from this section. Ouestions 2. 7. 12. 17: These questions determine the share of purchases made on credit. Each of these questions actually contains two separate questions: Do you purchase any [food item type] on credit? and If so, what percentage of [food item type] is purchased on credit? While it is efficient to merge these two questions, it is imperative that interviewers be well trained to ensure that zeros are written in when the household makes no such purchases. Here the difference between zero and blank is quite important; interviewer errors will be common if this is not made clear in their training and checked during data entry. Ouestions 3. 4. 5. 8. 9. 10. 13. 14. 15. 18. 19. 20: These trios of questions aim to determine what price differences exist between goods purchased with cash and goods purchased on credit. These questions have not been used in LSMS surveys, and there is limited experience on how important these questions are and how they should be phrased. Work done in Pakistan (Mansuri 1993) showed a five percent differential between cash and credit prices. Clearly pre-testing and other knowledge of the area where the survey is being carried out are needed. If it is determined that the price differential is negligible and/or that the difference is standard for certain products or by area, it may not be necessary to include these questions. Ouestions 4. 5. 9. 10. 14. 15. 19. 20: If it is determined that questions are needed on price differentials between cash and credit purchases, the difficulty will be in obtaining accurate information. The questions here try to find out the percentage markup for a category of purchases. The markup may not be standard across types of items, or the respondent may only be able to quote the price difference and not give a percentage figure. These possibilities are accounted for by the second of the two questions in each category, which ask for the explicit price difference between cash and credit costs of one item per category. The item used should be a commonly purchased item, or represent an important share of total purchases in the category. Given the complexities involved, it will be useful for interviewers to write any calculations they make (converting price differences to percentage terms, for example) in the side margin of the questionnaire page. Supervisors should check these calculations before passing the questionnaires to the data entry phase. Ouestions 6. 11. 16. 21: The list of purchase points will have to be adapted for each country. The key considerations are to be exhaustive and to focus on any types of sales points that might be of policy relevance. Government stores versus private stores may be an important distinction; specialty stores versus supermarkets might be another. If there is a policy interest in where credit is provided (as opposed to just its availability) it may be useful to add this list of purchase points to the community questionnaire to determine access. The small store in the answer lists for these questions could include more informal shops such as kioskos in the former Soviet Union or pulperias in Central America. The first use for credit data gathered in the agri- the interviewer should check the answers in the cred- cultural and the household enterprise modules is to it module given by that individual and determine if he link and cross-check loan data gathered in the credit or she received a loan for agriculture or a household module. In the process of identifying the appropriate enterprise. Questionnaire Example 4 provides an respondent for the agricultural or enterprise activities, example of the type of questions to add to the agri- 237 KINNON SCOTT Questionnaire Example 4: Credit Questions to Be Added to the Agriculture Module 1. [INTERVIEWER: WRITE IDENTIFICATION 2. [INTERVIEWER: CHECK QUESTION CODE OF HOUSEHOLD MEMBER INVOLVED IN NUMBER ** IN CREDIT MODULE: DID THIS AGRICULTURE:] PERSON STATE THAT RECEIVED LOAN FOR AGRICULTURE IN PAST 12 MONTHS? YES . 1l NO ..2 Note: Question numbers are included only to show sequencing and skip patterns. Ouestions 1. 2: The first two questions of the agricultural (or nonagricultural business) module will be filled in by the interviewer before the second-round interview. Question I identifies which household members to administer the agricultural (or nonagricultural business) questions to. Question 2 requires the interviewer to review the credit module and determine if the respondent reported borrowing for agricultural (or nonagricultural business) purposes. culture module; the same type of questions should be of product) to ensure that all supplier credit is captured. added to the household enterprise module. The prevalence and size of this source of credit justifies These added questions can be compared with the additional questions. Second, additional questions answers to questions on loans within the agriculture about the source, magnitude, and repayment terms of and nonagricultural enterprise modules to ensure that supplier credit are needed.This series of questions should no loans are missed. (This assumes that the agricultur- be repeated for all key inputs or groups of inputs, and an al and household enterprise modules contain a gener- additional question at the end of the series should ask if al question that associates a loan with a specific plot12 any other items were obtained on credit and, if so, which or household enterprise.) If a respondent claims to items, and for how much. have received credit but there is no sign of this credit The third credit issue that needs to be addressed in his or her answers to the questions in the credit in the agricultural and household enterprise modules module, the interviewer should try to reconcile this is the ownership of collateral such as real estate (land discrepancy. If necessary the interviewer should re- and buildings) and other assets. As with housing, it is administer the credit module to this person to ensure necessary to gather data on the ownership of land and that his or her credit data are accurate. nonresidential buildings as well as other assets at the The second issue to be addressed in the agricultur- individual level rather than at the household level. al and household enterprise modules is the issue of sup- The final credit-related issue that needs to be plier credit. Survey designers may decide to include addressed in the agricultural and household enterprise questions on this subject in these modules for the pur- modules is the issue of sales of outputs on credit. Like poses of analyzing agricultural and household enterprise the questions on supplier credit, questions concerning issues rather than credit issues.A discussion of these types sales on credit by the farmer or firm are already likely of questions can be found in Chapter 18 although not to be included in these modules for other analytical in Chapter 19. Detailed examples of such questions are reasons. Detailed information on how to formulate not provided here. (See Chapter 18 for ways to formu- these questions is provided in Chapters 18 and 19. late these questions.) However, there are two points that Again, survey designers need to ensure that the word- survey designers should bear in mind when assessing ing of the questions in these two modules meets the whether the questions about supplier credit in these two needs of credit analysis. modules are adequate for analyzing credit issues. First, while it would be possible to include an overall filter The Credit Module question for supplier credit for farm (or household The separate credit module is designed to capture enterprise) inputs in each of these modules, it would be information on all borrowing in monetary form-in better to include a series of detailed questions (by type other words, all credit excluding supplier credit and 238 CHAPTER 21 CREDIT the goods purchased on credit that are covered in The questions will provide the analyst with both the other modules-in the previous 12 months.The mod- total number of loans obtained in the period and the ule is broken down into four parts: use of credit, costs total amount of borrowing by the individual. All of and terms of credit obtained, credit history, and lend- this information will be broken down by source of ing. The module should be administered to all indi- funds. viduals of working age (without an upper age limit), The part also determines the last date a loan from regardless of whether they are actually working. a particular source was obtained. This helps demon- The placement of the credit module in the ques- strate which persons borrow (even if not in the past 12 tionnaire will come from a compromise between two months) and what credit sources are available to them. considerations. On the one hand, because credit Finally, the part contains information on total debt of questions are sensitive, it is better to place the mod- the individual, again broken down by source. ule as late in the interview as possible. The longer the interviewer and the respondent have interacted PART B: COSTS AND TERMS OF CREDET. While one before the credit questions are administered, the would, ideally, like detailed information on all loans more likely it is that trust will have been established held by individuals, obtaining such information between them-minimizing the likelihood that requires more questions and time than are feasible in respondents will refuse to respond or will give inter- an LSMS or other multitopic survey. The other viewers incorrect information. For this reason, in past extreme, collecting no detailed loan information at all, LSMS surveys credit modules have usually been would limit the types of analyses that could be done administered last. and in many cases would be unacceptably restrictive. On the other hand, since credit data are collected The part presented here represents a compromise at the individual level, they need to be collected dur- between these two extremes: data are collected only ing the individual-level phase of the interview. During on the loans that a person has most recently obtained. the first round of the survey the interviewer collects The data are collected by loan purpose: agriculture, information directly from all household members; due nonagricultural businesses, or consumption. This to the different schedules of each household member matches the data collected on purchases on credit in this often requires the interviewer to make a series of the three parts, enabling some assessment of total bor- visits to the household. Given the importance of get- rowing for each of these activities. ting credit information directly from those who are While Part B appears quite lengthy, in fact it com- responsible for that credit, and given the cost involved prises the same twenty-eight questions repeated for in interviewing each household member, it is impor- each of three possible loans. Since very few individu- tant that whenever possible, all individual level mod- als will have more than one type of loan from more ules are administered in one sitting. This means that than one source, it is expected that much of this part the credit module must be administered in the first will not be used by most respondents. Adding to this round. However, within this first round, it is recom- the fact that only a small minority of households has mended that the credit module be placed directly after loans, the average time devoted to this part will be the employment module, to take advantage of skip quite small. patterns (since the credit module is administered to all The range of questions that can be asked about a people of working age). This is a reasonable compro- specific loan is quite broad. It is clear that the draft mise as it means that the module will come late in the module proposed here is more limited than that of individual-level interviews while still ensuring that the surveys devoted entirely or primarily to credit. This is information is collected from the direct informant. a function of the tradeoffs required in doing a multi- topic survey as opposed to a single-topic one.As noted PART A: USE OF CREDIT. The questions in this part earlier, the data that can be collected in an LSMS or explicitly ask about loans from a wide variety of other multitopic survey may not be adequate for all sources, starting with the most informal (and most types of credit analyses. Examples of topics not likely to be missed) and moving to the most formal. addressed here are the long-term relation a borrower The questions are designed to elicit information on all has with a lender and the details on all short-term loan activity in the previous twelve-month period. credit obtained in a year. 239 KINNON SCOan PART C: PAST USE OF CREDIT. Part C of the credit mod- names and occupations. After Question 2 in Part B, a ule, on credit history, should be asked of all individuals new Question 3 should ask: "What is the main occu- regardless of whether they have received credit in the pation of this person [the lender]?" The question can previous 12 months. Part C also examines the issue of be coded with the occupational categories used in the credit access. Questions are asked about loan applica- employment module. tions that were refused, whether a person wanted cred- it even though he or she did not apply for it, and what The Short Version of the Credit Module a person thinks about his or her chances of obtaining a To shorten the credit module, survey designers must loan from the lenders in the market. These questions first determine which policy issues are a high priority enable analysts to examine the characteristics of people in the country. With these issues in mind, they can xvhose access to the credit market is limited-as well as decide which of the credit parts and questions can be of people who perceive that their access is limited. dropped and which must remain. For most general analysis, basic information is needed on use of credit, PART D: LENDING ACTIVITIES [OPTIONAL]. This part is access to credit, and ownership of collateral. If the sur- designed to identify individuals who lend money to vey is intended to evaluate a specific policy, appropri- others. Because such information is not available from ate questions may be added. any other source, this part is ver-y important. However, there are several drawbacks to including this part USE OF CREDIT. The use of credit could be covered in (which explain why it is considered optional here). a substantially reduced credit module that includes Part The most compelling argument for not including this A and some of Part B of the standard version of the module is that given the sample size of LSMS-type credit module. Even in the shortened version this mod- surveys, so few people may be identified that it may ule must be administered at the individual level, not the not be worthwhile to try to collect this information. household level. As in the draft short version of the Another argument against including Part D is that module in Volume 3, a series of 24 questions may be it touches on what may be a very sensitive area. Even used to determine the number of loans that the respon- though every attempt has been made to make the part dent has received in the past 12 months as well as some as unthreatening as possible to the lender, including key details about his or her most recent loan. This will such a part may not be possible in certain countries. If allow the analyst to determine the individual's total there is a great deal of negative press or bad feeling debt, but it will not provide data on the cost of obtain- about informal lenders, lenders may not wish to tell ing or repaying a loan, nor will it provide data on the interviewers about their lending operations. In such total inflow or outflow associated with credit use in any cases it is best to leave out this part to avoid the risk of given time period. To be sure to capture all credit use, obtaining poor-quality data and the risk of interviewees additional questions about the use of credit should be refusing to collaborate xvith the remainder of the survey. included in the consumption, agriculture, and house- If Part D is included, it is important to minimize hold enterprise modules. The series of questions about the extent to which respondents feel uncomfortable supplier credit and purchases "on credit" can be reduced with this part. This is done in the Part D of the draft to a simple "Yes/No" question in each module, phrased credit module. The respondent is never referred to as a as follows: "Did you, or anyone in your household, pur- lender, nor is the respondent asked about his or her chase any food (agricultural supplies/businesses inputs) lending activities. Instead the questions refer to who on credit during the past 12 months?" "owes you money" or "who borrowed from you." No However, it is important to keep in mind that questions are asked on interest rates charged and other these questions are included in these modules for spe- sensitive terms of lending.And the part has deliberate- cific analytical purposes not related to the analysis of ly been kept quite short. credit. Thus survey designers should consult the ana- If Part D is omitted analysts have to rely on the lytical objectives ofthe other modules before deciding detailed loan data from Part B for information on how to delete these questions. the informal market works. If survey designers decide to omit Part D, they may wish to gather additional ACCESS TO CREDIT. The questions that determine indi- information about informal lenders in Part B, such as viduals' use of credit also yield a great deal of infor- 240 CHAPTER 21 CREDIT mation about their access to credit. Additional ques- uals' communities. Community leaders are unlikely to tions that reveal credit access are in Part C of the stan- be able to identify lenders outside the community- dard credit module, on credit history; most of Part C again contributing to underestimation of the availabil- is incorporated into the short version of the module. ity of credit to community members. These questions are not controversial and can be A third limitation is that borrowers' physical answered quickly. access to lenders is not equivalent to actual access. Unlike education, for which the presence of a public COLLATERAL. As discussed above, the questions on school indicates that all children have access to collateral are placed in the housing, agriculture, and schooling if they so desire, the presence of a lender household enterprise modules. To shorten the ques- has little to do with whether a community member is tionnaire these questions can be simplified to "Who able to borrow. The number of lenders from which owns the asset?" and "What form of title does this any one person can borrow is limited by factors person hold?" The type of title is just as important as including his or her economic status, investment who owns the asset, so the latter question should not activity, and ethnic or religious status. To analyze peo- be dropped. Identifying individual ownership of ple's access to credit, it may be more relevant to col- goods may also be a priority for analysts of the pri- lect information on the borrowers' perceptions of mary topic of these other modules. The planning of their access than on any physical list of lenders in a these modules and the credit module will have to be given area (along the lines of the final question in Part closely coordinated. C of the credit module). Community-Level Data Facility-Level Data The community questionnaire can be used to gather If analysts want detailed information on lenders, a information on the pool of lenders in the communi- facility survey may need to be incorporated into the ty. These community-level data are needed to deter- overall survey, since information on lenders is usually mine the extent to which the location of lenders, as better obtained from service providers than from com- opposed to the characteristics of borrowers, may be munity leaders. However, a facility survey focusing on affecting access to credit. A simple series of questions credit providers may yield less useful information than in the community questionnaire can address this issue those that have typically been used in LSMS surveys (see Chapter 13 for details). focusing on schools or health facilities. Such a credit- However, there are three limitations to the useful- focused survey may also require significantly different ness of the community questionnaire for identifying fieldwork techniques. lenders. (These limitations do not apply to identifying On the positive side, formal lenders and even schools and other facilities.) The first difficulty is in iden- most semiformal lenders are willing to provide infor- tifying informal lenders. It is likely to be relatively easy mation on their lending policies and requirements (as to identify formal lenders in the community, as this this information is already in the public domain). information is usually available at the national level at the They are often willing to release annual reports indi- agency that registers banks.'3 Even many semiformal cating the total assets of their institutions. And semi- institutions can be identified this way.4 For identifying formal institutions often provide valuable information informal lenders no such lists exist. Other lenders in the on private lenders because they see private lenders as community can often supply information on informal their competitors. Such information may include lenders, but this usually only includes people who lend what interest rates are charged, what collateral is systematically.There is unlikely to be any source of infor- required, and what other fees are charged in the cred- mation on truly informal loans between relatives and/or it market.The information can be used to cross-check friends. It is not clear that the community questionnaire the household data and, if need be, to fill gaps in will be able to collect such information either; thus the individual-level data. community questionnaire will probably underestimate On the negative side, lenders of any type rarely pro- the availability of credit in the community. vide disaggregated information on the number of bor- The second limitation is that all of the lenders uti- rowers that they serve, the share of loans by lized by individuals may not be located in the individ- purpose/sector, the average loan size, or other similar 241 KINNON Scoan information.This is partly because of how lenders main- naire accurately captures all of the lending activities in tain their accounting and financial records (especially which individuals and households are engaged. lenders that are not computerized). But it is also due to their frequent unwillingness to share such information, Detailed Discussion of Questionnaire Modules which can often be an insurmountable problem. To boost the quality and quantity of data collect- This section contains detailed notes on the two credit ed on the informal sector, efforts must be made to modules inVolume 3. In addition to these specific sug- establish trust between the interviewer and the com- gestions, the questionnaire design team needs to keep munity-and especially between the interviewer and in mind that all questions will need to be adapted to moneylenders. In a Thailand study interviewers lived local circumstances. in communities for long periods of time to establish this trust (Siamwalla and others 1990).While poten- Standard Credit Module tially effective, this technique is probably not feasible in the context of a multitopic survey. PART A: USE OF CREDIT. As discussed previously, peo- ple may be sensitive about providing information on Other Sources of Data their borrowing activities. One way to make Part A Several other sources of data can be used to comple- less threatening to respondents is to replace the word ment the data gathered from households, communi- "borrow" with the word "lend"-asking how many ties, and facilities. times someone lent to the respondent rather than ask- ing how many times the respondent borrowed. REGISTRIES OF LENDERS. At the national level, data can This module should be administered to people of often be found at the superintendency of banks, asso- the same ages as the respondents for the labor ciations or federations of credit unions (or coopera- module-with no upper cutoff limit. tives), and similar institutions. These data sources can yield information about the location of all of the main Al-A2. These questions enable analysts to distinguish offices of such lenders and their branches throughout credit data provided by the individual borrower from the country. Information on total assets, the amount of indirect credit data (which are likely to be of much money loaned, and bad debts is also sometimes avail- lower quality). able from these sources. A3-A6, A7-AlO, All-A14, A15-A18, A19-A21, LEGAL SYSTEM. Analysts of credit issues are often inter- A22-A25. These groups of questions contain essentially ested in three types of laws. First, basic laws governing the same questions repeated for six different types of the formation of financial institutions can provide an lender: family or friends, employer or landlord, credit explanation for the observed form and distribution of union, cooperative or NGO, bank or government formal sector lenders in a given country. (Additional agency, rotating savings and credit association, and other. laws may exist that regulate semiformal lenders.) Second, information about contract law and the extent A3, A7, All, A15, A19, A22. For added clarity, these to which it is enforceable helps analysts understand the questions not only refer explicitly to loans but also availability and use of credit.Third, information will be explain exactly what is meant by adding the phrase needed on any laws that may affect credit use either "funds that you have to repay." (People tend to under- directly (say, by creating specific lending programs or report loans, especially loans from family members.) by banning discrimination) or indirectly (tax laws). A4, A8, A12, A16, A20, A23. These questions, used in OTHER STUDIES. Small sociological or anthropological conjunction with the first questions in each group, studies may exist that explore the attitudes of the pop- allow an estimation of average loan size by lender and ulation toward lending and borrowing and the various average loan size provided to each type of borrower. forms such activities take in a given country. These studies will be important sources of information for A5,A9,A13,A17,A21,A24. For respondents who have survey designers; they will ensure that the question- borrowed in the past 12 months, each of these ques- 242 CHAPTER 21 CREDIT tions provides the date of the most recent loan from B2. From this question on, data are collected only for one source. For respondents who have not borrowed the most recent loan for agriculture.This question asks in the past 12 months, these questions allow analysts to for the specific purpose of the loan. Only the main determine if the person has ever borrowed money, and purpose is collected here; it may be worthwhile if so, how recently. It is not feasible to attempt to col- adding an option for respondents to give two or three lect credit data beyond a limited time period (here, 12 purposes. Still, because money is fungible, adding more months). purposes may not significantly improve knowledge of how the loan is used. A6, A10, A14, A18, A25. The information collected in these questions includes borrowing outside the 12- B3. As the source of the loan is very important, one month reference period, and may differ from the total wants the list of options here to reflect all types of amount borrowed during this period. The series on lenders that exist. The list will need to be customized rotating savings and credit associations does not for the country where the survey is being adminis- include a question on the total debt of the borrower tered. A useful list will allow the analyst to determine as this is inappropriate for this type of lending if the source is formal, informal, or semiformal as well arrangement. as whether it is a government program (direct or spon- sored) or not. A26-A27. Cash loans may be obtained by pawning personal and household goods, especially in urban B4-B8. These questions are designed to determine the areas. transaction costs of obtaining a loan. The variety of costs charged to the borrower for the loan (Question A28. This is a filter question to determine if Part B of 38) will need to be customized for the specific setting, the module should be administered to the respondent. in terms of both content and the names by which cer- This is an awkward question, since reviewing the tain items are known. answers to A5, A9, A13, A17, A21, A24, and A26 can stop the flow of the interview. In training sessions, B9-B10. Many NGOs-and some banks and credit interviewers should be taught how to do this review unions-provide their clients and borrowers with of Part A quickly and accurately without losing the technical assistance. This assistance may take the form thread of the interview. of help in drawing up loan applications and designing projects, or the teaching of loan fund management and PART B: DETAILS or CREDIT. The respondents for this how to work with group borrowing schemes. Because part are all people of working age (with no upper age these services may have a major impact on the success limit) who borrowed in the 12 months prior to the rates of borrowers, determining exactly what services interview.This part is essentially three series of identi- have been received is critical. cal questions: the first on borrowing for agriculture, In the pilot test it will be important to deter- the second on borrowing for nonagricultural enter- mine how common this technical assistance is and prises, and the third on borrowing for consumption. whether it is focused on one topic or covers a vari- The following notes are based on the first series of ety of aspects of the loan process. If the assistance is questions, for agriculture, but are relevant to all three very rare or focuses mostly in one area, Question B10 series. is adequate as it stands. If, however, there is wide use of such assistance, or the assistance includes a variety B1. If the respondent has not borrowed any money for of aspects (such as project design and managing loan agricultural purposes in the past 12 months, the inter- funds), it might be worthwhile to allow multiple viewer skips to B29, which asks if the respondent has responses to B10. In any case, B10 will need to be borrowed for nonagricultural enterprise activity. If the customized. answer to B29 is no, the interviewer skips to B53, which asks if the respondent has borrowed for con- B11-B19. These questions all refer to collateral and sumption. If the answer to B53 is no, the interviewer how collateral is valued. Questions B11-B13 discuss moves on to Part C. labor that might be provided for a loan. 314-317 refer 243 KINNON SCOTT to the use of the harvest as collateral, and B18-B19 B23-B27. By determining the frequency, number, and refer to other types of collateral. value of loan payments, these questions provide fur- ther information on the cost of credit. Interest rate B11-B13. Question Bll determines the number of data that a respondent cannot provide may be calcu- days of labor, if any, the borrower had to provide for the lated from this series of questions. loan. If labor was not part of the credit arrangement, the interviewer skips to B1 4 on the use of harvests as B28. The module does not provide a complete look at collateral. If labor was a part of the arrangement, B12 the borrower's loan history with lenders and the and B13 allow analysts (in order to estimate the real extent to which loans have been paid off on time.This cost of credit) to distinguish arnong days of labor pro- question provides the only information on whether a vided free, days paid at the market rate, and days paid borrower is in arrears. below the market rate. Question B13 attempts to determine the value of the individual's labor in the B29-B52, B53-B76. Questions B29-B52 are the same normal market if no credit arrangement were involved. as Questions B1-B28 but focus on nonagricultural businesses rather than on agriculture. Questions B14-B17. Pledging part of one's harvest is a common B53-B76 follow the same pattern focusing on con- form of collateral for agriculture. Question B14 asks sumption.These two series are slightly shorter than the what quantity of the respondent's harvest is pledged series on agricultural loans because the three questions under the loan arrangements. If none of the respon- concerning the use of harvest as collateral have been dent's harvest is pledged, the interviewer proceeds to dropped. questions on other types of collateral. If part of the respondent's harvest is pledged, B15-B17 are needed PART C. CREDIT HISTORY. The respondents for this to determine how this harvest has been valued for part are all people of working age (with no upper age payment. Question B15 asks for the price that the limit) regardless of whether they borroxved in the 12 lender will pay for the harvest. If the agreement does months prior to the interview. not include a currency price but is instead a percent- age of some other price, B16 is needed to determine Cl.This question has a reference period of 12 months, whether the price is a percentage of a future price, and matching the information collected in Part A. While it what percentage. Question B17 asks what the value of might be interesting to expand this reference period, the respondent's harvest would be if the borrower the analyst has limited information about the borrow- were free to sell it outside of the credit agreement. er's characteristics more than 12 months prior to the Again, this information is needed to determine the survey, which reduces the usefulness of any informa- cost of credit. tion that could be obtained by expanding the refer- ence period. B18-B19. The values of personal guarantees, group lending, and cosigners cannot be directly estimated. C2. The reasons for refusal are important and can be linked with data in other parts of the questionnaire on B20. The total amount borrowed in the loan is need- total debt and on asset (collateral) ownership. ed to determine the cost of the loan and to determine how loan size relates to collateral, payback terms, and C3. This question is designed to determine xvhich other aspects of the loan. type of lender the borrower was not successful in bor- rowing from. B21-B22. This is an attempt to determine the interest rate a person is paying, if they are paying an interest C4. This question contains an awkward skip pattern rate. Question B22 may need to be revised to reflect that will need to be practiced by interviewers. the standard language or terms in use in a given coun- try. Or it may be necessary to add a question to probe C5. This question is for all people who either did not for interest rates. As discussed previously, many people request a loan in the previous 12 months or requested will be unable to answer B22 as it now stands. a loan and were turned down. 244 CHAPTER 21 CREDIT make this distinction, the question is useful-but not Box 21.1 CautionaryAdvice if it jeopardizes the rest of the information. How much of the draft module is new and unproven? D3-D10. These questions could be rewritten to ask Parts B. C and D of the standard module draw heavily on other credit surveys and some LSMS surveys. Part about a typical borrower rather than about the most A, which attempts to collecd data by source, has not recent borrower-for example, "Are most of the peo- been tried, and needs to be tested to determine both ple who borrow from you..." or "How much does the respondent's willingness to answer and the time the average person owe you?" While this approach involved in administering this section. could be useful, it has been avoided here because it * How well has the module worked in the past? Most of the emphasizes the respondent's status as a commercial data obtained in past versions of the module seems lender. reasonable, with the exception of interest rates. But because credit is a sensitive issue, some checking with other sources may be useful to ensure that underesti- Credit Module: Short Version mation does not occur The questions in this module are taken from Parts A, * Which parts of the module most need to be customized? B and C of the standard version. See the relevant dis- Specific questions on interest rates need to be cussions for each of the questions in the notes above. addressed most carefully, as do questions concerning the overall terms of loans. The language used to Notes describe these will be quite country-specific. In areas with minimal formal-sector or semiformal lenders, questions on total debt can be cut and fur-ther ques- The author has greatly benefited from the helpful comments and tions can be added to ensure a more accurate meas- suggestions of Paul Glewve, Margaret Grosh, Shahidur Khandker, urement of informal credit and its costs. Anjini Kochar, Ghazala Mansuri and Chris Udry. 1. Of course not all agricultural credit needs are for short term investments. Longer term investments, such as investments in infra- C6. The purpose of this question is to determine bor- structure and equipment, exhibit different patterns. rowers' perceptions of their access to credit.The ques- 2. The impact of credit on the costs of consumption smoothing tion differs substantially from other ones, as it is based is related to the degree to which savings are precautionary.This is, on opinion rather than fact. Although mixing togeth- in turn, a function of levels of risk and of the existence of alterna- er opinion-based and fact-based questions can be tive insurance mechanisms in the economy. It should also be noted problematic, this question is needed given the impor- that, to the extent that households save for precautionary motives, tance of borrowers' perceptions of their access to the supply of credit will affect overall savings levels as credit serves credit. as a substitute. Precautionary savings are covered in Chapter 20. 3. Creditors may also use positive incentives to encourage loan PART D. LENDING ACTIVITIES. The respondents for this repayment. Borrowers who do not default are often provided with part are all people of working age (with no upper age increased credit opportunities by their lenders (for example, by limit) regardless of whether they borrowed in the 12 credit unions and credit card companies). months prior to the interview. 4. A meeting of credit union managers in Tegucigalpa in 1991, attended by the author, discussed the fact that being a saver is con- Dl. The key issue in this question is to find the least sidered a 'good' and being a borrower 'bad'-and that underesti- threatening wording. The interviewer asks not "Have mation of credit xvas common. you lent money to someone?" but instead "Does 5. Data from the Kyrgyz study wvere not available for analysis at someone owe you money?" The idea is to avoid put- the time of writing. ting people in the position where they have to say that 6. The most that past household-level data sets revealed about they are a moneylender. individual credit use was who receives the benefit of credit for pro- ductive purposes. This was deduced by combining the information D2. Pilot testing may reveal that this question needs to on the purposes of the credit the household held (see Table 21.5 for be dropped, since it will identify the respondent as a the division of activities financed by household credit) with the moneylender rather than just a person who once lent household identification codes of the farmer or the owner of the a friend or relative some money. As one would like to household enterprise that are recorded. 245 KINNON ScOTa 7. Loan charges include application fees, compensatory bal- Anderson, Denis, and Farida Khambata. 1984. "Financing Small- ances, closing costs, and forced purchase of services from the lender. Scale Industry and Agriculture in Developing Countries: The Third parties receiving payments include such persons as agricul- Merits and Limitations of Commercial Policies." Economic tural extension agents who must fill out the loan form for the Development and Cultural Change 33 (2): 349-71. farmer or evaluate the land and collateral. Aryeetev, Ernest, and Hemamala Hettige, Machiko Nissanke, 8. This figure, calcilated from the 1991 Pakistan Integrated House- William Steel. 1997. "Financial Market Fragmentation and hold Survey may overestimate lack of knowledge of interest rates Reform in Ghana, Mala-wi, Nigeria and Tanzania." I/World Bank because in the survey database no distinction is made between missing Economic Revietv 11 (2): 195-218. data (refusals, omissions by the interviewer) and respondents' stating Bardhan, Pranab, ed. 1989. Thte Econosmic Tlheory of Agrarian that they did not know the answver to the question. The figure is based Institutions. Oxford: Clarendon Press. on the assumption that all of these answvers were "Do Not Know'" Barham, Bradford L., and Stephen Boucher, Michael R. Carter. This issue illustrates the importance of maintaining the distinc- 1996. "Credit Constraints, Credit Unions and Small-Scale tion between "refusals" and "do not know" answers, something Producers in Guatemala." lEorld Development 24 (5): LSMS data sets usually fail to do. Probably the most effective solu- 793-806. tion is to add an explicit question asking if there is an interest rate, Basu, Kaushik. 1989. "Rural Credit Markets: The Structure of as was done in the Vietnam LSMS. Interest Rates, Exploitation, and Efficiency" In Pranab 9. Unfortunately, the wording of the question on sources of Bardhan, ed., The Fcononmic Theory of Agrarian Institutions. credit makes it impossible to check whether all loans with zero Oxford: Clarendon Press. interest rates were from fansily or friends. Baydas, M. R. L. Meyers, and N. Aguilera-Alfred. 1994. 10. The results are surprisingly high for Ghana, where the "Discrimination against Women in Formal Credit Markets: Muslim population is proportionally much smaller than the Reality or Rhetoric?" WI'orld Development 22 (7): 1073-82. Muslim population in Pakistan. Bell, Clive. 1990. "Interactions between Institutional and Informal 11. This information was provided to the author during discus- Credit Agencies in Rural India." W'orld Bank Economic Revietw 4 sions with credit union managers in 1991. (3): 297-327. 12. A discussion of how to obtain plot-level data is contained in Besley, Tinaothy 1994. "How Do Market Failures Justify Chapter 19, along with an assessment of associated difficulties. Interventions in Rural Credit Markets?" 1lsorld Bank Research 13. In most countries this agency not only maintains lists of all Observer 9 (1): 27-47 banks (public and private) and their locations but also provides . 1995. "Savings, Credit and Insurance." In Jere Behrman information on total assets and habilities. and T. N. Srinivasan, eds., Handbook of Development Economics. 14. There are usually national federations of credit unions and Vol. 34. Amsterdam: Elsevier Science, B.V. other cooperatives; NGOs are also usually registered centrally Binswanger, Hans P, and Mark R. Rosenzwveig. 1986. "Behavioural Some such federations will only be able to provide information on and Material Determiiinants of Production Relations in the locations of such lenders, wvhile others will be able to provide Agriculture"Journal of Development Studies 22 (April): 503-39. details such as information on loans outstanding. Binswanger, Hans P,. Shahidur R. Khandker, and Mark R. Rosenzwveig. 1993. "Hoxv Infrastructure and Financial References Institutions Affect Agricultural Output and Investment in India." Journal of Development Economics 41 (August): 337-66. Adans, Dale. 1984. "Are the Arguments for Cheap Agricultural Bliss, C.J., and N. H.Stern. 1982. Palanpur: The Economy ofan Indian Credit Sound;" In Dale Adams, Douglas H. Graham, and J. D. Village. Oxford: Clarendon Press. Von Pishke, eds., tUindermining Rural Development with Cheap Bottomley, Anthony 1963. "The Premium for Risk as a Credit. Boulder, Col.:Westview Press. Determinant of Interest Rates in Underdeveloped Rural Adams, Dale, and G.I. Nehman. 1979. "Borrowing Costs and the Areas." QuarterlyJournal of Economics 77: 637-47. Demand for Rural Credit." TheJournal ofDevelopment Stuidies Braverman, Avishay, and Joseph E. Stiglitz. 1982. "Sharecropping 17 (2): 165-76. and the Interlinking of Agrarian Markets." Amnerican Economic Adams, Dale, and J. D.Von Pischke. 1992. "Microenterprise Credit Review 72 (4): 695-715. Programs: D6ja vu." World Developnsent 20 (10): 1463-70. Carter, Michael R., and Keith D.Wiebe. 1990. "Access to Capital Aleemi, Irfan. 1990. "Imperfect Information, Screening and the and Its Impact on Agrarian Structure and Productivity in Costs of Informal Lending: A Study of Rural Credit in Kenya." American Journal of Agricultural Econoinics 72 Pakistan." World Bank Economic Review 4 (3): 329-49. (December): 1146-50. 246 CHAPTER 21 CREDIT Choe, Chongwoo. 1996. "Incentive to Work versus Disincentive to .1985b. "Tests for Liquidity Constraints: A Critical Survey." Invest:The Case of China's Rural Reform, 1979-1984."Journal NBER Working Paper 1720. National Bureau of Economic of Comparative Economics 21 (une): 242-66. Research, Cambridge, Mass. Deaton, Angus. 1991. "Saving and Liquidity Constraints." Hodgman, Donald R. 1960. "Credit Risk and Credit Rationing." Econometrica 59 (5): 1221-48. QuarterlyJournal of Economics 74 (2): 227-57. .1992. Understanding Consumption. Oxford: Clarendon Press. Hoff, Karla, and Joseph E. Stiglitz. 1990. "Introduction: Imperfect de Meza, David, and David C.Webb. 1987. "Too Much Investment: Information and Rural Credit Markets-Puzzles and Policy A Problem of Asymmetric Information." Quarterly Journal of Perspectives." World Bank Economic Review 4 (3): 235-50. Economics 102 (May): 281-92. . 1993. "Moneylenders and Bankers: Fragmented Credit 1988. "Credit Market Efficiency and Tax Policy in the Markets with Monopolistic Competition."Working Paper 93- Presence of Screening Costs." Journal of Public Economics 36 10. University of Maryland, Department of Economics, June): 1-22. College Park, Md. Diamond, D.W 1989. "Reputation Acquisition in Debt Markets:" Jaffee, Dwight M. 1972. "A Theory and Test of Credit Rationing: Journal of Political Economy 97 (4): 828-62. Further Notes." American Econiomic Review 62 (3): 484-88. Eaton, Jonathan, and Mark Gersovitz, 1981. "Debt and Potential Jaffee, Dwight, and Franco Modigliani. 1969. "A Theory and Test Repudiations." Review of Economic Studies 48 (2): 289-309. of Credit Rationing." American Economic Review 59 (5): Eswaran, Mukesh, and Ashok Kotwa. 1990. "Implications of Credit 850-72. Constraints for Risk Behavior in Less Developed Economies." Jaffee, Dwight, and Thomas Russell. 1976. "Imperfect Information, Oxford Economic Papers 42 (April): 473-82. Uncertainty and Credit Rationing." Quarterly Journal of Feder, Gershon, Lawrence J. Lau, Justin Y Lin, and Luo Xiaopeng. Economics 90 (4): 651-66. 1989. "Agricultural Credit and Farm Performance in China." Jappelli, Tullio, and Marco Pagano 1988. "Liquidity Constrained Journal of Cormparative Econonmics 13 (December): 508-26. Households in an Italian Cross-Section." Discussion Paper 257. - 1990. "The Determinants of Farm Investment and Centre for Economic Policy Research, London. Residential Construction in Post-Reform China." Agriculture Jappelli, Tullio, Jorn-Steffen Pischke, and Nicholas Souleles. 1995. PolicyWorking Papers Series 471.World Bank,Agriculture and "Testing for Liquidity Constraints in Euler Equations with Rural Development Department,Washington, D.C. Complementary Data Sources." Discussion Paper 1138. Centre 1991. "Credit's Effect on Productivity in Chinese Agriculture: for Economic Policy Research, London. A Nicroeconomic Model of Disequilibrium."Agriculture Policy Keeton, William. 1979. Equilibrium Credit Rationing. New York: Working Papers Series 571.World Bank, Agriculture and Rural Garland Publishing Company. Development Department,Washington D.C. Khandker, Shahidur, and Hans Binsxvanger. 1989. "The Effect of Feder, Gershon, Lawrence J. Lau, Justin Y. Lin, and Z. Luo. 1990. Formal Credit on Output and Employment in Rural India:' "The Relationship between Credit and Productivity in Policy, Planning, and Research Working Paper 277. World Chinese Agriculture: A Microeconomic Model of Bank, Population and Human Resources Department, Disequilibrium." American Journal of Agricultural Economics 72 Washington, D.C. (5): 1153-57. Klein, Benjamin, Robert Crawford, and Armen Alchian. 1978. Ghose, A. K. 1980. "The Formation of Usurious Interest Rates; "Vertical Integration, Appropriable Rents, and the Notes and Coniments." Cambridge Journal of Economics 4 (2): Competitive Contracting Process."Journal of Law and Economics 169-72. 21 (2): 297-326. Guttentag,Jack, and Richard Herring. 1984. "Credit Rationing and Ladman,Jerry R. 1981. "Factors Impeding Borrowing:The Case of Financial Disorder." Journal of Finance 39 (December): the Bolivian Agricultural Bank's Small Farmer Credit 1359-82. Program:' Savings and Development 7 (3): 201-25. Hall, Robert E. 1978. "Stochastic Implications of the Life Cycle- Larson, Donald W 1990. "Honduras: Informal Financial Markets Permanent Income Hypothesis:Theory and Evidence." Joumral Assessment." Technical Report to USAID-Honduras. of Political Economy 86 (6): 971-87. Loria, Miguel, and Carlos E. Cuevas. 1984. "Basic Grains: Hart, Oliver, and John Moore. 1988. "Incomplete Contracts and Marketing Channels and Financing at the Farm andWholesale Renegotiation." Econometrica 56 (4): 119-39. Levels." Report to USAID/Honduras. Economics and Hayashi, Fumio. 1985a. " The Effect of Liquidity Constrains on Sociology Occasional Paper 1077. Ohio State University, Consumption: A Cross-Sectional Analysis." Quarterly Journal of Department of Agricultural Economics and Rural Sociology, Econonmics 100 (February): 183-206. Agricultural Finance Program, Columbus, Oh. 247 KINNON SCOTT McKinnon, Ronald I. 1973. .M1oney and Capital in Economic . 1983. "Incentive Effects of Termination: Applications to Development. Washington D.C.: Brookings Institunon. the Credit and Labor Markets." American Economic Review 73 Miller, Merton H. 1962. "Credit Risk and Credit Rationing: Further (5): 912-27. Comment." QuarterlyJournal of Economics 76 (3 ): 471-79 Thakor,AnjanV, and Richard Calloway. 1983. "Costly Information Montiel, Peter, Pierre-Richard Agenor, and Nadeem Ul Haque. Production Equilibria in the Bank Credit Market with 1993. Informal Financial Mlfarkets in Developing Countries: A Applications to Credit Rationing." Journal oJ Financial and Alkacroeconomic Analysis. Oxford: Blackwell. Quantitative Analysis 18 (2): 229-56. Pitt, Mark, and Shahidur Khandker. 1996. Household and Udry, Christopher. 1990. "Credit Markets in Northern Nigeria: Intrahousehold 7inpact of the Grameen Bank and Similar Targeted Credit as Insurance in a Rural Economv" World Bank Economic Credit Programs in Bangladesh. World Bank Discussion Paper Review 4 (3): 251-69. 320. Washington, D.C. . 1994. "Risk and Insurance in a Rural Credit Market: An Rao,Ju Mohan. 1980. "Interest Rates in BackwardAgriculture: Notes Empirical Investigation in Northern Nigeria." Review of and Comments." CambridgeJournal of Economics 4 (2): 159-67. Economic Studies 61 (3): 495-526. Ryder, Karl E. 1962. "Credit Risk and Credit Rationing." Quarterly U, Tun Wai. 1957. "Interest Rates Outside the Organized Money Journal of Economics 76 (3): 471-79. Market of Underdeveloped Countries." International Mfonetary Scott, Katherine MacKinnon. 1992. "Credit Unions and Fund StaffPapers 6 (1): 80-142 Participation:The Determinants of Credit Union Membership Vogel, Robert C, and Dale Adams. 1986. "Rural Financial Markets in Honduras." Ph.D. diss. University of Pittsburgh, Graduate in Low-Income Countries: Recent Controversies and School of Public and International Affairs, Pittsburgh, Penn. Lessons." World Development 14 (April): 477-87. ShaNv, Edward S. 1973. Financial Deepening in Economic Development. Wiens, Thomas. 1998. "Background Paper for Panama Poverty NewvYork: Oxford University Press. Assessment." World Bank, Latin America and the Caribbean, Sial, Maqbool H., and Michael R. Carter. 1996. "Financial Market Human Development,Washington, D.C. Efficiency in an Agrarian Economy: Microeconometric Williamson, Oliver. 1985. The Economic Institutions of Capitalism. Analysis ofthe Pakistani Punjab."Journal of Development Studies NewvYork: Free Press. 32 (5): 771-98. Williamson, Stephen. 1986. "Costly Monitoring, Financial Siamwalla, Ammar, Chirmsak Pinthong, Nipon Poapongsakorn, Intermediation, and Equilibrium Credit Rationing."Journal of Ploenpit Satsanguan, Prayong Nettayarak, Wanrak Monetary Economics 18 (September): 159-79. Minigmaneenakin, andYuavares Tubpun. 1990."TheThai Rural . 1987. "Financial Intermediation, Business Failures, and Credit System: Public Subsidies, Private Information, and Real Business Cycles." Journal of Political Econonmy 95 (6): Segimented Markets:" World Bank Economic Review 4 (3): 271-95. 1196-216. Stiglitz, Joseph E. 1987. "The Causes and Consequences of the Wolken,John D., and Frank Navratil. 1981."The Economic Impact Dependency of Quality on Price? Journal of Economic Literature of the Federal Credit Union Usury Ceiling."Journal of Finance 25 (1): 1-48. 36 (5): 1157-68. .1990. "Peer Monitoring and Credit Markets:" World Bank Yadav, S., K. Otsuka, and C. C. David. 1992. "Segmentation in Economic Review 4 (3): 351-67. Rural Financial Markets: The Case of Nepal." IWorrd Stiglitz, Joseph E., and Andrew Weiss. 1981. "Credit Rationing in Development 20 (3): 423-36. Markets with Imperfect Information." American Economic Zeldes, Stephen P 1989."Consumption and Liquidity Constraints." Revietv 71 (3): 393-410. Journal of Political Economy 97 (2): 305-46. 248 Time Use Andrew S. Harvey and Maria Elena Taylor The allocation of time is a crucial decision that influences many aspects of households' living standards. The way a household allocates its members' time among various economic activities is an important determinant of the level of its income. Low-income households often devote large amounts of time to activities to meet the basic needs of their members such as fetching water, gathering fuel for cooking, or processing food. Travel and waiting time are often an important part of the cost of using health, education, and other services, and leisure is a common human desire. All of these activities, and the way they are distributed among the different members of the household, are of interest to policymakers. Most previous LSMS surveys have included only a affects every member's time use. It would also make it limited number of questions on time use in the vari- possible to understand the contributions that some ous modules of the questionnaire, generally to study members, especially women, children, and the elderly, specific issues. In some past surveys the housing mod- make to household welfare through their nonmarket ule contained questions on the time that household activities such as collecting water and fuelwood and members spent gathering fuel or water, so that the growing crops for the household's own consumption. value of their consumption could be imputed. Because the economics profession is moving increas- Questions about time spent by household members at ingly toward models that do not take the household as work were sometimes included in the employment a single unit but that take into account intrahousehold module to help analysts calculate earnings rates. The allocations of time and resources, the demand for time education and health modules often contained ques- use data is growing. tions on travel time to facilities to help analysts to cal- A comprehensive time use module for all house- culate the implicit costs of these services. hold members might well be longer than any other Few previous LSMS surveys have tried to build a single module, and would therefore either displace comprehensive picture of time use in a household. others or add considerably to the length of the inter- While this would necessitate the inclusion of a sepa- view. However, these costs are partially offset because rate, lengthy module on time use, such data would a time use module would reduce the need for some extend considerably the range of issues relating to time question sequences capturing time use information in use that could be analyzed. Having such comprehen- other modules and would provide more accurate data. sive information would enable analysts to understand This chapter discusses the practical issues involved in how a change in the labor activities of one member including a separate, module on time use in an 249 ANDREW S. HARVEY AND MARIA ELENA TAYLOR LSMS-type household survey so that designers of The household time overhead is the minimum future surveys can evaluate their options.The first sec- number of hours that a household must spend on the tion describes the issues that can be analyzed with basic chores vital to the survival of the family. This time use data.The second section identifies the ancil- concept was first defined byVickery (1977) who posit- lary data that analysis of time use issues requires and ed that poverty is a function of time as well as money. discusses various ways to collect these data. The third The household time overhead consists of the number section introduces three drafts of a time use module of hours spent preparing meals, washing clothes, and (which are presented in Volume 3), and the fourth cleaning house. It also includes time spent fetching section provides explanatory notes for these draft water and gathering fuel for cooking and heating in modules. households that lack running water, electricity, or gas. In general, a household with low household time Policy Issues RegardingTime Allocation overhead will be better off than a household with a high time overhead. However, the impact of the Data on how much time individuals and households household time overhead also depends on the number allocate to performing various tasks can provide useful of adults and children available to assist in performing information to policymakers by improving analysts' these tasks. The impact of these costs on individual understanding of several areas: household members is determined by the distribution * Living standards. of household maintenance tasks among household * Human capital investment decisions. members. * The labor force. * Nutrition. Time Use As an Indicator of Human Capital Investment * Gender and age inequalities within households. Decisions * The use of public services. One of the primary uses of LSMS survey data is to * Social changes and households' quality of life. analyze households' decisions about their human cap- * Leisure and social interactions as indicators of ital investments such as health care and education. household welfare. Time spent in school and on homework are important Regional variations in time use may reflect variations measures of the investment that an individual and a in prevailing socioeconomic, environmental, or geo- household are making in education. Children's non- graphical conditions,- thus time use variations can pro- school activities, such as paid work or chores, may pre- vide policymakers with important insights into the vent them from benefiting fully from the time that causes of regional variations in household welfare. they spend in school. Time costs can be an important component of the Time Use As an Indicator of Living Standards cost of using education and health services. Therefore, Time is both an important asset and an important in order to analyze household decisions regarding component of the cost of various goods.Time is espe- schooling and health care, it is very important to have cially important in poor households, where capital is a measure of these costs. Having a measure of travel scarce. Empirical studies in developing countries have time is useful even if the analyst already knows the dis- shown that poor households make use of the time of tance to the facility in question, because not all house- all of their members, regardless of gender or age, to holds use the same mode of transportation. In the case provide for the basic needs of the household. Measures of health care facilities, it may also be important to of the amount of time spent by various household collect data on waiting time. The use of time data in members on basic household chores may be an impor- studying these sectors is explored in detail in Chapters tant indicator of household and individual living stan- 7 and 8 on education and health. dards. Also, the amount of time that individuals spend doing basic household chores may limit the time that Time Use Data in Labor Force Analysis they have available to work outside the home, to tend In industrialized market economies, most individuals to agricultural and other household production activ- work a specified number of hours each xveek in a des- ities, to obtain education, or to care for children and ignated location, for a specific employer and a well- elderly household members. defined period of time. However, in developing coun- 250 CHAPTER 22 TIME USE tries and in poor households everywhere, individuals In general, the extent to which time use data from tend to provide for their own needs and the needs of LSMS surveys can be used in nutritional analysis is their dependents through a combination of formal and limited. This is because, as discussed in Chapter 5 on informal activities rather than through formal con- consumption, LSMS surveys have never collected tracts with employers. Consequently, in order to information on individual caloric intakes and have obtain reliable indicators of labor availability and pro- only yielded very rough estimates of the availability of ductivity in surveys fielded in developing countries, calories for the household as a whole. In principle, the definition of work must reflect time spent in pro- having time use data would allow analysts to adjust ductive activities rather than just time spent at a job. their measurements of household food consumption, This means that labor market analysts must take into but this has not been done to date and future studies account the time spent in supplementary employ- will have to be conducted to explore whether it is ever ment, in self-employment, and in the production of worthwhile to do so. goods and services used by the household or traded in the market, as well as time spent in employment in a Time Use Data and IntrahouseholdAllocation and Gender primary workplace in the formal sector. The need to Inequality count all of these activities is reflected in the com- Time use data have become one of the most impor- plexity of the employment module discussed in tant tools of gender analysis. How tasks and resources Chapter 9. However, extreme cases of irregular labor are allocated within a household can have an impor- patterns may be more thoroughly captured in a sepa- tant impact on the well-being of the individuals in the rate time use module.Time use data can also be used household.This issue is discussed in detail in Chapter to measure productivity in household enterprises and 24. Time use data can help analysts study the nature agricultural activities (INSTRAW 1996). Labor mar- and extent of intrahousehold inequality and the ket policymakers need to know about the activities of impact of that inequality.Time can serve as a yardstick individuals who are not employed in the formal sec- that is equally able to measure market and nonmarket tor, what obstacles there may be to bringing people activities, women's and men's work, and the work of and activities into the formal sector from the informal the poor and the rich. The amount of time spent in sector, and how to measure the relative importance of paid activities outside of the home may also have an market and nonmarket production for a household's important impact on the influence that each individ- resources. ual in the household has on the decisions made by the Because time spent commuting is likely to enter household. into the labor market decisions of household members, When policyinakers have not had a complete pic- time use data can be useful in analyzing these decisions ture of what productive activities (whether market, and may provide an indication of the potential benefits nonmarket, or domestic) are needed to provide the of introducing or extending public transportation. household with its basic needs, this has often resulted in the adoption of inappropriate public policies. In Time Use Data in Nutritional Analysis particular, the differences in the contributions made by One measure of individuals' health and well-being is men and women to household time overhead have their nutritional status. A number of previous LSMS often been overlooked by policymakers. Because pub- surveys have gathered information on height and lic policies have long been based on market produc- weight as an indicator of nutritional outcomes. Fully tivity alone, they have underestimated the valuable understanding the determinants of these outcomes is contribution that women make to the household, to quite complex. One piece of the puzzle is the individ- society, and to the economy. One example can be ual's physical activities, since a person who works at a found in Berio (1980), which describes how a public desk every day requires far fewer calories than a per- policy that aimed to increase agricultural yields by son of the same age and sex who works every day encouraging the use of fertilizer on cash crops also doing strenuous physical labor. Having information on increased the growth of weeds and thus the amount of an individual's time use makes it possible for analysts time that farmers had to spend weeding. Since the cul- to calculate that individual's energy requirements tivation and weeding of the cash crop xvas done by (Berio 1980;James and Schofield 1990). women, households' balance of time allocation was 251 ANDREW S. HARVEY AND MARIA ELENA TAYLOR altered in ways detrimental to food production and of time that the children can devote to their schooling the care of children. and thus restrict their future job opportunities. Analyzing the Utilization of Public Services with Time Use Measuring Leisure and Social Interaction to Indicate Data Welfare Information on time use can inform policymakers The need to recuperate after a hard day's work and to about the utilization and impact of public services cultivate family and social relations is a basic human such as public transportation, schools, and electricity. need and is consequently an indicator of a person's Policymakers can factor this information into their standard of living. Yet researchers who have studied decisionmaking and evaluation processes. For exam- households living at subsistence level have tended to ple, when policymakers are estimating the impact of disregard the value of leisure. Poor people face great providing piped water, they should consider the demands on their time to produce the income amount of time that this will save for households as required for basic needs, and are thus unable to spend well as the likely health improvements that will result. enough time for rest and relaxation.This can have seri- Having information about the way time is allo- ous consequences for the health of household mem- cated among the various members of a household bers and for the potential of a household to expand its helps policymakers understand which household opportunities. members will be most directly affected by a policy Measuring leisure is difficult, not least because it modification (for example, one that changes the means different things to different people. Leisure may amount of time required to fetch xvater or to travel to include social interaction, relaxation, cultural activities, work). Furthermore, the price of using public services solitude, physical exercise, or participation in games or is often subsidized for low-income households to sports. The concept can vary from person to person, ensure that they have access to these services. culture to culture, and time to time. This idea has been However, the subsidized service may not actually ben- discussed by Ferge (1972), Skorzynski (1972),Young efit the intended households if it is located far away, as and Willmott (1973), and Shaw (1985). Ideally leisure this presents a substantial cost to the household in should only incorporate discretionary or voluntary terms of time. Thus having information on travel and time. It should not be overestimated by incorporating waiting times may indicate to policymakers the extent idle time or a lack of market activity. Typically in both to which a price subsidy may be eroded by such developed and developing countries leisure is equated opportunity costs. Similarly, if policymakers under- with a household's total time allocation to predefined stand how households organize their day, they can bet- discretionary activities. In a few surveys in industrial ter determine what hours of the day public services countries, leisure is identified by using a follow-up should be open to ensure maximum access for the question for each activity recorded in a time use diary intended target households. asking whether the activity was leisure. This method allows for the fact that definitions of leisure can vary Time Use Data for Evaluating Social Change and Quality at different times and among different people. To take of Life a Western example, women are usually responsible for If an LSMS survey is designed to be repeated over getting supper on the table and would call it a chore time or to collect panel data (as discussed in Chapter rather than leisure. If a woman's husband occasionally 23), the time use data that the survey collects may be cooks as a hobby and he fixes supper on the reference used to provide indicators of social change over time. day of the time use study, he may choose to define his In developing countries this can include how environ- cooking as leisure.This subjective approach is relative- mental degradation affects household time allocation ly time-consuming and thus cannot be done as part of and which economic activities increase the use of a very basic time use module.While subjective meas- child labor (Kumar and Hotchkiss 1988; Skoufias urement may be the ideal, objective measurement as 1994). Also, if adults are required to spend more time typically used has also proven robust and useful. on market activities, children may have to cover the Measuring social interaction helps analysts under- residual household time overhead by completing the stand the differences between paid and unpaid work adults' household tasks, which may reduce the amount and how members of families interrelate with each 252 CHAPrER 22 TIME USE other in both traditional and modern cultures. Many each episode is important; for example, four hours of of the changes that take place in the transition continuous strenuous labor places a greater burden on between subsistence and market economies restruc- an individual than two two-hour episodes. In other ture families' use of time in ways that reduce or alter cases the total duration of all episodes of the activity social interaction among their members. For example, may capture the relevant information; for example, the when women wash clothes in the river or plant crops total number of hours of television watched is more in the company of their children and friends, they are important than the number of hours watched in one interacting with each other and teaching their chil- sitting. dren necessary social skills. This is in marked contrast For some activities it may also be useful to gath- to people who work in factories, isolated from their er other information. The temnporal location-the time children, family, and friends. of day, week, month, or year when an activity is undertaken-may be important for understanding Data Requirements and Data Collection the rhythm of society and the time constraints with- Methods in which time allocation decisions are made. For example, the time of day at which people depart for This section explores the methodologies for gathering xvork, shop, eat, prepare meals, or get out of school the data needed to analyze the issues discussed in the may influence other household decisions. If so, this is previous section. First, various aspects of measuring useful information for decisionmakers to have as they time are discussed. This is followed by a discussion of decide what hours health clinics or community cen- four different ways to gather time use data in the con- ters will be open or what the peak times of day will text of an LSMS survey and the benefits and disad- be for the bus system. The activity sequence relates the vantages of each.The last part of this section discusses activity to activities that precede and follow it, help- further design considerations. ing analysts understand how individuals organize their day.This information can guide policymakers in Measuring Time locating services and in scheduling operating hours to Time use studies can be "exhaustive," in which case ensure that services are accessible. The context within they record all of a household's activities within a which activities take place can also be important; con- specified reference period (such as one hour, day, or textual infornutation includes the location of an activity, week). This is equivalent to continuous sampling and the other people present, the person for whom the captures all activity during a predefined period. activity was done, and any remuneration that may Alternatively, time studies may be "selective" in that have been received for the activity. Subjective infor- only the amount of time allocated to certain activities mation such as how much an individual enjoyed an is recorded. A third option is time sampling, in which activity or how much stress he or she felt may also be the respondent or observer records the behavior of useful. Finally, it is important for analysts to know household members at predefined points in time-for about any concurrent activities (for example, women example, at 10 a.m., noon, 4 p.m., and 7 p.m. The caring for children while preparing meals); failing to implications of these three sampling options are dis- control for these activities will cause analysts to great- cussed below. (For more details see Martin and ly underestimate the occurrence of secondary activi- Bateson 1986.) ties.' Secondary activities include listening to the radio, having conversations, talking to children, gar- WHAT TO MEASURE. In time use studies, the basic unit dening, and looking after pets (Harvey and is an episode, a single entry on the diary with all its Macdonald 1976). attendant dimensions. Time spent on an activity has A large percentage of these secondary activities two aspects: the numnber of episodes and the duration of tend to be underrecorded. For example, women tend the activity. For some activities, such as brushing one's to perceive domestic and personal care activities as of teeth, the salient feature is the number of episodes or no importance. Thus, unless an instrument is specifi- the frequency of their occurrence. For other activities, cally designed to capture activities that are not per- such as homework, the most important aspect is the ceived to be important, a large percentage of domestic duration of the activity. In some cases the duration of and personal care activities will remain unrecorded. 253 ANDREW S. HARVEY AND MARIA ELENA TAYLOR Information on participation in an activity and on their activities seem to be more predictable and rou- the activity's duration, frequency, and temporal loca- tine than those of people living in developed societies. tion is straightforward to collect using time diaries or Acharya (1982) wrote that subsistence societies: individual questions. Data on context, activity sequence, and concurrent activities are difficult to col- "... have their own time-scale that aids them for lect using questions, and generally require a diary for- daily activities. On visiting the fields in Asia dur- mat. Subjective information such as how a respondent ing cropping or harvesting season, one finds peo- felt during an activity can in general only be collected ple doing almost the same thing at the same hour using qualitative methods (see Chapter 25). every day. For example, in the Nepal Terai or in These basic pieces of information about episodes North India, during the harvesting period, people can be combined in different ways, using different wake up early before dawn, perform hygiene in units of analysis, depending on the issue to be studied the early hour and go to the fields. In the late (Harvey 1999). Analysis may focus on the whole pop- morning, around 7 or 8, it is usual to see the same ulation, the subset of people who participate in an persons carrying the food to the fields. One finds activity, an individual, a household, an activity, or an them having a mid-day meal at noon and resting episode. The whole population is the unit of analysis about one hour. Without a conception of daily for such issues as what proportion of the population is time distribution, it is hard to perform these activ- engaged in a particular activity and how much time ities with such accuracy." these people spend on it on the diary day. Analysts may also be interested in how often and for how long par- In order to collect time use data in such societies ticipants engage in a particular activity or how indi- or households, survey designers need to give special viduals or households allocate their total time. attention to translating the local perception of time Alternatively, analysts may wish to focus on an activi- into a standard 24-hour timetable. (The notion is sim- ty. When an episode is taken as the unit of analysis, it ilar to that of creaing a local calendar to help date becomes possible to determine the time of day at births, as discussed in the anthropometry chapter.) which episodes take place, the presence or absence of Table 22.1 illustrates one good example of a diary that individuals (such as the respondent's children or was developed to do this.Wagenbuur (1972) used such spouse), and the presence or absence of secondary a method in a time use study in southern Ghana.The activities. Using the episode as the unit of analysis also method was also used by Kennedy, Rubin, and shows how the respondent fits an episode of a given Alnwick (1991), who analyzed and compared three activity into a sequence of episodes during the day, time use studies in Kenya, two of which used the recall making it possible to analyze how people organize method in which time sheets were adapted to incor- their day. At the most fundamental research level it is porate the hours and time periods used in their area of necessary to know the amount of time allocated to study. In Islamic cultures, the five daily prayers can specific activities such as paid work, housework, child- provide markers for time during the day. care, and education. This yields information regarding Given that people's perception of time is based on the presence or absence of the activity, the amount of geographical conditions, religion, productive activities, time allocated to it, the frequency of its occurrence, and tradition, it is not possible to design a standard and the types of individuals and groups that engage in LSMS time-use module that can be used in all coun- it. tries. As with the other modules, its design must take into account and reflect the mores and traditions of the MEASURING TImE WITHOUT A CLOCK. In parts of the group to be surveyed. In this case it is necessary to world where people live at subsistence level, they often understand how the community identifies the hours of relate their activities not to hours as they appear on a the day and how they calculate the amount of time it clock face but to fluctuations of nature. These people takes them to perform an activity. This local under- may be more concerned about taking advantage of standing of time can be integrated into the time diary, daylight and the seasons than about coordinating their and it must be used to develop individual questions and activities.This does not mean that there is a complete answer codes. For example, it will be necessary to lack of organization in their lives. On the contrary, determine how respondents might answer the question 254 CHAPTER 22 TIME USE Table 22.1 Examples ofTimeTerminology in Southern Ghana Standard Time Time indications as used by majority of farmers English translation Activity Midnight Esuoom Deep darkness 1:00 a.m. Akok-kan First cock crow 2:00 a.m. 3:00 a.m. 4:00 a.m. 4:30 a.m. Otsia-ebaasa or Enyimaye-wona nyew-awo Third cock crow or Inability to recognize other faces 5:00 a.m. 5:30 a.m. 6:00 a.m. Anapa Morning 6:30 a.m. Akoffo-reko-haban-mu orAdze-Akye Farm-going period or Day is on 7:00 a.m. 7:30 a.m. 8:00 a.m. 8:30 a.m. 9:00 a.m. Win-awo Sky is dry 9:30 a.m. I 0:00 a.m. 10:30 a.m. Oka kakra ma wi-egyina Sun about to be still i 1:00 a.m. 11:30 a.m. Noon Wi-redan Sun still 12:30 p.m. 1:00 p.m. Wio-redan Sun turning 1:30 p.m. 2:00 p.m. Wi-adan Sun has turned 2:30 p.m. 3:00 p.m. Pon-aber-aso Closing time 3:30 p.m. 4:00 p.m. Abe-twa-ber Palm-wine tapping period 4:30 p.m. 5:00 p.m. Wireko or Osomfo-wia or Osomfo wi-asten Sun about to set 5:30 p.m. 6:00 p.m. Wi-ato Sunset 7:00 p.m. De dafo Sleeping agent 8:00 p.m. 9:00 p.m. Adze-asa Day is over 10:00 p.m. 1 1:00 p.m. Kurom-ater-dzinn Night is advanced, town is dead silent Midnight Source: Wagenbuur 1972. of "How much time do you usually spend fetching Instruments for Collecting Time Use Data water?" It will also be necessary to determine how to Traditionally, time use data in developing countries have translate certain answers into time. For example, when been collected by the interviewer observing the people women said that a given activity took them all morn- about whom the data are gathered. This has the advan- ing, the authors translated this as four hours (Kennedy, tage of providing extensive amounts of information on Rubin, and Alnwick 199 1).To minimize random inter- time use and on the cultural context within which pretation this translation must be done according to decisions about time use are made. However, observa- instructions given by the survey designers. tion is very costly and difficult to do for large samples. The importance of reflecting local perceptions of Therefore, this method is more often used in qualitative time in the module can be overemphasized. This is an studies of time use. In contrast, the discussion in this area in which qualitative studies may provide helpful chapter will focus on quantitative data collection meth- inputs into module design (see Chapter 25). At an ods that can be analyzed in conjunction with the other expert meeting in Osaka in 1994, representatives of household and individual data from an LSMS survey. seven Asian countries all agreed a diary was not partic- The activities that are of analytical interest and the ularly problematic with respect to time consciousness. type of information needed to analyze them will deter- 255 ANDREW S. HARVEY AND MARIA ELENA TAYLOR mine the instrument that is needed to gather time use below; explicit illustrations are given in the draft time data.A variety of quantitative instruments can be used to use module (introduced in the third section of this collect time use data; these instruments vary significant- chapter). These diaries should not be seen as ly in terms of the amount of time and expense that they "either/or" approaches; different kinds of diaries can require. So, as for all other modules of the survey, design- be combined within a single survey. In fact, within a ers have to make difficult decisions about the tradeoff single household a more elaborate diary may be used between additional or higher quality data and the extra for one respondent while a less elaborate diary is used expense and logistical burden of more detailed methods. for other household members. As long as underlying One way to collect limited time use data is to use design principles are adhered to, integrating the two random observation. The interviewer arrives at the types of diary should pose no significant problems. household at a time determined by the random draws of time periods specified as part of the sampling plan. Questions Included in Various Modules The interviewer can then observe a rich set of con- Very few previous LSMS surveys have included sepa- textual information about the activity-who is pres- rate time use modules. Most have placed individual ent, the quality of interaction among the people pres- questions on time use throughout the questionnaire. ent, any secondary activities, and how long the The housing module frequently contains ques- primary activity continues. This method has several tions about the amount of time spent gathering wood drawbacks. Because this method is more expensive and fetching water. In analysis of intrahousehold allo- than other methods, fewer households or less of their cation issues or barriers to children's enrollment in total time can be observed using this method. The school, the (often considerable) time dedicated to approach works best at times that are acceptable for these tasks can be counted as a contribution to the visiting people's homes-say, 7 a.m. to 7 p.m.-and household by the women and children who most thus causes the interviewer to miss any activities car- often do them. Some calculations of household con- ried out at dawn and in the evenings. It is difficult to sumption include imputations for the value of the capture any activities that take place very far from the time used in these tasks. Time spent gathering water home. And the timing of any observations made is dif- can also be used as an indicator of a household's con- ficult to reconcile with the data gathered on other tact with community and public services. activities in a multitopic survey such as the LSMS. Due The education module often includes questions to these drawbacks, the random observation method about time spent in school and time spent traveling to will not be examined further in this chapter. school. Questions about time spent doing homework The method used most often in previous LSMS are sometimes included to allow analysts to examine the surveys is to include questions about the household's balance between children's schooling and any domestic allocation of time to different activities in the various or paid work activities. Questions about time spent by modules of the household questionnaire. This other household members helping children with approach is not specifically illustrated in this chapter; homework, volunteering at their school, or helping instead, examples are given in the other chapters in them get to school are rarely included but can be use- this book that deal with the most common time use ful for rounding out an indicator of the actual invest- questions. However, when evaluated against more vig- ment a household makes in educating its children. orous collection approaches, evidence suggests that LSMS health modules often include questions on data generated by this approach are subject to signifi- the number of days lost due to illness in the previous cant reporting error. four weeks and on the amount of time spent traveling Another option is to design a special-purpose to, and waiting for care at, health facilities. In order to module in the form of a diary to capture data on all measure the total loss of welfare to the household, it time use during the reference period. The number of may also be important to have information on the activities specified in a time diary can range from a few amount of time that other household members devot- dozen to hundreds.2 The most complete type of diary ed to caring for the individual during his or her illness. can measure not only primary and secondary activities The employment module gathers information on but also social interaction and other contextual details. time use in various categories of employment.The ques- Three prototypes of time use diaries are described tions in this module are designed to determine labor 256 CHAPTER 22 TIME USE force participation rates, and data on the number of necessary data in a diary with a short reference period, hours worked are often used to calculate earnings rates. because this activity does not occur regularly. Questions may also be included to gather information on time spent commuting.A time diary can reveal details STYLIZED ACTIVITY LOG. The second type of diary, the about the allocation of time for self-employment or stylized activity log, captures not only the duration of work in household enterprises. Estimating labor force the activity but also the number of times it is done participation requires data only on the total number of (episodes). In the activity log, respondents note each hours worked, but, in reality, those hours may have been activity that they did in each 15-minute period of the divided among several enterprises. previous day. This approach requires the respondent to think through the day and remember transitions from Specific Time Use Modules activity to activity. With this information, analysts can When survey designers choose to include a separate understand how an individual's day is organized. This time use module in their survey questionnaire rather systematic recall process increases the accuracy and than distributing the time use questions among the completeness of the information that respondents give other relevant modules, they have several different about their time use. The same basic principles apply instruments from which to choose. to the log as to the list except that the time reference period for the log is fixed as a specific day and no STYLIZED ACTIVITY LIST. The stylized activity list (pre- information beyond that time is collected. If, as is rec- sented in Volume 3) collects information on the fre- ommended, diaries are obtained from all or most quency and duration of time spent on activities dur- members of the family, analysts should be able to study ing the previous 24 hours.This instrument is designed the tradeoffs made among members of the household. to gather data on the frequency and duration of time The stylized activity log yields information on the spent on a limited list of activities. For a list of 20-30 frequency of and duration of time spent only on activ- activities, respondents are asked whether or not they ities during the previous 24 or 48 hours. However, it participated in each activity in the previous six yields complete information on the timing, duration, months, in the previous week, on the previous day, or and number of episodes of those activities, as well as on the day before that. If the answer is yes for either on the activity sequence. Limited information on the the previous day or the day before that, the respon- location of and any remuneration for each activity dents are asked how many hours they have spent on may also be collected by asking the respondents about that activity during that day.The activities listed must those things in the stylized activity log. Secondary relate to a viable and internationally comparable set of activities may be noted by marking the appropriate activities and must capture all of the activities that the time periods using a different colored pen or pencil respondent does during a given day.The total duration than the one used for primary activities.And with the of all activities must equal 24 hours. Practically, this log, it is easy for the interviewer to check visually that forces the interviewer to interrupt the interview to do the whole day is accounted for, without having to the sum and possibly go back and correct times or interrupt the interview to do sums. Depending on log insert activities. (This awkxvard task is avoided in the design and equipment available, it may be possible to other formats of the time use module.) enter data using scanning rather than typing. The reference periods used-some combination of one year, six months, one month, one week, or one OPEN INTERVAL TIME DIARY. In an open interval time day-should reflect existing definitions of labor force diary, respondents either note or are asked what activ- participation and the periodicity of activities relevant ity they were doing when they began the day, what to the population in question. Analyzing several of the activity came next and at what time, and so on suc- issues mentioned in the first section of this chapter cessively through the day. For each principal activity requires having information on specific activities such and its start time, other data can be noted, including: as travel time to school or work; in such cases the activ- location; secondary activity; who else was present; the ity list might be expanded to include these activities. In person for whom the principal activity was per- other cases, such as time spent waiting at a health clin- formed; machines, equipment, or animals used; and ic during the last visit, it is not feasible to gather the remuneration received. 257 ANDREW S. HARVEY AND MARIA ELENA TAYLOR The open interval time diary is better than the time use per se. Some of these weaknesses can be stylized activity list or log in three xvays. First, respon- avoided in future surveys with minimal effort. Others dents are asked to recall their day in much more detail cannot be avoided except by including a separate time than in the log; being asked about where they went use module in the survey. and with whom triggers their memories of the previ- The most obvious limitation is that it is not pos- ous day's activities. Second, the answers are given in sible to acquire a full accounting of individuals' time the respondents' own words with no input from the use during the previous 24 hours, which limits the interviewer; after the respondent completes the diary, amount and range of analysis that can be done. It all responses are coded by trained data coders.3 Third, means, for example, that it is not possible to study how the open interval diary yields significantly more a change in one person's labor activities affects the use detailed information on each activity that it covers. of time by other household members. Nor is it possi- Time diaries kept for a 24-hour day or longer are ble to apply physical activity ratios to estimate energy the best way to obtain information on daily activities requirements or to compare energy use with energy or activities that occur regularly.This is particularly the intake. (Recall that dietary intake is not measured well case in situations where individual activities are hard in LSMS surveys anyway.) Moreover, without a 24- to distinguish from each other, as noted by Niemi hour accounting framework, time use specialists (1983). Both primary and secondary activities should believe that the resulting time use reports are often of be captured to reflect and measure daily activity pat- dubious accuracy. The lack of a 24-hour accounting terns fully. If information is not collected on second- time frame means that the analyst cannot tell whether ary activities, certain crucial data will be lost.Without respondents used those times for which they did not information on the location of the activity and on the mention having performed any activity to perform people accompanying the respondent, the other data other activities that the questionnaire did not ask are considerably less useful, as analysts are frequently about (such as personal care, sleeping, or leisure) or interested in finding out the extent to which people whether the respondents were doing something the are away from home and how much time they spend analysts would have liked to have known about. If the with, though not necessarily caring for, their children question is worded vaguely enough that respondents (Rigbers 1996). do not recognize all the possible variations in each cat- egory of activity, this can easily happen. The Advantages and Disadvantages of the Different Ways For example, the 1993 Jamaica Survey of Living to Coilect Time use Data Conditions, which used a modified activity list, yield- Among the range of options for collecting time use ed an average of only 50 hours' worth of activities per data presented in this subsection, there are significant week. The major activities missing from the list were differences in terms of cost and management com- personal care and sleeping, which could not be plexity, quality of data, and degree of detail of the data expected to account for more than 50 to 70 more collected.The general advantages and disadvantages of hours, so this left between 30 and 50 hours of the the various methods are set out in Table 22.2. An week that went unrecorded. Because the activity list important criterion in selecting a data collection did not have a summation to 24 hours like the draft method is the range of policy issues that can be illu- module introduced in this chapter does, the inter- minated by the data that will be collected with each viewers could not check to see that the respondents instrument. Table 22.3 rates each of the proposed were interpreting the activities in the way meant by instruments-individual questions, an activity list, an analysts. activity log, and an open interval diary-in terms of Another major drawback to the stylized approach how much valid, reliable, and in-depth information is that in certain occupations, such as agriculture and each instrument can collect for addressing different other primary sectors, it is difficult in aggregate to sep- analytical issues. arate activities over the day into work and nonwork To date, most LSMS surveys have collected time components. Thus stylized questions can lead to con- use data by including individual questions in various siderable reporting error. For example, in Canada, pri- modules.4 There have been several weaknesses in most mary sector workers' paid work per week was estimat- of these surveys from the point of view of studying ed at 22.9 hours in diary reports and 37.8 hours in 258 CHAPTER 22 TIME USE Table 22.2 Comparing Methods of Time Use Data Collection Direct Observation. The preferred method in developing countries has been direct observation-recording of episodes or events by an out- side observer Advantages Disadvantages * Gathers data in a systematic way (Kalfs 1993). High cost. * Eliminates the problem of illiteracy and t me measurement. Samples tend to be small, reducing the representativeness of the * Used in settings where seif-reporting, either through recall or diary data. keeping, is likely to produce highly questionable results-for example, * Since only one observer is assigned to a household, only one n situations where respondents do not have a clear sense of time. person can be read ly observed. * Can be useful where activities are unstructured and fractionated in - Knowing they are being observed, people tend to change their very sma I segments or where several activities are performed pattern of behavior choosing activities that are considered more simultaneously and respondents cannot allocate their time disposition. socially acceptable. * Helpful as a precursor to the development of tools and instruments for * Observers hnd it difficult to distinguish between time spent on understanding the sequence in which activities are performed. market and nonmarket activties. * An important tool to evaluate time inputs into an activity simultaneously engaged in by several individuals (Khan and others 1992). .....................................................1................................... .......................................................................................................................................... Random Observation. As the name indicates, this approach creates a sample of the time to be observed. Such a sample can take one of two forms: fixed-interval sampling or random-interval sampling.The most popular approach in time use studies is random time sampling, known as ran- dom spot check observation. Great care must be taken to insure that the sample is representative of the study population. It must capture a true cross-section of each respondent in each situation-location, time period, season. Advantages Disadvantages - Instantaneous sampling readily prov des data appropriate to * Time-consuming compared to the diary approach. estimating percentage of time spent in var ous activities (Altman 1974). Does not provide the detail provided by direct observation and - Gathers data in a systematic way (Kalfs 1993). time diaries. - Eliminates the problem of illiteracy and time measurement. * Limits the period when people may be observed to daylight or Used in settings where self-reporting, either through recall or diary nonsleeping hours a problem when measuring women's burden of keeping, is likely to produce highly questionable results-for example, work, especially that of young mothers whose work extends to night in situations where respondents do not have a clear sense of time. hours. - Can be useful where activities are unstructured and fractionated in * Observers find it difficult to distinguish between time spent on very small segments or where several activities are performed market and nonmarket activit es. simultaneously and respondents cannot allocate their time disposition (Khan and others 1992). * Reduces the cost of the survey and the impressionistic tendency in participants, relative to the direct observat,on approach. Questions Integrated Throughout Other Modules. This traditional paper and-pencil approach asks separate questions about time use in modu es that d scuss other topics.This is the approach most commonly used in LSMS surveys to date. Advantages Disadvantages * Reference periods may differ according to activity; facilitates the * Not very accurate. capture of rare events: can average out observations for common * Does not build comprehensive picture of individual or household but not daily actvities. time use. * Raises each topic at point in interview where other aspects of the * Difficult to adapt question wordings for non-clockwatching societies. ubject are covered. * Integrates well with rest of survey system in interviewer training, data entry requirements, and so on. * Inexpensive. ......................I............................................................................................................................................................................................................ Stylized Questions or Activity List. This traditional paper-and-pencil approach asks either open- or closed-ended questions on activity partici- pat on and use of time. Advantages Disadvantages * Reliable as far as ncidence and frequency of part cipation (Kafs 1993). * Tends to produce inaccurate estimates of duration (Kinsley and * Less costly to process than diary data. O'Donnell 1983; Robinson 1984). * If the list is designed to provide a place for any activity and is * Stylized questions are much more dependent on perception, or constrained to capture a 24-hour day, the accounting nature of subjective calculation, of time use, It is much more difficult to note time use studies is maintained. small changes in time allocation at an aggregate level. * Time-constrained stylized quest ons force people into a 24-hour day, assuming that a person does one thing at a time, Consequently, a large number of concurrent activities go unrecorded. (Table continues on next page.) 259 ANDREW S. HARVEY AND MARIA ELENA TAYLOR Table 22.2 Comparing Methods of Time Use Data Collection (continued) * Time-unconstrained stylized questions allow for the recording of concurrent activities. Ending with a day that has more than 24 hours, however, produces an irreconcilable problem at time of analysis: inability to determine if this is caused by an error of recording or recording of concurrent activities. * Wording questions in a stylized mode is extremely difficult and complicated, leading at times to misinterpretations as shown in the LSMS survey from Pakistan. * This method has limited facility for capturing dimensions other than the primary activity .................. ........................................................................................................................................................................................................... ...... Stylized Activity Log. This precoded diary form provides for recording all activities in the defined recording period, which is typically one day Advantoges Disodvantoges * Precoding considerably reduces coding time and cost. - Considerably reduces the level of detail possible for the activity * The instrument can be designed for mach ne capture. codes. * Using this log maintains the episodic structure of time use data. - Reduces the level of episode detail it is possible to capture. * Constrains time to a 24-hour day, maintaining the accounting nature of the data, * With the use of pictures, can be designed for use by an illiterate population. * Respondent or interviewer does not need to write in detail. Interviewer-Administered Time Diaries. These diary forms are for recording minimally primary activities but more generally for recording primary and secondary activity, location, other people present, and any additional episode (diary entry) data sought for a defined survey period (often 24 hours). Advantages Disadvantages * "A chrono ogical report comes closest to a reconstruction of life as - Design may be too complicated to follow, especially in developing it is experienced" (Scheuch 1972). countries where interviewers and respondents may have low levels of * Provides consistency in time activity data by following activity education. (However this problem of design can be overcome.) through the day and forcing a full accounting of time. * According to Kalfs (I1993), minimizes recall bias, wording problems of stylized questions, exaggeration of socially acceptable activity, and underreporting of socially undesirable activities. * Accurate and reliable data can be obtained. * Depending on design, may allow for recording of primary and concurrent activities as well as sequential, spatial, and social dimensions of the activity .............................................................................................................................................................................................................................. ...... Tomorrow or Left Behind Diaries. These are generally the same as the time diaries, except that the diaries are left for respondents to com- plete as the assigned diary day progresses.They are then collected and reviewed by the interviewer (preferable) or mailed back. Advantages Disodvantoges * Since events can be recorded as performed, the period recorded - Difficult to implement in an illiterate society unless diary is designed can be extended to a longer period of time than in other methods. especially for this purpose by including pictures of activities; * A less expensive method of data collection, unless staff have to spend considerable time may be spent training the respondents. a lot of time reviewing and correcting diaries. * In practice respondents do not record events immediately after their occurrence, reducing the reliability of the data. * There is a problem reporting the sequence of daily activity; sometimes interruptions, overlapping, or concurrent activities are not recorded. * Quality of recording may decline with time. Respondents tend to be conscious of recording at the beg nning of the project, but their enthusiasm drops as time passes. * Tomorrow diaries are expensive due to the repeated visit to the household. * Because respondents become aware of their activities, they may tend to change their habits. However since most activities in subsistence societies involve the fulfillment of a basic need, this may be less evident than in developed countries. Source. Authors summary of literature. 260 CHAPTER 22 TIME USE Table 22.3 Usefulness of Time Use Data Collection Tools in Addressing Policy Issues Modules from which additional data are Policy issue Questions List Log Diary required (data required) I Redefining poverty in terms of income Low Medium Medium High Income (general and time income data) 2 Measuring the ratio of market to nonmarket Medium Medium Medium High productive activity in the economy .........................................................................................................................................I......................................................................................... 3 Measuring the potentially employable Low Low Low High Education (education levels) labor force ................................................................................................................................................................................................................................... 4 Measuring the contributions of women, Medium Medium Medium High Employment, Household children, and the elderly to market and Enterprises, Agriculture nonmarket productive activities ................................................................................................................................................................................................................................... 5 Measuring the labor force participation Medium Low Medium High of children and the implications for their education 6 Assessing whether households can transfer Medium Low High Employment (wages), from nonmarket to market activities Community (prices), and meet all the household's Education (education levels) basic needs Household Enterprises, Agriculture ................................................................................................................................................................................................................................... 7. Identifying nonmarket activities that High Low Medium High Community would be unnecessary if (infrastructure) infrastructure were available (for example, fetching water) 8 Assessing how the time losttto ill High Low Medium High Health (health status) health affects the time individuals spend on productive activities 9 Evaluating how the burden of work is Low Low Medium High Time Use (diaries of multiple distributed among members of a household household members) ......................... ......................I.......................................................................................................................................................................... . .Recognizing the contribution women Low Medium Medium High make to satisfy the needs of the household I Assessing household and famiy Low Low Medium High Community (child care impact of women working outside available in the community) the homeseats .............................................................................................................................................................................................................................. .Assessing how access to public Medium Low Medium High Income (income levels), service affects household Community (level of fees, time allocation community services, infrastructure) 13 Assessing how family cohesion is -- Low High Time Use (data on affected when family members others present during increase the time dedicated to activities) productive activities .............................................................................................................................................................................................................................. 14 Assessing who looks after children - - High Time Use (data on and the elderly when families have others present during to increase their productive activities activities) i.5 Evaluating social behavior and - Low Medium High Time Use (diaries of household habits multiple household members) ............................................................................................................................................................................................................................. 16 Evaluating how the scheduling of public - - Medium High services affects the ability of groups, such as women, to benefit from them ......................... ............................................................................................................................................................................................... 17 Identifying voluntary activity in order - Low Low High Time Use (data on others to incorporate it into public policy and present during activities, national accounts recipients of activities) .Assessing differences in nutritional Low Medium High Anthropometry needs as a basis for nutritional (anthropometric data), monitoring and planning Health (health status) Source: Authors summary. 261 ANDREW S. HARVEY AND MARIA ELENA TAYLOR stylized questions on the labor force survey. In con- clinic in the four weeks prior to the survey are asked trast, for materials handling occupations these values how long they waited there. The reference period is were 32.8 and 32.9 hours. long because visits to clinics are not very frequent and Other drawbacks of using questions scattered are important enough to those involved that they can throughout several modules rather than a full time-use be remembered many days afterward. Moreover, since module can be avoided in principle, even if they have the information on waiting time is used to build up a not always been avoided in practice. In the Pakistan measure of the total implicit and explicit costs of seek- Integrated Household Survey (1991), for example, ing health care, it must have the same (four-week) ref- questions about nonmarket uses of time were erence period as the other data on health care costs. addressed only to women, so it was not possible to In other cases respondents are asked about the determine the extent to which men participated in amount of time spent on an activity, such as home- nonmarket activities, ruling out comparisons and work, in a normal day or week.5 In this case the ana- some intrahousehold analysis. Moreover, for women lyst is trying to understand how much investment is the amount of time and number of episodes spent being made in education.The implicit cost of a child's preparing meals were both lower than expected. An time doing homework, attending school, and com- answer to the question, "How many hours a day do muting back and forth from school may be added to you normally spend preparing meals?" yields no infor- explicit fees. mation on the number of episodes of meal preparation Customizing reference periods on time use ques- in a day and may cause respondents to underreport the tions will support the varying sector-specific analytical amount of time. When past surveys have included purposes for which time use data have been gathered questions about time spent gathering wood or fetch- in most LSMS surveys. However, it also complicates ing water, these questions have yielded data on the the job of trying to build a comprehensive picture of amount of time spent but not on the number of the way an individual or household uses time; such a episodes, nor on which household members are comprehensive picture may be required for accurate involved or on the time of day. It would, of course, be time use measurement or for analyses by time use possible to add some questions to cover these points, researchers. but it would too cumbersome to do so exhaustively. If The problems with the limited approach to time analysis of such points is an important goal of the use taken in LSMS surveys can be overcome by hav- LSMS survey, it may be necessary to include a separate ing a module that specifically covers time use. time use module with a diary to be completed by all However, the cost of including a separate module can household members. be significant. As Leones (1991) puts it: "Time alloca- Because it has sometimes been difficult for tion data win hands down as the most time-consurm- respondents to distinguish between primary and sec- ing and tedious household information to collect, ondary activities in specific questions on time use, there organize, and analyze." may have been substantial undercounting of some Using the diary approach represents a substantial activities in past surveys. For example, a woman may increase in interview time, which can be a burden to not report any time that she spends caring for her child respondents and increase the cost of the survey. There even if the child is with her all day as she minds a shop are no firm numbers, but anecdotal evidence from or works in the field. In surveys that are fielded in areas poor Latin Arnerican and African countries suggests where the local people are not accustomed to thinking that it takes an average of about 20-30 minutes per about time in hours and minutes, it is probably not person to fill out the diary in an interview in a devel- workable to include questions on time use in the vari- oping country.6 If all members of every household ous modules of the survey, as diaries may elicit more were to be interviewed, this would be one of the and better information from these populations. longest modules in the survey and, unlike some of the In LSMS surveys the questions included in vari- other long modules (such as agriculture or household ous modules have been designed according to the enterprises), the time use module would have to be needs of the particular modules. The reference periods administered to all of the households in the sample. have been chosen to suit the sector-specific analysis Having a separate time use module may add costs intended. For example, people who visited a health in less obvious ways depending on the nature of the 262 CHAPTER 22 TIME USE module used. A stylized activity list should make little and 5,000 households. This is on the high side but difference in data handling. However, if an activity log within the bounds of the usual LSMS sample size. or diary is used, the format is so different from that of the other modules that more time will be needed to AGE. The time use module should be administered to train interviewers and data entry operators. Data entry all family members who have reached the age at will take longer and data analysis will be more com- which a child might start to participate in household plicated. If an open interval diary is used, a large and or other unpaid work. This will differ from country to complex coding exercise will be required. Moreover, country. In general, the cutoff age should not be high- the resulting data will have a complex structure that er than it is for the employment module; the age may requires additional work to integrate them with the often be lower if children start to do a significant other information in the survey. number of chores before going to work outside the In summary, there are no "win-win" options for household. The age of 7 has been used in the draft dealing with time use. The traditional partial LSMS time use module inVolume 3. approach severely limits the range of analysis possible and does not conform to the practices that time use GEOGRAPHY. The sample for the time use module surveyors agree are required to get good quality data. should be a national sample and should be large and On the other hand, the most complete time use mod- diverse enough to enable analysis of significantly dif- ules will add a significant burden to the questionnaire; ferent economic or ecological areas. At a minimum, they are likely to take longer to fill out per household it should be possible to identify urban and rural than either the agriculture or household enterprise respondents. modules, and they must be filled out by all households. The cheapest alternative is the stylized activity list, RECALL PERIOD. As a rule of thumb, respondents should which can more efficiently and accurately provide not be expected to recall events that occurred more than time use data frequently collected in LSMS surveys, two days earlier. How the length of the recall period along with additional time use data. The activity list affects the quality of the data depends on which day of could conceivably be implemented at a lower cost the week respondents are asked to recall their activities. than previous time use question sequences. Days with a pattern of activities dissimilar from the one The activity log, while somewhat more expensive, on which the interview takes place are easier to remem- can provide for a limited number of activities and ber. Thus, in Western cultures with a Monday through attendant dimensions of all of the measures provided Friday workweek, respondents may be able to remem- by a time diary. This provides highly significant scope ber Saturdays activities fairly well on Tuesday and for analysis. The most data and analysis possibilities are Sunday's activities fairly xvell on Wednesday. However, provided by a diary, at significantly greater cost in time they may have trouble remembering Monday's activities and money. However, the diary provides nearly limit- on Thursday because there have been two other work less possibilities for analysis. days in between that had the same basic pattern of activ- ities.Juster (1985) found that there was little if any dete- Further Design Considerations rioration in the quality of data reported for Friday's A specialized time use module represents only one activities when they were reported on Saturday or component of the overall LSMS survey, but it is useful Sunday and only a 10 percent deterioration when they to list some design issues that pertain to single-purpose were reported on Monday to Thursday. Some of the surveys focusing on time use. Fortunately, most of recall periods used in the surveys that were exanmined for these design issues can easily be accommodated with- this study were as long as four to seven days. in the format of an LSMS survey. Klevmarken (reported in Keller and others 1982) studied 24 diaries, each of which covered 48 hours, SAMPLING. The sample design recommended for and found that some activities were underreported for LSMS surveys that include a nationally representative "the day before yesterday." However, he found the random sample of households is appropriate for the major drawback of the two-day diary was the fact that time use module. Most recent national time use sur- it took so long to complete-an average of 63 min- veys have chosen to have samples of between 3,500 utes. On the other hand, xvork by the Swedish Central 263 ANDREW S. HARVEY AND MARIA ELENA TAYLOR Bureau of Statistics (also in Keller and others 1982), groups of the sample (as there are fewer individuals in which took the average length of time reported on each group) and is likely to increase the number of various activities as a test of data quality, found that diaries that are only partially completed because of there was little difference between "yesterday" and increased respondent fatigue (Survey Research Center "day before yesterday" diaries. 1984).Thus, when collecting data by recall, it is virtu- ally impossible for survey designers to stipulate more NUMBER OF DAYS. Deciding how many days of time than a single day or at most two days unless the inter- use data to gather per respondent poses a dilemma viewer and the respondent meet several times. similar to the dilemma involved in gathering con- Given this fact, the diaries should refer to the day sumption data (see Chapter 5), but the dilemma is previous to the interview or to the day following the perhaps even more acute for time use. The fact that interview. The following-day diary is more flexible in people's memories of events start to fade after a day or terms of the number and timing of days because the two suggests that it would be wise to choose a short diary is left with the respondent for him or her to fill recall period. A recall period of a day or two is likely out on an agreed day. Thus the reporting day does not to capture activities that take place every day (such as have to be so tightly linked with the interviewer's cooking, fetching water, and feeding livestock) reason- schedule. The following-day diary, however, requires ably well, although some day-to-day variation is likely that the population be literate or that the interviewer even in these routine tasks. However, using a recall have some carefully designed illustrations that give period of a single day is less likely to capture activities respondents a clear visual explanation of how to fill that are frequent but not necessarily carried out daily, out the diary. If the diaries are to be left behind after such as washing clothes, shopping, working on house- the interview or dropped off before the interview, the hold enterprise activities, or (for children) doing logistics of this must be integrated into the fieldwork homework. Also, a one-day recall period is highly plan and the budget. It is highly recommended that unlikely to capture activities that are infrequent or respondents be asked to keep the time diary for at least highly seasonal, such as caring for the ill, working in two days. In an Australian study, 48-hour time diaries many types of agricultural or household enterprise were used (Australian Bureau of Statistics 1988) activities, and participating in festivals or ceremonies. because a pilot test found that the accuracy and detail If the fieldwork is spread appropriately over the of respondents' reporting increased on the second day, days of the week and year, gathering a single day's as they became more experienced at filling in details. information will give a reliable picture of society's Also, the 48-hour diary provided twice the number of time use and even the time use of some groups with- diary days as did a 24-hour diary, at only a very small in it. This will be sufficient for some analytical pur- increase in cost. Naturally, if the information xvere col- poses, such as looking at patterns of time use over lected in an interview rather than by a self-adminis- many years to examine social change or comparing tered diary, the increased cost of adding an extra day the time use patterns of urban workers in the formal would be much greater. sector with those of rural farmers. However, having The survey designer must thus make a difficult only one or even two days worth of data on house- choice among three options: to gather data on a day or holds' time use does not enable analysts to acquire a two of recall data and accept the risk of measurement very firm grasp of individuals' or households' time error; to gather data over a longer period even though budgets and the tradeoffs made within them, and such it is known that respondents' recall can be inaccurate a grasp is required for analyzing many of the issues dis- over such periods; or to administer the time use mod- cussed in the first section of this chapter. ule to the household several times on a series of visits One of the basic decisions that needs to be made and thus assemble a long period of data from multiple when designing the sample is about the tradeoff short recall periods. The third option of repeated vis- between number of respondents and number of recall its to the household is theoretically feasible but obvi- days per respondent (Survey Research Center 1984). ously more expensive. Increasing the number of recall days per respondent reduces important measurement errors, but it also DATA COLLECTION DAY. The data collection day is reduces the usefulness of the data for analyzing sub- usually randomly assigned to the household, so that 264 CHAPTER 22 TIME USE about one-seventh of the sample is interviewed each all days of the week. However, it is not necessary for day of the week (or two-sevenths are interviewed if the diary to run from midnight to midnight. Recent there is a two-day recall period). This is called the Canadian studies used the period 4 a.m. to 4 a.m., as "designated day" and is necessary because if the inter- previous work had suggested that 4 a.m. was the point viewers visited households simply according to their of minimum activity (and thus maximum sleep). A own convenience, the resulting data would probably recent Dominican Republic study ran from 7 a.m. to be biased. For example, if the interviewer were to take 7 a.m. a day off, the preceding day would be underrepresent- ed in the data. Moreover, if it is systematically easier to INTERVIEW MODES. Each of the time diaries can be find people at home on some days than on others (for administered by an interviewer or filled out by example, on weekends in areas with a lot of formal respondents for their activities either the day before or sector employment), the groups for which data are the day after the interview. Researchers at the collected on those days will be over-represented. University of Michigan compared data collected by a Experience suggests that while it is important to variety of methods Uuster and others 1983). They have a random distribution of reference days through- found that the quality of the data may be better if the out the sample, it is not critically important to ensure diary is left behind after the interview and filled in for that the plan be followed with absolute precision. In the following day because the respondents do not other words, in the inevitable instances when the need to rely on their memories. However, having the interviewer cannot keep to the schedule or finds that respondents fill in the diary generally requires that the respondent is not at home on the day planned, it they are literate or are given some training with illus- is better to change the reference day to the day pre- trated diaries and instructions. Also, Juster found that ceding the new interview date than to ask the respon- some respondents altered their normal activities if they dent to try to remember his or her activities a few days wrote each activity down. For these reasons, in most before or demand to make an extra visit on the same LSMS surveys it is better to have the diary informa- day the following week. A Swedish pilot study used a tion collected in an interview rather than to leave the designated day but allowed the interview to be post- diary to be filled in by the respondents themselves. poned to the same day in a later week (Lyberg 1989). The analysis suggested that there were no substantial OPEN OR FIXED INTERVAL DLARY. It is recommended differences in the duration of the activities reported in that survey designers allow respondents to record their the "on time" and the "delayed" diaries.The prime cri- activities as these activities change (an open interval teria for selecting the reference day are how the data diary) rather than expecting respondents to record will be used, the administrative burden, and the cost.A what they are doing every, say, 15 or 30 minutes (a study by Kinsley and O'Donnell (1983) found that the maximum fixed interval diary). Fifteen-minute blocks designated day approach was more demanding of the have been used in the most recent surveys in the research team compared to the convenient day Netherlands, Norway, Switzerland, and the United approach and, in the long run, yielded equally accurate Kingdom; a I 0-minute block was used in Finland's lat- data. est survey. In Canada and the United States, the open format has been used. Experts are divided about which TIME OF YEAR. Ideally, a full year should be covered in method is preferable. However, recent work suggests the survey, with respondents and diary days randomly that the open interval method may be more appropri- selected to ensure that an appropriate cross-section of ate for previous-day diaries and the fixed interval respondents is interviewed in each time period method may be more appropriate for leave-behind (Staikov 1982). If that is not possible, data should be diaries. It also appears that open interval diaries are collected that are representative of significantly differ- more efficient if the diary information is collected ent seasons and crops. from the respondent by the interviewer. TIMING OF THE REFERENCE DAY. To enable analysis of BUDGET. The budget required to collect valid and reli- daily and weekly cycles of activities, the data collected able time use data will depend on the desired level of must be representative of the full 24-hour cycle and of accuracy and detail. Founded on a few basic principles 265 ANDREW S. HARVEY AND MARIA ELENA TAYLOR of time use data collection, the activity list, the simplest from all individuals was put on a single grid rather form presented, is probably the cheapest of the mod- than on separate pages for each person. The advantage ule formats. The activity log will be more expensive of this approach was that the interviewers found it but will yield high-quality data with wide analytical more natural in the context of the rest of the ques- possibilities. If the designers of a future LSMS survey tionnaire. The disadvantage was that it made it more decide to include a full time-use module, they should awkward to calculate a sum across five pages of activ- be aware that this may substantially increase the inter- ities for each person to see that the time totaled 24 view time and complicate the survey's logistics. hours. The horizontal approach uses slightly fewer Administering full time-use diaries also significantly pages, with information for up to 12 household mem- increases the amount of data entry required and the bers contained on 5 pages rather than 12. In any case, complexity of data analysis.While the data gathered in the convenience to the interviewer and the probable time use diaries can support a substantial amount of reduction of errors in the data collection process are research, if all that analysts need to know is, say, the more important criteria by which to judge the design amount of time that people spend commuting to than the number of sheets of paper used. It should be work and if the analysts can tolerate some inaccuracy, noted that at the time of the pretest, the interviewers it may be easier to ask a direct question rather than use naturally had little training or familiarity with the complex time use diaries to extract that piece of infor- questionnaire. Survey designers interested in using an mation from respondents. activity list may wish to pretest both versions present- ed here to see which works best in their context. Draft Modules Also presented inVolume 3 are examples of a styl- ized activity log and an open-interval time diary. Only As discussed in the previous section, there are two the top part of a diary is shown because covering the main options for gathering time use data: specific full range of potential activities will usually require two questions scattered throughout the household ques- pages. For both the activity log and the open interval tionnaire and a separate time use module. Two varia- time diary a set of pages must be included in the ques- tions of the stylized activity list are presented in the tionnaire for each day for the maximum number of draft questionnaires in Volume 3. The first variation is household members expected. For example, if the log a generic example with some questions on periods of is to be filled out for one day and the rest of the ques- longer than a day. The second variation pares away tionnaire allows for up to 12 members in a household, those questions and reformats the remaining questions 12 of the log pages should be included. If this log is to into a layout similar to that used in other individual- be kept for two days, 24 log pages should be included. specific modules in LSMS surveys. The first format The log and diary shown here are generic instruments puts all activities for one individual on a single page. that should be modified to reflect the specific circum- Checking that the total amount of time reported adds stances in the country or region of the survey. up to 24 hours simply requires adding up one column of figures. The number of these sheets that should be Explanatory Notes included in the questionnaire is as many as is expect- ed to be the maximum household size-in LSMS sur- This section explains how to code the information veys, often 12 people. gathered in the draft modules presented in Part 3. Another alternative would be to use an activity list Information on primary activities, secondary activities, like the one used in the 1998 Nicaragua LSMS; such other people present during the activities, people for an activity list is also presented inVolume 3.When the whom the respondent carried out the activities, and first version of the Nicaragua activity list was pretest- location of activities can be precoded in the diary at ed, the interviewers found it awkward to switch from the survey design stage with some loss of detail. Such the grids in the rest of the questionnaire (in which the precoding can greatly speed up the collection, coding, information for each individual is listed horizontally) and entry of the data while preserving the integrity of to the vertically oriented activity list. So the survey the diary approach. planners flipped the matrix around: the activities Regardless of which coding system is used, there become columns rather than rows, and information is a consensus among time use experts that primary 266 CHAPrER 22 TIME USE precoding the primary activity greatly reduces the Box 22.1 CautionaryAdvice level of detail about the activity performed. For exam- • How much of the draft module is new and unproven? The ple, if an activity is coded as "child care," it could mean time use modules suggested have been used in several bathing and dressing a child or it could mean reading developing countries in single-purpose surveys. Such to a child. For some analytical purposes it may be modules have not, however, been used in the context important to distinguish reading from bathing and of LSMS surveys, with the exception of version 2 of the dressing. This distinction tends to get lost in the pre- activity list (which has been used only in a couple of coded term "child care." It is important that sufficient cases that have not yet been thoroughly evaluated). codes be available to distinguish activities of particular Many of the requirements for a good time use survey interest. It is also important that activities not precise- are usually already present in an LSMS survey-appro- ly defined can be clearly assigned to an appropriate priate sample size, well-trained interviewers, and so on. "other" category. Precoding has the same tradeoffs The main potential problem is that the module is quite here as elsewhere in surveying; it reduces costs but also long and will either displace others or increase total interview time significantly. Moreover, if the open inter- analytical possibilities. val time diary is used, a very large coding job will be What is needed is a hierarchical coding scheme that required after the interviewing, which is not usual in is sufficiently detailed to provide an unambiguous record LSMS surveys and will affect the logistics and speed of of activities. First, at least four categories of activity must data entry. be distinguishable: paid work; unpaid work; personal * How well hos the module worked in the past2 The pro- care; and free time. No "other" category should encom- posed modules appear to have worked well in the sin- pass activities that would far into more than one of the gle-purpose surveys from which they are drawn. pass actvie tha into more hancon ofst * What parts of the module most need to be customized? four. Second, the integrity of the coding hierarchy must The coding scheme will have to be carefully cus- be maintained in any precoded activity list. Regardless of tomized. If standard clock time is not a very salient cul- the collection instrument used, the data should be cod- tural concept in the population being surveyed, a spe- able into the four categories above and ideally into a cial conversion from local time terminology to the more extended set of at least 20-30 activities. Appendix 24-hour clock will need to be developed. 22.1 of this chapter includes two activity-coding schemes that can be used to guide the collection of activities must add up to 1,440 minutes per day. This LSMS data. The schemes reflect both the data needs significantly increases the accuracy and completeness specified in the Pakistan and Jamaica LSMS surveys and of reporting because it provides a check as to whether the general framework used in most national studies. the estimates of the duration of each activity were They were developed separately, with extensive interna- accurate and whether some activities were omitted. tional collaboration, by the Statistical Office of the European Communities (Eurostat) and by the United Activity Coding Nations Statistical Division for developing countries. To provide the most extensive data, respondents There is a wide variation in coding detail cross- should report on their participation in their primary nationally. The Japanese time use survey conducted by activities (in other words, what they consider to be the the Japanese Broadcasting Corporation (NHK) has the main things that they are doing at each time of day) in least-detailed coding system of all country surveys, their own words. The usual practice is to use a free- with only 32 activity codes. In contrast, the time use form diary that allows respondents to describe what surveys in Finland and Norway code 90 activities; the they were doing.This approach allows far greater flex- 1981 Canadian time use pilot survey coded 271 activ- ibility in recording and analyzing data than does pre- ities. If a diary is used, the level of detail in the coding coding. The flexibility does impose some burden on scheme makes relatively little difference to coding or analysts, since they must devise some kind of coding data capture yet provides considerable scope for analy- for themselves. Lingsom (1979) concluded that pre- sis. If precoded alternatives-an activity log or activity coding primary activities is not desirable from an ana- list-are used, the level of detail is constrained by the lytical viewpoint. instrument design and the interview situation. As noted The alternative to a free-form diary is a precoded above, the major cost choice is related to the choice of one. The major weakness of such precoding is that instrument rather than the instrument content. 267 ANDREW S. HARVEY AND MARIA ELENA TAYLOR 'With Whom" Coding Location Coding The open interval diary gathers information on who At a minimum, two categories of location should be was with the respondent while they were perfroming used: home and away from home. However, it is rec- the activity. The log can also be modified to gather ommended that survey designers include at least the such information or at least to indicate whether chil- following locations: inside the respondent's home; out- dren were present. Otherwise, data can be collected on side the respondent's home; at the workplace (away whether the respondent carried out the activity: alone; from home); at another person's home; elsewhere away with his or her spouse; with the children of the house- from home; traveling on foot; traveling by bicycle; hold; with other household members; with coworkers traveling by car; traveling by public transit; traveling in or schoolmates; with friends or relatives from outside another (or unknown) way; and at another (or the household; or with others outside the household. unknown) location. An alternative location coding in This information can be gathered to any degree of some cases may be: at home; in the community; and detail. outside the conmuunity. For clarity, a distinction must be made during Few existing data sets contain specific geographi- the collection or coding of the data between "being cal locations for activities, although in some surveys in the presence of" and "acting with." However, time conducted in industrialized cities, the addresses where use experts are still divided about whether it is the activities took place were collected. Undoubtedly, preferable to make this distinction during the collec- capturing precise geographical location data will be a tion or the coding of the data, and about how best to luxury in most cases and unsuitable in the context of ensure that the answers respondents give correspond multitopic surveys such as the LSMS. Nevertheless, with analysts' definitions of the terms. In the Finnish doing so is both feasible and analytically useful (Elliott, pilot time use survey it was found that only one-third Harvey, and Procos 1976). In LSMS surveys it may be of the respondents filled in the "with whom" infor- possible to capture a measure of the distance from the mation correctly in terms of being in the presence of home to the place where an activity took place in the others. One-third reported only the active involve- "where" or "comment" column of the most detailed ment of others, and other entries were ambiguous diary. (Niemi 1983). Therefore, it is important to give respondents explicit instructions about how to report Appendix 22.1 United Nations International this category. Classification for Time Use Activities "For Whom" Coding 1. Employment for establishments The "for whom" category of information has emerged 11 First job or employment as a legitimate and useful item of information for ana- 12 Second, third and other jobs lyzing voluntary activity and instrumental-material 13 Working in apprenticeship, internship and services that assist people, such as help with basic needs related positions like eating, washing, and going to see the doctor. In 14 Short breaks and interruptions from work Germany, Blanke and Schafer (1992) used four simple 15 Seeking employment and related activities categories to classify beneficiaries of respondents' 18 Travel to/from work and seeking employment activities: their household; another household; their in establishments household and another household at the same time; 19 Employment in establishments not elsewhere and voluntary organizations.The pilot time use survey classified in the Dominican Republic used more complex ben- 2. Primary production activities not for establishments eficiary categories consisting of: the respondent; the 21 Crop farming and market/kitchen gardening: respondent's employers; self-employed work in the planting, weeding, harvesting, picking, etc household; work on the respondent's home; another 22 Tending animals and fish farming household member; other social organizations or 23 Hunting, fishing, forestry and gathering of wild communities; work on other people's homes; and products work for the respondent's own and other households. 24 Digging, stone cutting, splitting and carving This categorization appeared to work well. 25 Collecting water 268 CHAPTER 22 TIME USE 26 Purchase of goods used for and sale of outputs 47 Pet care arising from these activities 48 Travel related to household maintenance, man- 28 Travel related to primary production activities agement and shopping (not for establishments) 49 Household maintenance, management and 29 Primary production activities (not for estab- shopping not elsewhere classified lishments) not elsewhere classified 5. Care for children, the sick, elderly and disabled for 3. Services for income and other production of goods own household not for establishments* 51 Physical care of children: washing, dressing, * In each activity buying of inputs and selling the feeding products are included, and may be disaggregat- 52 Teaching, training and instruction of own children ed at the third digit level 53 Accompanying children to places: school, 31 Food processing and preservation activities: sports, lessons, etc grain processing, butchering, preserving, curing 54 Physical care of the sick, disabled, elderly 32 Preparing and selling food and beverage prepara- household members: washing, dressing, feed- tion, baking, confectionery and related activities ing, helping 33 Making and selling textile, leather and related 55 Accompanying adults to receive personal care craft: weaving, knitting, sewing, shoemaking, services: such as hairdresser's, therapy sessions, etc. tanning, products of wood 56 Supervising children and adults needing care 34 Building and extensions of dwelling: laying 58 Travel related to care of children, the sick, eld- bricks, plastering, thatch, bamboo, cutting glass, erly and disabled in the household plumbing, painting, carpentering, electric wiring 59 Care of children, the sick, elderly and disabled 35 Petty trading, street/door-to-door vending, in the household not elsewhere classified shoe-cleaning and other 6. Community services and help to other households 36 Fitting, installing, tool setting, maintaining and 61 Community organized construction and repairing tools and machinery repairs: buildings, roads, dams, wells, etc. 37 Provision of services for income such as com- 6' Community organized work: cooking for col- puter services, transport, hairdressing, cosmetic lective celebrations, etc. treatment, baby-sitting, massages, prostitution 63 Volunteering with for an organization (which 38 Travel related to services for income and other does not involve working directly for individuals) production of goods (not for establishments) 64 Volunteering with for an organization 39 Services for income and other production of (which does not involve working directly goods (not for establishments) not elsewhere for individuals) classified 65 Participation in meetings of local and informal 4. Household maintenance, management and shop- groups/cast, tribes, professional associations, ping for own household union, fraternal and political organizations 41 Cooking, making drinks, setting and serving 66 Involvement in civic and related responsibili- tables ties: voting, rallies, etc. 42 Cleaning and upkeep of dwelling and sur- 67 Informal help to other households roundings 68 Travel related to community services 43 Care of textiles: sorting, mending, washing, 69 Community services not elsexvhere classified ironing and ordering clothes and linen 7. Learning 44 Shopping for goods and non-personal services: 71 General education: school/university atten- capital goods, household appliances, equip- dance ment, food and various household supplies 72 Studies, homework and course review related 45 Household management: planning, supervis- to general education ing, paying bills, etc. 73 Additional study, non-formal education and 46 Do-it-yourself home improvements and main- courses during free time tenance, installation, servicing and repair of 74 Work-related training personal and household goods 78 Travel related to learning 269 ANDREW S. HARVEY AND MARIA ELENA TAYLOR 79 Learning not elsewhere classified 1. Employment 8. Social, cultural and recreational activities 11 Main job 81 Participating in cultural activities, weddings, 12 Second job(s) funerals, births, and other celebrations 13 Time connected with own employment 82 Participating in religious activities: church 2. Study services, religious ceremonies, practices, 21 School/university rehearsals, etc. 22 Free time study 83 Socializing at home and outside the home 3. Household and Family Care concerning own 84 Arts, making music, hobbies and related courses Household 85 Indoor and outdoor sports participation and 31 Food preparation related courses 32 Household upkeep 86 Games and other pass-time activities 33 Making and care FOR Textiles 87 Spectator to sports, exhibitions/museums, 34 Gardening and pet care cinema/theatre/concerts and other perform- 35 Construction and repairs ances and events 36 Shopping and services 88 Travel related to social, cultural and recreation- 37 Household management al activities 38 Child-care 89 Social, cultural and recreational activities not 39 Adult assistance and care elsewhere classified 4. VolunteerWork and Meetings 9. Mass media use 41 Organisational work 91 Reading 42 Informal help to other households 92 Watching television and video 43 Participatory activities 93 Listening to music/radio 5. Social life and Entertainment 94 Accessing information by computing 51 Socialising 95 Visiting library 52 Entertainment and culture 98 Travel related to mass media use and entertain- 53 Resting-time out ment 6. Sports participation 99 Mass media use and entertainment not else- 61 Physical exercise where classified 62 Productive exercise 0. Personal care and self-maintenance 63 Sports related activities 01 Sleep and related activities 7. Hobbies and Games 02 Eating and drinking 71 Arts 03 Personal hygiene and health 72 Technical hobbies 04 Receiving medical and personal care from pro- 73 Games fessionals and household members 74 Other hobbies 05 Doing nothing, rest and relaxation 8. Mass Media 06 Individual religious practices and meditation 81 Reading 08 Travel related to personal care and self- 82 TV andVIDEO maintenance 83 Radio/music 09 Personal care and self-maintenance not else- 9. Travel and unspecified time use where classified 90 Travel by purpose Appendix 22.2 EUROSTAT Survey on Time Notes Use Activity Coding List The authors xvould like to thank Lynn Brown, Margaret Grosh, and 0. Personal Care Raylvnn Oliver for suggestions. 01 Sleep 1. This was confirned in the 1981 Canadian Time-use Pilot 02 Eating Study, which found that more than one activity was occurring in 03 Other personal care 74 percent of the episodes reported. 270 CHAPTER 22 TIME USE 2. Appendix 22.1 contains a preliminary activity list relevant to Altman, Jeanne. 1971. "Space-Time Budgets and Activity Studies LSMS surveys. Appendix 22.2 contains a coding scheme developed in Urban Geography and Planning." Environment and Planning by the Statistical Office of the European Communitites (Eurostat) 3. for a number of planned cross-national comparable studies. Australia, Bureau of Statistics. 1988. "Information Paper, Time Use 3. LSMS survey designers try to avoid such ex post data cod- Pilot Survey." Catalogue 4111.1. Sydney ing as much as possible because this can be a significant source Berio,Ann-Jaquelne. 1980. "The Analysis of Time Allocation and of nonsampling error, it can add significantly to the costs of the Activity Patterns in Nutrition and Rural Development survey, and it can enormously extend the period between the Planning." Food and .Nutrition Bulletin 6 (1): 53-68. end of the fieldwork and the time when data become available Blanke, Karen, and Dieter Schafer. 1992. "What for Whom? for analysis. Another option is to have the interviewers code the Experience from the Diaries of the Pretest of the 1991/92 answers, either during the interview or before turning the ques- Time Budget Survey in Germany." Presented at the tionnaires in for data entry This requires extensive training of International Association for Time use Research, National the interviewers as well as a very clear coding scheme. In a sur- Statistic Institute of Italy June 15-18, Rome. vey in the Netherlands that used a written diary, the respondents Elliott, David H., Andrew S. Harvey, and Dimitri Procos. 1976. "An themselves coded the activities. Choosing this option requires Overview of the Halifax Time-Budget Study" Society and both a literate population and very clear instructions and cod- Leisure 3: 145-59. ing lists. Eurostat. 1996. "Pilot Survey on Time Use 1996: Coding List and 4. The 1998 Nicaraguan LSMS had the most extensive time use Coding Diary." (DOCE2/TU/PILOT/11+12/96). module, but results are not vet available.The 1993 Jamaica Survey of Ferge, Susan C. 1972. "Social Differentiation in Leisure Activity Living Conditions contained a time use module that incorporated a Choices: An Unfinished Experiment:" In Alexander Szalai, ed. more comprehensive range of activities and more questions relating The U )se of Time: Daily Activities of U rban and Suburban Pop- to unpaid work than most LSMS surveys, wvithout, however, cover- ulations in Tu'elve Countries. The Hague: Mouton. ing all possible activities. The 1991 Pakistan Integrated Household Harvey Andrew S. 1999. "Guidelines for Time Use Data survey had a series of questions on the nonmarket use of women's Collection and Analysis:' In Wendy Pentland, Andrew S. time. None of these surveys used the generic diaries referred to in Harvey M. Powell Lawton, and Mary Ann McColl, eds., Tinie this chapter; instead, all of them used a series of questions such as Use Researchl in the Social Sciences. New York: Klewer/Plenum "How much time did you spend yesterday on ... [activity] ... ?" with Publishers. the list of activities comprising a dozen or so items that included Harvey, Andrew S., and W Stephen Macdonald. 1976. "Time various market, nonmarket, domestic, and leisure activities. Diaries and Time Data for Extension of Economic Accounts:' 5. It is unclear how valid a measure such a question wvill pro- Social Indicators Reseachl 3 (1): 21-35. duce. Some respondents may respond w,vith a normal amount INSTRAW (United Nations International Research and Training throughout their school lives while others respond with an average Institute for the Advancement of Women). 1994. Thle Dou,in- for the past wveek even if the wveek wvas in some wvay abnormal. ican Republic Pilot Time Use Study Santo Domingo, Dominican 6. In a Canadian survey conducted by telephone, the time use Republic. diary took about 12 minutes of interview time. It asked only about . 1996. Valuation of Houselhold Production and thie Satel- primary activity, location, and who was present, and did not include lite 4ccounts. Santo Domingo, Dominican Republic. all the follow-up questions in the draft modules introduced by this James, William Philip Trehearne P. T., and E. C. Schofield. 1990. chapter. Canada probably also has a larger share of the population Human Energy Requiremzents: A Manual for Planners and in formal-sector jobs with continuous hours that, together with Nutritionists. Oxford: Oxford University Press. commuting, account for a good deal of a person's active day. Juster, E Thomas. 1985. "The Validity and Quality of Time use Moreover, Canada is a "time-conscious" society Both of these cir- Estimates Obtained from Recall Diaries." In E Thomas Juster cumstances would lead to average response times lower than might and F. P. Stafford, eds., Time, Goods, anid Well-Being. Ann Arbor, be expected in many developing countries. Mich.: Institute for Social Research, Survey Research Center, University of Michigan. References Juster, F Thomas, Martha S. Hill, Frank P Stafford, and Jaquelynne E. Parsons. 1983. "Study Description: 1975-81 Time Use Acharya, Meena. 1982. Tisne use Data and thie Living Standards Longitudinal Panel Study," Project 466066. University of M11easurement Stidy. Living Standards Measurement Study Michigan, Institute for Social Research, Survey Research Working Paper 18.Washington. D.C.:World Bank. Center, Ann Arbor, Mich. 271 ANDREW S. HARVEY AND MARIA ELENA TAYLOR Kalfs, Nelly. 1993. Hour by Hour: Effects oflMlodes ofData Collection in Niemi, Iiris. 1983. The 1979 Time Use Study Method. Helsinki: Time Use Research. Amsterdam: University of Amsterdam. Central Statistical Office of Finland. Keller, Janet, Dorothy Kempter, Susan GeoffTimmer, and Linda Rigbers, Anke. 1996. Personal correspondence. Karlsruhe, Young-DeMarco. 1982. Proceedinigs of the International Time use Germany. Workshop. Ann Arbor, Mich.: Institute for Social Research, Robinsonjohn P 1984. Free Time in Western Countries. University of University of Michigan. Maryland, Survey Research Center, College Park, Md. Kennedy, Eileen T, Deborah Rubin, and D. Alnwick. 1991. "A Scheuch, Erwin K. 1971. "The Time Budget Interview" In Comparison ofTime Allocation Methods and Implications for Alexander Szalai, ed., TlDe Use of Time: Daily Activities of Urban Child Nutrition." International Food Policy Research and Suburban Populations in Twelve Countries. The Hague: Institute, Washington, D.C. Mouton. Khan, M. E., Richard Anker, S. K. Ghosh Dastidar, and Bella C. Shaw, Susan M. 1985. "The Meaning of Leisure in Evervday Life." Patel. 1994 "Methodological Issues in Collecting Time Use Leisure Sciences 7 (1). Data for Female Labour Force." Die Indian Journal of Labour Skoufias, Emmanuel. 1994 "Market Wages, Family Composition Economics 35 (1): 55-72. and the Time Allocation of Children in Agricultural Kinsley, Brian L., and Terry O'Donnell. 1983. Marking Time: Households."Journal of Development Studies 30 (2): 335-60. Explorations in Time Use. Vol. 1. Otta-wa: Canada Department of Sk6rzyfiski, Zygmunt. 1972. "The Use of Free Time in Torun, Communications, Employment and Immigration Canada. Maribor, and Jackson." In Alexander Szalai, ed., nie Use of Klevmarken, N.Anders. 1982. "Household Market and Nonmarket Time: Daily Activities of Urban and Suburban Populations in Tvelve Activities (Hus): A Pilot Study." University of Goteborg, Countries. The Hague: Mouton. Department of Statistics, Goteborg, Sweden. Staikov, Zahari. 1982. "Problems of the Comparison of Time- Kumar, Shubh K., and David Hotchkiss. 1988. "Consequences of budgets." In Zahari Staikov, ed., It's About Time. Sophia, Deforestation for Women's Time Allocation, Agricultural Bulgaria: Institute of Sociology at the Bulgarian Academy of Production, and Nutrition in Hill Areas of Nepal." Sciences and Bulgarian Sociological Association. International Food Pohcy Research Institute, Washington, Survey Research Center. 1984. "Addendum to a proposal to the D. C. National Science Foundation for Funding" for "A 1985-86 Study Leones, Julie P. 1991. "Rural Household Data Collection in ofTime Allocation among American Households." University of Developing Countries: Designing Instruments and Methods for Michigan, Institute for Social Research, Ann Arbor, Mich. Collecting Time Allocation Data." Working Papers in United Nations Secretariat, Statistical Division. 1997. "Expert Agricultural Economics 91-16. Cornell University, Department Group Meeting on Trial International Classification for Time- of Agricultural Economics and Cornell Food and Nutrition Use Activities." October 13-16, NewYork. Policy Program, Ithaca, NewYork. Vickery, Clair. 1977. "The Time Poor: A New Look at Poverty." Lingsom, Susan. 1979. Advantages and Disadvantages ofAlternative Time Journal of Human Resources 13 (1): 27-48. Diaries:A Workitig Paper. Oslo: Central Bureau of Statistics. Wagenbuur, Harry TM. 1972. Labor and Development:An Analysis of Lyberg, Ingrid. 1989. "Sampling, Non-response, and Measurement the Time Budget and of the Production and Productivity of Lime Issues in the 1984/85 Swedish Time Budget Survey." Paper Farmers in Southern Ghana. The Hague, Netherlands: Institute of prepared for U.S. Bureau of the Census Fifth Annual Research Social Studies. Conference (ARC 4), March, Washington, D.C. Young, Michael, and Willmott, Peter. 1973. The Symmetrical Family: Martin Paul, and Patrick Bateson. 1986. Mfeasuring Behavior: An A Study of Work and Leisure in the London Region. London: Introductory Guide. Cambridge: Cambridge University Press. Routledge & Kegan Paul Ltd. 272 Part 4 pecial Topics 3q < Recommendations for Collecting Panel Data 2 *) Paul Glewwe and Hanan Jacoby When survey planners expect to implement a series of surveys, either every year (as in the core and rotating module design discussed in Chapter 3) or every three to five years, the question nat- urally arises whether to interview the same households in each survey and thus collect "panel" data. The answer to this question is not a simple choice between "yes" and "no" because there are many different ways to collect panel data.When planners are deciding whether to collect panel data, they must take into account the resources available and the main objectives of the survey. A typical LSMS survey has many objectives because it is a multitopic survey, so survey planners have to make compromises and rank research priorities.The details regarding which kinds of panel data are most useful for research on specific topics can be found in the individual chapters in this book. The aim of this chapter is to provide a broad overview of the advantages and disad- vantages of collecting panel data in LSMS-type surveys, then to offer some practical advice on when and how to collect such data. This chapter is organized into four sections. The first survey.The data collected using this approach are often section assesses the advantages of panel data over data referred to as repeated cross-sectional data. For con- from a series of surveys in which different households ducting research and analysis, panel data have several are interviewed in each survey.The second section dis- potential advantages over repeated cross-sectional data. cusses the disadvantages of collecting and using panel Whether all of these advantages can be realized is a data. The third section reviews the experience of col- matter of contention among economists and other lecting panel data in developing countries, focusing on data analysts, as will be discussed further below. This the issue of sample attrition. The fourth section con- section reviews all of the potential advantages of panel cludes with a set of general recommendations for col- data, beginning with those that are the clearest and lecting panel data. least controversial and concluding with those on which a consensus has not been reached. Advantages of Collecting Panel Data Estimating Changes in Variable Means More Precisely The most common alternative to collecting panel data Policymakers often want to know whether certain is the implementation of two or more surveys in characteristics of individuals or households have which different households are interviewed in each changed over time. For example, they may want to 275 PAUL GLEWWE AND HANAN JACOBY know whether the incidence of poverty or the rate of net changes in the population. However, repeated unemployment have changed. Such policy questions cross-sectional data reveal nothing about the move- also arise with respect to health and nutrition indica- ments of individuals or households over time, often tors, education outcomes, agricultural activities, referred to as gross changes, because different house- income levels, and many other topics, as discussed in holds and individuals are interviewed in each survey. more detail in other chapters of this book.These issues Gross changes can only be addressed using panel data. can be investigated whenever comparable household Before turning to the next topic, it is important to surveys have been implemented at two or more points point out that measurement of gross changes over time in time. can be biased by random measurement error in the When household survey data are used to investi- characteristic of interest. For example, random meas- gate these kinds of questions, it is necessary to verify urement error in household income or expenditures that the measured changes are statistically significant will exaggerate movements into and out of poverty and thus unlikely to be caused by chance alone. Two over time. This problem is potentially very serious, and surveys at different points in time may show that the is not easily solved without reliable information on the percentage of the population classified as poor has extent of measurement error in the data. increased by one percentage point, but this increase may not be statistically significant. In many cases esti- ProvidingAccurate Data on Past Events mates of changes in mean values of individual or A third advantage of panel data is that they record past household characteristics that use panel data are more events more accurately than do retrospective questions precise than estimates based on data from cross-sec- used in a single cross-sectional survey. Research has tional surveys, which increases the likelihood that a shown that survey respondents often do not remem- given measured change will be statistically significant. ber the timing of past events very precisely, and in As long as the characteristic in question is positively some cases past events are completely forgotten. If cur- correlated over time (for example, households that are rent information is gathered from the same respon- poor during one time period have a greater-than- dents at txvo or more points in time, the data from the average probability of being poor during the next earlier survey are likely to be more accurate than data period), the standard error associated with the estimat- collected in a single more recent survey that asks ed change in the mean value of that characteristic is respondents to recall past events. Analysts are often smaller for a given sample size when the change is cal- interested in the relationship betxveen a past and pres- culated using panel data.1 This is an important advan- ent phenomenon-for example, whether children tage that panel data have over repeated cross-sectional who began their schooling at a relatively late age sev- data. eral years ago are more likely to drop out of primary school today. To investigate this using only a single Estimating Changes at the Individuol Level cross-sectional survey, respondents must be asked to Policymakers are often interested in the persistence of recall how old their children were when they started certain characteristics of individuals or households primary school. Unfortunately, the respondents'mem- over time. For example, they often want to know what ories are very often inexact, which leads them to give proportion of the people found to be poor in one year erroneous answers to questions posed by interviewers. are still poor several years later-that is, xvhether These errors in the data will lead to inaccurate corre- poverty is a chronic or temporary condition. Similarly, lations between past and present events. In contrast, if policymakers may wish to know whether farm house- two surveys had been done that interviewed the same holds that have low productivity in a given year suffer households, the first around the time the children the same fate every vear or whether this phenomenon entered primary school and the second after the chil- tends to affect different farm households in different dren completed primary school, the data on the rela- years. tionship between age of entry and current enrollment Repeated cross-sectional surveys can be used to would be much more accurate. make comparisons over time across broad groups For some kinds of data households may not be (rural versus urban or skilled versus unskilled). These able to provide information about past events. For comparisons are often referred to as measurement of example, researchers may want to know how likely it 276 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA is that a child who was malnourished at, say, two years nomic outcomes and estimating "structural" models of of age will be enrolled in school at the time of the sur- behavior (including dynamic models). Estimating the vey. Nutritional status can often be measured in terms impact of government programs is discussed in this of weight or height (see Chapter 10), but it is virtual- subsection and estimating structural models of behav- ly impossible for household members to remember ior is discussed in the following subsection. In both precisely the heights or weights of their children in subsections the uses of panel data are illustrated with past years. In such cases panel data are essential to specific examples using data that are typically collect- allow researchers to relate past socioeconomic phe- ed in LSMS surveys. nomena to current household and individual charac- One important use of household survey data is to teristics. estimate the impact of government programs and services on households' living standards. In some cases Reducing Costs specialized surveys are designed to collect information A fourth advantage of collecting panel data is the that can be used to study the impact of a specific pro- reduced cost of such data relative to repeated cross- gram or project. LSMS surveys are not usually imple- sectional surveys.When the same households are inter- mented for such purposes; instead they typically have viewed in each survey, it is not necessary after the first many objectives and are nationwide in scope. Even so, survey to ask about characteristics of the household LSMS-type surveys can be used in certain circum- that cannot change, such as dates of birth, characteris- stances to evaluate government programs. In particu- tics of the parents of household members, or educa- lar, many government programs, including schools, tional attainment of adults. In contrast, if a new sam- health clinics, and agricultural extension services, ple of households is interviewed each time, these operate at the community level.The community ques- questions must be asked each time the survey is field- tionnaire in LSMS surveys (see Chapter 13) can be ed. Reducing the number of questions asked reduces used to collect information on the existence and char- the amount of time it takes to conduct the interview, acteristics of these government services. Combining which in turn reduces the costs of interviewing a community data with household data makes it possi- given number of households. ble to analyze the likely impact of improving or However, for LSMS-type surveys this advantage expanding these programs. may be relatively small. The great majority of ques- When evaluating programs it is important to dis- tions asked in LSMS and similar surveys concern tinguish between experimental and nonexperimental household characteristics that usually change between data. Experimental data are collected when a program surveys. For example, most answers to questions in the or service is provided to a randomly selected "treat- consumption module, which is a major part of any ment" group and withheld from a randomly selected LSMS questionnaire, would change from year to year. "control" group. The impact of the program is meas- Moreover, it is good practice to ask at least some ured by comparing the outcomes for both groups (see redundant questions to double-check the responses Newman, Rawlings, and Gertler 1994). In contrast, that were given in the initial survey. nonexperimental data are collected on "real world" variation in both program participation and the out- Estimating the Impact of Government Programs comes of interest; access to the program is not con- The advantages of panel data discussed so far are clear trolled in any way by the survey team that collects and and thus do not require detailed explanation. (For fur- analyzes the data. The impact of the program is ther discussion of these advantages see Duncan and inferred from econometric analyses of the relationship Kalton 1987 and Kalton and Citro 1993.) The remain- between access to the program and outcomes of inter- ing two advantages are more complex and contentious est. In this chapter, only nonexperimental data are since they involve the application of advanced econo- considered since the typical LSMS survey requires metric and statistical techniques to panel data. In the nationwide samples; it is usually feasible to collect authors' opinion, the two most important ways in experimental data only on a much smaller scale which panel data can be used to carry out policy-rel- because of the costs involved. (See Newman, evant econometric analysis are estimating the impact Rawlings, and Gertler 1994 for a recent discussion of of government programs and services on socioeco- analysis based on experimental data). 277 PAUL GLEWWE AND HANAN JACOBY The main analytical issue regarding nonexperi- about their children's health than do other households. mental data is how to estimate the impact of a given If this is the case, the communities that have the health program accurately-in other words, how to obtain an program will contain households that, on average, care unbiased (in the statistical sense) estimate of the true more about their children's health than households in impact. To make the discussion less abstract, consider communities that do not have the program. The assessing the impact of a newv child health program extent to which households care about child health is using cross-sectional data (for example, Thomas, Lavy, difficult to measure in household surveys. Thus this and Strauss 1996).2 A single LSMS-type survey usual- household characteristic will be both unobserved and ly provides data on height-for-age (an indicator of positively correlated with the presence of the health cumulative health status; see Chapter 10) for a random program, which implies that ordinary least squares sample of children, as well as data on local health pro- estimates will overestimate the program's impact on grams from a community questionnaire. The simplest child health. way to estimate the impact of a program on child How can panel data solve these two problems? health is by using ordinary least squares multiple Suppose two surveys of the same households (and regression analysis. This can be done by regressing children) are conducted in different years and that child height-for-age on a variable indicating the exis- between these years a child health program was imple- tence of the program and on other variables that could mented only in some communities. Assume further affect child health. that the unobserved factors that bias ordinary least Txvo problems arise with ordinary least squares squares estimation (such as unobserved community estimation based on cross-sectional data: nonrandom characteristics that affect child health) do not change program placement and selective migration.The prob- over time. Most importantly, assume that by the time lem of nonrandom program placement, which was of the second survey the new health program has first discussed in detail by Rosenzweig and Wolpin existed long enough to significantly affect child health. (1986), arises because governments do not assign For communities that benefit from the new health health services to communities at random. Instead, the program, the change in child health in response to the provision of health services to communities is deter- change in the availability of the program measures the mined by several factors, some of which are not impact that the program has on child health. If the observed by researchers and may be correlated with unobserved factors that biased the ordinary least child height. For example, if the government first pro- squares estimates do not change over time, they can- vides services to the communities with the sickliest not affect (and thus cannot cause biases in) estimates children, ordinary least squares regression results may based on changes. The mathematics behind this intu- show that services are associated with worse health itive argument is presented in Appendix 23.1. The outcomes.This is not really the effect of the program simplest version of this estimation method is called but instead comes about from the rules that guide the fixed effects estimation, since it is based on the placement of the program. In general, when such pro- assumption that the unobserved factors that can lead grams are targeted to worse-off areas, ordinary least to bias in ordinary least squares estimates are fixed-in squares regressions will underestimate the program other words, do not change-over time. effect. Similarly, if better-off areas are targeted (perhaps Before concluding that fixed effects estimates for political reasons), ordinary least squares results will based on panel data are superior to ordinary least overestimate program impact. squares (or any other) estimates based on cross-sec- The second problem is selective migration (see tional data, three points need to be made. First, the Rosenzweig and Wolpin 1988). Internal migration is assumptions that must be made for the fixed effects common in many developing countries, and individu- procedure to produce unbiased estimates of program als or households may base their decisions to migrate effects are rather stringent, and thus may be false. In on the amenities offered in the communities they are particular, the unobserved factors that can affect the considering as potential destinations (see Chapter 16 placement of the program may not be fixed over time. for a full discussion of migration). For example, house- Indeed, Heckman and Robb (1985) argue that panel holds that move to communities with government data are not a panacea for estimating the impact of health programs may do so because they care more programs because the statistical assumptions required 278 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA to justify fixed effects estimation may be no more sickly first child may choose to migrate to communi- plausible than those required to justify other alterna- ties that benefit from the health program. tives to ordinary least squares estimation that, unlike A final concern is panel attrition, which is dis- fixed effects estimation, require only cross-sectional cussed further in later sections of this chapter. If two data to deal with nonrandom program placement and surveys are fielded four or five years apart, in the sec- selective migration. (See Appendix 23.1 for a more ond survey it is possible-even likely-that a sizable detailed technical discussion of this point.) fraction of households interviewed in the first survey Second, fixed effects procedures do not necessari- cannot be found. Random attrition does not lead to ly require individual-level panel data. Nonrandom bias in the estimates of the impact of government pro- program placement involves unobserved community- grams, but panel attrition is rarely random. For exam- level factors, so two surveys covering the same com- ple, individuals who are more willing to take risks are munities but different households can be used. Pitt, more likely to migrate to find better employment Rosenzweig, and Gibbons (1993) used community- opportunities, and thus more likely not to be found in level panel data from Indonesia to study the effects of the second survey. Nonrandom panel attrition has the a government program on school attendance, fertility, same effect as selective migration; households leave the and child mortality. community, and thus the sample, in part because of Community-level data can also be used to deal government programs. Attrition will lead to bias if the with selective migration if recent migrants can be indi- households or individuals that remain in the sample vidually identified.3 As long as community-level aver- differ in unobserved ways from those that have left. If ages of household variables remain constant over time all the characteristics of individuals or households that for the nonmigrant population, the fixed effects proce- determine attrition are time-invariant, the fixed esti- dure will yield unbiased estimates.4 Unfortunately, this mator will be unbiased, for the same reason that it is procedure has a serious disadvantage, which is that to unbiased under selective migration. While this may obtain accurate estimates of the impact of a program it appear to be good news, such an assumption may be is generally necessary to collect data on a very large false. number of households-probably tens of thousands. LSMS surveys usually have considerably smaller Estimating Structural Models of Behavior samples. The other potential advantage of collecting panel data Another example of fixed effects estimation that that involves advanced econometric and statistical does not require individual-level panel data is estima- techniques is the estimation of structural models of tion based on retrospective data. LSMS and similar behavior-in other words, of relationships between surveys often contain retrospective information on variables that are not affected by changes in policy.5 labor force participation, migration, and fertility. Even Estimates from such models can be used to evaluate height-for-age data contain information on past child not just the impact of one specific program but also of nutrition and health. Suppose a survey collected a variety of policies, including ones that have not yet anthropometric data on all children in each house- been implemented, by simulating households' behav- hold, along with community-level information on ioral responses to these policies. For example, a labor when a new health program was set up. It would be supply elasticity (the percentage change in labor sup- possible to estimate the program's impact by subtract- ply given a percentage change in wage rate) can be ing the average difference in the height-for-age of sib- used to simulate the impact of various tax policies and ling pairs in households with one child born before welfare programs. Similarly, the parameters of an agri- the program was introduced and another born after cultural production function (or input demands) can the program began from the average difference in be used to assess how price subsidies or other policies height-for-age of sibling pairs in households where that affect prices influence households' agricultural the program was never introduced or already existed production decisions. Another example of a structural before the first child was born (see Appendix 23.1 for model of behavior is a "dynamic decision rule," which a formal presentation). However, a problem with this is the relationship between an individual's current "within-household" estimator is that it does not cor- decision (say, to enroll his or her child in school or to rect for selective migration; parents with an unusually participate in the labor force) and the current value of 279 PAUL GLEWWE AND HANAN JACOBY a set of "state" variables (such as the household's cur- labor and capital markets are active and that their rent assets or the health status of individual household prices vary enough to allow for precise estimation.Yet members). even this strategy may be flawved, because community- A key advantage of panel data in this context is level prices may be correlated with unobserved com- that they control for unobserved variation ("hetero- munity infrastructure, which in turn may affect pro- geneity") among individuals or households that could ductivity directly. For example, input prices may be lead to biased estimates of structural parameters. This higher in villages with no paved roads, and a lack of idea goes back to the literature on "management bias" paved roads may directly lower agricultural output, in estimates of agricultural production functions. A holding constant labor and capital inputs. farmer's management ability cannot be observed by an An alternative to using instrumental variables analyst, but it can affect (and thus be correlated with) would be to use household panel data over two or the farmer's output and input choices. Hence, an ordi- more agricultural cycles. Unobserved farmer manage- nary least squares regression of farm output on agri- ment ability could be eliminated by regressing changes cultural inputs may produce biased estimates of the in output over time on changes in observed inputs. technological relationship between inputs and out- The intuition behind this is that because farmer man- puts. For example, farmers who are good managers agement ability is assumed not to change over time, may also be more likely than other farmers to use this variable does not appear in the regression based on modern hybrid seeds (as opposed to traditional seed changes in variables. However, other problems may varieties), in which case part of the observed positive arise. Bad weather (such as floods or drought) and correlation between the use of hybrid seeds and out- other unforeseen shocks (such as crop-eating pests) put xvould be due to unobserved management ability. could directly affect output and also affect inputs that With farm-level panel data, Mundlak's (1961) fixed are used after planting time, including labor and appli- effects estimator removes this management bias.Yet, as cations of fertilizer. It is hard to collect data on such emphasized in the previous subsection, fixed effects unforeseen shocks, and most LSMS surveys do not do estimation requires the strong assumption that the so. (See Chapter 1 9 on the agricultural module.) unobserved factors that affect household behavior Harvest labor is particularly likely to be correlated (more specifically, that are correlated with the regres- with almost all current-year shocks, since there is less sors in the equation of interest) are fixed over time. crop to harvest when output is low. Because these The remainder of this subsection explains why this unobserved shocks affect both inputs and outputs, they assumption is unlikely to be true and suggests what to will lead to biased estimates even in regressions based do about it. on changes, since the error term in those regressions To illustrate the general issues involved in struc- will be correlated with some of the regressors. tural estimation, consider the estimation of a (linear) The situation with capital is also problematic. agricultural production function, which is feasible Investment may be low in the year immediately fol- using data from the standard or expanded version of lowing a poor harvest.This reduces the capital stock in the agricultural module presented in Chapter 19.6 the next year, inducing correlation between changes in Assume that a farmer's management ability cannot be capital stock and changes in unobserved shocks. Thus observed.There may also be other unobserved factors, fixed effect estimates would, in general, be biased. One such as land quality and community infrastructure xvay to deal with this problem is to apply instrumental variables. As explained above, ordinary least squares variable methods to panel data; see Appendix 23.1 for estimation of such an equation using data from a sin- some examples. gle cross-sectional survey of agricultural households An additional problem can arise when the data on may lead to biased estimates of the impact of the observed farm inputs are inaccurate.This is known as observed variables. One solution to this problem measurement error. Measurement error can bias ordi- would be to use instrumental variables to predict nary least squares estimates. If measurement error is observed variables.7 This would require only cross-sec- random and uncorrelated over time, fixed effects esti- tiona] data.Wages and the rental price of capital in the mation will exacerbate this bias (see Griliches and community may be good instrumental variables (pre- Hausman 1986). The standard way to remove meas- dictors) for labor and capital inputs, assuming that urement error bias is to apply instrumental variable 280 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA techniques.8 An alternative approach using panel data for example, a person's current employment depends was suggested by Griliches and Hausman (1986), but on his or her past labor market experience. It is hard to this approach requires at least three rounds of data. see how anything other than panel data could be used Are household-level panel data necessary to to estimate such structural relationships. Consider the obtain unbiased estimates of the agricultural produc- child height example; using repeated cross-sectional tion function? A retrospective survey covering multi- data on children at different ages to follow cohorts of ple agricultural cycles would almost certainly not be children over time would not yield enough observa- feasible because there are simply too many variables tions for a practical estimator, and aggregating at the for households to remember. What about repeated community level to exploit cross-sectional variation in cross-sectional surveys with different households in height would work only if there were significant dif- the same communities (that is, a community-level ferences in mean height across communities, which is panel)? Unbiased estimates based on community often not the case. Moreover, collecting retrospective means can be obtained if there are a large number of data for several different time periods on income, communities and if one assumes that all communities expenditures, and even labor force status is likely to use the same agricultural technology. However, this yield very unreliable data. estimator is unlikely to be very precise, since it uses On the other hand, estimation of dynamic struc- only intercommunity variation in agricultural output tural relationships using panel data is not straightfor- and inputs. ward. First, there are difficulties in finding credible In other applications, repeated cross-section data instrumental variables and in correctly specifying the have proven to be quite useful for estimating structur- unobserved heterogeneity. (For a more detailed discus- al relationships. One application is research on con- sion see Appendix 23.1.) Second, an important prob- sumption and labor supply behavior over the life lem in the analysis of income dynamics is that it is cycle. In this case, by averaging, say, hours worked and generally impossible to distinguish between transitory wages within an age-cohort, it is possible to follow a income shocks and measurement error in the income given cohort of individuals or households over time data, even with panel data.The only way to distinguish using data from repeated cross-sectional surveys (see between these two phenomena is by using ancillary Browning, Deaton, and Irish 1985). Such panel data information to calculate the variance of the measure- sets constructed from a series of representative cross- ment error.'0 Even this information is insufficient if sectional surveys have the important advantage of measurement error is serially correlated, in which case eliminating nonrandom attrition, which often plagues it is not possible to distinguish the serial correlation in panel data. Measurement error bias may also be the transitory component of income from the serial reduced by using average values of variables for each correlation in the measurement error (see Ashenfelter, cohort, since random measurement errors tend to can- Deaton, and Solon 1986). On a more optimistic note, cel each other out when averages of mismeasured the relative impact of measurement error declines as variables are calculated. the length of time being examined increases, and The last topic on using panel data to estimate measurement errors are sometimes less of a problem structural relationships concerns dynamics. Sometimes, when examining transitions between discrete events, current production depends on past output. For exam- such as movements into and out of unemployment or ple, a child's current height or weight is likely to welfare participation. A third problem is panel attri- depend directly on the child's height or weight in a tion. Individuals who leave the sample are unlikely to previous time period, and a student's level of literacy in be random with respect to changes in their living stan- a given year probably depends directly on his or her dards over time, which means that the sample that level of literacy in the previous year (see Hanushek remains will yield misleading estimates; for example, 1986). More generally, policymakers are often interest- households that lose an important source of income ed in the dynamic relationships concerning income, may move to a less expensive dwelling. The third sec- earnings, consumption, labor force status, welfare par- tion of this chapter discusses the extent of sample ticipation, and other socioeconomic phenomena.9 In attrition in panel data sets from developing countries, each case, current decisions made by people or house- and the fourth section provides recommendations on holds depend on the outcomes of their past decisions; how to reduce this attrition. 281 PAUL GLEWWE AND HANAN JACOBY Overall, these problems are potentially serious, but error, although repeated contemporaneous observa- recognizing them when the questionnaire is being tions can be used to minimize this difficulty. (See designed can reduce many of them. For example, as Appendix 23.1 for further details.) explained in Appendix 23.1, collecting data on assets owned by the household prior to the first agricultural Summary cycle (or perhaps retrospective data on inherited assets) Panel data have several advantages over series of cross- and on prices and wages from community and price sectional surveys. They allow policymakers and questionnaires provides useful instruments for estimat- researchers to estimate changes in variables' means ing agricultural production functions. Also, if a survey more precisely by reducing the standard errors of can collect repeated contemporaneous measures of those estimates. Only panel data can be used to exam- specific outcomes of interest, such as child height and ine changes at the household and individual levels over xweight or scores on academic achievement tests in dif- time, and thus reveal whether certain characteristics, ferent subjects, researchers can use this information to such as being poor, are transient or permanent. Panel correct for measurement error. For further advice on data can provide more accurate data on past events, how to collect panel data with an eye to minimizing and they may be able to reduce survey costs, although these problems, see the individual chapters in this such cost savings are probably minimal for LSMS-type book on each module. surveys. Finally, a more general set of difficulties impinges Panel data also allow researchers to use more on the use of fixed effects procedures for estimating sophisticated econometric and statistical techniques to both structural models of behavior and the impacts of analyze the impacts of government programs and to government programs. First, fixed effects estimates estimate structural relationships. While these tech- reduce statistical degrees of freedom. In a two-year niques are subject to serious criticism, and researchers panel roughly half of the degrees of freedom are lost disagree on the merits of panel data, access to such by differencing (estimating relationships using data on data gives analysts a wider variety of methods with changes in variables). Other things being equal, this which to estimate program effects and structural rela- means that standard errors on the estimated coeffi- tionships. Two points are particularly important cients will increase by a factor of approximately ,\2 . regarding estimation of program effects. First, panel On the other hand, longer panels have lower propor- data are most useful when the interval between sur- tionate losses in degrees of freedom, and hence lower veys is long enough for the program or policy in ques- increases in the standard errors in the estimated coef- tion to have had an impact; indeed, the interval must ficients; estimation with three years of data loses only be long enough for some policy variation to occur. In one-third of the degrees of freedom, and so on. several past LSMS surveys that collected panel data Second, one typically loses the ability to estimate the (described in the third section of this chapter) the time coefficients on time-invariant variables (such as the Z,, between surveys was only one year, greatly diminish- variables in equation 3 in Appendix 23. 1).A third lim- ing their usefulness for evaluating programs. itation is that simple transformations of the data to Second, community data are usually needed to remove fixed effects, such as differencing, are usually evaluate community-level programs. This typically possible only in linear models. In contrast, it is difficult requires administration of a community questionnaire, to account for fixed effects and discrete or censored which must contain sufficiently detailed information dependent variables (which usually require nonlinear to distinguish between programs that failed because specifications), although fixed effects are sonmetimes they were never really implemented and programs that tractable in exponential models (for example, logit were implemented and still failed. Regarding estima- models) and have recently become feasible for tobit- tion of structural models, panel data are most useful to type models (Honore 1992). At present, fixed effects correct for unobserved heterogeneity at the household are also incompatible with most semiparametric esti- or individual level. Simple fixed effects methods are mation techniques, which are increasingly being used the easiest way to use panel data to estimate structural in the development literature (for example, models, but problems can arise with these methods. Subramanian and Deaton 1996). Lastly, differencing Designing the questionnaires to collect certain addi- can greatly exacerbate the problem of measurement tional data can reduce some of these problems. 282 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA Disadvantages of Collecting Panel Data dwelling is picked) or a random sample of households (in which case it is possible that only one of the two While panel data have several advantages over data households would be selected). from a series of cross-sectional surveys, these advan- Sample designs become more complicated when tages must be weighed against any disadvantages.This the aim is to follow households or individuals over section discusses two potential disadvantages of col- time. A sample of dwellings is not the same as a sam- lecting panel data: higher costs and sample attrition. ple of households or of individuals, since households These two problems are related; attempts to reduce and individuals may move but dwellings generally do sample attrition will usually increase costs. Moreover, not.12 Thus the decision to collect panel data requires the extent of both problems is directly related to the a more specific decision about whether the sample sample design used. To put these issues in perspective, over time should be a sample of dwellings, of house- this section begins with a discussion of sample designs, holds, or of individuals. and then considers each of the problems in turn. In general, attempting to follow households over time leads to both practical and conceptual problems Sample Designs because households, unlike dwellings and individuals, Almost all household surveys collect information from split up and regroup in many different ways. Children a sample of households that represents a much larger grow up and move away, adult household members population of households. In almost every LSMS sur- leave to look for work, and new brides move into their vey the population of interest to analysts is the entire husbands' households. Consequently, there are always population of the country, so the households inter- cases in which it is unclear which households are the viewed must be representative of the current national successors of the original households. Simple rules can population. The method used to choose the sample of be applied, but they generally result in a sample grad- households is known as the sample design.An essential ually becoming unrepresentative over time. (See the objective of any sample design is to ensure that the discussion of the ARIS-REDS survey in India in the households selected are representative of the larger third section of this chapter.) More complicated rules population of households. can be applied (see, for example, McMillan and Most household surveys-and almost all LSMS Herriet 1985), but sooner or later problematic cases surveys-initially choose a sample of households based are encountered that defy resolution. Indeed, some on a sample of dwellings. The sampling procedure survey statisticians question whether a panel survey of adopted usually includes two stages, although some households is a well-defined concept (see Duncan and have three stages. In the first stage (or the second stage Hill 1985). In view of these problems, LSMS-type sur- if a three-stage sample is used), many small, well- veys should not attempt to collect panel data based on defined geographical areas, known as primary sam- a sample design that tries to follow households over pling units, are randomly chosen, after which a list is time. made all of the dwellings in each primary sampling Thus there are really only two practical options unit and of all the households living in those for collecting panel data: following dwellings over dwellings.11 In the last stage, a sample of dwellings (and time and following individuals over time. Following by extension of households) is randomly chosen from dwellings is the simplest option, because dwellings this list. In the simplest case, one household lives in almost never move. If a dwelling has been torn down each dwelling so a random sample of dwellings is or has otherwise disappeared, it can be dropped from equivalent to a random sample of households. Since the sample, since no households in the population are each individual is assumed to belong to one house- currently associated with it. A more serious issue is hold, this also constitutes a random sample of individ- that new dwellings are built over time, and the house- uals. (See Chapter 6 for discussion of cases in which holds that live in them are likely to be different from the household to which an individual belongs is the households that live in older dwellings. Because ambiguous.) If two or more households live in one LSMS-type surveys almost always require a sample dwelling, this can be easily handled after the listing representative of the current population at each point operation is complete by drawing either a random in time that the survey is done, new dwellings (and the sample of dwellings (taking both households if such a households that live in them) must be added to the 283 PAUL GLEWWE AND HANAN JACOBY sample. Specifically, when a survey team returns to an point in time (a goal of virtually all LSMS surveys), area where a survey was conducted in a previous year, new people must be added to the sample to represent the team should make an updated list of all the those who are new members of the population- dxvellings in the area. This list will include all new either newborns or immigrants. (Deceased individuals dxvellings (and the households that live in them); a and emigrants must also be dropped from the sample, sample drawn from the new dwellings can then be but this is automatic because people who die or emi- added to the sample of dwellings drawn in the previ- grate cannot be interviewed).'4 Devising rules for ous survey to ensure that the new survey is represen- adding new members of the population is rather comn- tative of the current population.'3 plex. For example, if a xwoman in the sample marries, Following a sample of people is a more difficult her new husband is unlikely to have been included in option, since some people may move great distances the original sample.When such couples have children, from their original dwvelling places. Moreover, when should all of their children be added to the sample or following people over time, there are two distinct con- perhaps only half of them? One common approach is cepts of a representative sample; one refers to the cur- to add both the spouse and the children to the sample; rent population at each point in time and the other over time this can significantly increase the number of refers to a particular population cohort that ages over individuals (and households) interviewed-thereby time. A survey that aims to represent the current pop- also significantly increasing costs. ulation at each point in time should not only attempt to follow the people who were interviewed in the Higher Costs of Collecting Panel Data previous survey or surveys but should also try to add Although collection of panel data may reduce costs nexv people to represent new members of the popula- because some questions need not be asked again in tion (specifically, new births and immigrants). One folloxv-up surveys, other aspects of collecting panel example of such a survey is the U.S. Panel Study of data may raise costs. This subsection focuses on the Income Dynamics, which regularly adds new births to costs of conducting a panel survey relative to con- the sample and has more recently attempted to add ducting a series of cross-sectional surveys. immigrants. A survey that aims to represent a particu- The previous subsection argued that there are lar population cohort over time simply follows the only two practical sample designs for collecting panel original population of interest as its members grow data: panels of dwellings and panels of individuals. older and is not concerned with new entrants into the Panels of dwellings are both less expensive and easier population. An example of this is the U.S. National to implement than panels of individuals, but panels of Longitudinal Survey ofYouth, which follows a cohort dxvellings have the disadvantage that people who move of individuals who were ages 14-22 in 1979. to another dwelling are lost from a sample even Sample designs that follow the original individu- though they are still members of the population. (This als over time and also add new individuals to the sam- is the problem of sample attrition, which is discussed ple to represent new members in the population, such in the next subsection.) In most cases the cost of as the Panel Study of Income Dynamics, are represen- implementing a panel survey that follows the same tative samples in both senses (although, when using dwellings over time is not much more than the cost of the data, analysts should include only the individuals implementing a series of cross-sectional surveys. belonging to the type of sample required for the There are two additional costs of returning to the analysis). same dwellings in subsequent years. First, a slightly The main problem with implementing a panel larger questionnaire is needed to ask a small number of survey that follows individuals over time is locating questions that help match the people who currently individuals who have moved. Not only is this poten- live in the dwelling with the people who lived in the tially very expensive, but also it is not usually possible dwelling at the time of the previous survey (Such to find everyone who has moved. Since people who additions to the questionnaire are discussed in detail in move or are otherwise difficult to find are likely to dif- the fourth section.) Second, some additional adminis- fer from the rest of the population, losing them would trative costs are incurred by storing the names and lead to an unrepresentative sample. In addition, if a addresses of households interviewed in previous sur- representative sample must be maintained at each veys and transferring these names to the new ques- 284 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA tionnaires. In money terms, such extra costs are likely previous sample of households. This means that these to constitute a very small percentage of the total costs households and their members must be added to the of implementing any given survey. In fact, as pointed sample, increasing the total number of households to out in the first section of this chapter, there may be be interviewed. other savings involved in implementing a panel survey, Unfortunately, it is difficult to quantify the costs savings that offset part or even all of these additional involved in following individuals who move because costs (although in LSMS-type surveys these savings are these costs depend on many specific details. Indeed, in likely to be small). the case of existing panel surveys that followed indi- Perhaps the most serious issue is not money costs viduals, it is often difficult to say what the costs would but the capacity of the organization implementing the have been if a set of repeated cross-sectional surveys survey to perform the additional administrative work had been performed instead.Yet it is possible to get an correctly. A statistical organization in a poor develop- idea of the magnitude of the problem by examining ing country with weak administrative capacity may past panel surveys to see how many people moved out find it difficult to carry out even a modest number of of their original dwellings over time; such an exami- additional administrative tasks. However, in many nation is performed in the third section of this chap- cases, hiring one or two external consultants should ter. In addition, it is important to realize that no ensure that these tasks are performed correctly; such amount of money would make it possible to track consultants can also provide some training to increase down all of the individuals who participated in the local capacity. original survey. Therefore, survey planners must weigh The above discussion did not include the cost of the benefits of each additional effort to find more of performing a lsting operation to find new dwellings the original individuals against the costs involved in in subsequent survey years as an additional cost of col- making that effort. lecting panel data. This is because listing operations One low-cost option would be to follow only should also be done each time a cross-sectional survey individuals who moved within the same village or is implemented.i5 Thus the number of listing opera- local community, but the disadvantage of this would tions is the same for a panel survey based on a sample be the loss of those people who moved outside of the of dwellings (that updates the sample to include new township or community-which leaves a sample that dwellings) as for a series of repeated cross-sectional is no longer representative of the population original- surveys based on a sample of dwellings, so no extra ly sampled.16 A very high-cost option would be to listing costs are incurred in the panel survey. search for everyone who has not died, emigrated or The fact that few additional costs are associated entered an institutional living arrangement, but even with collecting panel data on a sample of dwellings after making elaborate efforts and spending large sums changes the focus of the discussion.The relevant ques- of money, it is still not possible to avoid missing a siz- tion becomes: what additional costs are incurred if able number of individuals. (See the specific experi- panel data are collected on a sample of individuals as ences discussed in the third section of this chapter.) opposed to a sample of dwellings? The decision to fol- There are also intermediate options such as attempt- low all individuals interviewed in the initial survey ing to locate all individuals who moved less than, say, who are still part of the population being sampled (in 50 miles from their previous location. In any case, other words, to follow all individuals except those who planners of panel surveys of individuals cannot avoid have died, emigrated, or moved into an institutional the fact that subsequent surveys will have samples that living arrangement) can involve substantial costs, for are no longer representative of either the original pop- two reasons. First, it can be very expensive to locate ulation or the current population. This problem of and interview people who have moved, because some attrition is explained in more detail in the following of them may have moved hundreds or even thousands subsection. of miles away, and in general they will be scattered throughout the population as opposed to being locat- Sample Attrition Problems ed in a relatively small number of primary sampling As mentioned above, panel surveys that attempt to fol- units. Second, many of the people who have moved low the same individuals over time cannot avoid los- now belong to households that were not part of the ing some individuals in the original sample because of 285 PAUL GLEWWE AND HANAN JACOBY the difficulties in following people who move. be lost from the sample. Such attrition can be thought Moreover, panel surveys based on a dwelling sample of as one type of "mismanagement attrition)' a phe- design lose all of the individuals who move out of the nomenon that also affects cross-sectional surveys. original dwellings; while this does not lead to samples Mismanagement attrition has plagued many past sur- that are unrepresentative of the current population veys, but it can usually be avoided through careful (assuming that newly built dwellings are added to the administrative xvork. sample, as explained above), it does lead to samples Once the (surviving) dwellings that were sampled that, as time goes by, become less representative of the in the previous survey have been located, adding a individuals originally sampled who are still members sample of newly constructed dwellings will ensure that of the current population. The fundamental problem the sample is a random sample of all dwellings that in both cases is that the people who drop out of the currently exist and thus a random sample of the cur- sample do not do so randomly; they tend to possess rent population. A final problem is that once a previ- characteristics that are different from the characteris- ously surveved dwelling has been located, attrition can tics of people who remain in the sample. This implies occur if the household members who still live in the that the people who remain are an increasingly unrep- dwelling refuse to participate in the new survey. resentative sample of the population that was original- Refusal rates vary by country, but they are quite low ly surveyed. 17 in most of the developing countries where panel sur- In fact there are two different sample attrition veys have been done; refusal rates after a first success- problems: attrition that results in the remaining sample ful interview are usually only 1-2 percent (see the not being representative of the current population and third section of this chapter).While refusal rates for the attrition that results in the remaining sample not being initial interview tend to be higher,'8 these rates affect representative of the (surviving) members of the pop- panel and nonpanel surveys equally. ulation originally sampled. Repeated cross-sectional Overall, panel surveys based on a sample of surveys avoid the first type of nonrandom attrition, dwellings that incorporates newly built dwellings and the second type is irrelevant since by definition should be representative of the current population at these surveys do not attempt to follow individuals over each point in time.While people who move are often time (though they can be used to follow population lost from the sample (the second type of sample attri- cohorts over time, as explained in the first section of tion), this problem does not put panel surveys at a dis- this chapter). advantage compared to a series of repeated cross-sec- As with the problem of additional costs, it is use- tional surveys, since such dwelling-based panel surveys ful to begin the discussion of sample attrition with the do not follow anyone over time. case of a sample design based on dwellings. Panel sur- Now consider panel surveys based on samples of veys that follow dwellings over time usually have no individuals.These surveys can suffer from both types of serious problems of the first type of sample attrition (a attrition because it is virtually impossible to locate all sample that over time is increasingly unrepresentative of the individuals who were interviewed in the initial of the current population) as long as newly built survey. Turn first to the problem that over time the dwellings are added to the sample. That is, if it is pos- sample is increasingly unrepresentative of the current sible to locate the dwellings covered in the previous population. Ideally, a panel survey that follows a sam- survey (or, in the case of dwellings that no longer exist, ple of individuals will locate every person interviewed the places where they used to be) there should be no in the previous survey who has not died, emigrated, or problem of attrition of dwellings. A recent innovation moved into an institutional living arrangemenit. In this that can help locate dwellings is Global Positioning case the sample would continue to be representative of System technology, which can record the geographical the (surviving) members of the population originally position of dwellings within 100 feet. The fourth sec- sampled. In order to ensure that the sample also repre- tion of this chapter provides useful guidelines on how sents individuals added to the population during the to track down dwellings visited in previous surveys. Of time between the previous and current surveys, new- course, if the location of the dwelling was poorly borns and immigrants must be added to the sample. recorded in the initial survey (or if the information However, even if these new population members are was lost after the survey was completed), dwellings will adequately represented (and adding them to the sam- 286 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA ple is not a trivial task), the inability to find all the panel data in developing countries. In particular, it (surviving) people interviewed in the previous survey appears that implementing a panel survey based on a means that those who can be found are not a repre- sample of dwellings is relatively simple, entails few sentative sample of the "old" members of the current additional costs (relative to repeated cross-sectional population. surveys), and can ensure that the sample is representa- The extent to which panel surveys based on sam- tive at each point in time as long as newly built ples of individuals suffer from these attrition problems dwellings are added to the sample. When such a depends on the mobility of the population and the dwelling-based sample is used, panel surveys will also amount of effort used to locate individuals who move. yield data on a large sample of individuals because Mobility varies widely; mobility was relatively low in most individuals in developing countries do not move a well-known survey in rural India implemented by over periods of five or even ten years.The existence of the International Crops Research Institute for the such panel data presents analysts with the opportunity Semi-Arid Tropics (ICRISAT), while mobility was to exploit the advantages of panel data discussed in the quite high in the 1990 LSMS survey in urban Peru. first section, although the advantages that are based on (These surveys are discussed in the third section of this advanced econometric techniques are still a matter of chapter.) A recent panel survey in Malaysia (the debate. Malaysian Family Life Survey, also discussed in the The discussion of the previous paragraph suggests third section) shows what can be done if substantial that whenever a series of LSMS-type surveys is imple- resources are available to follow migrants. Of a sample mented over time, panel data should be collected of women who were first interviewed in 1976,72 per- based on a sampling of dwellings that incorporates cent were found and reinterviewed 12 years later.Yet newly built dwellings. The reasoning behind this rec- even this rate of attrition is high, since the 72 percent ommendation is very simple: there is little cost to who were found are unlikely to be a representative doing so, and there may be substantial benefits. If this sample of the original women. Finally, a very recent advice is followed, survey designers should be aware survey in Indonesia (the Indonesian Family Life that there are two ways to incorporate new dwellings. Survey, also discussed in the third section) also made Many, if not most, new dwellings are built in areas great efforts to follow individuals and households that where a large number of older dwellings already exist- moved. About 91 percent of the individuals inter- ed. These can be added to the sample of dwellings viewed in 1993 were located and reinterviewed in from each area by conducting a new listing operation 1997. in that area, as explained above. Dwellings constructed In general, while mobility varies across countries in areas where no dwellings have existed before, such and some surveys have been more successful than oth- as on the outskirts of existing cities, may necessitate ers in following individuals over time, almost all panel the addition of new primary sampling units from these surveys that are based on a sample of individuals will areas to the sample. Whether and how to add such gradually become unrepresentative of the current sampling units will depend on the details of the sam- population.This is a serious problem for LSMS surveys ple design of the original survey. A sampling expert because governments almost always want these surveys should be consulted to provide specific advice on this to be representative of the current population each matter. time they are implemented. One advantage that panel Collecting panel data based on a dwelling sample surveys based on samples of individuals have over design has several advantages over using a series of panel surveys based on samples of dwellings is that repeated cross-sectional surveys; however, many types they more effectively follow individuals over time. of analysis would be better served if the panel survey followed some or all of the individuals who moved out A Suggested Approach to Dealing with Cost and Attrition of their original dwellings between surveys.19 This can Problems be done by adding the dwellings in which these indi- While the fourth section provides detailed advice on viduals now live to the original sample of dwellings. when and how to collect panel data as part of a house- When this is done the survey will be representative in hold survey, the discussion in the two previous subsec- two different senses, which correspond to the two dif- tions already suggests a general approach to collecting ferent (but overlapping) data sets contained in the 287 PAUL GLEWWE AND HANAN JACOBY overall survey. First, by retaining the dwellings that viduals after they have been in the sample for a certain were in the original sample and adding new ones to number of years.This leads to the issue of sample rota- represent newly constructed dwellings, the survey will tion, which will be discussed further in the fourth sec- still be representative of the current population, as dis- tion of this chapter. cussed above. (Of course, the data set to use for any analysis that is to be representative of the current pop- Experiences Collecting Panel Data in ulation must omit dwellings added for the purpose of Developing Countries following people who moved out of their original dwellings.) Second, the effort to follow individuals There have been many attempts to collect panel data who moved will produce data that can be used to in developing countries, especially during the past 20 study a cohort of individuals over time. Studies that years. This section reviews these attempts, with partic- follow such a cohort over time should exclude from ular emphasis on experiences with sample attrition the data all individuals not interviewed in the original and tracking households that move. Some reference is survey. also made to the experiences that developed coun- The question then becomes: How much effort tries, particularly the United States, have had in col- and resources should be devoted to locating and inter- lecting panel data. Information on the surveys from viewing individuals who move? The answer to this developing countries discussed in this section is sum- question can vary from "none" to "as much as is nec- marized in Table 23.1. essary to find and interview all household members who have moved." Ultimately, the choice made along Surveys Based on Samples of Dwellings or Households this spectrum will depend on the objectives of the sur- Most panel surveys in developing countries, including vey, the degree of mobility in the population, and the several LSMS surveys, have used a sample design that resources available, so it is not possible to give general follows dwellings or households over time. In many of recommendations on the extent to which movers these surveys, sizable sample attrition has occurred; should be followed. However, the experiences of many individuals interviewed in the initial survey no developing countries in collecting panel data, which longer lived in the dwelling at the time of the second are presented in the next section, give an idea of what survey. This subsection reviews the experience of both is feasible. LSMS and other household surveys, beginning with There are several ways to reduce costs when LSMS surveys. tracking individuals who have moved. One option is One of the earliest attempts to collect panel data to follow only individuals who have moved within in an LSMS survey wvas the 1990 LSMS survey in their original village or local community. While this Lima, Peru, in which survey teams attempted to revis- will not eliminate attrition in a sample of individuals, it households in 1,200 dwellings that were originally it will reduce such attrition, and the costs of locating surveved in 1985-86. (The 1985-86 sample was and interviewing nearby individuals are much lower nationwide while the 1990 follow-up survey covered than the costs of locating people who have moved only Lima, the capital city.) The sample design was hundreds or even thousands of kilometers away. A sec- based on a panel of dwellings, so no attempt was made ond option is to follow all individuals who moved but to follow individuals who had moved out of their to do this only for a random subset of the sampling original dwellings. Interviews were completed in areas (primary sampling units) or a random subset of about 82 percent of the original dwellings, with most the households in each area.A third option is to com- of the missing dwellings due to refusals (8 percent) or bine these two approaches so that the survey team to dwellings being vacant or no longer existent (6 per- attempts to follow most or all of the people who have cent).2' This refusal rate is very high for a developing moved short distances but a smaller proportion of country and undoubtedly reflects the social turmoil those who have moved longer distances. (When draw- (terrorist insurgency and an extremely sharp decline in ing random subsamples, survey designers must calcu- GDP per capita) that prevailed in Peru during that late sampling weights that reflect the relative probabil- time; it also may reflect that this was an urban popula- ity of being retained in the sample.) A final xvay to tion, since urban populations are typically both more economize on resources is to stop following all indi- mobile and more likely to refuse to participate. In 288 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA Table 23.1 Panel Data Collection in Developing Countries Time Sample between Years of size surveys/ Sample Country Survey survey (households) rounds design Attrition Comment Peru LSMS 1985, 1990 1,200 5 years Dwelling 43 percent of households Urban areas only, time of after 5 years social unrest Cote LSMS 1985-88 800 I year Dwelling 13 percent of households d'lvoire after I year Ghana LSMS 1987/88, 1,600 I year Dwelling About 50 percent of Did not explicitly ask 1988/89 households after I year about previous household members in follow-up survey ................................................................................................................................................................................................................................... Jamaica LSMS 1988- 2.000 I year Dwelling About 50 percent of Did not explicitly ask present households after I year about previous household members in follow-up survey Tanzania L5M5 1i99 i -94 800 6 months Household i O percent of househoids Mostly in rurai areas, one (Kagera but also after 2 years region only Region) followed individuals Vietnam LSMS 1992 93 i4,800 5 years Dweliing, About 9 percent of households Relatively low mobility 1997-98 but followed and 24 percent of individuals due to national households after 5 years restrictions on migration that moved short distances India ARiS i 968/69- 5, i 5 I year Househod 9 percent of hous ehods Rurai are as only 1970/71 after 2 years India REDS 1970/71, 4,756 11 years Household 34 percent of households Rural areas only 1981/82 after II years India ICRISAT 1975-84 120 Multiple Household i 3 percent of households Rural areas only, interviews after 10 years interviewers living in per year villages, incentives provided to households to remain in the survey Russia RLMS 1992- 6,334 6 mon hs Household 4 i percent of househoids present after 15 months Poland HBS 1993, 1996 8,000 3 years Household 37 percent of housenolds after 3 years Malaysia MFLS 1976, i;262 12 years Individual 28 percent of women 1988/89 after 12 years Philippines CLHNS 1983- 3,085 Varies Individual 9 percent of children after One region only present (see text) 12 years Indonesia IFLS 1993, 7,200 4 years Household, 6 percent of households 13 provinces (out of 27) 1997, 1998 (up to 1997), but also and 9 percent of individuals 5 years followed after 4 years (up to 1998) individuals Note: ARIS is Additional Rural Incomes Survey REDS is Rural Economic and Demographic Survey ICRISAT s International Crops Research Institute for the Semi-Arid Tropics. RLMS is Russian Longitudinal Monitoring Survey HBS is Household Budget Survey MFLS iS Malaysian Family Life Survey CLHNS is Cebu Longitudinal Health and Nutrition Survey IFLS s Indonesan Family Life Survey Source. Authors' summary about 31 percent of the dwellings where follow-up found and reinterviewed is almost certainly lower than interviews took place, none of the individuals who 57 percent, since household members in many of the had lived in the dwellings in 1985-86 remained (for reinterviewed households had moved away. While this details see Glewwe and Hall 1998). Only about 57 is a very high rate of sample attrition, the 1990 Peru percent of the households interviewed in 1985-86 LSMS survey may be an extreme case because the sur- were found in their previous dwellings and consented vey took place in an urban area during a time of to be reinterviewed. The proportion of individuals extreme economic and social instability. 289 PAUL GLEWWE AND HANAN JACOBY The Cote d'Ivoire LSMS survey also had a panel The most recent LSMS survey that collected element. The data gathered in the four consecutive panel data was the Vietnam Living Standards Survey, surveys done from 1985-88 can be linked to form which was implemented in 1992-93 and 1997-98. three successive two-year panels. As in Peru, the sam- The first survey covered 4,800 households, while the ple design was based on dwellings rather than on indi- second covered 6,000 households. The second survey viduals, and no attempt was made to follow individu- attempted to reinterview 4,704 of the 4,800 house- als who had moved. About 87 percent of the dwellings holds sampled in the first survey by returning to the sampled in the first year contained the same household original dwellings of those 4,704 households. in the second year, in the sense that at least one adult Households that moved within the village or com- household member in the first year was still a house- mune were followed and reinterviewed, but those that hold menmber in the second year. While this attrition left the commune were replaced.The 1997-98 survey rate may seem relatively low, it is important to note successfully reinterviewed 4,301 of the 4,704 house- that it refers to a time period of only one year. After holds, which implies an attrition rate of only 8.6 per- several years, the attrition rate would certainly be cent. The definition of a household being the same much higher. Moreover, the attrition rate of individu- was that at least one member in the original house- als is almost certainly higher than the household attri- hold remained in 1997-98. A more stringent defini- tion rate; in many households, several individual mem- tion-that 50 percent or more of individuals who bers left but the household was considered to be the were members in either the first or the second survey same because some inembers from the first year still were members in both years-implies a much higher remained. attrition rate of 18.2 percent. Two other LSMS surveys, those in Ghana and Finally, the percentage of individuals who were Jamaica, also had a panel structure. Half of the successfully reinterviewed is much less than the per- dwellings surveyed in the 1987-88 Ghana LSMS sur- centage of households using either definition. Attrition vey were surveyed again in 1988-89. The Jamaican for individuals was 24.0 percent; 7.6 percent,was due to LSMS survey has been implemented annually since the entire household leaving the sample, 2.2 percent 1988, and in some years dwellings surveyed in one was due to deaths, and 14.2 percent resulted fromin di- year were surveyed again in the following year. In viduals who left their households. Overall, theVietnam Ghana, and in the first few years of the Jamaica survey, LSMS was much more successful at retaining house- no attempt was made to follow individuals or even to holds than earlier LSMS panel surveys. Much of its suc- determine whether the people currently living in the cess reflects the high quality of work done by the resurveyed dxvelling were the same as those inter- General Statistical Office ofVietnam, but another fac- viewed in the previous survey. Thus researchers who tor is the relatively low mobility of the population due were interested in panel data had to match households to government restrictions on migration. and individuals in the data sets for the different years In none of the LSMS surveys discussed so far were on the basis of the names, ages, and sexes of the house- steps taken in the initial survey to collect information hold members in the computer files,2' which was a that would make it possible to match households in time-consuming, messy, and inexact process. Because the initial survey with the households surveyed in later of this, in Ghana and in the first years of the Jamaica surveys. Nor, with the exception of the Vietnam sur- survey only about one-half of the individuals surveyed vey, did any of these surveys attempt to follow house- in two consecutive years could be matched with a rea- holds or individuals who had moved. However, one sonable degree of certainty (although in later years of LSMS survey, the Kagera Health and lDevelopment the Jamaican survey this matching process was more Survey, did both. This survey collected four rounds of successful due to improved office procedures and bet- data over a period of two years (1 991-93) from about ter questionnaire design). The lesson here is that 800 households in the Kagera region of northwest explicit questions must be included in all follow-up Tanzania.The survey was designed to study the impact surveys to allow data analysts to match the individuals of adult mortality (specifically AIDS mortality) on in one survey with those in another. The fourth sec- household welfare. It was representative of the Kagera tion of this chapter contains detailed recommenda- region but not of the nation. Special efforts were made tions on how to design such questions. to track households during the course of the survey. In 290 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA the end, the sample attrition of households after two the household head died and the remaining household years (four rounds of interviews) was 9.6 percent, members split into different households, the house- mostly due to migration out of the region (see hold was considered no longer to exist and none of Ainsworth, Ghosh, and Semali 1995). As with C6te the "split" households were followed. This demon- d'Ivoire, the rate of attrition for individuals was almost strates the practical problems of using a sample design certainly considerably higher, because some of the that tries to follow households, as opposed to people individuals in the households that stayed in the sample or dwellings, over time. left their households and were not followed. Only 1.5 A final set of panel surveys from developing coun- percent of the households interviewed in the first tries based on household or dwelling sampling frames round refused to be interviewed in a later round, is a number of small-scale, geographically focused ini- despite the high level of adult mortality in the sample. tiatives. Probably the best known of these among This low attrition after two years suggests that some of economists is the ICRISAT survey conducted in the panel attrition in earlier LSMS surveys might have India.22 This survey was designed primarily to study been substantially reduced by better planning, agricultural production, employment, and income although it must also be recognized that the mobility dynamics in several drought-prone villages of Central of the population varies considerably from country to and Southern India. In the three villages that were fol- country and that the Kagera survey primarily covered lowed continuously for 10 years, 87 percent of the rural areas, where mobility is typically lower. households (104 out of an original 120) were inter- Turn now to the experience of non-LSMS sur- viewed every year. As in the ARIS-REDS surveys, the veys. To the authors' knowledge, the first household follow-up rules focused on the household head, so survey that attempted to collect a nationally represen- that the attrition rate for individuals was presumably tative panel data set (though only in rural areas) in a much higher. Attrition was primarily due to migration developing country is the Additional Rural Incomes out of the village and to the death of the household Survey (ARIS) administered by the National Council head (Walker and Ryan 1990, p.16). This remarkably ofApplied Economic Research in India from 1968/69 low attrition rate is partly due to the fact that the pop- to 1981/82 (National Council of Applied Economic ulation being studied was rural and thus relatively Research 1986).This survey attempted to use a house- immobile. In addition, the interviewers lived in the hold-based sample design to collect panel data over villages continuously during the ten-year period, thus time. If households split, the part that contained the developing a rapport with the villagers, and incentives head of the original household was defined to be the to remain in the survey were periodically provided to "successor" of the original household. The initial sur- the households. vey covered 250 villages from 1968/69 to 1970/71. It Finally, since LSMS surveys have recently been began with 5,115 households in 1968/69, of which implemented in some countries in Eastern Europe 4,118 were interviewed in all three years-an attrition and the former Soviet Union, it is useful to examine rate of 19 percent over two years.The attrition rate for attempts to collect panel data in these countries even individuals was almost certainly higher, because indi- though they are not usually considered to be develop- viduals who left the sampled households were dropped ing countries. Many of these countries have a long tra- from the survey. dition of collecting panel data, but the sampling meth- The panel data from 1970/71 to 1981/82 are ods used have had numerous deficiencies and panel more interesting. The 1970/71 survey covered 4,756 attrition has been high. For example, in the Ukraine households. In 1981/82 an attempt was made to find the sampling was not at all random, and each year and reinterview all of these households.The 1981/82 between 10 and 20 percent of the households dropped survey, known as the Rural Economic and Demo- out of the sample. Similar problems exist in panel data graphic Survey (REDS), successfully found and rein- that have been collected for decades in Russia. terviewed 3,139 of the original households-implying A more recent attempt to collect panel data in an attrition rate of 34 percent over 11 years. Again, the these countries is the Russian Longitudinal attrition rate for individuals was almost certainly high- Monitoring Survey. The first round of this survey er. One reason why so many households were lost duly-October 1992) covered 6,334 households; 7,200 stems from the follow-up rules used. In particular, if households were selected, but about 11 percent 291 PAUL GLEWWE AND HANAN JACOBY refused to participate in the first interview (Russian (Malaysia has two distinct geographical sections, Longitudinal Monitoring Survey 1999). In the second Peninsular Malaysia, where the majority of the popula- round (December 1992-March 1993), 6,068 of these tion lives, and the Sarawak and Sabah regions on the households were still in the sample, which implies an island of Borneo.) The second Malaysian Family Life attrition rate of 4.2 percent over about five months. Survey, carried out in 1988-89, attempted to reinter- (Households or individuals who had moved were not view all of the 1,262 women who were still living in followed.) By the fourth round (October Peninsular Malaysia. The survey also attempted to fol- 1993-January 1994) only 5,473 households were still low a random sample of each woman's children, includ- in the survey, so the attrition rate between the first and ing those who no longer lived with their mothers. fourth rounds was 13.6 percent over a period of about The follow-up efforts in the 1988-89 MFLS wvere 15 months. fairly successful (Haaga and others 1993). Of the orig- Another recent example is the modified inal 1,262 women,31 were known to have died and 2 Household Budget Survey in Poland (Okrasa 1999).A were known to have left Peninsular Malaysia. Of the new sample of about 16,000 households was drawn at 1,229 who remained, 889 (72 percent) were located the end of 1992. Of these, about 8,000 were followed and reinterviewed. Another 13 (1 percent) refused, and from 1993 to 1996. After three years, about 5,000 20 (2 percent) were located but were away from home households were still in the sample, an attrition rate during the days when the interviewers were working near 37 percent. In both this survey and the Russian in their areas.The other 306 could not be located; pre- Longitudinal Monitoring Survey, the attrition rate of sumably most of them moved within Peninsular individuals was presumably much larger than the Malaysia, but a few may have died or left the peninsu- household attrition rate, and a substantial proportion la. In some cases the addresses that had been recorded of the attrition is thought to have occurred because of in the original survey were inadequate, so the original poor management. dwelling could not be found. Overall, about half of the original respondents were known or believed to have Surveys Based on Samples of Individuals moved since 1976, and 51 percent of those who had Almost none of the panel surveys reviewed in the pre- moved were found and successfully reinterviewed. Of vious subsection (the sole exception being the LSMS the children selected to be in the new sample who no survey in Kagera, Tanzania) attempted to locate and longer lived with their mothers, 63 percent were reinterview individuals who moved away from their located and interviexved. Given that 12 years had original dwelling (or, in surveys that attempted to fol- elapsed since the initial survey, a 72 percent success low a sample of households, individuals who moved rate was a marked improvement over the attrition rates away from their original household). Thus these sur- of many of the surveys discussed in the previous sub- veys inevitably lost sizable fractions of their original section. Yet even this success rate could be improved samples of individuals. In general, this is what one upon, since the initial survey in 1976 was not designed would expect given a sample design based on dwellings for the purposes of following households in later years. (or on households). There appear to be only three Methods to improve the design of the initial survey for large-scale surveys in developing countries that were follow-up purposes are discussed in the fourth section based on sample designs of individuals and thus made of this chapter. major efforts to locate individuals who moved as long Another large-scale panel data set based on an as they xvere still thought to be part of the population individual sample design is the Cebu Longitudinal originally surveved (in other words, they had not died, Health and Nutrition Survey conducted in Cebu, the emigrated, or moved into an institutional living second largest metropolitan area in the Philippines. arrangement). This subsection discusses these surveys. The initial survey began with a sample of 3,085 chil- The Malaysian Family Life Survey was the first dren born in the Cebu metropolitan area betxveen panel survey in a developing country that attempted to May 1983 and April 1984, along with the children's maintain an individual sample design on something mothers. Each mother was interviewed about herself approaching a nationwide scale. The first MFLS xwas and her child every two months for a period of two conducted in 1976. It successfully interviewed 1,262 years. As with the Malaysian Family Life Survey, the "ever-married" women in Peninsular Malaysia. children and their mothers were followed if they 292 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA moved as long as they still lived in the Cebu metro- sentative of 13 provinces that contained 83 percent of politan area. After the first two years of life, 318 of the the total population of Indonesia. Four years later, in children (10 percent of the sample) had moved out of 1997, the Indonesian Family Life Survey was fielded metropolitan Cebu, 156 (5 percent) had died, and the again. All of the original households were to be revis- mothers of 50 (2 percent) refused to continue partici- ited, and individuals or households that had moved pating in the survey. Every child in the sample was were to be followed as long as they still lived in one of accounted for. In 1991, when the children were about the 13 provinces surveyed. In 1998 a 25 percent sub- 8 years old, a follow-up survey was conducted of the sample of the original households was visited again to 2,561 remaining children. Of these children, 155 (6 assess the short-term impact of the financial crisis that percent) had migrated out of metropolitan Cebu, 38 began in late 1997. (1 percent) had died, and 137 (5 percent) could not be The Indonesian Family Life Survey was quite suc- located. In 1995, the children who were located in cessful in following individuals and households that 1991 were surveyed for a third time. In this case, after had moved.About 94 percent of the households inter- a gap of three to four years, 98 percent of the children viewed in 1993 were successfully reinterviewed four were found. years later in 1997, and 96 percent were reinterviewed These retention rates are remarkably high given five years later in 1998.23 The retention rate for indi- that those individuals who died or emigrated from viduals was lower-about 91 percent in 1997-as one metropolitan Cebu were (correctly) dropped from the would expect since finding a household does not sample since they no longer resided in the metropoli- imply finding all of the members of that household. tan Cebu area. Ignoring such individuals yields an Most of the attrition was due to failure to locate attrition rate over the first two years of only 2 percent households and individuals that moved; of the 6 per- and attrition rates of only 5 and 2 percent during the cent attrition in households from 1993 to 1997 only periods when the children were ages 2-8 and 8-11, one percentage point was due to households who respectively. These exceptionally high retention rates refused to participate in the second survey (after may have been due to the competence of the survey agreeing to participate in the first survey) while five team and the small size of metropolitan Cebu relative percentage points were from inability to find house- to, for example, Peninsular Malaysia. (None of the holds that had participated in the first survey. mothers and children who remained in metropolitan One particularly interesting piece of information Cebu had moved very far away from their original from the Indonesian Family Life Survey is an estimate dwelling; conversely, those who did move far away no of the cost of tracking people who had moved. longer lived in metropolitan Cebu and thus, by defini- Thomas, Frankenberg, and Smith (1999) report that tion, did not need to be found.) the cost of tracking people added only about 20 per- The high retention rates may also have been due cent to the cost of the survey. However, this figure rep- to questionnaire design and survey methods.The orig- resents only the added cost of "long-distance" track- inal questionnaire was designed to collect information ing; "local" tracking (following households and that would be useful for conducting follow-up sur- individuals that moved relatively short distances) was veys. Interviews were repeated every two months in done at the same time as visits to the households that the first two years of the survey, helping interviewers had not moved, so the additional costs of local track- remember the location of the dwellings and presum- ing were not included in this figure-suggesting that ably building trust between interviewers and house- the 20 percent additional cost ought to be treated as a holds. Interviewers were assigned to the same house- lower bound. For more information on the holds that they had covered in previous surveys. Indonesian panel data see Frankenberg, Thomas, and The Indonesian Family Life Survey, which drew Beegle (1999) and Thomas, Frankenberg, and Smith on the experience of the Malaysian Family Life Survey (1999). (and like the Malaysia Survey was designed and imple- mented by the Rand Corporation), also followed indi- Lessons from Experience viduals over time. The Indonesian Family Life Survey This review of past experience of panel data collection covered about 7,200 households in 1993. While the in developing countries provides several lessons. First, a sample was not nationally representative, it was repre- panel survey that uses a dwelling sample design and 293 PAUL GLEWWE AND HANAN JACOBY does not attempt to follow individuals who move viewed in the previous survey.The importance of this should expect to lose a substantial proportion of the can be seen by comparing the results of the Ghana and original sample of individuals over time. After one or Cote d'Ivoire LSMS surveys. In Ghana the need to two years, survey planners should expect to lose at least match "by hand" resulted in a loss of about half of the 1 0 percent of the individuals in the original sample and potential panel, many of whom almost certainly still perhaps as many as 20 percent. Over five to ten years, lived in the originally sampled dwelling but had been as many as 30-50 percent of the individuals in the given an erroneous age, sex, or name in the initial sur- original sample can be expected to move, as was seen vey. In contrast, in Cote d'lvoire about 87 percent of in Peru and Malaysia.24 The degree of mobility will the households could be matched. vary, but will generally be lower in rural areas. A final A final conclusion is that refusals by people in the point regarding mobility is that individuals often do original sample to participate in subsequent surveys not move very far. Of the movers who were located in have been encouragingly low-around 1 percent in the 1988 Malaysian Family Life Survey, 76 percent had the Kagera, Cebu, Malaysia and Indonesia surveys.The moved within the same district and another 11 percent refusal rate was much higher in the Peru survey, but had moved to a different district within the same state. much of this could be due to the exceptional circum- Second, attempts to folloxv individuals over time, stances in late-1980s Peru. even if the individuals move, can be reasonably suc- cessful if they are well executed.The Indonesia survey Recommendations for Collecting Panel Data in was able to reinterview 91 percent of the individuals LSMS-Type Surveys who were surveyed four years earlier. The Malaysia survey successfully interviewed 72 percent of the indi- This section draxvs upon the previous sections to pro- viduals from the sample of 12 years earlier, although vide recommendations for collecting panel data in this figure drops to about one-half when only people LSMS and similar surveys. It begins with general rec- who move are considered. The survey in the ommendations, moves to some more detailed aspects Philippines retained over 90 percent of surviving orig- of sample design, and concludes with specific practical inally sampled children over a period of 11 years, advice, including draft questionnaire pages. although the fact that the area sampled was only one metropolitan area, as opposed to an entire country, General Recommendations may have been an important reason for this high rate. Whenever any household survey is implemented there Examples from the United States are also is usually at least some possibility that subsequent sur- instructive. After 15 years, the National Longitudinal veys will be done in later years, providing a possible Survey of Youth retained 89 percent of the youths opportunity to collect panel data. In such situations ages 14-22 who were originally interviewed in there are three possible choices: fielding a series of 1979. On the other hand, in 1992 the Panel Study of cross-sectional surveys in which a different sample of Income Dynamics retained only about 50 percent of households is interviewed each time, collecting panel the respondents who had entered the survey 24 years data based on a sample of dwellings, and collecting earlier; in the first year alone 14 percent of the panel data based on a sample of individuals. The dis- households dropped out after completing the initial cussion at the end of the second section of this chap- interview.25 Taking the Malaysian Family Life Survey ter leads to a first recommendation: LSMS-tvpe sur- as a benchmark, nationwide surveys that devote sub- veys should collect panel data based on a sample of stantial resources to following individuals who move dwellings. As explained in the second section of this ought to be able to find about half of them 10 years chapter, the sample of dwellings must be updated each later and perhaps three-quarters of them 5 years time the survey is implemented by adding a sample of later. newly built dwellings, which will ensure that the sam- A third conclusion is that good questionnaire ple is representative of the current population. design can greatly reduce sample attrition. For exam- There are two reasons for this recommendation. ple, substantial attrition can be avoided by including a First, there are potentially significant advantages to page in the questionnaire of the follow-up survey that reinterviewing the households currently living in the asks what happened to each household member inter- dwellings that were sampled in the initial survey, and 294 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA there appear to be no serious disadvantages to doing dwelling. Thus about 7-12 percent of the original so as long as newly built dwellings are added to the household members will have moved to dwellings that sample. Second, for a sample design that attempts to are close to their original dwellings. follow individuals over time, substantial sample attri- If these new households were added to the sam- tion is virtually unavoidable. For example, even though ple, the field costs of including these extra interviews the Indonesian Family Life Survey was very successful in the sample (adding a few percentage points to at reinterviewing households and individuals after four account for additional travel expenses) would be about years, about 9 percent of individuals could not be rein- 10-15 percent higher than the field costs of a panel terviewed. As a result, a panel data set based on a sam- survey based on dwellings that does not try to follow ple of individuals becomes representative of neither any individuals who move. Over five years, about the current population in subsequent surveys nor the 30-50 percent of the individuals who were inter- cohort of individuals covered in the initial survey. This viewed in the initial survey are likely to move to is a serious problem for LSMS surveys, since in almost another dwelling. About 75 percent of these would all cases governments want these surveys to be repre- have moved a relatively short distance from their orig- sentative of the current population each time they are inal dwelling, so about 25-40 percent more house- fielded. holds would have to be added to the survey, which The recommendation to collect panel data in would raise the field costs of the survey 30-50 percent household surveys based on a sample of dwellings, as (again, after adding a few percentage points for addi- opposed to a sample of individuals, limits the potential tional transportation expenses). In LSMS surveys field usefulness of the data because many types of analysis costs are typically 40-50 percent of total costs (see that use panel data are better served by a survey that Grosh and Munoz 1996), so the total cost of an LSMS follows individuals when they move. However, it is survey would increase by about 5-10 percent if the possible, though more expensive, to accommodate time between surveys were one or two years and these analytical needs by implementing a survey that 15-25 percent if the space between surveys wvere five uses a dwelling sample design and also follows indi- years. viduals who move out of their dwellings. Such a sam- If survey planners choose to follow individuals pling procedure would produce two distinct, although who move even greater distances, field costs will be overlapping, data sets: a representative sample of the higher-especially for transportation and "search" current population at each point in time and a sample costs. One can only speculate what field costs would that, apart from the inevitable attrition problems, rep- be for a nationwide survey that seriously attempted to resents a cohort of individuals who are followed over follow all individuals who had not died, emigrated, or time. moved into an institutional living arrangement; such The fundamental issue then is whether the ana- costs might be 25-50 percent more than those of a lytical benefits of following individuals who move out simple dwelling-based panel survey if the time of the original sample of dwellings are worth the cost. between surveys were one to two years and 50-100 This is a matter of judgment, and there are a number percent more if the time interval were five years or of different options available to survey designers. In longer.26 These increased field costs imply increased countries with weak survey implementation capacity, total costs of 10-25 and 20-50 percent, respectively. the safest course may be not to follow individuals at Even with such expenditures, it will not be possible to all. For other countries, some attempt to follow indi- find all of the individuals who moved after the first viduals may be worthwhile. survey. Based on the experience of the 1988 Malaysian A relatively unambitious option is to follow only Family Life Survey, as many as 50 percent of individ- individuals who have moved a relatively short uals who move may be lost, although the success of the distance say, within the same township or within 1997 and 1998 Indonesian Family Life Surveys in 10-20 kilometers. In the space of one to two years, locating movers suggests that most movers can be about 10-20 percent of the original household mem- found. In general, looking for individuals who move bers are likely to move. The experience of the 1988 outside the local community could significantly Malaysian Family Life Survey suggests that most of increase the costs of the survey and will be only par- them will stay within 10-20 miles of their original tially successful. 295 PAUL GLEWWE AND HANAN JACOBY The costs of following individuals who move may work can be decisive for determining whether a panel appear to outweigh the benefits, especially in the case survey is a success or a failure. In particular, supervisors of individuals who have moved long distances. should check carefully when interviewers claim that However, as discussed at the end of the second section, they have not been able to find the dwelling previous- these costs can be reduced by following only a subset ly surveyed, and supervisors should revisit at least some of the individuals who move, especially for individuals of the dwellings that interviewers claim are the ones who move long distances. For example, survey plan- that were covered in the previous survey. Monetary ners may decide to follow only individuals in a subset incentives could be offered to interviewers who find of the sampling areas or only a subset of households in the correct panel dwellings; this approach was taken in each area. Another possibility would be to follow a the Indonesian Family Life Survey. For a more gener- randomly chosen subsample of people who have al discussion of supervision issues see Grosh and moved to a different state or region. A disadvantage of Mufioz (1996). following only a subset of migrants is that there will be little data for in-depth analysis of long-distance migra- Recommendations for Sample Rotation in Panel Surveys tion and associated phenomena.Yet such a subset may When a decision is made to collect panel data, survey provide information that will be useful for measuring planners must consider how long to keep the original the extent of attrition bias. sample before starting over with a new sample of Another option for containing costs is to follow dwellings or individuals. In principle, panel surveys individuals only for a certain number of vears. The based on a sample of dwellings can follow the same longer one attempts to follow individuals who move, dwellings for decades by continuing to add newly built the more expensive the survey will be and the less rep- dwellings to the sample and dropping dwellings that resentative the sample will be of the individuals cov- cease to exist from the original sample. On the other ered in the initial survey. For LSMS-type surveys, hand, surveys could replace the initial sample with a which generally have multiple objectives, there may be completely new sample after five or ten years. The little reason to follow individuals for more than 10 same issue arises when trying to follow individuals years; any analytical objective that requires observa- who have moved out of the dwellings since the initial tions on the same individuals for more than 10 years survey: how long should the survey attempt to follow should probably be carried out using a specialized sur- the original sample before starting over with a new set vey.27 Some analytical objectives may even allow sur- of individuals? veys to stop following movers after only five years.This Consider the case of a panel survey based solely gets into the issue of sample rotation, which is dis- on a sample of dwellings.There are three practical rea- cussed in the next subsection. sons not to follow the same dwellings for many Before turning to more specific suggestions, a decades. First, each time a survey is fielded, a small per- final general point to consider is that collection of centage of the dwellings involved will be dropped panel data will probably require more careful supervi- from the sample because the dwellings' occupants sion of interviewers. Even in cross-sectional surveys, refuse to continue participating in the survey. Over interviewers are often tempted to cut corners to make many years, the percentage of the original dwellings their work easier. In panel surveys there will be addi- no longer in the sample due to refusals will increase, tional temptations, such as the temptation not to steadily eroding the representativeness of the sample. search very hard to find the dwelling that was surveyed Second, there could be errors in the procedure that in a previous year. Indeed, some analysts have adds new dwellings to the sample. For example, the informed the authors that they suspect that matching listing operation in a follow-up survey may select the problems in some panel surveys were due to inter- wrong "newly built" dwellings to add to the sample if viewers substituting a random dwelling or household the data on age of dwellings contain errors. for the panel dwelling or household that they were Third, household members who have been inter- supposed to interview. viewed many times over several years may behave in Additional supervision procedures must be taken ways that compromise data accuracy. Some may grow to prevent interviewers from succumbing to this and weary of the long interviews typical of LSMS-type sur- other temptations, because quality control in field- veys and decide to give misleading answers that their 296 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA experience has shown them will shorten the interview The first sample rotation option in Figure 23.1 is (claiming, say, that they no longer operate a household a repeated panel without overlap. Each dwelling sur- business when in fact they do). A related problem is veyed in the first year is surveyed again in the second Hawthorne effects, in which people's behaviors are and third years. In the fourth year all of the dwellings affected by being observed over time. For example, if that were interviewed in the first three years are interviewers ask women each year whether they know dropped and an entirely new sample is selected for the of, or have used, a particular contraceptive method, the fourth, fifth, and sixth years, and so on every three women interviewed may decide to find out more years. When newly built dwellings are added in the about the method and possibly use it. second year, these dwellings are surveyed in the third The discussion in the previous paragraph suggests year but dropped in the fourth year; any newly built that panel surveys based on samples of dwellings need dwellings added in the third year are also dropped in a system to rotate the sample-that is, a system to the fourth year. replace dwellings that have been followed for several There are three disadvantages to using repeated years with new dwellings. To provide background for panels without overlap. First, this could lead to mis- recommendations on sample rotation, a brief review of leading trends over time because the number of several basic options is needed. (For more detailed dwellings added to the sample varies widely from year information see Duncan and Kalton 1987 and the ref- to year. For example, consider a survey in which only erences they cite.) Figure 23.1 shows three different newly constructed dwellings are added in the second ways to follow dwellings over time. In the first two and third years, while all dwellings are replaced in the cases, after a dwelling has initially been surveyed it is fourth year. If wealthy households are more likely to resurveyed in the two subsequent time periods (here- refuse to participate in subsequent surveys (perhaps after referred to as years, although they could be inter- because monetary incentives used to win households' vals of several years), after which it is replaced by a cooperation are less enticing to them), sample attrition "fresh" dwelling. The presence of an arrow between due to refusals will reduce measured mean income in any two boxes in Figure 23.1 indicates that the same the second and third years and increase it in the fourth dwelling (or, more generally, the same group of year.28 Second, since it is never the case in any year dwellings) is retained in the following year; the absence that a random subsample of dwellings is retained while of an arrow indicates that a dwelling has been dropped others are replaced, it is not possible-at least in the from the sample and replaced with a new one. first three years-to check for attrition bias by com- Figure 23.1 Repeated, Rotating, and Split Panels Year 1 2 3 4 5 6 7 8 9 Repeated panel R = etc. without overlap j j j Rotating panel, _ - - etc. even rotation j Split rotating - - i etc. panel - l 297 PAUL GLEWWE AND HANAN JACOBY paring the replacement dwellings with dwellings terviewed. However, this is not a major problem in retained from the previous year. Third, it may be that most countries, and even if it is, the extent of possible changes at the household and individual levels (gross bias can be investigated within a rotating design with- changes) that span the third and fourth years are of out a split, as explained above. particular interest to analysts, but these changes cannot In Figure 23.1, dwellings are replaced after only be examined if this rotation scheme is used because all three years, but this is just for the purposes of illustra- dwellings are replaced after the third survey year. tion. If a survey is fielded every four to five years, there The second design shown in Figure 23.1 is a should be no problem in following the same dwellings rotating panel witb even rotation, in which one-third over three surveys (in other words, over eight to ten of the sample dwellings are replaced each year by a years) as long as the proportion of households that new set of dwellings. This design overcomes the three refuse to be reinterviewed is well under 5 percent in disadvantages of the repeated panel without overlap. In the second year. If 5 percent or more refuse to be rein- this scheme, in any given year after the first two years terviewed, survey planners should seriously consider of the survey, one-third of the sample will consist of either replacing dwellings after they have participated new dwellings, one-third will be interviewed for the in only two surveys or using a split panel rotating second time, and one-third will be interviewed for a panel scheme.There is probably little benefit to retain- third time.Thus the first disadvantage discussed in the ing dwellings for longer than 10 years, which means previous paragraph (spurious time trends) is avoided. that it is reasonable to rotate one-third of the dwellings The second disadvantage is also circumvented because out of the sample with each survey. each year one-third of the sampled dwellings are a When a survey is performed annually, in most new random sample of existing dwellings, so each year countries it is feasible to use the same dwellings for at one can check whether there is any attrition bias due least five years and, if there is a strong interest in ana- to refusals (or any other causes). Finally, this scheme lyzing panel data, for up to ten years. If the cumula- yields panel data for any two- to three-year period, tive number of refusals of follow-up interviews reach- resolving the third problem. es 5 percent or more of the original samiple of In both of the panel rotations discussed so far, all dwellings after, say, five years, survey planners should of the dwellings are part of the panel rotation scheme. consider rotating dwellings out of the samnple after This does not have to be the case. The third and last five years. scheme shoxvn in Figure 23.1 is a split rotating panel. The recommendations of the previous paragraphs One-fourth of the sampled dwellings (the top row of assume that the decision on whether to do the survey blocks) are not part of any panel, and three-fourths are annually or less frequently (such as every 3-5 years) part of a three-year panel.Thus half of all of the house- has already been made. For a general discussion of how holds in the sample are new each year. This leads to a to make this decision, see Chapter 2. Factors in this higher proportion of new households than in the decision that are specific to the analysis of panel data evenly rotating panel (which had one-third), reducing were presented in the first section of this chapter. In the problem of the sample becoming unrepresentative particular, this decision should consider the analytical due to refusals by respondents to participate in follow- objectives of the survey. For example, attemnpts to esti- up interviews. On the other hand, such refusals are mate structural models using panel data are better usually rare, so this advantage of split-panel rotation served by a survey that collects panel data every year schemes (relative to a "full" rotating panel) may be for at least five years (and preferable longer than that). outweighed by the disadvantage that panel data are In contrast, using panel data for program evaluation collected for only part of the sample. often requires only two surveys, but these surveys The above discussion suggests that survey design- should be several years apart to allow new programs to ers should use a full (as opposed to a split) sample rota- have an effect on the outcomes of interest. tion scheme, with even rotation to replace dwellings in These recommendations for sample rotation also the original sample. One exception to this recom- apply to surveys that attempt to follow individuals mendation is that it may be better to use a split rotat- who move out of their original dwellings. In particu- ing panel in cases in which a significant percentage (5 lar, there is no point in following people whose orig- percent or more) of households may refuse to be rein- inal dwellings have been rotated out of the sample. 298 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA Because the sample of individuals followed over time System readings are recorded and securely stored for is a combination of movers and nonmovers, when future use, survey teams in later years should be able to nonmovers are dropped, the movers by themselves will locate virtually all of the dwellings covered in an ini- not be representative of any initial population. If the tial survey, as long as these dwellings still exist. For costs of following movers are high, it may be best to dwellings that no longer exist, the team should be able drop the movers before dropping the dwellings from to determine with a high degree of certainty that which they moved, especially in the case of those who those dwellings are in fact gone. have moved far away from their original dwelling and Collecting information on each individual house- are therefore more expensive to follow. The sugges- hold member in the initial survey also helps inter- tions made in the previous subsection about when it viewers in subsequent surveys to match individuals makes sense to follow movers still hold for the from different survey rounds after their dwellings have dwelling sample rotation scheme recommended in this been found. Interviewers should obtain not only the section. formal names of household members but also any informal names by which they are known in the com- Specific Recommendations munity. If the names of the persons are written in a This subsection provides specific recommendations for different alphabet or language than the one used for collecting panel data. Survey planners can use these the survey questionnaire, interviewers should record recommendations to minimize sample attrition and these names on the household roster both in that ensure that the panel data they collect are as accurate alphabet or language as well as in the alphabet or lan- as possible. guage used to fill out the survey questionnaire. Finally, in countries where individuals have national identifi- T1EIE FIRST SURvEY. As long as there is a possibility, cation codes (such as social security numbers), it may however remote, that an initial survey will be followed be possible to collect this information to match data by another survey, planners should take a few simple on individuals from different survey rounds. However, steps in the first survey that will improve the ability to national identification codes may be sensitive informa- locate dwellings in the future and thus reduce unnec- tion, so interviewers should not insist that individuals essary sample attrition if subsequent surveys are provide them; doing so may reduce respondents' level implemented. First, the addresses of dwellings should of cooperation. be recorded in the initial survey in as much detail as possible, so that they can be easily located if something FOLLOW-UP SuRvEYS. If an initial survey collects all of changes (for example, if the dwelling is painted a dif- the information described above, it should not be dif- ferent color, if the name of the street changes, or if the ficult to reinterview people who have remained in the numbering system changes). Second, detailed maps of same dwelling during the intervening time period.The where the households are located should be drawn up most important task is to record the information clear- in the initial survey and the maps should be stored in ly so that it will be possible to create a computer file a safe place for use in future years. that matches dwelling and individual ID codes from Third, Global Positioning System equipment the first survey with such codes from the second sur- should be used at the time of the first survey to obtain vey. In the case of dwelling codes, the simplest approach the latitude, longitude and even altitude of each is to use the same code in both years.29 If a dwelling no dwelling. Global positioning devices are relatively longer exists or is vacant, or if the occupants refuse to inexpensive (a few hundred dollars) and quite small cooperate, this should be recorded as well. (A ques- (about the size of a cellular telephone). Fourth, and tionnaire page that collects this information-provided perhaps most importantly, all of the above information in Volume 3-is introduced in the following subsec- should be stored for future use, as paper copies, in elec- tion.) To match individuals across surveys, certain basic tronic form, or preferably both; in some countries information on the individuals interviewed in the first panel surveys could not be carried out because ques- survey must be recorded and transcribed onto the tionnaires and all other records with information from questionnaire for the subsequent survey before the the previous survey had been thrown away. When interview begins-specifically, the name (including any detailed addresses, maps, and Global Positioning nicknames), age, and sex of the individual and the ID 299 PAUL GLEWWE AND HANAN JACOBY code used in the initial survey. It may also be useful to team the information necessary to find and interview record information on the person's relationship to the that person. This procedure worked quite well in the head of household, the person's occupation, and, if pos- 1988 Malaysia Family Life Survey and the 1997 sible, any national identification code. The following Indonesia Family Life Survey. Finally, the importance subsection discusses in detail two different ways to of finding individuals and households that move must record such information. be stressed during the training of the interviewers, and Two things can be done to minimize refusals in planners should consider giving incentives, such as follow-up surveys. First, the survey team can use the cash bonuses, to interviewers who locate individuals same interviewers that were used in previous surveys, and households who have moved, with higher bonus- assigning them to the same households that they inter- es for more distant movers. Such incentives were used viewed before. This can help because a familiar inter- in the 1997 Indonesia survey, and may partially explain viewer is more likely than a stranger to induce a reluc- that survey's success in finding movers. tant household to cooperate. The survey team that implemented the Cebu Longitudinal Health and Implications for Questionnaire Design Nutrition survey in the Philippines, which was quite This subsection introduces two questionnaire pages successful in minimizing attrition, attributes part of its that allow interviewers to match data on households success to that practice. However, there is one poten- and individuals in a follow-up survey with data on the tial problem with this; analysts will not be able to dis- same households and individuals from a previous sur- tinguish (time invariant) measurement error specific to vey. (The questionnaire pages themselves are found in a given interviewer from (time invariant) measure- Volume 3.) This subsection also discusses changes that ment error specific to a given household. Second, may be needed in other modules of the household planners should seriously consider offering small gifts questionnaire in surveys that collect panel data. The or monetary incentives to households to participate in first page determines whether the dwelling was the follow-up survey. While some statistical agencies included in the sample for the previous survey and, if are quite comfortable with such payments, others are so, whether the dwelling was successfully located and not; each country will have to decide based on past its inhabitants reinterviewed in the subsequent survey. survey experience. This page also contains some information on dwellings that are being added to the sample in an FOLLOWING INDnIDUALS WuO MOvE. The discussion so effort to follow individuals from the initial survey who far has focused on household members who do not have moved; for these dwellings, information is move. If survey planners decide to follow some or all of recorded on the household member being followed. the household members who move after the first survey, The second page matches the individuals current- several additional steps can be taken during the first sur- ly in the dwelling with the individuals who occupied vey and in subsequent surveys that will make it easier to that dwelling during the previous survey. Using ID locate these members in later years. First, in the initial codes, this page enables the survey team to match data survey it is useful to ask the head of household the from the previous survey on all the original inhabi- names and addresses of one or two people in the imme- tants that still live in a dwelling with data on these diate vicinity who would be most likely to know at individuals in the current (follow-up) survey.This page some later date where the household may have moved. is also used to collect information on what happened Second, sometimes local officials can provide informa- to any former occupants who are no longer in a tion on the current location of people who move. For dwelling. The following notes explain how to use example, in the 1988 Malaysia Family Life Survey, postal these two questionnaire pages. delivery workers were often able to provide the survey team with information about people who had moved. ADDITIONAL QUESTIONS FOR THE HOUSEHOLD Third, it is important to establish reliable lines of IDENTIFICATION AND CONTROL PAGE. The Metadata communication between the survey teams working in module (introduced by Chapter 4) contains a question- different areas. Thus, when one team discovers that an naire page entitled "Household Identification and individual has moved to an area that is in the vicinity Control Information." This page collects basic informa- of another team, the first team can give the second tion about a dwelling and the household in it, including 300 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA the address of the household, the ID code for the pri- However, if household or cluster ID codes already mary sampling unit, and the ID code for the household. convey this information, it is not necessary to make The version of this page presented inVolume 3 does not this distinction in question 2. consider the possibility of collecting panel data. The first page introduced here, "Additional Q2-Q4. These three questions can be dropped if the Questions for the Household Identification and survey does not attempt to follow individuals who Information Control Page," provides questions that move. should be added to the household identification page of the metadata module. These questions should be Q3-Q4. These two questions, for people who have included only in follow-up surveys, and not in an ini- moved to new dwellings, should be filled out before tial survey. the interview by either the office staff or the field team For purposes of collecting panel data, there are supervisor. This information allows the data analyst to three kinds of dwellings: dwellings that were covered match the individuals in the current dwelling with the in the previous survey, dwellings that have been con- same individuals who, at the time of the first survey, structed since the last survey or added as part of a lived in a dwelling that was part of the sample for the dwelling sample rotation scheme, and dwellings that first survey. The personal ID codes for these people have been added because people who moved into should be filled in here as well as on the "identifica- them occupied a dwelling covered by the original sur- tion of persons interviewed in the previous survey" vey. If the survey does not follow individuals who page (discussed below). This is necessary to show move, only the first two kinds of dwellings will be part which individuals currently living in the dwelling are of the survey, and Questions 2-4 of the "additional the movers being followed. questions" page should be removed. The following The basic scenario for filling out these questions notes provide detailed information on the design of is as follows. An interviewer goes to a dwelling that the "additional questions" page. was part of the previous survey and finds out that Two general changes need to be made to the someone who was interviewed in that survey has since household identification page when adding these extra moved out of the dwelling. By asking the current questions concerning panel data. First, a distinction occupants of that dwelling (or neighbors or local offi- must be made between the dwelling and the house- cials), the interviexver discovers where that person is hold, and ID codes must be assigned to both. Second, currently living. Someone must then interview the the questions about replacing households should be household members (including the person who has moved toward the end of the page, and should come moved) living in the mover's current dwelling. Before after the questions presented in the "additional ques- the interview in the new dwelling is conducted, a new tions" page of this chapter. The following paragraphs questionnaire is prepared. The cluster and household provide more detailed information on the additional ID codes of the dwelling of origin are entered in questions for the household identification page. response to Question 3, and the personal ID code (at the dwelling of origin at the time of the previous sur- Q1-Q4. These four questions should be filled out by vey) of the person who moved is entered in response office staff or the field team supervisor before the to Question 4. Finally, some basic information on this interviewer visits the dwelling. person needs to be added on the "identification of persons interviewed in the previous survey" page (dis- Ql-Q2. These two questions identify dwellings that cussed below), in the line that corresponds to that per- are in the sample as part of an effort to follow people son's personal ID code in the previous survey. This who move. procedure works equally well if several people move from a dwelling included in the first survey to anoth- Q2. Interviewers can also distinguish (by having sepa- er dwelling. rate response codes for each case) between dwellings surveyed for the first time because they are new Q5-Q7. These are standard questions to ask about any dwellings and dwellings added because they are replac- dwelling in which a household is being interviewed, ing dwellings that were rotated out of the sample. regardless of the status of the dwelling.The interview- 301 PAUL GLEWWE AND HANAN JACOBY er should ask Question 7 for all cases in which an and disadvantages, and the one to use will depend in interview did not take place, because the appropriate part on the nature of the overall survey and the action for the interviewer will depend on the answer amount of time between the previous survey and the to that question. current survey. This chapter presents draft question- naire pages for both methods. Q7. This question determines what should be done if The main advantage of the same ID code method it is not possible to conduct an interview in the is that individuals retain the same personal ID code as dwelling. If the dwelling was covered in the previous long as they remain in the dwelling where they were survey, it may still be possible to get some information originally interviewed.This helps data analysts manip- from friends and neighbors; in this case the interview- ulate the data, and reduces any errors that may be er should go on to the "identification of persons inter- caused by the interviewer or data entry operator in viewed in the previous survey" page (discussed below). assigning new ID codes to the same person. Having If the dwelling is included in the survey because it is the same ID code is particularly advantageous when inhabited by an individual who has moved and is households and individuals are interviewed many being followed, there is little that can be done and the times, such as when they are interviewed annually over interview should end. If the dwelling is being added to a period of five or ten years. the sample as part of a sample rotation scheme or The main disadvantage of the same ID code because it is a newly built dwelling, it should be treat- method is that it becomes more cumbersome when ed exactly as one would treat a dwelling that cannot there has been substantial mobility in and out of the be interviewed in a cross-sectional survey. Question 7 dwelling, which is more likely to occur when a long uses a skip code that assumes that when such a house- time has passed since the previous survey. In particu- hold cannot be interviewed another household will lar, if most or all of the previous inhabitants have replace it. Another option in such cases is not to moved out of the dwelling, all pages of the question- replace the household, in which case the skip code for naire that collect individual-specific questions in rows the last response should be changed to "END OF lined up with the household roster (see the discussion INTERVIEW" in Chapter 3) will have a blank row for each person who has left. For example, suppose there were seven PAGE FOR IDENTIFYING PEOPLE INTERVIEWED IN THE individuals in a dwelling in the previous survey and all PREvious SuRvEY. This page should be included in of them left. Since this method reserves lines 1-7 of all follow-up surveys but not in the first survey. Its basic the individual-specific modules for those people, all purpose is to match the individuals who lived in the seven lines must be left blank in each of the modules dwelling in the previous survey with the same indi- of the questionnaire used in the follow-up survey.This viduals still living in the dwelling at the time of the disadvantage becomes particularly cumbersome if follow-up survey. For people who have left the adding new people exceeds the number of lines dwelling, this page gathers information on where they allowed for household members; in this case two are now and why they left the original dwelling. household questionnaires have to be used, one with There are two ways to collect this information. some members of the household and another with the The first way is to start by collecting information other members. For example, if seven other people about individuals currently living in the dwelling, and currently live in the dwelling mentioned above, they then ask whether these are the same people that were must be allocated to lines 8-14, and if the question- living in the dwelling in the previous survey. For rea- naire has only 12 lines (as is the case for the modules sons that will become clear below, this will be called in this book) a second household questionnaire must the "new ID code" method. The second way is to ask be filled out for persons 13 and 14. about the household members who lived in the Another disadvantage of the same ID code dwelling in the previous survey and record who is still method is that it becomes more cumbersome when in the household and who has left, then to ask about the decision is made to follow individuals who move any individuals who have moved into the dwelling out of their previous household into new dwellings. At since the previous survey.This will be referred to as the the new dwelling there is little reason to prepare a "same ID code" method. Each method has advantages questionnaire that has the names of all the household 302 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA members of the mover's previous dwelling, and it may vey.The basic idea is that the household roster page in also be awkward to try to retain the same ID code for the first survey is detachable and can be inserted into the mover in his or her new dwelling. In most cases it a slot prepared for it in all subsequent surveys. For an is probably more sensible to have new household ros- example of how this worked using the same ID code ters for new dwellings (those added to the sample to method see Ainsworth and others (1992) and KDHS follow people who move) than to retain the rosters Research Team (1999). A similar approach could be used for dwellings that are part of the original sample. used with the new ID code method; the roster from This may imply having two types of household ques- the initial survey would be inserted in such a way as tionnaires, one for following individuals who move replace Questions 2-5 in the "Identification of Persons and one for dwellings of origin; this was the approach Interviewed in the Previous Survey" page. followed in the Kagera Health and Development The following paragraphs explain how to use the Survey (see KDHS Research Team 1999). "new ID code" version of the questionnaire page that The advantages and disadvantages of the new ID matches individuals across surveys.This page should be code method are the inverse of those for the same ID added to the end of the household roster module, code method. The main disadvantage of the new ID which provides a list of all individuals who are cur- code method is that individuals do not retain the same rently members of the household. (For more detail on ID codes over time, even if they remain in the same the household roster see Chapter 6.) Questions 2-5 of dwelling. This is a disadvantage because it makes the this page are filled out by office staff or the field team work of data analysts a bit more complicated and per- supervisor before the household is interviewed, based haps more error-prone, and it also increases the possi- on information copied from the questionnaire used in bility that the wrong ID code is entered at the data the previous survey (or from electronically stored data entry stage. The main advantage is that using the new files). An alternative to copying this information onto ID code method it is easier to fill out the question- the new questionnaire is to have a detachable house- naire for dwellings in which most or all former house- hold roster in the previous survey, as explained above. hold members have moved out. The new ID code Copying can also be avoided if the form is printed out method is also more convenient to use when the sur- from electronic files from the previous survey; the vey attempts to follow individuals who have moved form can then be stapled into the appropriate place in out of their households. the new questionnaire. Both methods have been used in LSMS surveys. This questionnaire page is designed primarily for The same ID code method was used in the Kagera use when an interviewer is going to a dwelling that Health and Development Survey, and the new ID was part of the sample in the previous survey. code method was used in Cote d'Jvoire, Peru, and However, it can also be used when the interviewer is Vietnam. In general, short time intervals between sur- following people who have moved out of a dwelling veys favor the same ID code method while the new sampled in a previous survey. In the second case, only ID code method has more advantages when the time people who moved from the original dwelling to the interval between surveys is longer. The new ID code current dwelling should have their names entered in method is more convenient when following individu- column 2. (The same applies to the accompanying als who move. However, survey designers must decide information in columns 3, 4, and 5). One row needs to for themselves which method to use. The decision be filled in on this form for each mover-the row cor- regarding which method to use should be made care- responding to the mover's personal ID code in the fully, and it may be wise to try both methods in a field previous survey. (In many cases this means only one test before choosing between them. row will be filled out for a dwelling.) For such movers A final general comment regarding both methods the answer to Question 6 will always be "yes," the link of matching individuals across surveys is that it may be with the ID code in the current household will be useful to build removable household roster pages in made in Question 7, and Questions 8-14 will never be the questionnaires. This eliminates the tedious (and filled out. potentially error-prone) process of copying informa- tion from the household roster of the previous survey Ql. If the answer to this question is "NO ONE," the into the questionnaire to be used in the current sur- interviewer should proceed to the next section. A 303 PAUL GLEWWE AND HANAN JACOBY response of "NO ONE" indicates that a completely Q12. This question aims to determine whether the per- new household now lives in the dwelling and neither son is still in the population being sampled (assuming the members of this household nor anyone else in the that people who move into institutional living arrange- area have any information about the former occu- ments are excluded from the sample population). pants, all of whom have left. Q14. If mortality is of particular interest to analysts, Q2-Q5. These four questions contain information to questions to collect additional information on mortality, assist in identifying people who are interviewed across such as cause of death, can be added after this question. two or more surveys. One could add more informa- tion for the same purpose. For example, information Turn now to the "same ID code" version of the ques- could be added on each person's occupation or on tionnaire page that matches individuals across surveys. some kind of national identification number (such as a There are two pages for this version; the first consists of social security number). Individuals who were not instructions for filling out the second. The instruction household members in the initial survey will not have page is needed because this version of the "identification any information filled in for these questions, and none of persons previously interviewed" page completely of the questions on this page should be asked regard- replaces Part A of the household roster module, which ing those individuals. also has a page of instructions. (For more detail on the household roster see Chapter 6.) Questions 1-3 are filled Q5. In almost all cases the codes in Question 5 should out by office staff or the field team supervisor before the be exactly the same as the codes used for the same household is interviewed, based on information copied question in the household roster of the previous sur- from the questionnaire used in the previous survey (or vey The only exceptions are if the head of household from electronically stored data files). An alternative to has changed or if the relationship is one that can copying this information onto the new questionnaire is change, such as spouse (in the event of a divorce). to have a detachable household roster in the previous survey, as explained above. Copying can also be avoided Q7. The ID codes collected in this question, when if the form is printed out frouii electroniic files from the paired with the previous survey ID codes (which are previous survey; the form can then be stapled into the in the column at the far left of the questionnaire page), appropriate place in the new questionnaire. provide the crucial inforniation that allows data users In theory, one could use this questionnaire page to to match the data from the two different surveys for follow a person who has moved out of a dwelling sam- any individual who was included in both surveys, pled in a previous survey, but doing this is rather cum- Supervisors must give extra attention to this data, both bersome. For example, when one individual is being when checking the work of the interviewers and followed, questions 1-3 could be prepared just for that when checking the work of the data entry operators. person, assigning him or her the same individual ID code he or she had in the previous survey. All other Q8. The second response code for this question refers to lines could be used for other members of the new a situation in which a household "splits" into two or dwelling. On the other hand, it is probably clearer not more households but no member actually moves out of to insist that this person receive the same individual ID the original dwelling; such a situation is common in code (that is, the same line number) that he or she had some East Asian countries. One questiornaire shiould be in the dwelling of origin. But if the ID code is filled out for each household, and when the interviewer changed, a new question needs to be asked to deter- is filling out a questionnaire for the original household, mine which person in this new dwelling is the one the people who live in a newly founded household in being followed. There are other difficulties as well, so the dwelling should be given the code "2." Completing it may be best to design a different questionnaire for a questionnaire for a newly founded household should following individuals who moved. Alternatively, one follow the same procedure used for a household that was could use the "new ID code" method described above. added to the survey for purpose of following individuals A final general comment is that one may want to who moved. In countries where such household "splits" ask a question similar to Question 1 of the "new ID do not occur, the second response code can be dropped. code" page to indicate who is providing answers to 304 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA questions in cases in which none of the original house- household should follow the same procedure used for hold members remain in the dwelling. This could be a household that was added to the survey for purpose done across the top of the page, as done on the "new of following individuals who moved. ID code" page, or as the last question of the page. Q19. This question aims to determine whether the per- Q1-Q3. These three questions gather information to son is still in the population being sampled (assuming assist in identifying people who are interviewed across that people who move into institutional living arrange- two or more surveys. One could gather more infor- ments are excluded from the sample population.) mation for the same purpose. For example, informa- tion could be collected on each person's occupation or Q21. If mortality is of particular interest to analysts, on some kind of national identification number (such questions to collect additional information on mortality, as a social security number). such as cause of death, can be added after this question. Q4. In almost all cases the codes in this question CHANGES TO OTHER QUESTIONNAIRE MODULES. A final should be exactly the same as the codes used for this task when collecting panel data is to review other question in the household roster of the previous sur- modules of the survey to see whether changes need to vey. The only exceptions are if the head of household be made. In general, this only occurs for follow-up has changed or if the relationship is one that can surveys, not for the initial survey.There are two reasons change, such as spouse (in the event of a divorce). why such changes may need to be made.The first is to avoid collecting redundant information. The fertility Q5-Q6. These two questions are crucial for determin- and migration modules, and to a lesser extent the ing who is currently a household member and who is employment, education, and housing modules, collect not. Household members proceed to Questions 8-14, information about the past that should never change, while nonmembers who were in the previous survey and there is no reason to collect such information each go to Questions 15-21. Individuals who are not mem- time an individual or household is interviewed (except bers and were not in the first survey need not be asked to check the consistency of responses). any further questions. (The two different kinds of The fertility module collects birth history infor- nonmembers are distinguished in Question 7.) In mation from all women of childbearing age and the some cases survey designers may want to ask migration module collects information on the place of Questions 8-14 to the second kind of nonmember. In birth and on migration that took place many years these cases Question 7 can be deleted. ago.The employment module asks about employment five years ago, and some of the information collected Q8-Q14. These questions are identical to a set of ques- for adults (especially adults 30 or older) in the educa- tions in the household roster module. See Chapter 6 tion module-such as grade repetition and grade for notes on those questions. attainment-will almost never change. For each of these modules it may be useful to change the design so Q15-Q21. These are essentially the same as Questions as not to collect redundant information. However, it is 8-14 in the "new ID code" page discussed above. very important to keep in mind that no information will be redundant for new household members; provi- Q15. The second response code for this question refers sion must be made for the full set of information to be to a situation in which households "split" into two or collected from anyone who was not interviewed in a more households but no member actually moves out previous survey. For an example of how this was done of the original dwelling; such a situation is common in for fertility and migration data see the descriptions of some East Asian countries. One questionnaire should the Kagera Health and Development Survey in be filled out for each household, and when the inter- Tanzania in Ainsworth and others (1992) and KDHS viewer is filling out a questionnaire for the original Research Team (1999). household, the people who live in a newly founded A second reason to modify other questionnaire household in the dwelling should be given the code modules is to collect information on changes in assets "2." Completing a questionnaire for a newly founded or the use of those assets. For example, if one would 305 PAUL GLEWWE AND HANAN JACOBY like to estimate an agricultural production function age in communities without the program. That is, using panel data from farming households, one might & OLS = HS'() - HF() where RI(I) = E[H;I I D = 1] want to match the ID codes of specific plots of land andfl(0) = E[I-.5 I DCS = 0], and E[. I] denotes the con- across different surveys. Similarly, one might want to ditional mean. match the ID codes of specific household businesses, If the placement of health programs is nonran- or perhaps even the assets owned by these businesses. dom, a OLS captures both the impact of the program A final example is studies of households' responses to and the rule by which the government places pro- income or other shocks; researchers may want to grams to communities. For example, if programs are match households' durable goods with other assets implemented first in the villages with the sickliest over time, to see, for example, which items were sold children, a OLS wil underestimate the program impact by households that experienced a severe loss of (unless in equation 1 the analyst can control for all income. Such data were collected for major assets in health conditions that governments consider when the Kagera Health and Development Survey in placing the program). If selective migration exists, Tanzania; see Ainsworth and others (1992) and KDHS either because more "caring" families are more likely Research Team (1999) for further explanation. to migrate into a community with the program or are less likely to migrate out of such a community, a OLS Appendix 23.1 Technical Discussion of will overestimate the program impact. Econometric Analysis of Panel Data To solve these two problems, suppose that a sec- ond survey of the same households (and children) is This appendix provides a technical presentation of the conducted after year s, in year t, and that between these material discussed in the fifth and sixth subsections of two years the health program is expanded to new Part I of this chapter. The discussion is in the same communities. Suppose further that by year t the new order as those subsections: program evaluation comes health programs have existed long enough to signifi- first,30 followed by estimation of structural models. candy affect child height. Finally, assume that equation 1 also applies in period t and that the error term has Program Evaluation three additive components: Consider the example in the text on how to assess the impact of a new health program on child health. A (2) UiC = VI + W; + ey.l for T = s, t household survey conducted in year s provides infor- mation on height-for-age for a random sample of chil- where vI represents unobserved community-level dren. The community data indicate which communi- determinants of child height that are constant over ties have implemented the program. Let H. be the time, w. is a child- or household-specific time-invari- height-for-age of child i in community c in year s and ant component reflecting parental "tastes" for child let D., be a dummy variable indicating whether the health and/or a child's "health endowment" (innate community has the health program in year s. (D = 0 healthiness), and eja represents all time-varying com- if community c does not have the health program in ponents of u;C (such as "shocks" or measurement error year s and DEc = 1 if it does.) Let a be the impact of in His). Nonrandom program placement leads to cor- the health program on height-for-age, which for sim- relation between v and DC, while selective migration plicity we assume to be the same for every child.31 leads to correlation between w and D, (for example, Ignoring the constant term and other regressors, parents with high w, which could represent tastes for we have the following simple regression equation: child health, migrate to communities with the pro- gram). These correlations imply that & OLS is a biased (1) HzrH = atDls + uici estimate of the true a. Consider the fixed effects estimator, or "differ- where uil represents unobserved factors determining ences-in-differences" estimator, U FE = (H(l) -HFM) - child height. The ordinary least squares estimate of aX, (HO() - NsJ)), where the superscript now denotes denoted by & OLS' is equivalent to the difference values of DC' - DCS.32 Intuitively, & FE is the difference between the mean height-for-age of children in com- between the change in health status over time in munities with the program and the mean height-for- communities where a new program was started, 306 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA Hj'M - Hi1), and the change in the health status over year. Suppose there are two successive cross-sectional time in communities where nothing changed, household surveys, the first conducted in year s cover- H,(0) -H 0'. As long as the placement of the program is ing a randomly selected set of communities indexed uncorrelated with the time-varying component of ui., by c and the second conducted in year t covering a (formally, E[e. - ea, I Dc, - D] =0), a FE is a consistent random set of communities indexed by (. If these two estimator of a. By differencing out vcy & FE reflects the sets of communities overlap, either partially or entire- true impact of the health program, not the program ly, one can construct a community-level panel as dis- placement rule. Similarly, because t FE uses differences cussed in the text. If there is no overlap, a community in height for the same children, variation across chil- questionnaire is needed that was administered in year dren in either endowments or parental preferences, t and that collects information on the presence of the which are embodied in uw, are also eliminated. health program and the date it was established in com- The discussion in the text casts doubt on this use munities c', as well as the same information in year t of panel data to "solve" both problems. Heckman and from all the original communities c. Using this Robb (1985) argue that alternative cross-sectional information from two successive cross-sectional sur- estimation methods exist that, unlike ordinary least veys, the first term in the expression for 5^ FE given squares, deal with both problems. For example. one above,I410) -Fi(1), is formed as the difference in aver- might argue that the placement of the new health pro- age height-for-age of children living in communities gram in communities is based in part on the proximi- where a health program was started between years s ty of the community to the capital city, and that the and t.The second term, HI() -H 0, is the difference in proximity of a household to the capital does not affect average height-for-age of children living in conmmuni- child height, given the presence or absence of the pro- ties where the health program existed either in both gram. In this case one could use selectivity correction years or in neither year. If equation 2 is valid, repeated or instrumental variables procedures on cross-section- cross-sectional data solve the nonrandom allocation al data to eliminate the influence of the correlation problem since, with successive representative samples between clinic placement and the unobservables. Of of communities, E[vg - vt] = 0. Note, however, the effi- course, the choice of instrumental variables can be ciency advantage that panel data have over repeated controversial; for example, selective migration may cross-sectional data. Using a panel, I,; is purged with imply that proximity to the capital is correlated with probability 1, while in repeated cross-sectional data it WI. The point is simply that it is not clear which is is purged only in expectation. Thus, with repeated worse, a "bad" instrument or a "bad" assumption about cross-sectional data there is an additional source of the error term. sampling variance, the difference in the mean values of An example of a "bad" assumption about the error vc. and I. This source of variance is negligible only in term is the case in which the fixed effect is not really very large samples of communities, so that the panel fixed. Consider a simple example in which placement data fixed effects estimator is more efficient.34 of the new health program is determined not only by Without information on the placement of the the level of, but also by trends or shocks in, health program in the original communities, the fixed unobserved determinants of child health, as reprcsent- effects estimator using cross-sectional data is infeasible, ed by e, in equation 2. In this case new program since neither H (1) nor HI(0) can be estimated.35 placements are correlated with changes in eCi (that is, Repeated cross-sectional data can still be used to esti- E[e,,, - e1, I D, - D] 0), so that 5^ FE is a biased esti- mate a, but this requires different assumptions about mate of ax. One potential solution to this problem is to the unobservables than those embodied in equation 2, use initial program status, D, as an instrumental vari- and generally requires more than two cross-sectional able for the change in program status, D,5 - D,5.33 As surveys. For example, assume that tt, in equation 2 with any instrumental variables procedure, the cost of exhibits first-order autocorrelation: uicf+ = Pu"o + ec+11, this approach is a loss in statistical efficiency. for t>s, p 1. Heckman and Robb show that a can Turn next to Heckman and Robb's (1985) argu- then be estimated using three successive cross-section- ment that fixed effects estimates can be obtained from al surveys done after the health program started-that repeated cross-sectional data if one has information on is, during and after year t. There is no reason to believe the presence of the health program in each survey that this specification of the error term is less plausible 307 PAUL GLEWWE AND HANAN JACOBY than equation 2. See Heckman and Robb (1985) for Structural Models of Behavior details. Following the text, consider estimation of an agricul- Selective migration is hard to handle using tural production function.A common log-linear spec- repeated cross-sectional data, regardless of whether ification is the sets of communities surveyed in successive years overlap (that is, whether a community-level panel (3) logYh, = alogL,,, + 3logK6, + ylogZh + I'h + Eh, can be formed). Because a random sample of resi- dents of any community in year t is not a random where h denotes household, t denotes time, Y is out- sample of the year s population of that community, put, L is labor, K is the capital stock, and Z is a time the fixed effects procedure outlined in the previous invariant input (such as education of the farmer).38 paragraph is not robust to selective migration.36 Thus The error term Rth1 represents the farmer fixed effect, having two (or more) cross-sections is no better than which, in addition to management ability, may reflect having one when the problem is selective migration; unobserved land quality and aspects of the commu- in both cases one must resort to correcting ordinary nity infrastructure, while %h, represents a transitory least squares regressions for selectivity bias using one "shock" to output due to, say, weather or pests. As of the many available procedures (for example, explained in the text, estimating equation 3 on a sin- Heckman 1979). The key advantage of panel data is gle cross-sectional data set using ordinary least that one knows an individual's status-that is, squares would lead to biased estimates of a, ,B and y whether the individual has access to the health pro- due to the correlation between jh and the three gram and whether he or she migrates-in both sur- input variables. vey rounds. Fixed effects estimation based on repeat- One could try to correct this problem using ed cross-sectional data can overcome the problem of instrumental variable methods, but the use of input nonrandom clinic allocation-but not selective prices for this purpose could be criticized. An alterna- migration-if one knows the community's status in tive solution is to use household panel data to elimi- both survey rounds. If one views selective migration nate the fixed effect by taking first differences of equa- as a minor problem, this is reassuring. However, tion 3 across years: Rosenzweig and Wolpin (1988) argue that selective migration may be a serious problem. (4) AlogYh, = otAIogLh, + J3AlogK,, + AEh, Finally, consider the discussion in the text of fixed effects estimation based on retrospective data. For where AlogYh, = logYh, - logY,, and so forth, for years households with at least two children, one (child i) t and s (t > s).The time-invariant terms logZh and Rh born before the program was started and another drop out of equation 4, so y cannot be estimated (childj) born after it was started, a can be estimated (except under strong assumptions; see Hausman and by subtracting the average difference in the height- Taylor 1981). for-age of such sibling pairs from the average differ- Estimating equation 4 by ordinary least squares ence in height-for-age of sibling pairs in communities yields unbiased estimates of a and fi only if there is no with no change in program status across siblings. In correlation between the time-differences in the inputs terms of equation 2, this estimator differences out v, (AlogLh, and AlogKh6) and the time-differences in the but not necessarily wu, since siblings may have different transitory "shock" term (Ae6h).As discussed in the text, height endowments. The problem with this "within- this is unlikely to be the case.To get around this endo- household" estimator is that parents with an especial- geneity problem, one could use instrumental variables ly sickly first child (a low wtV) will be more likely to to predict AlogLh, and AlogK8,. Valid instruments migrate to the community with the health program, so would include wages and prices (perhaps collected it does not correct for selective migration. Excluding from community-level surveys),39 as well as the period recent migrants from the sample would not help s capital stock and other permanent characteristics of unless one also corrected for the resulting selection the farm that were determined prior to the realization bias. Alternatively, one could use program status prior of £hs. If more than two rounds of data are available, to the birth of child i as an instrumental variable, as values of L and K lagged two or more periods can also discussed above.37 be used as instruments. 308 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA Finally, consider estimation of dynamic relation- a concern. In a low-income setting, child mortality ships. An example is child height (or weight), for which can be high, so that many of the original children in a a production function might be specified as follows: sample may die. Child death is a form of nonrandom attrition if it is correlated with the child health (5) H., = pH1,_ + yXiN,- + v, +U¾,,. endowment v. and unobserved health "shocks" w i'41 Once again, differencing the data removes the former Height in period t depends on height in period t-1, error component but not the latter, so that panel attri- nutrition inputs in period t-1 (Xil), and an error term tion may induce selection bias; on average, children that we have decomposed into permanent and transi- who survive may have a different distribution of health tory components (see, among others, Blau, Guilkey, shocks than children who die.42 Appropriate methods and Popkin 1996). Almost any dynamic decision rule, to deal with this bias are currently an active area of such as the choice of X, as a function of Hi, l fits into research. this framework. A final dynamics issue does not concern structur- When using panel data to estimate dynamic rela- al models, but is of interest nevertheless: the persist- tionships, the appropriate estimation procedure ence of poverty over time. Consider movements into depends on whether one believes v. is correlated with or out of poverty, defined as a minimal level of annu- the other regressors in equation 5. If so, we can again al household per capita expenditures. Let C , be the take first-differences to eliminate vi, so that the term per capita expenditures of household h in year t, and AHt,= H=_ - Hj,_2 appears on the right-hand side of let z be the poverty line.The probability that a house- the equation, along with AX,,, and Awit. Three rounds hold with a given per capita expenditure level today of panel data are needed now, since the regression will fall below the poverty line next year, p is involves Hi,, Hj,_, and Hi,2' 40 Also note that AH,,_ is Pr[Ch,+i < z I Ch]. A reduced form equation for the correlated with Awu, by virtue of the fact that H,_1 is determinants of expenditures is: correlated with w.t,_. Moreover, since the choice of Xj, l may depend on the realization of wit_1, AX2tt_may (6) C,,t = 5'Xh, + uhf also be correlated with the error term. The standard solution is to use Hi, 2 and Xi,-2 as instrumental vari- where the X variables are observable household char- ables, as these are correlated with AH and AX acteristics. To estimate ph we need to know what respectively, but uncorrelated with Aw,,. One potential fraction of the variance of uh, is permanent (unob- problem with this strategy is that these instrumental served endowments of the household) and what frac- variables may have weak predictive power for AH,,_ tion is transitory (luck in any given year).This sort of and AX,t_,. decomposition is impossible with cross-sectional data The presence of measurement error in H and X but can be done with panel data. For example, let raises the data requirements for unbiased estimation of uht = +h I e1,t, with e,t + peht-I + vht, which allows for equation 5. If measurement error is serially uncorre- persistence in the transitory component (see, among lated, H,,,3 and Xjt_3 are sensible instruments, but this others, Lillard and Willis 1978). For simplicity, suppose requires at least four rounds of data. Alternatively, as well that the consumption shock, vt, is independent repeated measures of child anthropometric status and identically distributed normal. Using panel data it taken at the same point in time can be used to deal is straightforward to estimate the household-specific with measurement error. For example, rather than permanent component (w/,), the shock variance (cv '), using lagged height (H.,-,) as an instrument for AH,,_, and the persistence parameter (p).With these estimates one can use lagged weight, since weight is measured one can calculate differently than height but is strongly correlated with it. There is no analogous approach to measurement ( - )(8x,+t^ -C. error in X, although prices lagged two periods back (7) it-,t = could be used as instruments (as in the agricultural production function example above). Given that a multiround panel is required to esti- where (D is the standard normal cumulative density mate equation 5, nonrandom panel attrition becomes function and hats denote estimated parameters. 309 PAUL GLEWWE AND HANAN JACOBY Unfortunately, while this type of analysis appears to be 8. All of the potential instruments mentioned in Appendix 23.1 an important advantage of panel data, there is one seri- would be valid in the case of (serially uncorrelated) measurement ous problem. As mentioned in the text, measurement error in both inputs, with the exception of the first-period capital error in C0, (which is quite likely, as explained in stock. Chapter 5), prevents unbiased estimation of 6so,, so 9. Some of these dynamic relationships are not structural ones, that p >, cannot be estimated. but the general point still holds that policymakers are quite often interested in a wide variety of dynamic relationships. Notes 10. In the United States both the Panel Study of Income Dsnamics and the Current Population Survey have conducted val- The authors would like to thank the following people for their idation studies of earnings reports based on matched data from very useful comments on previous drafts of this paper: Martha employer or tax records. Unfortunately, these validation methods Ains-worth, Julie Davanzo, Angus Deaton, Elizabeth Frankenberg, would be of limited use in most developing countries, where wage Margaret Grosh, John Haaga, Graham Kalton, Fiona Maclntosh, income is less prevalent and recordkeeping by employers and tax Alberto Martini, Juan Munoz. Mark Rosenzwveig, Kalanidhi authorities is less reliable. Subbarao, and Duncan Thomas. 11. For simphcity, the discussion here assumes that each house- 1. This follows directly from the formula for the variance of the hold lives in a dwelling, which implies that there is no homeless or difference of two means, y, and y-, which is Var(71) + Var(y,) - nomadic population. In most developing countries the vast major- 2p(Var(7,)Var(7y))05, where p is the correlation between y1 and 72 ity of households live in some kind of fixed dwelling, hoxvever flim- (which equals the correlation betveen y, and y2). In repeated cross- sy. However, in some countries substantial homeless or nomadic sectional surveys p 0 O because the two samples are independent of populations may exist. The difficulties associated with sampling each other. Note that this argument assumes no measurement such populations occur whether or not panel data are collected, so error. Although isieasuremiient error comphcates the analysis, in issues concerning homeless or nomadic populations are not con- general it does not alter the result that for a given sample size, panel sidered further in this chapter. See United Nations (1993) for a use- data yield more precise estimiiates than repeated cross-sectional sur- ful discussion of how to draw samples and conduct surveys for such veys (see Ashenfelter, Deaton, and Solon 1986). populations. 2. The use of this specific example does not imply that one can 12. Examples of dwellings that may move are houseboats, tent- analyze only new programs. Changes in the characteristics of exist- like dwellings used by nomads, and very simple shantytown ing programs, such as improvements in the quality of services, can dwellings constructed of boards, cardboard, sheet metal, and other also be studied in the same way. scrap materials. In most countries such dwellings are either rare (for 3. All versions of the migration module presented in Chapter example, houseboats) or seldom move (for example, shantytown 16 identifW recent migrants. dwellings).Thus problems arising from dwellings that move will not 4. In some cases panel data are not even necessary at the com- be discussed further in this chapter. moiuity level to implement the fixed effects estimator. Heckman 13. This listing operation will also provide information about and Robb (1985) argue that fixed effects estimates can be obtained rates of population growth since the previous survey in each pri- from repeated cross-sectional data if there is information on the mary sampling unit.This information may be needed to adjust the presence of commnunity prograins in each survey. Yet selective samphng weights associated with each household. Such adjustments iigration is harder to handle with repeated cross-sectional data, are straightforward and will not be discussed further in this chapter. even if a community-level panel can be formed. See Appendix 23.1 14. The other way an individual can leave the sample popula- for further details. tion is by entering an institutional living arrangement, such as a 5. A inore precise definition in terus of economic theory is that military barracks, a nursing home, a college dormitory, or a prison. the paranmeters of structural models are the fundamental compo- Due to the difficulties of covering such populations, most house- nents of either technological relationships (such as production hold surveys deliberately exclude them. A problem that arises when ftsnctions) or household (or individual) preferences. survey planners want to follow the same individuals over time is 6. Most of this discussion also applies to the estimation of con- how to find them again after they leave the institutional living ditional profit funictions and the analysis of nonagricultural arrangement and rejoin the general population. enterprises. 15. In some cases, especially those in which cross-sectional sur- 7. See Chapter 26 for a brief explanation of instrumental vari- veys are done annually, listing operations are done only every 3-5 able methods in econometrics. For a more thorough exposition see years. This practice is unwise because it can lead to unrepresenta- a standard econometrics textbook such as Greene (2000). tive samples (except in societies with very loxv mobility). 310 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA 16. Actually such migrants are not completely lost. Household may not be so costly;Thomas, Frankenberg, and Smith (1999) report members, friends. and neighbors who remain in the area can pro- that the cost of "completing a case" (finding and reinterviewing an vide some information about these individuals, including their cur- individual or a household) of a long-distance mover is only about rent place of residence and their current economic activity; such 50-60 percent higher than the cost of completing a case for a information may keep more distant migrants in the sample for household that did not move or moved only a short distance. some purposes. 27. Here large economies of scale may be possible. In particular, 17. Calculating adjusted sampling weights can reduce bias due after an LSMS survey stops following individuals, it is possible to to attrition, but it cannot completely remove such bias because implement a separate, specialized survey that attempts to follow households and individuals who drop out of the sample will almost them for a further period of time. This would be much less expen- certainly differ from those who remain in the sample in terms of sive than implementing a new survey that does not build on any both observed and unobserved characteristics. Reweighting can previous survey adjust for differences in observed characteristics but not for differ- 28. This problem could be avoided by looking only at those ences in unobserved characteristics. dwellings that are in the panel in all three survey years. However, 18. Examples of initial refusal rates from previous LSMS surveys this means that it x,vould be necessary to wait three years before are: 0.9 percent (Cote d'lvoire), 0.6 percent (Pakistan), 1.3-3.3 per- examining the data (since it is not possible to know in advance cent (Peru), and 0.8 percent (Vietnam).Two important exceptions to which dwellings will stay in the survey through the third year). these generally low refusal rates are inJamaica, where refusal rates have 29. One potential problem with this system is that if a new ranged from 5.7 to 10.4 percent (primarily due to use of non-LSMS household replaces an old household in the same dw,elling, retain- fieldwork methods) and in LSMS-type surveys recendy done in ing the same dwelling number may cause data analysts to mistak- countries of Eastern Europe and the former Soviet Union-such as enly assume that the new residents are the same household as the the Kyrgyz Republic, which had a refusal rate of 3.8 percent in 1996. old residents. This problem is best avoided by making clear in the 19. Specific examples were discussed in the first section of this survey documentation that the only way to verify that the same chapter. One general benefit of following individuals who move is people live in a given dwelling at txvo points in time is to check the that it reduces the magnitude of any bias caused by sample attrition. data collected expressly for the purpose of making such matches. 20. To account for new dwellings in Lima, the original sample 30. This discussion of program evaluation is at an introductory of dwellings wvas supplemented using a new sample of dwellings level. Two recent and very thorough expositions are Manski and drawn from previously uninhabited areas that had been settled by Garfinkel (1992) and Heckman, LaLonde, and Smith (1999). migrants between 1985 and 1990. 31. The more general specification, xvhere a varies across the 21. While the Ghana computer files included all three variables population, raises issues in interpreting the program impact. See (name, sex, and age), the Jamaica computer files included only age Heckman and Robb (1985) for a thorough discussion. and sex. 32. Assume that the program is never dismantled, so that 22. Another ICRISAT panel survey wvas done in Burkina Faso but D,, - D). -1. continued for only three years. (See Fafchamps 1993 for a descrip- 33. This is the fixed effect-instrumental variables estimator, tion.) In addition, several small-scale panel data sets were collected in which can be written as the 1980s and 1990s by researchers at the International Food Policy Research Institute. Most had small samples, but a few of these panels 6FE IV = 1-s 4( ) - i(O -(H-(v) - m() contained more than 500 households, including surveys begun in the Pt Ps mid 1980s in the Philippines and in Pakistan (Yohannes 1994). where superscripts denote values of D and p, is the proportion of 23. Many of the households interviewed in 1993 that wvere not children xvho had access to a program in year r = t,s. Intuitively, this found in 1997 were found in 1998. estimator compares the groxvth of children who could potentially 24. The 1990 Peru LSMS survey lost nearly 43 percent of sam- have had newfound access to the health program (D7 = 0) with those pled households; the percentage of individuals lost was probably xvho could never have had newNfound access since they already had the more than 50 percent. program in their community (D7 = 1), and then adjusts for the prob- 25. The figures for the NLSY were obtained from the website ability that the program was actually implemented. Foster and Rosen- maintained for that survey, http://stat.bls.gov/nishome.htm. The zxveig (1996) use a similar procedure to estimate a school enrollment PSID information is taken from Brown, Duncan, and Stafford decision rule as a function of school availability in the village. (1996). 34. In principle, the fixed effect-instrumental variables estima- 26. Recent evidence from the 1997 Indonesian Family Life tor discussed in the previous note can also be implemented on Survey suggests that following individuals who move great distances repeated cross-sectional data. 3 1 1 PAUL GLEWWE AND HANAN JACOBY 35. In the unlikely event that the environment is stable (so that Kagera Region, Tanzania." World Bank, Policy Research average child health will not change between years s and t for any Department,Washington, D.C. reason other than the introduction of health program) repeated Ainsworth, Martha, Godlike Koda, George Lwvihula, Phare Mujinja, cross-sectional data can be used to estimate a given one addition- Mead Over, and Innocent Semali. 1992. Mlleasuring the Impact of al piece of information. If we know (or can estimate) the propor- Fatal Adult Illness in Sub-Saharan Africa:An Annotated Household tion of children in the population, Pt, who had access to the health Questionnaire. Living Standards Measurement Study Working program in each survey year, ac can be estimated as Paper 90.Washington, D.C.:World Bank. (H,-H)/(p, - p), where the numerator is the average difference in Ashenfelter, Orley, Angus Deaton, and Gary Solon. 1986. Collecting child height-for-age across survey years. Panel Data in Developing Countries: Does it M1fake Sense? Living 36. One could drop communities that had no health program Standards Measurement Study Working Paper 23.Washington, by year t under the assumption that out-migration to take advan- D.C.: World Bank. tage of the program in other villages (between years s and t) is Blau, David M., David K. Guilkey, and Barry M. Popkin. 1996. important only in communities with no program to begin with.Yet "Infant Health and the Labor Supply of Mothers." Journal of this does not remove bias in the fixed effects estimator because it Human Resources 31 (1): 90-139. ignores the possibility that implementng the program may dis- Brown, Charles, Greg Duncan, and Frank Stafford. 1996. "Data courage households from (selectively) migrating out of a commu- Watch: The Panel Study of Income Dynamics." Journal of nity when they otherwise would have. Economic Perspectives 10 (2): 155-68. 37. Rosenzweig andWolpin (1986) criticize the within-house- Browning, Martin, Angus Deaton, and Margaret Irish. 1985. "A hold estimator for a related reason. Suppose one exploits variation Profitable Approach to Labor Supply and Commodity in the length of exposure of each child to a health clinic. Sibling Demands over the Life-Cycle." Econometrica 53 (3): 503-43. differences in length of exposure depend on the age differences Burtless, Gary 1995. "The Case for Randomized Field Trials in between siblings, which in turn depend on child spacing behavior Economic and Policy Research."Journal of Economic Perspectives of parents. Rosenzweig and Wolpin argue that because parents may 9 (2): 63-84. adjust birth intervals in response to the realization of child-specific Duncan, Greg, and Martha Hill. 1985. "Conceptions of endowmnents, length of exposure to the clinic may be correlated Longitudinal Households: Fertile or Futile?"Journal of Economic with these endowments.Therefore, the within-household estimator and Social Measurement 13 (3-4): 361-75. would produce biased estimates of aX. Duncan, Greg, and Graham Kalton. 1987. "Issues of Design and 38. This equation is derived by taking logarithms of both sides Analysis of Surveys Across Time." International Statistical Review of the following Cobb-Douglas production function: 55 (1): 97-117. Y= LxK Z2'et". Fafchamps, Marcel. 1993. "Sequential Labor Decisions Under 39. Wages and prices are valid instruments here, unlike in equa- Uncertainty: An Estimable Household Model ofWest-African tion 3, because unobserved community infrastructure, embodied in Farmers." Econometrica 61 (5): 1173-97. g,,, does not appear in equation 4. Foster, Andrew, and Mark Rosenzweig. 1996. "Technical Change 40. For more complicated dynamics, such as second-order auto- and Human Capital Returns and Investments: Evidence from correlation in height, more rounds are needed. the Green Revolution." American Economic Review 86 (4): 41. A 1997 paper by Lee, Pitt, and Rosenzweig did not find evi- 931-53. dence of bias due to attrition in the form of child mortality. Frankenberg, Elizabeth, Duncan Thomas, and Kathleen Beegle. 42. A similar problem arises xvith education production func- 1999. "The Real Costs of Indonesia's Economic Crisis: tions in that children who perform poorly in school tend to drop Preliminary Findings from the Indonesian Family Life out. In this case, hoNvever, school dropouts can be followed and test- Surveys." Labor and Population Program Working Paper Series ed. See Chapter 8 on education. 99-04. RAND Corporation, Santa Monica, Cal. Glewvve, Paul, and Gillette Hall. 1998. "Are Some Groups More References Vulnerable to Macroeconomic Shocks than Others? Hypothesis Tests Based on Panel Data from Peru." Journal of Ainsworth, Martha, ed. 1999. "Using the Kagera Health and Development Economics 56 (1): 181-206. Development Survey Data Sets."World Bank, Policy Research Greene, William. 2000. Econometric Analysis. Upper Saddle River, Department,Washington, D.C. NY: Prentice Hall. Ainsworth, Martha, Susmita Ghosh, and Innocent Semali. 1995. Griliches, Zvi, and Jerry A. Hausman. 1986. "Errors inVariables in "The Impact of Adult Deaths on Household Composition in Panel Data."Journal of Econometrics 31 (1): 93-118. 312 CHAPTER 23 RECOMMENDATIONS FOR COLLECTING PANEL DATA Grosh, Margaret, and Juan Mufioz. 1996. A Manualfor Planning and McMillan, D., and R. Herriot. 1985. "Toward a Longitudinal Implementint tihe Living Standards Mleasurement Study Survey. Definition of Households." Journial of Economic and Social Living Standards Measurement Study Working Paper 126. Mleasurement 13 (3-4): 349-60. Washington, D.C.:World Bank. Mundlak, Yair. 1961. "Empirical Production Functions Free of Grossman,Jean Baldwin. 1994. "Evaluating Social Policies: Principles Management Bias."Journal of Farmit Economics 43 (1): 45-56. and U.S.Experience." World Bank Researchl Observer9 (2): 159-80. National Council ofApplied Economic Research. 1986. Saving and Haaga,John,Julie DaVanzo, Christine Peterson, and Tey Nai Peng. Investment Beliaviour of Rural Households in India:A Longitudinal 1993. "Txvelve-Year Follow-Up of Respondents and Their Study 1970-71 and 1981-2. New Delhi. Children in a Panel Survey in Peninsular Malaysia." Labor and Newman John, Laura Rawlings, and Paul Gertler. 1994. "Using Population Program Working Paper Series 93-19. RAND Randomized Control Designs in Evaluating Social Sector Corporation, Santa Monica, Cal. Programs in Developing Countries." World Bank Researci Hanushek, Eric A. 1986. "The Economics of Schooling: Production Observer 9 (2): 181-201. and Efficiency in Public Schools."Jourmal of Economic Literature Okrasa, Wlodek. 1999. "Who Avoids and Who Escapes from 24 (3): 1141-77. Poverty during the Transition? Evidence from Polish Panel Hausman, Jerry A., and William E. Taylor. 1981. "Panel Data and Data 1993-96." Policy Research Working Paper 2218. World Unobservable Individual Effects." Econometrica 49 (6): 1377-98. Bank, Poverty and Human Resources, Development Research Hausman, Jerry A., and David A. Wise. 1979. "Attrition Bias in Group,Washington, DC. Experimental and Panel Data: The Gary Income Maintenance Pitt, Mark M., Mark R. Rosenzweig, and Donna M. Gibbons. Experiment." Econonietrica 47 (2): 455-73. 1993. "The Determinants and Consequences of the Placement Heckman, James J. 1979. "Sample Selection Bias as a Specification of Government Programs in Indonesia:' Wiorld Bank Econiomic Error." Ecotiometrica 47 (1): 153-61. Review 7 (3): 319-48. Heckman, James J., and Richard Robb, Jr. 1985. "Alternative Rosenz-weig, Mark R., and Kenneth Wolpin. 1986. "Evaluating the Methods for Evaluating the Impact of Interventions." In Effects of Optimally Distributed Public Programs: Child James J. Heckman and Burton Singer, eds., Longitudinal Health and Family Planning Interventions." American Economic Analysis of Labor MVarket Data. New York: Cambridge Revieu' 76 (3): 470-82. University Press. . 1988. "Migration Selectivity and the Effects of Public Heckman,JamesJ.,andJeffreyA. Smith. 1995."Assessing the Case for Programs".Journal of PUblic Economics 37: 265-89. Social Experiments."Journal of Economic Perspectives 9 (2): 85-1 10. Russian Longitudinal Monitoring Survey. 1999. "Sampling Heckman, James, Robert LaLonde, and Jeffrey Smith. 1999. "The Overview." From the website of the Carolina Population Center Economics and Econometrics of Active Labor Market at the University of North Carolina at Chapel Hill: Programs." In 0. Ashenfelter and D. Card, eds., Hatidbook of http://wxvw.cpc.unc.edu/projects/rlms/project/sampling.htn1. Labor Economiics. Amsterdam: North Holland. Subramanian, Shankar, and Angus Deaton. 1996. "The Demand for Honore, Bo. 1992. "Trinumed LAD and Least Squares Estimation of Food and Calories."Journal of Political Econoniy 104 (1): 133-62. Truncated and Censored Regression Models with Fixed Thomas, Duncais, Elizabeth Frankenberg, and James Smith. 1999. Etfects" Econometrica 60 (3): 533-65. "Lost but not Forgotten: Attrition in the Indonesian Family Kalton, Graham, and Constance Citro. 1993. "Panel Surveys: Adding Life Survey." RAND Corporation. Santa Monica, Cal. the Fourth Dimension." Survey Methodology 19 (2): 205-15. Thomas, Duncan, Victor Lavy, and John Strauss. 1996. "Public KHDS (Kagera Health and Development Survey) Research Team. Policy and Anthropometric Outcomes in the Cote d'Ivoire." 1999. "Using the Kagera Health and Development Survey Journal of Public Economics 61 (2): 155-92. Data Sets" World Bank, Development Research Group, United Nations. 1993. Sampling Rare amid Elusive Populatiomis. Washington, D.C. National Househlold Survey Capability Programn. Department for Lee, Lung-Fei, Mark Pitt, and Mark Rosenzweig. 1997. "The Economic and Social Information and Policy Analysis. Nexv Effects of Improved Nutrition, Sanitation, and Water Quality York. on Child Health in High-Mortality Populations." Joumral of Walker, T., and J. Ryan. 1990. Village and Houselhold Ecoiotumies in Econometrics 77 (1): 209-35. India's Senmi-Arid Tropics. Baltimore, Md.: Johns Hopkins Lillard, Lee, and Robert Willis. 1978. "Dynamic Aspects of Earnings University Press. Mobility." Econometrica 46 (5): 985-1012. World Bank. 1999. "Basic Documentation: Kagera Health and Manski, Charles, and Irvin Garfinkel. 1992. Evaluatitig We!fare and Development Survey." World Bank, Development Research Training Programmns. Cambridge, Mass.: Harvard University Press. Group,Washington, DC. 313 PAUL GLEWWE AND HANAN JACOBY Yohannes,Yisehac. 1994. "Documentation and Storage of FCND International Food Policy Research Institute, Food and Data sets Collected from Primary Sample Surveys as of 1982." Consumption Division, Washington, DC. 314 Intrahousehold Analysis / 4 Nobuhiko Fuwa, Shahidur R. Khandker,Andrew D. Mason, and Tara Vishwanath Why perform intrahousehold analysis? The efforts of developing country governments to reduce poverty are often based on analyses of the determinants of poverty and of their effects on eco- nomic growth and development. Many of these studies use the household as the unit of analysis based on the assumption that the household is the appropriate level at which to target income transfers or other policies to enhance welfare, promote growth, or narrow income differentials. However, researchers and policy analysts are increasingly recognizing that, under certain circum- stances, the outcomes of policies can have unforeseen consequences if patterns of intrahousehold resource allocation are not taken into account in the design of policies and projects. Analyzing how a household's resources are allocated hold aggregate measures can result in a distorted pic- among its members requires individual-level data on ture of poverty if some households whose average per each household member. Analysis of these data is fun- capita expenditures are above the poverty line contain damentally useful to policymakers in two ways. First, members (for example, women or the elderly) whose analysis of individual welfare outcomes (for example, standard of living actually falls below the poverty line. in terms of consumption or health) can signal to pol- Moreover, data from many countries around the world icymakers whether there is a need for interventions indicate that there are significant (and sometimes per- targeted to specific groups or individuals within sistent) gender disparities in human capital, access to households. Second, such analysis can help policymak- productive resources such as credit and agricultural ers assess whether existing policy interventions that are extension services, and participation in economic targeted to individuals or groups within households activities (World Bank 1995). Thus understanding the (such as children or the elderly) are succeeding or fail- intrahousehold distribution of basic goods and servic- ing. These points are illustrated in turn below with a es such as food, health inputs, education, and credit can few examples. help policymakers set strategic policy priorities. Intrahousehold resource allocation analysis can be Analysis of intrahousehold resource allocation is critical to the measurement of poverty and inequality. also important for understanding how policy inter- Haddad and Kanbur (1990) reported from Philippines ventions affect the welfare of different household that estimates of poverty can differ quite members. Recent studies suggest that women and significantly-by as much as 30 percent-when meth- men differ in how they allocate resources among ods using household-level and individual-level data are household members and that women may tend to compared.' Their findings suggest that using house- allocate more resources to children than men do.2 If 315 NOBUHIKO FUWA,SHAHIDUR R. KHANDKER,ANDREW D. MASON,ANDTARAVISHWANATH that is in fact the case, policies and programs that to take control of rice cultivation from the women. As increase women's control of resources within the a result, women's share of household income actually household may be more effective in improving chil- declined, which was contrary to the project's intention dren's welfare than policies and programs that target (von Braun and Webb 1989). In this context, house- resources at the level of the household as a whole. hold members who have something to lose (in this Moreover, policy initiatives can have unintended case, the women) may focus their productive effort on consequences if policymakers are unaware of "cross- preserving their personal control over their income person effects" in which the behavior of an individual rather than on maximizing farm profits at the house- within a household depends on behavior of other hold level (also see Jones 1986 for evidence from members. As households have their own ways of allo- Cameroon). cating resources, it is possible that household behavior These examples make clear that understanding may alter or offset the intended impact of a policy ini- how resources are allocated within the household and tiative (Schultz 1989). One classic example of this type how changes in the policy environment affect these of outcome is in school feeding programs. Sometimes allocations is critical if policymakers are to design and when children receive meals at school, their food allo- implement policies that achieve their objectives and cations at home are reduced and reallocated to other are effective in improving individuals' welfare. (See household members. This reduces the effectiveness of Alderman and others 1995 and Haddad, Hoddinott, such programs, the aim of which is to increase chil- and Alderman 1997 for more detailed discussions of dren's food intake (Beaton and Ghassemi 1982). the policy significance of intrahousehold resource Similarly, microcredit programs that are intro- allocation.) duced to increase women's involvement in market- The primary purpose of this chapter is to discuss based production may have a variety of unanticipated how the draft LSMS questionnaire modules, which effects on the family. For example, as women under- appear in earlier chapters of this book, can be designed take more market-oriented activities, girls' housework to support intrahousehold analysis.The first section of responsibilities often increase, as they are often respon- the chapter discusses the modeling and empirical sible for taking over the housework formerly done by issues involved in intrahousehold analysis. The second their mothers. In analyzing microcredit programs in section exanines the definition of a household and of Bangladesh, Pitt and Khandker (1998) found that giv- headship and looks at changes in household composi- ing women access to credit tends to increase fertility tion. The third section describes how intrahousehold and decrease contraceptive use, while giving credit to analysis can be supported by the data that would be men tends to have exactly the opposite effect. While collected in the draft LSMS questionnaire modules giving women access to credit increases the schooling presented in this book. The third section also high- of both boys and girls, the gender impact is not sym- lights the limits of LSMS-type surveys in pursuing nmetrical, with the impact being much more pro- intrahousehold analysis and shows how designs that go nounced for boys than for girls. Thus, since the alloca- beyond regular LSMS-type surveys can gather the data tion of time and the allocation of other resources necessary to address more specific intrahousehold within the household are interrelated, it is necessary issues. The fourth and final section summarizes the for analysts to examine not only the direct impact of chapter and advises those who will implement LSMS- policies on the outcomes in which they are interested type surveys in the future. but also any indirect impacts of the same policies. In some extreme cases, even interventions direct- Modeling Intrahousehold Resource Allocation ly targeted to women may backfire because of the way Behavior households allocate their resources among their mem- bers. For example, a recent project on rice cultivation This section reviews the conceptual and empirical in Gambia was designed to increase the productivity issues involved in intrahousehold analysis.3 It begins of rice, which was cultivated by women, by giving by discussing how to model household resource allo- these women incentives to produce for the market cation behavior and goes on to consider the issues rather than for home consumption. Rice yields involved in defining household membership and increased as a result of the project, but this caused men headship. The section concludes with a brief exami- 316 CHAPTER 24 INTRAHOUSEHOLD ANALYSIS nation of issues involved in empirical econometric Consequently, in the last decade, other models of estimation. household decisionmaking have increasingly been developed in the literature. These more general How to Model Intrahousehold Resource Allocation Behavior approaches, called "collective" models, incorporate the In developing countries, particularly in rural areas, the preferences and relative decisionmaking power (or vast majority of economic activity takes place within bargaining power) of different household members in the household. Decisions about education, labor force the models of household decisionmaking (McElroy participation, savings and asset accumulation, invest- and Horney 1981; Manser and Brown 1980; ment, marriage, and fertility, which are made within Chiappori 1988; Lundberg and Pollack 1993; Carter households, critically affect growth and the distribu- and Katz 1997). tion of income across households. Therefore, it is very As long as different models produce the same pre- important for analysts and policymakers to understand dictions about the relationship between policy instru- the structure of the household and how it varies across ments and intrahousehold outcomes, which model is households and over time.4 used in any analysis will not be of great significance to In the vast majority of the theoretical literature, policymakers in terms of how they design policy the household is treated as a collection of individuals interventions (Hoddinott 1992). For example, who behave as if they are in complete agreement on Rosenzweig and Schultz (1982) observed that the how best to combine their time and the other probability of survival in rural India differs by gender resources at their disposal to produce goods at home and that the probability of survival is higher for and in the market for attaining the household's maxi- females in areas with greater opportunities for female mum possible welfare. Under this framework, house- employment. Under the unitary model framework, hold behavior is analyzed as if "the household" is a sin- this observation can be interpreted as a result of gle decisionmaker (Becker 1991; Samuelson 1956). households responding to their perception of the dif- This unitary model has been used to analyze how cer- ferent returns yielded by investments in boys as tain types of intrahousehold outcomes in resource opposed to girls. Under the collective model frame- allocation, including gender disparities, arise (Berhman work, the same observation can be seen as the result of 1988a, 1988b; Rosenzweig and Schultz 1982; Pitt, household bargaining between the husband and the Rosenzweig, and Hassan 1990). This theoretical wife. When women have higher earning potential, framework has also been used to analyze some impor- they are more able to influence how the household's tant cross-person effects such as the interdependence resources are allocated (Folbre 1984; Rosenzweig and between a husband's and his wife's respective decisions Schultz 1984). However, regardless of the theoretical on whether and how much to participate in the labor framework, the policy implication is the same-that market. increasing women's employment opportunities However, analysts are increasingly recognizing sig- reduces the gap between boys and girls in the proba- nificant limitations of the unitary model approach to bility of survival. household decisionmaking. First, the theoretical Collective models do have some additional policy assumptions that underlie the unitary model-such as implications that the unitary model does not suggest. that all members share the same preferences or that a For example, collective models predict, as the unitary single member dictates all intrahousehold model does not, that which household member allocations-are quite restrictive and thus may not receives the income affects how that income is subse- accord with reality (Bergstrom 1989; Folbre 1986). quently allocated among alternative consumption Second, certain phenomena such as domestic violence goods (Browning and others 1994; Bourguignon and seem to contradict the unitary view of household others 1993; Thomas 1990, 1993; Lundberg, Pollack, decisionmaking. Third, an increasing amount of and Wales 1997). econometric testing of some underlying assumptions Furthermore, collective models show that any of the unitary model has rejected those assumptions institutions or social norms that affect the access of (Bourguignon and others 1993; Browning and others different household members to economic resources 1994; Thomas 1990, 1993; Thomas and Chen 1994; can significantly affect intrahousehold resource alloca- Schultz 1990).5 tion. These include the rules, norms, and laws that 317 NOBUHIKO FUWA, SHAHIDUR R. KHANDKER,ANDREW D. MASON, AND TARA VISHWANATH govern marriage (McElroy 1992), inheritance, access member's status within the household, the higher the to common property resources in communities productivity of his or her plots (after controlling for (Haddad and Kanbur 1992), and use rights for agri- other factors).7 cultural land (Platteau 1996; Vishwanath and others However, for analysts to test the validity of these 1996). alternative models of household behavior, they need Yet another major implication of some collective data on income and assets at the individual level rather models is that the outcomes of the intrahousehold than at the household level. They may also require resource allocation process may not be efficient. If information on the variables (including marriage and resource allocation within households is indeed ineffi- inheritance customs as well as rules about land access cient, this could signal the need for policy interven- or common property access) that can affect the intra- tions.While the assumption of efficiency in the alloca- household allocation of household resources. tion of consumption has not generally been challenged in the empirical literature, it has been chal- Empirical Issues lenged on the production side. It has been document- There are three types of econometric models for ana- ed that there is a great deal of gender disparity in pro- lyzing intrahousehold behavior: structural models, duction, which varies widely across different areas. reduced form models, and conditional demand mod- Men and women may grow different crops and differ- els (see Chapter 26 of this book on econometrics; ent amounts of these crops. In Sub-Saharan Africa Strauss and Thomas 1995; Behrman 1992; Pitt, women grow almost 80 percent of all food. Rosenzweig, and Hassan 1990; and Pitt 1997). One issue that has been widely discussed is whether or not inputs for production are allocated STRUCTURAL MODELS. Estimating "structural" relation- efficiently across different plots. Some recent studies ships makes it possible for analysts to infer whether have found that the intrahousehold allocation of pro- household members favor the boys or the girls of the ductive resources between men and women leads to household. If there is a preference, analysts can assess production inefficiencies. For example, Jones (1986) whether this preference (as reflected in the intrahouse- examined the work done by Massa xvomen in North hold allocation of schooling and nutrients) affects the Cameroon. She found that because women controlled children's earnings and health outcomes (Behrman the income from their own plots but not from pro- 1992, 1988a, 1988b).Analysts can also discover to what duction on their husbands' plots, they spent more time extent preferences and income maximization motives on their plots and less time on their husbands' plots (which are usually assumed to be based on a household than would have been optimal if they were trying to consensus) explain any inequality in food intake among maximize total household income. Thus, as suggested members of the same household (Pitt, Rosenzweig, earlier, they were more interested in preserving their and Hassan 1990). However, using this method involves own personal income than in maximizing total house- a few difficulties. For example, it is necessary to control hold income. the endogeneity of right-side variables, and it is diffi- Many studies have supported this finding in other cult to measure and costly to collect some of the cru- countries. Although some of these studies have been cial input variables (such as individual-level data on qualitative, they strongly suggest that labor is not food intake and time use) usually required to make a pooled within households and that labor allocations structural estimate of intrahousehold outcomes. across plots are inefficient. Udry (1996) examined yield differentials for identical crops from men's and REDUCED FORM MODELS. Reduced form relationships women's fields in Burkina Faso and found that are mostly easier to estimate than structural form rela- women's fields had lower yields than men's yields and tionships and are appropriate for examining certain that the allocation of inputs among these fields was types of intrahousehold cross-person links, such as how inefficient. This implies that the value of household a change in the wife's wage rate affects the participa- production could be increased significantly simply by tion of her husband and children in the labor force. reallocating factors of production that are currently used on men's plots to women's plots.6Vishwanath and CONDITIONAL DEMAND MODELS. Conditional others (1996) also suggest that the higher a household demand functions can be used to analyze the 318 CHAPTER 24 INTRAHOUSEHOLD ANALYSIS reduced form demand relations that depend on (or What Constitutes a Household? Iare conditional on; using an econometric jargon) behavioral variables such as the utilization of or par- No matter how the decisionmaking process is mod- ticipation in government programs (for example, eled for analyzing intrahousehold resource allocations, schooling, health services, or credit schemes). In esti- a more fundamental question is who makes up the mating conditional demand functions, price vari- household and why. Should the definition of a house- ables are commonly used as identifying instruments hold be a group of individuals who live together under to control for the endogeneity of right-side behav- one roof and share a common kitchen or cooking pot, ioral variables (see Chapter 26 on econometrics). as assumed in most household surveys? The definition However, in the case of intrahousehold analysis, esti- of a household can be very complex and is often cul- mating conditional demand functions at the individ- ture- and location-specific. Households are formed ual level, the analyst may find it difficult or even partly by economic incentives but often also reflect impossible to find a sufficient number of suitable social and cultural norms.The household is not a stat- identifying instruments. See Pitt (1997) for a ic institution, as its structure and composition varies in detailed discussion of the econometric estimation of response to internal and external changes. The com- conditional demand functions in the context of plexity is best stated in Guyer (1980): "A household is intrahousehold analysis. a particularly dense center in a network of exchange relationships."As such, any fixed definition of a house- SuMMARY. As discussed in detail in Chapter 26 (on hold may be an oversimplification and may, therefore, econometrics), evaluating the effects of policy inter- lead analysts to draw erroneous conclusions when ventions is a complex issue that not only involves the evaluating the intrahousehold impact of policies and design of surveys and methods of analysis but must also projects. take into account the design and the nature of the pro- The definition of the household that is usually grams to be evaluated. In cases where the main aim of used in LSMS surveys can impose serious constraints a particular household survey is to evaluate an inter- on empirical analysis because of a lack of information vention, the ideal scenario would be to incorporate the about other individuals or families who influence survey into the implementation schedule of the pro- decisions and outcomes within the household. For gram itself. Likewise, the program might incorporate example, in many households in developing countries, some randomized experimental components (see extended families live within the same household, and Newman, Rawlings, and Gertler 1994 for examples). several households live within the same compound. In such circumstances it may not be vital to collect a Guyer (1986) provides examples from West Africa in set of potential instrumental variables in the household which households within a kinship group may farm survey. separately but consume many goods jointly. When neither panel data nor randomized experi- mental data are available, another approach may be to Changes in Household Composition look for exogeneity in the eligibility criteria associat- The composition of households can change over time ed with the program. In such cases, even when a base- as a result of marriage, divorce, migration, partition, or line survey is not available, it is often possible to eval- the combining or linking of previously separate uate the impact of a program if there is some random households (see Foster 1993 who uses data from or predetermined criterion that restricts participation Bangladesh).9 Moreover, there are a number of factors in the program or eligibility for program benefits.8 that can influence people's decisions to come togeth- This approach is the so-called "quasi-experimental er as a household or to alter household composition survey method" (see Pitt and Khandker 1998). over time. For example, social norms and rules regard- However, in the case of general multitopic household ing inheritance-and thus access to resources-can surveys like the LSMS, it may be prudent to collect affect individuals' incentives to live together. The data on a set of possible "instruments"-variables that inability of people to borrow may cause them to share can be used in the analysis to identify the impact that a household with their parents or siblings. The nature a given program has on household and intrahousehold of land and labor contracts may affect people's deci- outcomes. sions to extend their households (Walther and Nugent 319 NOBUHIKO FUWA, SHAHIDUR R. KHANDKER,ANDREW D. MASON, AND TARA VISHWANATH 1988). And households may choose to send some of While this definition fulfills its original accounting their members to live elsewhere to insure themselves purpose, problems can arise when this definition of against market or institutional failures or to mitigate headship is used in analysis. Some of the ways in which the impact of economic shocks. the concept of household headship are used in analy- sis are in studying the incidence and depth of poverty Selecting Marriage Partners in female-headed households and the different ways in Taking household composition as given also implies which female- and male-headed households allocate that households have no control over the characteris- resources among their members. tics of the spouses of household members. On the If female-headed households are over-represented contrary, household members often choose their among the poor, it may be appropriate for policymak- spouses by the attributes that the spouses can bring to ers to specify female headship as a targeting criterion the household. Ignoring this fact in analysis may result for poverty reduction programs.12 It is often claimed in flawed interpretations of household behavior and, that female-headed households bear a triple burden: a subsequently, wrongly targeted policies. (See Weiss female breadwinner is likely to have fewer job oppor- 1997 for a general review of the economic literature tunities, earn less per hour, and have less access to eco- on marriage.) For example, if a man values educated nomic resources than a male head would; female heads children, he may choose to marry an educated tend to have a heavy time burden because they have woman. Ignoring this endogenous selection in analy- both productive and "reproductive" responsibilities sis would overestimate the influence of the wife's (such as child rearing and household chores); and schooling on their children's educational attainment, being the sole economic supporter means that a To correct for the endogenous choice of partners, female head has more people to support with her analysts need data on the characteristics of the mar- income than she would in a jointly headed household riage market10 at the time and place of the couple's (Rosenhouse 1989). marriage.11 To understand how spouses are selected in Yet it is critical to recognize that self-reported the marriage market, analysts also need data on the female-headed households can actually vary widely in characteristics of individuals who enter the marriage terms of the marital status of the head and of who market with particular endowments such as education economically supports the household. For the pur- or assets, as well as data on their parents.This informa- pose of identifying poor or vulnerable households, tion will help analysts understand what factors influ- analysts have attempted to disaggregate self-reported ence bargaining power within households, as well as female-headed households by the presence or absence postmarriage outcomes including patterns of intra- of male partners13 or to redefine female-headed household resource allocation. Because the character- households according to the degree to which the istics of the partners forming a household cannot be female head is the sole economic supporter of the regarded as exogenous, analysts must explicitly take household.14The empirical evidence on the question the selection of the marriage partner into account in of the poverty of female-headed households is mixed. their modeling. Such analysis requires more data than A recent literature review by Buvinic and Gupta is usually collected when the conventional definition (1997) concluded that evidence generally points to a ofthe household is used in a survey. positive association between female headship and poverty. However, Quisumbing and others (1995) Household Headship analyzed consumption expenditure data from 10 An important related issue is the definition of the developing countries15 and found that, based on the "household head." The original reason for identifying self-reported headship definition, female-headed the household head in household surveys was to households were not obviously poorer than house- account for all of the household members and to avoid holds headed jointly or by men, except in two cases double-counting any members in the household ros- (rural Ghana and Bangladesh).16 ter by assigning one member to be a reference person Another common analytical use of the house- (Rosenhouse 1989). Typically in LSMS surveys, the hold headship concept is in the analysis of how male household head is identified by the survey respondent, and female heads differ in terms of how they allocate who may or may not also be the household head. their household's resources (Bruce and Lloyd 1997; 320 CHAPTER 24 I NTRAHOUSEHOLD ANALYSIS Handa 1996b; Rogers 1995). For such analysis the Implications for Data Collection household head needs to be defined as the person who has decisionmaking authority over the alloca- The discussion so far has a number of implications tion of household resources (commonly proxied for for the collection of data that can support the analy- by the person's economic contribution to the house- sis of intrahousehold issues. There are implications hold). Researchers often infer that the differences not only for the types of information collected but between male and female heads in terms of their also for the quality of the information and, hence, resource allocation behavior are due to systematic how the information is to be collected.This section differences in preferences between men and women. focuses on a set of issues concerning the collection of However, perhaps a better approach for this kind of these data, including the definition of the household inquiry would be to analyze the allocation patterns and the household roster, the scope and limitations of of consumption expenditures, controlling for the dif- collecting data at the individual and plot level, the ferent economic contributions made by women and collection of data on suitable 'instruments' to be used men to the household and the extent to which they in econometric analyses, and various methods of data have control over the allocation of resources collection. (Hoddinott and Haddad 1991; Thomas 1990, 1993, It must be emphasized that analysts have only 1994). recently given much attention to intrahousehold Even when alternative definitions of headship resource allocation. As a result, some of the recom- rather than self-reported headship are used for analyt- mendations in this section concerning data collection ical purposes, there can still be problems in using a have not necessarily been well tested in the field so far. given definition of headship and analyzing the effects Also, some of the information collected at the indi- of female headship on various household or intra- vidual level rather than the household level is costly household outcomes using survey data (Bruce and (for example, data on time use and food consump- Lloyd 1997; Strauss and Beegle 1996).When compar- tion). Therefore, it is extremely important for survey ing female-headed households with households head- designers to be aware of the policy priorities so they ed jointly or by men, analysts usually assume (albeit can be selective in determining what kind of informa- implicitly) that household headship and membership tion they need to collect at the individual level. are exogenous with respect to the outcomes in which they are interested.17 However, while some women Household Roster, Household History, and lnterhousehold may become the household head because of an Links exogenous event (for example, the death of their hus- How households are formed, their composition, and band), others often become head as a result of delib- how they ought to be defined are very culture- and erate choices made by individuals such as a marriage, location-specific. Although many household surveys, a common union, or a household split. Thus the sys- including most LSMS surveys, have typically defined a tematic differences that are often observed between household as "people living or eating together," this female-headed households and those headed jointly definition imposes serious constraints on empirical or by men may not be the direct result of female analysis because it fails to give any information about headship but instead arise from other factors that other individuals or families who influence decisions jointly determine both the woman's headship and and outcomes within a household. The shortfalls of these outcomes. this definition have at least two important implications These potential problems with regard to endoge- for collecting data for intrahousehold analysis. First, it nous female headship come back to the issues of the is vital in the household roster and migration modules definition of the household, the formation and disso- to collect information on factors that cause changes in lution of the household, and changes in the house- a household's composition, on the migration history hold's composition.'8 For this and other reasons, of the household's members, and on any interhouse- despite the wide attention that female-headed house- hold networks. Second, within certain parameters it is holds have been given in policy discussions, the use- important for survey designers to take local conditions fulness of some studies using the concept of household into account when designing the questions in the headship may be questionable. household roster. 321 NOBUHIKO FUWA, SHAHIDUR R. KHANDKER,ANDREW D. MASON, AND TARA VISHWANATH Several types of information are particularly valu- holds include more than two generations) to make able for intrahousehold analysis. sure that the questionnaire records which children belong to the different sons and daughters within the HOUSEHOLD HEAD. As was discussed in the previous household, as is intended in the draft household roster section, a major problem with traditional headship introduced by Chapter 6. analysis is that since the "head" of the household is self-reported, it is often not clear what "household RECENT MOVEMENTS OF HOUSEHOLD MEMBERS. All headship" means for each household. One way of three versions of the draft migration modules intro- addressing this problem is to collect information on duced by Chapter 16 collect information on the how each household defines its household head. After migration history of household members, which can asking the respondent to identify the household head, be useful for studying how households form and additional questions could be asked such as: "What is change. For example, the modules ask questions about the main reason for this particular person to be iden- any recent movements by the members of the house- tified as the head?" Based on pretesting of survey hold (and by the household as a whole if all the mem- instruments, a possible set of answvers could be precod- bers have moved recently) including the timing of ed, including: he or she makes important input and those moves ("how long has it been since you came to output decisions; he or she is older than other house- stay here?") and the person's original location. Once hold members; he or she controls most of the house- previous locations have been identified in the migra- hold resources; he or she contributes most of the tion module, information on those localities (and the household income; and so on. distance from those localities to the household's cur- rent place of residence) can be gathered from sources CHARACTERISTICS OF THE HOUSEHOLD HEAD'S other than household surveys. NONRESIDENT RELATIVES. As we have seen in the pre- All of the versions of the draft migration module vious section, relatives who are not members of the also ask about the birthplace of each household mem- household by the conventional household definition ber, and both the standard and the expanded versions can still be vital players in the allocation of resources obtain an almost complete migration history for each within the household. For this reason it is very useful member since his or her birth. After identifying the to collect information on the age, education level, and locations of members' past residences (particularly occupation of the household head's parents and chil- their places of birth and childhood residence), it is dren as xvell as his or her spouse's parents, as is pro- possible to gather information on the characteristics of posed in the household roster chapter (Chapter 6). In such places, including rainfall variance and level, from addition, if the parents of household head and spouse sources other than household surveys. These data in do not live in the same household as the head and his turn can be used to control for household/individual or her spouse, it may be useful to add a question ask- heterogeneity in analyses of intrahousehold resource ing the rough distance between their own and their allocations. parents' residences.This can be seen as one measure of potential financial links betwveen the households. If the CONTEXT-SPECIFIC INFORMATION. Certain context- parents are no longer living, similar data should be col- specific information about the structure of the house- lected on the nearest extended family member who hold can also be important to support analysis of intra- does not live in the same household, such as a sibling. household outcomes. For example, beyond simply It can also be useful to analysts to have data on the noting whether a household member is male or assets held by these nonresident relatives, particularly if female, in some cases it may be important to collect it appears that interhousehold transfers are important information about each individual's standing within and potentially affect intrahousehold resource alloca- the household. In certain contexts in Sub-Saharan tion. In this case the household roster can be expand- Africa it may be important to disaggregate the gender ed to include similar information on relatives within a category according to the person's social status or sen- certain range, which can be determined through iority within the household or within the extended pretesting or qualitative research preceding the house- family or kinship network. This may be particularly hold survey. It is also important (especially if house- important in polygamous societies or in societies with 322 CHAPTER 24 INTRAHOUSEHOLD ANALYSIS clan and tribal structures.19 Gender differences in the place during a specific reference period (the previous allocation of resources and in other outcomes of inter- 12 months), the household may maintain interhouse- est may also vary significantly across ethnic groups, hold links with some potential donor or recipient even within a given country. This makes it crucial to households, whether or not such transfers actually take collect data on the ethnic and religious affiliation of place during the survey period. Collecting separate household members, as is proposed in the draft house- information on potential providers of transfers can hold roster introduced by Chapter 6. The extent to help analysts understand the patterns of interhouse- which collecting these or similar context-specific data hold links. For example, Pitt and Khandker (1998) col- is important in a given country context can be deter- lected and used data on potential providers in a survey mined by using qualitative methods of data collection that focused on the impact of microcredit schemes on prior to development of the survey itself (see below intrahousehold resource allocations in Bangladesh. and also Chapter 25 on qualitative methods). Individual-Level Resource Allocation Outcomes Potential Providers of Transfers In order to analyze individual outcomes, personal Given the potential importance of understanding interdependencies, and cross-person effects within interhousehold transfers in order to understand intra- households, it is necessary to collect some key data at household outcomes, one possibility is to add questions the level of the individual rather than the household. asking the household head and his or her spouse about The kinds of data most commonly collected at the potential (as opposed to actual) providers of transfers to individual level are twofold: first, who brings in the household. Although this is not proposed in income to the household and controls these resources, Chapter 6 on the household roster nor in Chapter 11 and second, to whom these resources are allocated. In on transfers and other nonlabor income, one possibili- many respects, collecting more individual-level data ty would be to ask both the head of the household and on inputs and outcomes would make it easier for ana- his or her spouse to list up to three individuals or lysts to explore intrahousehold resource allocation in households who might provide transfers to the house- greater detail. This kind of information is important hold if it were to experience an economic shock.The partly because most analytical work so far has focused head and spouse could also be asked to give informa- on how resources are allocated between sons and tion on the age, education, occupation, location daughters and, to some extent, between older and (including whether rural or urban), and land holdings younger children, but there is work to be done on the of each of these potential providers of transfers. allocation outcomes of the elderly and the disabled as Gathering this additional information would have well. several advantages for analysts. First, they could use This does not imply that multitopic household these data as potential instrumental variables for iden- surveys need to collect all data on an individual basis. tifying household behavioral variables such as demand Some individual-level variables are extremely difficult for credit or participation in the labor market. Second, to measure and costly to collect. Survey designers by putting these questions to the household head and should decide how much data to collect at the indi- to his or her spouse separately, these data could also be vidual level in light of the relative benefits and costs of used as individual-specific variables, which are rela- collecting such data. tively rare. For example, it is possible that the husband Among the variables that capture intrahousehold and the wife may expect to count on different people resource allocation outcomes, many of them are inher- to provide transfers during a time of crisis. Such dif- ently individual-level variables and are thus already ferences can be exploited econometrically to enhance collected at that level in most household surveys, the intrahousehold analysis. including the LSMS. These variables include data on Third, asking about potential providers of transfers anthropometric, educational, labor market, and health would complement the questions about actual transfer outcomes. As recommended in Chapter 10, data on incomes (see modules on migration and on transfers anthropometric outcomes should be collected for and other nonlabor income) in the analysis of inter- household members of all ages, not just for children. household transfers. While the questions about actual As discussed in Chapter 8, individual-level data on the transfers capture the transactions that happened to take utilization of health services and goods are also critical 323 NOBUHIKO FUWA, SHAHIDUR R. KHANDKER,ANDREW D. MASON, AND TARA VISHWANATH for understanding intrahousehold allocations and can consumption allocations between girls and boys- also be collected relatively easily in household surveys. even in areas where a strong bias towards boys is On the other hand, some outcome data are very believed to exist (see Ahmad and Morduch 1993 on difficult and costly to collect at the individual level and Bangladesh, Subramanian 1994 on northern India). thus should not be gathered on a routine basis in Some experts would argue that these results are an nationally representative, multitopic household surveys indication that such indirect approaches are not cap- like the LSMS. Collecting individual-level data on all turing the full and true picture of the way resources consumption goods would be problematic. For one are allocated within households.22 thing, not all consumption items can be assigned to In low-income countries, a large proportion of individuals since some of them (such as housing, util- household consumption expenditures go toward food. ities, and furniture) are public goods, consumed joint- Thus, in principle, it might seem desirable-even ly by all household members. essential-to collect individual-level data on food This means that data on these consumption items consumption. In practice, however, collecting accurate can only be collected at the household level. However, data on individual food consumption is very difficult. some key private goods are related to human capital Among the most serious issues is the potential for development and are relatively easy to assign to indi- measurement error (see Behrman 1992 for a more vidual beneficiaries, and data on these items have detailed discussion). Given the relatively short refer- already been collected at the individual level in most ence period typically used in collecting individual previous LSMS surveys.This is reflected in the relevant food intake data (for example, the previous 24 hours), draft modules presented in Volume 3 (for example, the a large variation in the food consumption of individ- modules on education and health). Furthermore, some ual household members could lead to random meas- previous LSMS surveys have differentiated between urement errors. At the same time, more intrusive adult men and women and between boys and girls in methods of collecting food intake data, such as the collecting data on expenditures on some private goods direct weighing of food at mealtimes, may cause such as clothing and footwear. Being able to identify respondents to alter their eating behavior from their the goods consumed only by adults enabled Deaton normal pattern towards what they consider to be the (1987) and others to make inferences about the possi- norm-thereby causing systematic measurement ble gender biases in consumption expenditure alloca- errors.23 tions between boys and girls. This study was one that Collecting individual food consumption data is did not require a full set of individual-level consump- also costly in terms of time and money.The time costs tion data in order to make some inferences about are made clear by the International Food Policy intrahousehold allocations of consumption. Other Research Institute's Bukidnon study in the approaches have been taken in Chiappori (1988), Philippines, for which collecting individual food Browning and others (1994), and Bourguignon, intake data (based on 24-hour recall interviews, typi- Browning, and Chiappori (1995). These studies used cally with wives within the household) required about both individual-level data on control of income or an hour of interview time (Bouis 1997). In monetary assets (see below for a discussion of the difficulties terms it was found that collecting food intake data at involved in collecting such data) and aggregate house- the individual level through 24-hour food weighing hold consumption data to make inferences about the was about four times more expensive than collecting "sharing rules" by which intrahousehold allocations household-level data on food acquisition using a take place.20 seven-day recall period (Garcia and Senauer 1992). There has been some debate about the usefulness While one of the major purposes of collecting of using indirect methods to infer patterns of intra- household consumption expenditure data is to meas- household consumption allocations when only house- ure and monitor the changes in welfare of household hold-level data are available.While Deaton's method of members, household-level consumption measures can using information on the consumption of adult goods be inadequate for monitoring poverty and inequality has been used to analyze intrahousehold consumption at the individual level. At the same time, given the dif- allocations in several developing countries,21 little evi- ficulties and costs of collecting individual-level food dence has been found of gender discrimination in consumption data, welfare monitoring at the individ- 324 CHAPTER 24 INTRAHOUSEHOLD ANALYSIS ual level may be better achieved by collecting and ana- market-related work that the mother would otherwise lyzing data on anthropometrics, health and morbidity, have done. and education, which are all inherently individual- Another policy issue often raised is whether or to level measures, and are collected in LSMS-type surveys what extent the time women spend on household on a regular basis. A distinct advantage of using these chores such as cooking, child care, and collecting water outcome measures for monitoring the welfare of indi- and firewood prevents them from spending more time viduals is that there is less likelihood of measurement on productive activities. The "double day" burden on error in the collection of these measures than in the women (in that they have both productive and house- collection of individual-level food intake data hold chores) raises the question of the relative lack of (Behrman 1992). leisure time that women have compared to men. If For these reasons, a multitopic household survey leisure time is seen as a welfare measure, data on the such as LSMS is probably not an appropriate instru- leisure time of individual household members can be ment for collecting individual-level food consumption used as an indicator of one dimension of well-being. data on a regular basis. However, there may be a To analyze these policy issues, it is important to rationale for collecting such data under particular cir- develop a comprehensive time profile that accounts cumstances, in a more specialized survey, or in one of completely for the activities of the previous 24 hours the rotating modules in a multiyear LSMS survey and thus captures both the market and nonmarket (possibly on a smaller scale than the typical LSMS or activities of household members, including childcare using a subset of the full survey sample). For example, and all types of household chores (see Chapter 22 on when outcome measures, such as anthropometric or time use).The time allocated to leisure can be derived heath measures, indicate biases related to gender, age, as a residual-in other words, considered equivalent to or birth order in the allocation of resources among whatever time has not otherwise been accounted for. household members, survey designers may want to In order to examine cross-persons effects, it is also consider conducting a small-scale, specialized survey important to collect such information for all individ- to collect individual-level consumption data, including uals in the household (above about age 6), due to the food intake data, on an experimental basis. In particu- fact that the way children and adults dispose of their lar, designers may want to consider collecting individ- time is linked. ual-level food intake data in areas where there is severe However, as discussed in more detail in Chapter 22 and chronic malnutrition or in those countries or on time use, it should be recognized that collecting regions within a country (for example, northern India) time use data is subject to several potential complica- where gender-specific mortality rates or population tions. For example, it is sometimes difficult to obtain sex ratios are extremely out of balance. accurate information on how individuals use their time Other individual-level outcome data that are rel- when they carry out several activities concurrently. atively difficult to collect are data on the time alloca- This is generally more of a problem with women who tions of individual household members; these data can may be involved in multiple activities during a single, be critical for shedding light on intrahousehold discrete time period (for example, looking after a child resource allocations. As discussed earlier, changes in and preparing food both for human consumption and wages, prices, or access to productive resources such as for livestock production). Moreover, collecting accurate credit usually result in a reallocation of time and tasks data on time use is often confounded by ingrained cul- among individuals within the households, which can tural definitions and perceptions about work. Women, have serious implications not only for the adults with- for example, may perceive that they are not working in the household but also for the children. For exam- when involved in activities related to home production ple, households in developing countries often consid- even though this may be considered work in other cul- er children's labor to be a substitute for the labor of an tural contexts. Also it can be very difficult to obtain adult (often their mother).This means that if women's accurate measures of how individuals allocate their wages go up, the adult women of the household are time among various tasks (including the intensity of more likely to seek work outside the home, which that work effort-in other words, how much work is may mean that their children will be expected to stay accomplished during a given time period) when they home from school and do the home chores and carry out several activities at the same time. 325 NOBUHIKO FUWA, SHAHIDUR R. KHANDKER,ANDREW D. MASON, AND TARA VISHWANATH Apart from such measurement issues, another used gather data that will be adequate for analyzing major problem with collecting comprehensive time these questions? For example, in the Nicaragua LSMS use profiles for all household members is the fact it in 1998, high-priority policy issues included the issue takes a long time to collect such information. The of child labor and the question of whether or not the amount of time required for administering a time use time that women spend on household chores prevents module differs depending on various factors including them from participating in the labor force. Therefore, how the module is designed, the degree of detail in the designers of the LSMS decided to include a time the precoded activity categories (see Chapter 22 for use module in the survey. However, due to budget more details on available options), and the different constraints they decided to administer it to a random- kinds of lifestyles of people in different circumstances. ly selected subsample (roughly half) of the total sam- (Lifestyle differences may be clearly differentiated ple of households, and they also reduced the number betrveen adults and children, between urban residents of questions in the other modules to some extent. and rural residents, and between people with full-time Because of the need to capture all activities including occupations and people with a combinationi of self- household chores, the mnodule was designed to yield a employment jobs.) The average time required for con- comprehensive time profile. Because of the need to ducting a time use module with a recall period of the analyze cross-person effects among household mem- previous 24 hours can vary from 10 minutes per per- bers (including children), the designers chose to col- son in a developed country (generally by telephone lect time use data for all members of the households in interview; Harvey 1997) to 20-30 minutes per person the subsample. Other examples of surveys in which in a rural areas in a developing country (Bouis 1997). time use data were collected in developing countries Given that an interviewer would need to spend this are some of the surveys conducted by the amount of time with every household member above, International Food Policy Research Institute and the say, age 6, this could add up to a significant amount of International Crops Research Institute for the Semi- time in the administration of the household survey Arid Tropics. questionnaire. Therefore, despite the potential usefulness to ana- Individual Control of Resources lysts of having comprehensive time use data, it may not Analysts also need data on individuals' control over the be practical to include such a module in an LSMS- household's resources to study how resources are allo- type survey on a routine basis except when the study cated among household members. Knowing who of time use is a high analytical priority in the country. controls how much of the resources available to the However, there are some alternatives to including a household is more crucial for understanding intra- comprehensive time use module in an LSMS-type household resource allocations than knowing the survey. For example, a partial time use profile could be amount of resources available to the household as a developed, in which respondents are asked how much whole. In the empirical literature, such information time they spent on specific set of activities without has been crucial in testing the "income pooling" having to account for the entire 24-hour period. (See, hypothesis as xvell as the Pareto efficiency of house- however, Chapter 22 for a discussion of potential hold resource allocations (Schultz 1990;Thomas 1990, drawbacks of such options.) Alternatively, the time use 1993;Thomas and Chen 1994) and in inferring "shar- nmodule could be administered to only a subset of ing rules" between husband and wife (Browning and individuals (such as adult women and all children) others 1994).Testing the efficiency of input allocations rather than to all individuals.24 Yet another possibility in agricultural production requires data on individuals' might be to collect time use data using qualitative access to land (Udry 1996; Vishwanath and others methods such as random spot-checking, often con- 1996). ducted by anthropologists (Acharya and Bennett Therefore, data on individually oxvned assets-in 1981). other words, individual-level data on asset ownership In choosing among these options, survey design- within the household-are critical for undertaking ers should consider the priority policy issues in the intrahousehold analysis. These include data on land- country studied. What policy questions can be ana- holdings, which should be collected in the agricultur- lyzed using time use data? Can the collection methods al module (see the next section), and data on other 326 CHAPTER 24 INTRAHOUSEHOLD ANALYSIS physical assets including those used for nonagricultur- mating income from joint production activities, al enterprises that the household may have. The imputing the value of in-kind income (for example. expanded version of the draft household enterprise meals received by agricultural laborers), and the fact module introduced by Chapter 18 identifies which that specific individuals (often women) may prefer to household member is in charge of each household underreport their income if their earnings are not enterprise and which member owns each asset used under their full control. There may be other gender- for the enterprise including land, buildings, equip- specific differences in reporting the value of income. ment, and machinery and vehicles. As discussed in For example, Lampietti (1999) found in non-LSMS Chapter 18, from an intrahousehold analysis point of household survey data from Ethiopia that female view it is important to identify the owner of major respondents systematically reported lower levels of asset items used for household enterprises separately agricultural income than male respondents for house- from the manager of the business, since the day-to-day holds with otherwise similar (observable) characteris- manager of a household enterprise may not actually tics.27 However, all this is not to say that no individual- own the assets used for his or her business. Owning level data on income and earnings should be collected. assets may be more important than the day-to-day use It is particularly useful to collect data on individual or management of such assets in determining the bar- wage income (see Chapter 9 on employment). In fact, gaining power of household members and thus the if accurate data on individual's time use can be col- intrahousehold allocation of resources. lected (see above), a potentially "cleaner" measure of It can also be very useful for intrahousehold individuals' income can then be imputed. analysis to collect information on which household A number of researchers have recently tried to member owns any assets that were inherited or owned analyze the links between how much control men and prior to the formation of the household.25 This cate- women have over household income and a variety of gory of assets is less likely to have been affected by the intrahousehold outcomes (Schultz 1990;Thomas 1990, decision that the individual made in the past about 1993; Hoddinott and Haddad 1995; Wang 1997). As participating in the labor market or investing and thus McKay notes in Chapter 17 of this book, if this is an may be particularly useful as instruments for condi- analytical priority, it may become necessary to collect tioning variables. Therefore, it would be useful to data on either total earned income or "unearned" identify any assets that were owned individually by (nonlabor) income at the individual level. However, adult members of the household before they were beyond problems of measurement, there are other married. (Individual ownership generally means that problems associated with this approach. One specific the owner has a right to sell or dispose of the assets problem with using income as an explanatory variable without the consent ofher or his spouse.) Which assets in regression analyses is that income is generally an should be listed in the questionnaire will depend on endogenous variable dependent on both individual and the country and culture of the study. It may be neces- joint (household) choices regarding the allocation of sary to use pretesting or qualitative research to identi- members' time to different tasks (see Chapter 26 on fy the assets most often brought into the household at econometrics for a discussion of endogeneity prob- the time of marriage-for example, land, a house, a lems).Various efforts have been made in the empirical housing lot, animals, or personal items such as jewelry. literature to control for the endogeneity of income. Once these items have been identified, a brief set of Schultz (1990) andThomas (1990, 1993) analyzed data questions can be added to the durable goods section of on unearned or nonlabor income since this is taken to the consumption module.26 be (at least) largely unaffected by individual labor sup- Despite the general desirability of collecting data ply decisions.28 However, even this measure has some on individuals' control over the economic resources of problems since nonlabor income, as a function of asset the household, collecting data on individual income or investment income, may partly reflect a person's past from all sources can be problematic. First, as with indi- decisions about participating in the labor market.While vidual-level consumption and time use data, there are it is possible to use instrumental variables to control for potentially serious problems of measurement error. In the endogeneity of income, finding appropriate instru- the case of income data these can stem from a range of ments that affect income but do not affect the outcome problems including difficulties associated with esti- of interest is often quite difficult (see below). 327 NOBUHIKO FUWA, SHAHIDUR R. KHANDKER,ANDREW D. MASON, AND TARA VISHWANATH Plot-Level Agriculture Data In countries where policymakers are concerned To facilitate intrahousehold analysis, in many countries about the effect of the intrahousehold distribution of it is most productive to collect agricultural data at the land tenure security on agricultural productivity, it plot level rather than at the household or individual may be necessary to gather panel data to analyze land level.This is the methodology followed in the expand- use patterns over time and the changing nature of ed version of the agricultural module introduced by local land institutions. However, this kind of informa- Chapter 19. Particularly in Sub-Saharan Africa, where tion is better gathered by qualitative research methods men and women farm autonomously on separate or in a community questionnaire than in the house- plots, having detailed data on agricultural inputs and hold questionnaire of an LSMS-type survey. outputs by plot will help analysts study the intra- household allocation of resources for agricultural pro- Instruments duction. It will help them not only to analyze the effi- To properly measure the effects of endogenous condi- ciency of production but also to assess the tioning variables on a dependent variable in an cost-effectiveness of different potential interventions. econometric analysis, it is essential to find appropriate Given the importance of collecting information on variables (for example, "instruments" or "instrumental individuals' control over economic resources, as dis- variables" in an econometric jargon) that influence the cussed above, it would also be desirable to collect indi- conditioning variable (whether this is income or par- vidual-level data on who owns the plot, who has use ticipation in a program) but do not affect the outcome rights to it, and who manages the plot, for each plot of primary interest (see Chapter 26 on econometrics separately, whenever it is appropriate and possible to for a discussion of how to use instrumental variables in do so. econometric estimations). As noted above, data on Moreover, although it may not be necessary to individually owned assets can be important instru- collect such data on a routine basis in multitopic ments for identifying the effects of conditioning vari- household surveys, questions on how individual plots ables (such as income) in estimations. Data on land- were obtained and under what tenurial and contractu- holdings at the individual level and male and female al arrangements may be quite useful in some contexts. shares of household business capital and of other types For example, in some Sub-Saharan African countries, of physical (nonlabor) assets may also be useful in this the intrahousehold allocation of land rights and the regard. It can be difficult to assign the ownership of effect of this allocation on agricultural productivity is a some assets to individuals if the assets are jointly major policy concern. Some questions dealing with owned by all of the household members. However, it this issue are in the expanded version of the draft agri- is generally useful to collect data on assets whose own- cultural module (for example, "How was the plot ership can be assigned to individuals. Moreover, data obtained, from whom, and, if purchased, at what on inherited wealth, dowry-related assets, and premar- price?"). Nevertheless, if this policy issue is important, ital assets can also serve as important instruments that survey designers may also wish to include questions are unrelated to a person's decision about participating such as: When did vou obtain the plot? If it is bor- in the labor market or to other current choice vari- rowed, what are your contractual arrangements? Were ables.29 However, even these variables are not perfect, any payments related to the plot made this year? Do in that they can contain recall errors (as the respon- you expect to have to make any payments over the dents may have acquired their inheritances or premar- next year? If you lend land out, have you had any dis- ital assets many years earlier) and it can be difficult to putes with the borrower? Have there been any disputes calculate the current values of these assets. over the control of the plot within your family? Have The prices of conditioning variables can also be any disputes resulted in an intervention by the village potential instruments for identifying the effects of head or by other parties? Have there been or are there such variables. For this reason, it is critical that the any court cases regarding the plot? Is there any written community and price questionnaire gather detailed documentation regarding the plot, its ownership, or its information on the market prices of commodities at current tenurial status? Do you have any plans to trans- the community level. The issue of using prices as fer or bequest any of the land that you own or control instruments is complicated in the context of intra- to your children, relatives, or others? household analysis since observed market prices often 328 CHAPTER 24 INTRAHOUSEHOLD ANALYSIS influence both the conditioning variable and the final focuses on these key issues (see Chapter 25 on col- outcome of interest. For this reason it is also critical to lecting qualitative data). collect data on individual-specific prices such as indi- vidual wage levels. As discussed above, the occupation COLLECTING INFORMATION ON AND FROM MALES AND and education level of the respondent's parents and the FEMALES WTHIN THE HOUSEHOLD. In many country characteristics of his or her spouse may provide other and cultural contexts, collecting information on instruments if they are not observed to affect produc- women from a male household head will lead to seri- tivity of the person in question. ous problems of measurement error. In some contexts Depending upon the specific focus of the analysis, this may also be the case when asking women to there may be a variety of variables that would serve as respond to questions in the presence of men. Men may suitable instruments to control for the endogeneity of not have accurate information about their spouses but such other variables as participation in a program. For may nonetheless insist on responding to questions by example, proxies for "social capital" might be used as male interviewers. Similarly, women may want to con- identifying instruments30 for evaluating a credit pro- ceal certain information from their husbands (for gram. In the case of group-based lending, these prox- example, by underreporting their income if their hus- ies might include the characteristics of the group's bands control how at least some of that income is members (for example, their age, education level, and spent). Therefore, for the purposes of intrahousehold production activities) or of the organizers of the cred- analysis, it is critical to collect data directly from the it program (for example, their age, education level, and individual in question to the extent possible. In some family background). Data on numbers and character- cultural contexts it will also be necessary to collect data istics of potential providers of transfers may also be from women in the absence of men.This may also have appropriate instruments for identifying the effects of important implications for the size and gender compo- participation in a particular program in a number of sition of the survey interview team.31 It may mean that contexts (see Pitt and Khandker 1998). enumeration teams need to be larger and include a higher proportion of women than has typically been Methods of Data Collection the case in past LSMS surveys. Therefore, as with the The factors affecting intrahousehold resource alloca- design of the questionnaire, it will often be necessary to tion and thus the types of data needed to carry out gather preliminary information using context-specific intrahousehold analysis can be very context specific.To qualitative methods to investigate the most appropriate determine which variables are likely to be most ways to compose and train enumeration teams and important for intrahousehold analysis, how best to conduct the survey interviews within households. word survey questions, and which household mem- bers to ask specific questions (for example, men or Conclusions women), it will often be important to collect qualita- tive and other context-specific data before developing In recent years both analysts and survey planners have the questionnaire and assembling and training the increasingly come to recognize that understanding enumeration team. how resources are allocated within the household can be critical for designing, implementing, and assessing QUALITATIVE AND CONTEXT-SPECIFIC DATA. Much of poverty reduction policies and projects. This chapter the information useful for intrahousehold analysis is has cited some recent theoretical and empirical litera- specific to a given culture and location. This includes ture on intrahousehold resource allocation and distri- such issues as how men and women perceive and bution in support of this view. Rigorous empirical define work (which affects the approach used to col- testing of theories about how intrahousehold decisions lect time use data) and how men and women differ in are made is still in an early stage, but work is current- their approach to household budgeting and expendi- ly being done in turning these theories into testable ture (which affects the way household behavior is hypotheses, as well as in collecting innovative data to modeled). For these reasons, it is important that the support intrahousehold policy analysis. design of the survey questionnaire be preceded and Combined with the guidelines presented in this informed by context-specific qualitative analysis that chapter, the draft LSMS modules in this book reflect 329 NOBUHIKO FUWA, SHAHIDUR R. KHANDKER,ANDREW D. MASON, AND TARA VISHWANATH many of the recent advances in understanding what data estimated overall poverty level may have differed, possibly substan- are needed to undertake intrahousehold analysis and in tially, from Haddad and Kanbur's results if individual needs based on collecting these data. Nevertheless, the data requirements activity or body mass index had been taken into account. for analyzing intrahousehold resource allocation can be 2. Thomas (1990) found that children's health outcomes differ quite extensive, given the need for individual-level data, significantly depending on whether (unearned) income is con- and particularly when information on individuals out- trolled by mothers or fathers. A recent study of the Grameen Bank side the household is needed to enable analysts to inves- in Bangladesh (Pitt and Khandker 1998) also suggested that the tigate interhousehold links. Furthermore, some of the impact of borrowTing on a number of outcomes (including boys' data about individual household members (for example, and girls' schooling and nutritional status) differs significantly on consumption and time use) can be costly to collect. depending on the sex of the borrower. (See also Behrman and Given the limited amount of time available for conduct- Deolalikar 1990; Deolalikar 1991, 1993; and Echevarria and Merlo ing a survey interview with each household, it is not 1996.) However, as will be discussed in later sections, drawing these practical to collect all household information at the indi- conclusions is not alxvays straightforward because of various data vidual level in multitopic surveys such as the LSMS. problems and econometric issues. Therefore, it is important that the designers of multi- 3. A recent paper by the International Food Policy Research topic household surveys set clear policy priorities for Institute (1996) covered a similar set of intrahousehold policy and analysis with respect to intrahousehold resource alloca- data collection issues. The paper focused on the collection and tion and that they are selective about which data to col- analysis of clata from special purpose and small sample surveys lect at the individual level to make their analysis both aimed at answering specific policy questions, such as food security feasible and cost-effective. In closing, it is worth noting In contrast, this chapter focuses on integrating data requirements that given the extensive data needed for some types of for intrahousehold analysis within the context of nationally repre- intrahousehold analysis, a multitopic household survey sentative, multitopic surveys like the LSMS surveys that aim to such as the LSMS may not always be the most effective answer multidimensional policy questions. However, it also discuss- vehicle for collecting these data. In cases where con- es situations in which a prototype LSMS survey may not be appro- ducting intrahousehold analysis is the outstanding prior- priate but a special-purpose survey might be fielded instead to ity, policy analysts may consider fielding special, focused gather data specifically for analyzing intrahousehold issues. surveys to ensure that all of the necessary variables can 4. It should be noted here that the empirical evidence cited in be collected in a practical, cost-effective way. this section, as xvell as in the introductory section above, comes mainly from Africa and South Asia, where there is a rich literature Notes on intrahousehold resource allocation.This should not be taken to imply that intrahousehold issues are unimportant in other areas such The authors would hke to thank Jere Behrman, Mark Pitt, John as Latin America or the transition economies of Eastern Europe and Strauss, Shankar Subramanian, and the participants in the LSMS the former Soviet Union. It is only that there is relatively less empir- authors' vorkshop for their valuable comments on earlier drafts. ical analysis in these regions and that a whole range of different 1. It can be argued that if poverty or inequality levels are under- issues may be relevant to intrahousehold analysis in these areas. estimated by the same degree (for example, across countries, across 5. Bourguignon and others (1993), BroNvning and others (1994) regions, across any other categories of population, or over time), the and Thomas and Chen (1994) used data from France, Canada, and ranking among these areas or groups will not be affected. However, Taiwan respectively to test the efficiency of household consumption this is an empirical argument for wvhich there is relatively little evi- allocations and the unitary model assumption. They all rejected the dence so far. Haddad and Kanbur (1990) found that despite poten- unitary model assumption but accepted the efficiency assumption. tially large differences in the estimated levels of inequality and In earlier tests that did not examine the efficiency assumption, the poverty between individual- and household-level measures, unitary model was also rejected by Thomas (1993,1990) and Schultz inequality and poverty patterns (in other words, the ranking of dis- (1990), using Brazilian and Thai data sets respectively tributions among different social categories of people) are mostly 6. Udry also notes, however, that the effects of intrahousehold unaffected. However, they found a few cases where the poverty inefficiency may not be very large compared to those of inter- rankings of men and women were reversed when individual-level household differences. measures were used instead of household-level measures. It should 7. This issue of access to land among different household menm- be noted that Haddad and Kanbur's results were based on particu- bers can have major policy implications, for example, in the imple- lar assumptions about individuals' needs for food consumption.The mentation of land titling programs. Based on his review of litera- 330 CHAPTER 24 INTRAHOUSEHOLD ANALYSIS ture ona land rights in Sub-Saharani Africa, Platteau (1996) noted anid male-headed households that are pour.While the poverty of that there could be serious negative consequences when land tide both women and men hving in female-headed households can is granted to the male heads of households and when the custom- often be related to the disadvantages of the female heads, focusing ary use rights of women (as well as other groups such as pastoral- on female-headed households wvili not necessarily shed light on ists, hunter-gatherers, casted people, former slaves, and serfs) to land poor women living in male-headed households. are not taken into account. Since women are vital agricultural pro- 13. The most common disaggregation of this type is between ducers in these areas, he argued, there may be negative efficiency as de facto female-headed households (where the reported female well as equity consequences due to such land tiding. head has a spouse or partner who is physically absent from the 8. For example, the rule that only those with less than 0.5 acres household most of the time, possibly as a migrant worker, but still of land are eligible to participate in the Grameen Bank microcred- economically maintains the household and exercises some deci- it program makes eligibility a discontinuous function of landhold- sionmaking authority) and de jure female-headed households ing. Since landholding is a variable that is not hkely to change in (where the reported female head does not have a steady male part- the medium-term in rural Bangladesh, it is reasonably considered ner and has either never been married or is widowed, divorced, or to be exogenous, in other words, out of the control of the house- separated). hold. Similarly, other types of"discontinuities" in program eligibil- 14. Such "economic" definitions of household headship include ity rules include caps on inconie or wealth in means-tested pro- the earner of the largest cash income (the cash head), the earner of grams, minimum or maximum education levels of household the largest labor income (see Rogers 1995), and the person who members, and employment histories. Pitt and Khandker (1998) contributes the most productive labor time to the household (the used as an identifying device the existence in the survey sample of working head; see Rosenhouse 1989). households who were located in the program villages but who did 15. These countries were Botswana, Cote d'lvoire, Ethiopia, not have the option of participating in the credit programs due to Ghana, Madagascar, Rwanda, Bangladesh, Indonesia, Nepal, and the landholding criterion. In other words, if it is assumed that the Honduras. behavior of the households (other than the fact that some house- 16. In addition, many country studies suggest that the relation- holds are elgible to participate in the program) with more or less ship between female headship and poverty may differ significantly than 0.5 acres of land is the same, the differences in the household depending on such factors as the level of dlisaggregation of the data outcomes of interests (for example, household expenditure and on reported headship by marital status and other demographic time use) may be attributable to the effects of the program without characteristics, regional disaggregation (for example, into urban and any concerns about the endogeneitv of participation (since their rural), the use of an "economic definition" of headship rather than participation or nonparticipation is assumed to be exogenous) or the self-reported definition, and the adjustment of per capita con- any community-level heterogeneity bias. sumption expenditure measures by adult-equivalent scales and 9. Foster fouLnd that the distribution of resources between economies of scale. For examples see Bushan and Chao (1997) on linked households living within the same compound or bari is Ghana; Dreze and Srinivasan (1997) on India; Fuwa (forthcoming) important in the sense that the educational attainment of children on Panama; Handa (1994) on Jamaica; Louat, Grosh, and van der in partitioned households is positively associated with land owner- Gaag (1992) on Jamaica; Rogers (1995) on the Domunican ship and widt the education of the head of the joinit houseliold. Republic; and Rosenhouse (1989) on Peru. 10. The term "marriage market" is loosely used to mean the 17. One exception is Handa (1996a) who controlled for the social space where men and women meet potential partners (by endogeneity of female headship. arrangement, self-selection, or other-wise) and eventually agree to 18. A recent study (Handa 1996b) that used data from the be married. "The phrase 'marriage market' is used metaphorically Jamaican LSMS survey directly addressed the issue of endogenous and signifies that the mating of human populations is highly sys- female headship. Handa's results suggest that some unobserved tematic and structured" (Becker 1991). characteristics of wvomen (related either to their preferences or to 11. Foster (1996) examined the effect of selecting a partner in their innate abilities) appear to be a determinant both of headship the marriage market on the formation of human capital in rural and of some aspects of household behavior that are often studied in Bangladesh. He found that the selection of a partner has a signifi- relation to headship (such as household income or the likelihood cant influence on how parents' characteristics affect children's of favoring children's consumption over adults' consumption). schoohng outcomes. While Handa (1996b) used a cross-section of data, another poten- 12. The identification of poor female-headed households, how- tial approach to analyzing endogenous household headship would ever, cannot be used as a proxy for the identification of poor be to use panel data; this would enable analysts to observe changes women. Obviously there are both men and women in both female- in the headship of the same set of households over time. 331 NOBUHIKO FUWA, SHAHIDUR R. KHANDKER,ANDREW D. MASON, AND TARA VISHWANATH 19. Using data from Burkina Faso, Vishwanath and others allocation of resources Nvithin the household. The Institute for (1996) found that a person's "within-sex" status (for example, Rural Development Studies (IFLS) data have explicit measures of among females) in the household had important implications for the number of resources brought into marriage by the partners as agricultural productivity when other factors were controlled for. well as other information on the children's health status and other Warner et al. (1997), analyzing data from Ghana, found that with- social and demographic characteristics of the children and their in-sex differences in status in the household significantly affected an parents.The analysis addresses the different sources of measurement individual's control and ownership of assets. error in the values of assets and suggests some analytical approach- 20. In order to identify "sharing rules," Brow.Aning and others es to correct for them. (1994) had to be sure that at least one consumption good was 26. A section on individually owned assets svas included in the "assignable" (in other words, the consumption of the good by indi- household survey focusing on the intrahousehold impact of micro- vidual members could be observed).They used the consumption of credit schemes in Bangladesh (Pitt and Khandker 1998). An alter- cotlioing (which svas separately observed for the husband anid for native approach would be to add a question asking which house- the wife) as this assignable good. A more recent version of their hold member owns each item in the durable goods submodule of model in Bourguignon, Browning, and Chiappori (1995), howev- the consumption module. This approach was taken in the 1998 er, does not need to have an assignable consumption good. The Nicaragua LSMS. empirical application of the "sharing rule" approach has so far been 27. Lampietti (1999) postulated that these differences are attrib- limited to developed countries (for example, Canada and France). utable to different gender roles in agricultural production and mar- 91. These countries include Thailand and Cote d'lvoire keting, xvith men being predominantly responsible for produce sales (Deaton 1989), India (Subramanian and Deaton 1991; Subranmanian and cash transactions in the market. 1994), Burkina Faso (Haddad and Readon 1993), Bangladesh 28. Chapter 11 on transfers and other nonlabor income makes (Ahmad and Morduch 1993), rural China (Burgess and Zhuang at least partial provision for these income endogeneity problems in 1996), and Pakistan (Deaton 1997). the standard module by recommending that survey designers col- 22. See, for example, Ahmnad and Morduch (1993), Browning lect data on public transfers at the individual level. If this were a pri- (1992), Strauss and Beegle (1996), and Subramanian (1994). mary analytical concern, however, it would be useful to collect data Hosvever, another interpretation is that the method is not flawed on all categories of unearned income, including various forms of but rather discrimination against girls takes a form not detectable rental income. For the reasons indicated above, such an approach is in the data on allocation of consumption expenditures, such as the generally preferable to having a single respondent indicate whether allocation of a mothers' time (rather than money) or bias in certain a particular component of rental income accrued to a specific indi- critical interventions like taking a child to a doctor wshen the child vidual within the household, as recomaended in the standard is sick (Deaton 1997). module on transfers and other nonlabor income. 23. However, other potential sources ofsvsteIlatic iiieasurenrieiit 29. Sorne ailialvsts have eveii argued againisL using inheritalnce errors in typical household-level food expenditure data have been and dowry variables as instruments because of potential links also pointed out, such as the omission of food served to non-house- between such bequests and characteristics of the individuals being hold members, which tends to increase with the household income analvzed that are not observed in the data (Hoddinott, Alderman, level. Furthermore, whether food availability data (typically derived and Haddad 1997). from household-level consumption expenditure data) or food 30. If group members are self-selected, data on these character- intake data (derived from individual- or household-level observa- istics may not provide valid instruments. tions of food consumption either with a recall period of the previ- 31. Blanc and Croft (1997) examined the effects of the sex of ous 24 hours or by direct weighing at meal time) are more reliable the interviewer on responses, using data from the Demographic sources of food consumption data has been much debated. (See, for Health Survey in Ghana. example, Bouis and Haddad 1991, Subramanian and Deaton 1996, and Strauss and Thomas 1995.) References 24. Such an approach, however, would inevitably limit the analysis of cross-persons effects within intrahousehold allocation. Acharya, Meena, and Lynn Bennett. 1981. "Rural Women of 25. A recent study (Thomas, Contreras, and Frankenberg 1997) Nepal: An Aggregate Analysis and Summary of 8 Village focuses on a different measure of power-the value of assets Studies." Tribhuvan University, Centre for Economic brought into the nmarriage by the husband and the wife-and Development and Administration, Kathmandu, Nepal. examines whether women who own considerable assets at the time Ahmad,Asif, andJonathan Morduch. 1993. "Identifying Sex Bias in of their marriage are more able than other xvomen to influence the the Allocation of Household Resources: Evidence from Linked 332 CHAPTER 24 INTRAHOUSEHOLD ANALYSIS Household Surveys from Bangladesh." Development Allocation of Consumption: Some Evidence on French Data:" Discussion Paper 463. Harvard Institute for International Annales d'Econonuie et de Statistique 29:137-56. Development, Cambridge, Mass. Browning, Martin. 1992. "Children and Household Economic Alderman, Harold, Pierre-Andre Chiappori, Lawrence Haddad, Behavior:" Jourtnal of Economiiic Literature 30 (September): John Hoddinott, and Ravi Kanbur. 1995. "Unitary versus 1434-75. Collective Models of the Household: Is It Time to Shift the Browning, Martin, Francois Bourguignon, Pierre-Andre Burden of Proof?" World Bank Researcht Observer 10 (1): 1-19. Chlappori, and Valerie Lechene. 1994. "Income and Beaton, George H., and Hossein Ghassemi. 1982. "Supplementary Outcomes: A Structural Model of Intra-Household Feeding Programmes for Young Children in Developing Allocation."Journal of Political Econtomy 102 (6): 1067-96. Countries." Anierican Journial of Clinical Nutrition 34 (4): supple- Bruce, Judith, and Cynthia Lloyd. 1997. "Finding the Ties That ment. Bind: Beyond Headship and Household." In Lawrence Becker, Gary. 1991. A Treatise on the Fatnily. Cambridge, Mass.: Haddad, John Hoddinott, and Harold Alderman, eds., Harvard University Press. Intrahousehold Resource Allocation in Developing Countries: Behrman, Jere. 1988a. "Intrahousehold Allocation of Nutrients in Mlletlhods, Models, and Policy. Baltimore, Md.: Johns Hopkins Rural India: Are Boys Favored? Do Parents Exhibit Inequality University Press. Aversion?" Oxford Economnic Papers 40 (1): 32-54. Burgess. Robin, and Juzhong Zhuang. 1996. "Dimensions of -. 1988b. "Nutrition, Health, Birth Order and Seasonality: Gender Bias in Intrahousehold Allocation in Rural China:' Intrahousehold Allocation in Rural India." Journal oJ London School of Ecoionaics, Suiotory anid Toyota Development Economics 28 (1): 43-63. International Center for Economics and Related Disciplines, -. 1992. "Intra-Household Allocation of Nutrients and London. Gender Effects: A Survey of Structural and Reduced-Form Buvnic, Mayra, and Geeta Rao Gupta. 1997. "Female-Headed Estimates." In S. R. Osmani, ed., Nutrition and Policy. Oxford: Households and Female-Maintained Families: Are They Clarendon Press. Worth Targeting to Reduce Poverty in Developing 1997. "Intrahousehold Distribution and the Family." In Countries?" Economic Developmeent and Cultural Clhange 45 (2): Mark R. Rosenzweig and Oded Stark, eds., Handbook of 259-81. Population and Family Economtics. NewYork: Elsevier. Carter, Michael R., and Elizabeth Katz. 1997 "Separate Spheres and Behrman, Jere, and Anil Deolalikar. 1990. "The Intra-Household the Conjugal Contract: Understanding the Impact of Gender- Demand for Nutrients in Rural South India.?Joturnal of Human Based Development." In Lawrence Haddad, John Hoddinott, Resources 25 (4): 665-96. and Harold Alderman, eds., Intralhouselhold Resource Allocation in Bergstrom, Theodore. 1989. "A Fresh Look at the Rotten Kid Developing Countries: Alethods, Models, and Policy Baltimore, Theorem and Other Household Mysteries."Journal of Political Md.: Johns Hopkins University Press. Economy 97 (5): 1138-59. Chiappori, Pierre-Andre. 1988. "Rational Household Labor Bhushan, Indu, and Shiyan Chao. 1997. "Measurement Issues in Supply." Econometrica 56 (1): 63-89. Gender-Based Poverty Comparisons: Lessons from Ghana." Deaton, Angus. 1987. The 41location of Goods withlin the Hotusehold: World Bank, Africa Region, Washington, D.C. Adults, Cihildren, anzd Gender Living Standards Measurement Blanc, Ann, and Trevor Croft. 1997. "The Effects of the Sex of the StudyWorking Paper 39.Washington, D.C.:World Bank. Interviewer on Responses in Fertility Surveys: The Case of - 1989. "Looking for Boy-Girl Discrimination in Ghana." IRD/Macro International, Columbia, Md. Household Expenditure Data." World Banlk Economiiic Revie,v 3 Bouis, Howarth. 1997. Personal communication.Washington, D. C. (1): 1-15. Bouis, Howarth, and Lawrence Haddad. 1991. "Are Estimates of - 1997. The Analysis of Houselhold Surveys:A Alicroeconometric Calorie Income Elasticities too High?: A Recalibration of the .4pproacht to Development Policy. Baltiimiore, Md.: Johris Hopkiins Plausible Range." Journal of Development Economics 39 (4): University Press. 333-64. Deolalikar, Anil. 1991. "Intrahousehold Allocation of Health Inputs Bourguignon, Francois, Martin Browvning, and Pierre-Andre and Distribution of Health Outcomes among Indonesian Chiappori. 1995. "The Collective Approach to Household Children." In Robert MclNaniara Fellou,v:hip Program: 10th Behavior?" Queen Elizabeth House Development Studies Anniversary Publication. Washington, D.C.:World Bank. Working Papers 79. Oxford University, Oxford. . 1993. "Gender Differences in the Returns to Schooling Bourguignon, Francois, Martin Browning, Pierre-Andre and School Enrollment Rates in Indonesia.?Jouroal of Hunman Chiappori, and Valerie Lechene. 1993. "Intrahousehold Resources 28 (4): 899-932. 333 NOBUHIKO FUWA, SHAHIDUR R. KHANDKER, ANDREW D. MASON, AND TARA VISHWANATH Dey, Jennie. 1981. "Gambian Women: Unequal Partners in Rice Handa, Sudhanshu. 1994. "Gender, Headship and Intrahousehold Development Projects?"Journal