45496 Margaret Grosh, Carlo del Ninno, Emil Tesliuc, and Azedine Ouerghi Transient po Tr ent poor Poverty line Pove Chronic poor hroni oor The Design and Implementation of I E Effective Safety Nets For Protection &Promotion For Protection and Promotion THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS For Protection and Promotion THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Margaret Grosh, Carlo del Ninno, Emil Tesliuc, and Azedine Ouerghi with the assistance of Annamaria Milazzo and Christine Weigand Washington, D.C. © 2008 e International Bank for Reconstruction and Development / e World Bank 1818 H Street NW Washington DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org E-mail: feedback@worldbank.org All rights reserved. 1 2 3 4 11 10 09 08 is volume is a product of the staff of the International Bank for Reconstruction and Development/ e World Bank. e findings, interpretations, and conclusions expressed in this volume do not necessarily reflect the views of the Executive Directors of e World Bank or the governments they represent. e World Bank does not guarantee the accuracy of the data included in this work. e boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of e World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions e material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. e International Bank for Reconstruction and Development/ e World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; telephone: 978-750-8400; fax: 978-750-4470; Internet: www.copyright.com. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, e World Bank, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2422; e-mail: pubrights@worldbank.org. IISBN-13: 978-0-8213-7581-5 eISBN: 13: 978-0-8213-7582-2 DOI: 10.1596/978-0-8213/7581-5 Library of Congress Cataloging-in-Publication Data For protection and promotion: the design and implementation of effective safety nets / Margaret Grosh, Carlo del Ninno, Emil Tesliuc, Azedine Ouerghi p. cm. ISBN 978-0-8213-7581-5 (alk. paper) -- ISBN 978-0-8213-7582-2 1. Economic assistance, Domestic. 2. Poverty--Prevention. 3. Social policy. 4. Social planning. 5. Poor--Services for. I. Grosh, Margaret E. HC79.P63F67 2008 362.5'6--dc22 2008032155 Cover design by Drew Fasick. ( Contents Acknowledgments...................................................................................... xiii Preface.......................................................................................................... xv Abbreviations............................................................................................. xvii 1. Introduction............................................................................................. 1 1.1 How Do Safety Nets Contribute to Development Policy?...........................1 1.2 What Is a Good Safety Net?........................................................................2 1.3 What Is a Safety Net?..................................................................................4 1.4 How Is is Book Organized? ....................................................................6 2. The Case for Safety Nets ..................................................................... 11 2.1 Why Should Countries Have Safety Nets? ................................................12 2.2 How Do Safety Nets Fit in Wider Development Policy? ...........................26 2.3 What Are the Challenges to Safety Nets? ..................................................30 Notes....................................................................................................................43 3. Financing of and Spending on Safety Nets........................................ 45 3.1 e eory on Expenditure Allocation .....................................................46 3.2 Sources of Financing for Safety Nets.........................................................51 3.3 In Search of Countercyclical Financing for Safety Nets.............................55 3.4 e Cost of the Welfare State in Developed Countries..............................58 3.5 Levels and Patterns of Safety Net Spending in Developing and Transition Countries..................................................................................................61 3.6 Delivering Safety Nets in a Decentralized World.......................................68 Notes....................................................................................................................82 4. Enrolling the Client:Targeting, Eligibility, and Intake........................ 85 4.1 Basic Concepts of Targeting......................................................................86 4.2 Results of Targeting ..................................................................................89 4.3 Targeting Options ....................................................................................99 4.4 Implementation Matters for Targeting....................................................105 Notes..................................................................................................................125 5. Benefit Levels and Delivery Mechanisms ........................................ 127 5.1 Determining Benefit Levels in eory and Practice.................................128 5.2 Managing Work Disincentives................................................................137 v vi FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS 5.3 Enhancing Safety Net Programs to Promote Household Independence...145 5.4 Managing Payments................................................................................156 Notes..................................................................................................................173 Annex: Generosity of Safety Net Programs of Last Resort in OECD Countries.....175 6. Using Monitoring and Evaluation to Improve Programs................. 181 6.1 e Value of Good Monitoring and Evaluation ......................................181 6.2 Distinct, but Complementary, Tools.......................................................183 6.3 Development of an M&E System...........................................................186 6.4 Monitoring.............................................................................................199 6.5 Evaluation ..............................................................................................213 Notes..................................................................................................................238 Annex 6.1: Sample M&E Indicators for Typical Safety Net Interventions..............239 Annex 6.2: Monitoring of Oportunidades, Mexico...............................................244 Annex 6.3: A Summary of Experimental Methods................................................250 7. Understanding Common Interventions ............................................ 253 7.1 Cash and Near Cash Transfers ................................................................256 7.2 In-Kind Food Transfers and Other Food-Based Programs.......................268 7.3 General Subsidies....................................................................................283 7.4 Workfare.................................................................................................297 7.5 Conditional Cash Transfers.....................................................................312 7.6 Fee Waivers, Exemptions, and Scholarships ............................................324 7.7 Conclusion.............................................................................................333 Notes..................................................................................................................338 Annex: Coverage of School Feeding Programs Sponsored by the World Food Programme as of 2005............................................................................341 8. Assisting Traditionally Vulnerable Groups ....................................... 345 8.1 Income Support for the Elderly ..............................................................347 8.2 Income Support for Orphans and Vulnerable Children...........................357 8.3 Income Support for People with Disabilities...........................................364 Notes..................................................................................................................372 9. Weaving the Safety Net ...................................................................... 373 9.1 What Is a Good Safety Net?....................................................................373 9.2 Know Your Target Group........................................................................374 9.3 Assessing the Performance of Individual Programs..................................384 9.4 Assessing the Performance of the Entire Safety Net System.....................394 9.5 From Diagnosis to Action.......................................................................398 9.6 How Often Should Safety Net Strategies Be Revised? .............................409 Notes..................................................................................................................409 Annex: Administrative Costs by Type of Intervention.........................................411 10. Customizing Safety Nets for Different Contexts.............................. 415 10.1 Safety Nets in Very Low-Income Countries ............................................417 10.2 Safety Nets in Middle-Income Countries................................................422 10.3 Safety Nets for an Economic Crisis Situation..........................................431 CONTENTS vii 10.4 Safety Nets after Natural Disasters..........................................................435 10.5 Safety Nets to Facilitate Reforms.............................................................440 10.6 Safety Nets for Rising Food Prices ..........................................................443 10.7 Summary................................................................................................451 Notes..................................................................................................................452 Appendixes A. Basic Concepts of Poverty and Social Risk Management.........................453 B. Main Features of Selected Safety Net Programs.......................................465 Glossary ..................................................................................................... 507 References ................................................................................................. 517 Index ........................................................................................................... 567 Boxes 1.1 Definitions of Safety Nets and Social Assistance..........................................5 2.1 e Motivating Force of Educational Stipends: e Bangladesh Female Secondary School Assistance Program.......................................................16 2.2 e Loss of an Ox.....................................................................................18 2.3 A Poverty Trap in Shinyanga.....................................................................19 2.4 A Policy Maker's Take on Growth, Equality, and Policy ............................22 2.5 e Developmental Effects of the Elizabethan Poor Laws .........................23 2.6 Is Dependency Always Bad?......................................................................35 2.7 Women's Reactions to Questions about Transfers and Fertility..................40 3.1 Okun's Leaky Bucket ................................................................................47 3.2 What Tax Instruments Should Governments Use to Support Safety Nets?...53 3.3 Fiscal Responsibility Laws.........................................................................57 3.4 Literature on Safety Net Spending Levels Based on the IMF's Government Finance Statistics..................................................................63 3.5 Financing Arrangements and Incentives in the Netherlands......................74 3.6 Quality Assurance of Eligibility Determination in the U.S. Food Stamp Program....................................................................................................75 4.1 How Narrow Should Targeting Be?...........................................................91 4.2 Universalism versus Targeting ...................................................................98 4.3 Resources on Different Targeting Methods ...............................................99 4.4 How Do the Reasons for Adopting a Program Affect Targeting Choices? 105 4.5 Communications and Transparency in Program Intake...........................113 5.1 e Value of a CCT Program's Education Grant: From eory to Practice ...129 5.2 e Benefit Formula for Brazil's Bolsa Familia Program..........................134 5.3 Lack of Applicability of High Labor Disincentives in the AFDC to Safety Nets in Middle- and Low-Income Countries................................139 5.4 Labor Disincentives in Very Low-Income Countries: e Kalomo District Pilot Social Cash Transfer Scheme, Zambia................................141 5.5 Is Reduced Work Effort Really So Bad?...................................................143 5.6 Managing Work Disincentives in Romania's GMI Program....................144 5.7 Bulgaria's From Social Assistance to Employment Program.....................146 viii FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS 5.8 Global Experience with Active Labor Market Programs..........................147 5.9 Cash Incentives Aimed at Behavioral Change Rather an Income Support..149 5.10 Introduction of a One-Window Approach in Arzamas, Russia................155 5.11 A Transfer Program with Strong Psychosocial Support for Families: Chile Solidario........................................................................................157 5.12 Delivering In-kind Transfers ...................................................................158 5.13 ATMs and POS Machines: Conducting Transactions Remotely Using Electronic Cards .....................................................................................161 5.14 Managing the Contracting Process..........................................................164 5.15 Processing Payments Using a Smart Card System....................................170 5.16 Delivering Cash in an Emergency...........................................................171 6.1 Key Uses of Performance Indicators........................................................205 6.2 Tracking Error, Fraud, and Corruption...................................................206 6.3 Example of Supranational Monitoring System: e European Union's Mutual Information System on Social Protection ...................................212 6.4 Cost-Benefit and Cost-Effectiveness Analysis..........................................214 6.5 Misleading with Targeting Statistics........................................................222 6.6 Ignorance Has High Costs......................................................................232 6.7 e Problem of Bias................................................................................234 7.1 Classification of Types of Programs Covered...........................................255 7.2 e South African Child Support Grant.................................................258 7.3 Use and Effectiveness of Cash Transfers in Emergencies..........................266 7.4 e Vulnerable Group Development Program, Bangladesh.....................270 7.5 From Universal to Targeted Distribution, India and Indonesia................271 7.6 Cash and In-Kind Transfers: Alternatives or Complements?....................272 7.7 Inferior Commodities and Inframarginal Consumption .........................275 7.8 Universal Food Subsidy System for Bread and Flour, Egypt ....................285 7.9 Reforming Food Price Subsidies, Tunisia.................................................287 7.10 Lowering the Cost of Rice by Adjusting Import Tariffs versus Targeted Cash Transfers, Madagascar ....................................................................290 7.11 Public Works Program, Korea.................................................................298 7.12 Factors Underlying the Success of Argentina's Trabajar Program..............301 7.13 e Changing Nature of Public Works Programs, India..........................304 7.14 e PROGRESA/Oportunidades Program, Mexico ...............................314 7.15 e Debate: Conditional versus Unconditional Cash Transfers...............316 7.16 Heath Fee Waiver Program, Armenia......................................................327 7.17 Elimination of User Fees and Waivers, South Africa and Uganda............332 8.1 Including Ethnic Minorities in Safety Net Programs...............................347 8.2 e Complexities of Measuring Poverty among Different Age Groups....349 8.3 e Political Economy of Old-Age Support............................................355 9.1 Seasonality of Vulnerability in Two Localities in Mozambique................379 9.2 e Public Expenditure Review Lens: Analysis of Individual Programs...386 9.3 A Proposal for Benchmarking Administrative Costs................................392 9.4 e Public Expenditure Review Lens: Sectorwide View ..........................395 9.5 Common Pitfalls in Reforming Safety Net Systems ................................397 CONTENTS ix 9.6 Communication Strategy: A Key Component of Reform........................406 10.1 Eastern European Safety Nets: From Central Planning through Transition to Accession ..........................................................................416 Figures 1.1 Processes and Stakeholders Involved in a Safety Net....................................3 2.1 Where Safety Nets Fit in Larger Development Policy................................12 2.2 Perceptions of Fairness of Country Income Distribution in Latin America...14 2.3 Distribution of General Revenue­Financed Transfers for Selected Countries by Population Quintile.............................................................33 3.1 Societal Attitudes about Poverty and Spending on Social Welfare .............59 3.2 Safety Net Expenditures as a Percentage of GDP, Selected Countries and Years.........................................................................................................64 3.3 Social Assistance and Social Insurance as a Percentage of GDP by Region, Selected Years............................................................................................66 3.4 Social Assistance, Social Insurance, and Social Sector Spending by Region, Selected Years.................................................................................................66 3.5 Spending, Income, and Public Attitudes...................................................68 4.1 Share of Population by Quintile at Received Benefits in Selected Safety Net Programs, India, Fiscal 2004/05 ..............................................88 4.2 Share of Benefits Accruing to Each Quintile for Selected Safety Net Programs, India, Fiscal 2004/05 ..............................................88 4.3 Errors of Inclusion, Selected Programs and Countries...............................90 4.4 Conceptualizing Administrative Costs ......................................................93 4.5 Targeting Performance by Targeting Method ..........................................103 4.6 From Population to Beneficiary: e Stages of Targeting ........................106 4.7 Fraction of the Social Assistance Budget Captured by Each Quintile, Armenia, 1998 and 1999........................................................................123 5.1 Reconciling Needs with Budget Constraints ...........................................131 5.2 Generosity of Selected Safety Net Programs, Europe and Central Asia and Latin America and the Caribbean, Selected Years, 2001­4 ...............136 5.3 Generosity of Selected CCT Programs in Selected Latin American and Caribbean Countries, Various Years ........................................................136 6.1 A Typical M&E System ..........................................................................185 6.2 Logical Framework for Ethiopia's Productive Safety Net Program...........187 6.3 M&E Strategy of Oportunidades............................................................197 6.4 Framework for Distinguishing among Different Types of Performance Indicators ...............................................................................................204 6.5 Incidence of Social Protection Transfers Depends on the Assumed Pretransfer per Capita Consumption, Republic of Yemen, 1999 .............227 6.6 Ex Ante Estimation of the Average Net Impact of Reforming Fuel Subsidies in Indonesia.............................................................................229 7.1 Designing and Implementing Public Works Programs ............................311 7.2 Coverage by Decile, Selected CCT Programs and Years ..........................318 7.3 Share of Benefits Accruing by Decile, Selected CCT Programs and Years ...319 x FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS 8.1 Sub-Saharan African Countries with Orphaning Rates of 15 Percent or Greater, 2003..........................................................................................358 8.2 Percentage of Orphans by Age, Asia, Latin America and the Caribbean, and Sub-Saharan Africa, 2003 ................................................................358 8.3 Prevalence of Disability by Age Group, Selected East European Countries and Years................................................................................365 8.4 Disability Rates by Poverty Status, Selected East European Countries and Years................................................................................366 9.1 Losses by Households at Experienced Shocks, Guatemala, 2000.........381 9.2 Share of Administrative Costs in Program Budget, Median Value by Type of Program, Selected Programs and Years........................................391 10.1 Incidence of Diesel, Gasoline, and Kerosene Subsidies, Indonesia, 2004 ....442 10.2 Coverage and Incidence of the Unconditional Cash Transfer Program, Indonesia, 2005......................................................................................443 10.3 Food Prices.............................................................................................444 10.4 Types of Mitigating Policies Adopted by Selected Countries, Early March 2008...................................................................................448 A.1 Households in Chronic versus Transient Poverty.....................................455 A.2 Poverty as a Disease ................................................................................457 Tables 2.1 Safety Nets for Protection and Promotion.................................................25 2.2 Possible Target Groups, the Role of Safety Nets, and Complementary Policies .....................................................................................................28 2.3 Examples of Social Protection Programs by Life Cycle ..............................29 2.4 Summary of How to Handle Challenges to Safety Nets............................43 3.1 Cost-Benefit Estimates of the Education Effects of the Bono de Desarrollo Humano Program, Ecuador.....................................................49 3.2 Options for Increasing Safety Net Budgets: Advantages and Disadvantages..51 3.3 Perceptions of Poverty in the United States, Europe, and Latin America, 1995­7 (percentage of respondents) .........................................................60 3.4 Correlations between Spending on Social Sectors and Other Factors ........67 3.5 Advantages and Disadvantages of Financing Sources for Subnational Governments............................................................................................72 3.6 Options for Managing Local Governments' Incentives to Use Lax Entry Criteria for Centrally Financed Safety Net Programs.................................73 3.7 An Example of a Functional Analysis of a Program and Responsible Levels of Government: e Village Full Employment Program, India.......78 3.8 Advantages of Alternative Allocations of Institutional Responsibilities for Household Targeting Systems ...................................................................79 3.9 Management Solutions to Implementation Challenges in a Decentralized Context, Bolsa Familia Program, Brazil.....................................................82 4.1 Errors of Inclusion and Exclusion.............................................................87 4.2 Administrative Costs of Targeting for Selected Means-Tested and Proxy Means-Tested Programs, Various Years......................................................94 CONTENTS xi 4.3 Estimated Total and Annual Costs of Household Targeting Systems, Selected Countries, 2002..........................................................................95 4.4 Approaches to Household Registration...................................................109 4.5 Institutional Consolidation, Family Poverty Benefits Program, Armenia, 1999­2004.............................................................................................124 5.1 Generosity of Selected Cash Transfer Programs in Selected East European and Central Asian Countries, Selected Years 2001­4.......137 5.2 Distributing Agencies and Instruments...................................................159 5.3 Costs of Delivering Benefits, Selected Countries and Programs...............172 A5.1 Social Assistance Benefits, Selected OECD Countries, 2005...................176 6.1 Key Differences between Monitoring and Evaluation..............................184 6.2 Logical Framework and Selected Output and Outcome Indicators for Ethiopia's PSNP......................................................................................188 6.3 Advantages and Disadvantages of Using Administrative Data to Monitor and Evaluate Programs..............................................................192 6.4 Sample Indicator Specification Sheet Adapted from the Bolsa Familia Program, Brazil.......................................................................................201 6.5 Guidelines for Collecting Outcome Indicators........................................203 6.6 Selected Targets for Panama's Red de Oportunidades..............................208 6.7 Examples of Disaggregation Subgroups and Benchmarks........................209 6.8 Examples of Questions Addressed by Process Evaluations of Safety Net Programs ................................................................................................217 6.9 Common Targeting Measures.................................................................220 6.10 Some of the Main Types of Targeting Assessments ..................................226 6.11 When to Conduct an Impact Evaluation: A Checklist ............................237 A6.2.1 Selected Monitoring, Evaluation, and Performance Indicators Used by Oportunidades, Mexico, 1999­2003......................................................244 A6.2.2 Indicators Collected from the Sample of Sentinel Points, by Area of Residence, in Percentages, Oportunidades, Mexico, 2000-02..................247 7.1 Comparison of Delivery Options for Supplementary Foods....................273 7.2 Types of Interventions in Education by Country ....................................329 7.3 Characteristics of Safety Net Interventions..............................................335 8.1 Percentage of the Population Living in Poverty by Age, Selected Sub-Saharan African Countries and Years ...............................................350 8.2 Percentage of the Population Living in Poverty by Age, Selected Eastern European Countries, Selected Years 1993­5 ...........................................351 8.3 Characteristics of Social Pension Programs, Selected Countries and Years...357 8.4 Variations in Orphans' School Enrollment, Selected Countries and Years...360 8.5 Targeting Criteria for OVC by Setting: Burundi, Malawi, and Selected Countries Served by World Vision..........................................................362 8.6 Social or Family Assistance Programs for People with Disabilities, Selected Countries..................................................................................367 8.7 Advantages and Disadvantages of Alternative Approaches to Operational Definitions of Disability in Medical Assessments.................368 9.1 Failure to Reduce Indigence Despite Growth, Chile, Selected Years ........376 9.2 Percentage Share of Income by Quintile, Indonesia, 1996 ......................376 9.3 Percentage of Population Moving Into and Out of Poverty, Rural Ethiopia, 1989 and 1995 ..............................................................377 9.4 Types of Data Required for Different Types of Diagnostics.....................383 9.5 Cost-Efficiency of Various School Feeding Programs, Panama, 2005 ......389 9.6 Cost-Effectiveness of Transfers to Reduce Poverty, Guatemala, 2000.......390 9.7 Examples of Recent Reforms by Type .....................................................400 9.8 Estimated Annual Costs and Expected Coverage of a Proposed Safety Net Reform Package, Pakistan, 2006­10......................................402 9.9 Options for Institutional Locations for Safety Nets.................................408 10.1 Sample of Public Works Supported under the PSNP, 2007.....................421 10.2 Targeting Outcomes of the GMI Program, Bulgaria, 2003 .....................429 A.1 Examples of Social Risk Management Strategies and Arrangements........461 B.1 Cash and Near Cash Programs................................................................465 B.2 In-Kind Food Transfers and Other Food-Based Programs (Targeted Programs) ...............................................................................475 B.3 General Subsidy Programs ......................................................................480 B.4 Public Works Programs...........................................................................486 B.5 Conditional Cash Transfer Programs.......................................................493 B.6 Fee Waivers for Health and Education....................................................502 Acknowledgments In writing this book we benefited from the forbearance of our director, Robert Holzmann, and from the contributions of many colleagues. The Safety Net Primer steering commit- tee provided useful input into the concept note and early outline of the study. Helpful comments were received from a quality enhancement review panel chaired by John Hod- dinott and consisting of Van Adams, Harold Alderman, Tamar Manueylan Atinc, Kathy Lindert, Michael Lipton, Mamta Murti, Sandor Sipos, and Kalanidhi Subbarao. Jehan Arulpragasam, Tamar Manuelyan Atinc, Francisco Ferreira, John Hoddinott, and Kathy Lindert provided indispensable comments in their role as official peer reviewers. Com- ments on various chapters or sections were received from Harold Alderman, Francisco Ayala, Jeanine Braithwaite, Bénédicte de la Briere, David Coady, Aline Couduel, Indermit Gill, Markus Goldstein, Theresa Jones, Valerie Kozel, Maureen Lewis, Jennie Litvack, Daniel Mont, Demetra Nigthingale, Harry Patrinos, Robert Prouty, Carolina Sanchez- Paramo, Kalanidhi Subbarao, Cornelia Tesliuc, Steven Webb, and William Wiseman. John Blomquist was an important member of the team in the early years of work regard- ing the related suite of safety net primer papers and training, which helped to cement the knowledge that underlies the whole work, especially chapter 7. Emanuele Baldacci wrote a background note for a section of chapter 3; material from it is incorporated in table 3.2, and boxes 3.2, 3.3 and 3.4. Jane Hoddinott did much of the research for and first drafting of chapter 5, section 4. Lucian Pop contributed the analysis of the World Values Survey data. Asmara Achcar, Deni Sanchez, and Kaleb Tamiru helped locate background materials for chapters 9 and 10. Manorama Rani was in charge of document processing. Alice Faintich edited the document, Nita Congress laid it out, and Bob Elwood did the indexing. Paola Scalabrin arranged translations into Russian, Spanish, and Vietnamese and oversaw the production process along with Elizabeth Kline. xiii Preface This book is part of a larger Safety Net Primer work program supported by the Social Protection Unit of the Human Development Network. The full work program comprises a variety of activities all designed to share information on the design and implementation of safety net programs around the world. Safety Nets Primer Papers are good starting points to learn about an important aspect of safety nets. Some summarize international "good practices" on which there is a great deal of consensus. This is especially the case for papers on the most common types of interventions. Some primers present new research, such as the work commissioned on targeting. Other primers represent the views of individual thinkers, moving forward the international debate on themes where there is not yet firm consensus--for example, on issues of institutions, political economy or the role of safety nets in development policy. Still other primers offer case studies, such as the work on social safety net assessments in Latin America. Most Safety Nets Primer Papers are commissioned by the team, often with the support of the World Bank Institute or regional partners, to fill gaps in knowledge or as teaching tools. Other papers are the results of jointly sponsored work or work done by others that provides useful information to the primer's audience. Primer Notes are brief pamphlets that distill the main messages of the underlying primer papers. They serve as a briefing for those who do not have time to read compre- hensively on all aspects of safety nets, and as an introduction to those who will want to read more on the specific theme. A list of primer papers and notes is provided at the back of this book. The team offers a variety of training, workshops, and conferences either based on primer materials or contributing to them. Together with the World Bank Institute, the team jointly offers an annual two-week, Washington, D.C.-based course, "For Protection and Promotion: The Design and Implementation of Effective Safety Nets," and periodi- cally offers distance versions in English, French, Russian, and Spanish. A second week- long course, "How the Rich Protect Their Poor: Social Safety Nets in the Organisation for Economic Co-operation and Development," is offered on demand for small groups of countries. Occasional multiday workshops are organized on themes of importance. These have included three international conferences on conditional cash transfer programs, two workshops on protection for orphans and vulnerable children, and one workshop on food aid. Also, there are periodic internal training events and a brown-bag-lunch seminar series for World Bank staff. xv xvi FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS The Web site of the Safety Nets team, www.worldbank.org/safetynets, is a compre- hensive guide to its knowledge on safety nets. It contains primer papers and notes, often in several languages. It archives agendas, presentations, and papers from all the training events, workshops, and brown-bag seminars for the last several years. The Web site also contains a catalogue of World Bank projects and analytic work on safety nets. And it provides references to much other work--abstracts of, links to, and copies of hundreds of other papers on safety net programs or themes. Abbreviations AFDC Aid for Families with Dependent Children (United States) ATM automated teller machine BULOG National Food Logistics Agency (Indonesia) CCT conditional cash transfer EFC error, fraud, and corruption GDP gross domestic product GMI guaranteed minimum income JPS Jaringan Pengaman Sosial (Social Safety Net) (Indonesia) JRY Jawahar Rozgar Yojana (Jawahar Employment Program) (India) M&E monitoring and evaluation MDS Ministry for Social Development and the Fight against Hunger (Brazil) MEGS Maharashtra Employment Guarantee Scheme (India) MIS management information system MISSOC Mutual Information System on Social Protection MPCT marginal propensity to consume out of the transfer NGO nongovernmental organization OECD Organisation for Economic Co-operation and Development PDS public distribution system POS point of service PROGRESA Programa de Educación, Salud y Alimentación (Education, Health, and Nutrition Program) (Mexico) PSNP Productive Safety Net Program (Ethiopia) VAT value added tax WFP World Food Programme xvii CHAPTER 1 Introduction A ll countries fund safety net programs for the protection of their people. Though an increasing number of safety net programs are extremely well thought out, adroitly implemented, and demonstrably effective, many others are not. This book aims to assist those concerned with social policy to understand why countries need social assistance, what kind of safety programs will serve them best and how to develop such programs for maximum effectiveness. 1.1 How Do Safety Nets Contribute to Development Policy? Safety nets are part of a broader poverty reduction strategy--interacting with and work- ing alongside of social insurance; health, education, and financial services; the provision of utilities and roads; and other policies aimed at reducing poverty and managing risk. Safety net programs can play four roles in development policy: · Safety nets redistribute income to the poorest and most vulnerable, with an immediate impact on poverty and inequality. Most societies hold strong con- victions that adequate provision for the poor is required, though they may differ in how this should be achieved. · Safety nets can enable households to make better investments in their fu- ture. In this role, safety nets basically act to remedy credit market failures, allow- ing households to take up investment opportunities that they would otherwise miss--both in the human capital of their children and in the livelihoods of the earners. · Safety nets help households manage risk. At minimum, safety net programs help households facing hard times avoid irreversible losses, allowing them to maintain the household and business assets on which their livelihoods are based, and to adequately nourish and school their children. At best, they can provide an insurance element that lets households make choices about livelihoods that yield higher earnings. Safety nets thus both protect households and promote their independence. · Safety nets allow governments to make choices that support efficiency and growth. An adequate permanent social assistance system can fulfill whatever re- distributive goals the society has, freeing other sectors from the role and letting them concentrate on efficient provision of services. Thus, for example, energy 1 2 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS sectors can price for efficiency, and trade policy can focus on growth rather than job protection. Short-term safety net programs can compensate those negatively affected by needed reforms or who may oppose and stall these reforms. Though useful, safety nets are not a panacea, and there are real concerns over wheth- er they are affordable and administratively feasible or desirable in light of the various nega- tive incentives they might create. In most settings where there is political will to do so, such concerns can be managed through a number of prudent design and implementation features. Much information and innovation exist on these topics; this book summarizes, references, and builds on this knowledge base to promote well-crafted safety nets and safety net policy. 1.2 What Is a Good Safety Net? Safety net systems are usually woven of several programs, ideally complementing each other as well as complementing other public or social policies. A good safety net system is more than a collection of well-designed and well-implemented programs, however; it also exhibits the following attributes. · Appropriate. The range of programs used and the balance between them and with the other elements of public policy should respond to the particular needs of the country. Each program should be customized for best fit with the circum- stances. · Adequate. The safety net system overall covers the various groups in need of assistance--the chronic poor, the transient poor, those affected by reforms, and all the various subsets of these groups. Individual programs should provide full coverage and meaningful benefits to whichever subset of the population they are meant to assist. · Equitable. The safety net should treat beneficiaries in a fair and equitable way. In particular, it should aim to provide the same benefits to individuals or households that are equal in all important respects (horizontal equity) and may provide more generous benefits to the poorest beneficiaries (vertical equity). · Cost-effective. Cost-effective programs channel most program resources to their intended target group. They also economize the administrative resources required to implement the program in two ways. First, at the level of the whole safety net system, they avoid fragmentation and the subsequent need to develop adminis- trative systems without realizing economies of scale. Second, they run efficiently with the minimum resources required to achieve the desired impact, but with sufficient resources to carry out all program functions well. · Incentive compatible. Safety nets can change households' behavior, for better or worse. To ensure that the balance of changes is positive, the role of safety nets should be kept to the minimum consistent with adequacy. The safety net system often may include programs that explicitly help build assets or incomes of their individual clients or communities by linking transfers to required or voluntary program elements. Public works programs can provide physical assets to commu- 1. INTRODUCTION 3 nities. Conditional cash transfer programs build the human capital of households. Links to financial, job search, training, or social care services may help households raise their incomes. · Sustainable. Prudent safety net systems are financially sustainable, in that they are pursued in a balanced manner with other aspects of government expenditure. Individual programs should be both financially and politically sustainable so that stop/start cycles of programs are avoided, as these result in enormous lost oppor- tunities for efficient administration and the achievement of programs' promotive aspects. In low-income countries, programs started with donor support are gradu- ally incorporated into the public sector. · Dynamic. A good safety net system will evolve over time.The appropriate balance of programs will change as the economy grows and changes, as other elements of policy develop, or when shocks occur. The management of specific programs should also evolve as problems are solved and new standards set. Much of the quality of a safety net is in the details of its implementation (figure 1.1). An adequate transfer program involves at the least a system to register clients, pay them, and eventually take them off the rolls. An exceptional program can entail much more-- the minimal registry of clients is supplemented with strong outreach campaigns to ensure that errors of exclusion are low, a strong screening mechanism to ensure that ineligible people do not register, a mechanism to handle grievances, periodic monitoring of targeting FIGURE 1.1 Processes and Stakeholders Involved in a Safety Net Information providers and consumers Government Potential Service Financial Civil institutions beneficiaries providers institutions society (ministries) Program process Beneficiary identification Registry Conditions Payment Control Targeting Database Data collection Eligibility Grievances Registration Validation Verification Payment Processes Graduation Updates Penalization Reconciliation Impact SOURCE: Adapted from Arribas-Baños and Baldeón 2007. 4 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS outcomes, and so on. The payment process may become more complex with differenti- ated payments, more convenient or sophisticated payment mechanisms, more attention to fraud and error control, and the like. A range of noncash benefits or requirements may be added to the program to help households improve their earnings. Monitoring and evalua- tion functions will take on increased importance the more complex the program becomes or the larger and more long lasting. Each of these functions requires systems, data, and interactions among different agencies or groups. Figure 1.1 shows some of these interac- tions. In recent years, there has been a great deal of innovation and learning in safety nets and allied programs. This book focuses on these program "how to" aspects in their myriad details. Two overarching and linked lessons are stressed: · The quality of implementation is vital. Good intentions are not sufficient; real working systems need to be developed. A badly implemented program is not worth doing. While numerous good examples exist to show that worthwhile pro- grams are possible in many settings, there are still more programs that do not deliver all they could, and some do not deliver enough to be worth the money spent. · Good safety net programs require investments in their administrative sys- tems. Excessively high overheads are obviously undesirable--but so too are insuf- ficient systems. Developing systems that allow programs to become their most effective and deliver the most value for the money will require some investment. An important part of that investment is development over time by self-critical and proactive managers. 1.3 What Is a Safety Net? In this book, the terms "safety nets" or "social assistance" is used to refer to noncontributory transfer programs targeted in some manner to the poor or vulnerable; this is a fairly commonly accepted definition (box 1.1). Some writers, especially in the United States, equate this with welfare. Sometimes, especially in Europe, social assistance connotes only means-tested cash transfer programs, but we use the term much more broadly and often substitute the term safety nets to recognize the varied forms the programs take in the developing world. We recognize that safety net is not a particularly apt metaphor. In the circus, a safety net catches those who are falling from a height; in social policy, safety net programs are meant both to help catch those falling downward economically before they land into destitution and to provide assistance or a minimum income to those more permanently poor. The programs we here include as common elements in a safety net follow: · Cash transfers or food stamps, whether means tested or categorical as in child al- lowances or social pensions · In-kind transfers, with food via school feeding programs or mother/child supple- ment programs being the most common, but also of take-home food rations, school supplies and uniforms, and so on 1. INTRODUCTION 5 BOX 1.1 Definitions of Safety Nets and Social Assistance · The Asian Development Bank defines social assistance as programs designed to assist the most vulnerable individuals, households, and communities meet a subsistence floor and improve living standards (Howell 2001). · The U.K. Department for International Development defines social assistance as noncontributory transfers to those deemed eligible by society on the basis of their vulner- ability or poverty. Examples include social transfers and initiatives such as fee waivers for education and health, and school meals (DFID 2005). · The International Labour Organization defines social assistance as tax-financed ben- efits to those with low incomes (ILO 2000). · The International Monetary Fund defines safety nets as instruments aimed at mitigating possible adverse effects of reform measures on the poor (Chu and Gupta 1998a). · The Organisation for Economic Co-operation and Development defines social as- sistance as support targeted to households that are clustered within the lower segment of the income distribution and provided to prevent extreme hardship among those with no other resources, reduce social exclusion, minimize disincentives to paid employment, and promote self-sufficiency (Adema 2006). · The Food and Agriculture Organization defines social safety nets as cash or in-kind transfer programs that seek to reduce poverty by redistributing wealth and/or protect households against income shocks. Social safety nets seek to ensure a minimum level of well-being, a minimum level of nutrition, or help households manage risk (FAO 2003). · Price subsidies meant to benefit households, often for food or energy · Jobs on labor-intensive public works schemes, sometimes called workfare · In-cash or in-kind transfers to poor households, subject to compliance to specific conditionalities on education or health · Fee waivers for essential services, health care, schooling, utilities, or transport The following further clarifies what this book does and does not consider under the rubric of safety nets. · Social protection. As used here, safety nets do not include the rest of social pro- tection--that is, social insurance programs such as pensions and unemployment insurance.To the extent that these schemes deliver benefits based on contributions of their own members, they are not safety nets; rather, they might be thought of as deferred compensation packages for affiliated employees. · Labor. The extensive regulatory aspects of labor are separate from safety nets. Active labor market policies and income support to the unemployed are closely related to--and, indeed, sometimes directly overlap with--safety nets, but most of the programs used to these purposes are well covered elsewhere and are not discussed here. 6 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS · Social policy. In our nomenclature, safety nets are complemented by social in- surance contributory programs such as pensions and unemployment insurance, and more broadly by the rest of social policy, especially in health and education, sometimes with important elements of housing or utility policy. Because we define safety nets rather narrowly, their costs are lower than some people associate with safety nets. In Uruguay, for example, total social sector expenditure (social assistance, social insurance, health, education, and other) is quite high--accounting for be- tween 20 and 25 percent of gross domestic product (GDP) between 2000 and 2005--but expenditures on safety nets per se are only 0.5 percent of GDP (World Bank 2007g). On average, expenditures on safety nets as we define them account for 1 to 2 percent of GDP, though sometimes much less or much more. Finally, note that our definition concentrates on publicly financed safety nets--that is, those funded by national or local government or by official international aid. Most often, such safety nets are delivered by the state, although nongovernmental organizations may be used as well and certain functions contracted to the private sector. Even though private action via interhousehold transfers, community support arrangements, private re- ligious contributions, private contributions to nongovernmental organizations, and other forms of charity may involve substantial flows of resources (indeed, sometimes exceeding public funds), and while the policy maker must understand the scope and shape of these privately financed safety nets, the main realm of public action is via publicly financed programs. Thus, this book focuses on the public sector. A last point on terminology: Throughout the book, it is presumed that readers are familiar with policy areas such as social protection, social risk management, or poverty reduction, and the distinct but related concepts of poverty, vulnerability to poverty, vul- nerable groups; social risk management instruments or arrangements; safety nets, social protection or social policy. For those who are not, a briefing is provided in appendix A. 1.4 How Is This Book Organized? In designing and implementing effective safety nets, the big picture and the details have to fit together and so must both be kept in mind simultaneously. The traditional meta- phor for this is to look at the big picture and then to zoom in on some detail within it. In the case of safety nets, the more appropriate metaphor might be a "picture in a picture" computer display where the big picture is kept crisp while the display with the finer detail is equally crisp as well. This presents something of a dilemma in writing a book which must be presented, if not necessarily read, linearly. We have chosen to present the big picture at the beginning and end of the book, in chapters 2, 3, 9, and 10. Chapters 2 and 3 make the case for safety nets and their financ- ing. Chapters 4 through 6 are the "how to" key processes of all safety nets; chapters 7 and 8 summarize design features and choices of specific interventions. Thus chapters 4 through 8 supply information that can help in assessing choices, culminating in a treatment in chapter 9 of principals to be used in "weaving" the safety net and fitting it into broader social policy. Chapter 10 illustrates how those principals lead to different variations of safety net systems and programs in different country circumstances. While this structure 1. INTRODUCTION 7 may make it seem that the big picture is initially left incomplete, the big decisions cannot be made wisely without a good understanding of the details involved. We recognize that some readers will approach the material sequentially while others may sample different chapters or subchapters according to their interests. To assist both sets of readers, key messages for each chapter are presented at its beginning, and the fol- lowing describes the main themes of each: Chapter 2: The Case for Safety Nets. This chapter describes and illustrates the reasons for having safety nets--how they provide immediate redistribution and poverty reduction, how they allow households to invest in their children and their livelihoods, how they help households manage risk, and how the provision of safety nets can handle redistributive concerns thoroughly, thereby enabling governments to make more efficient policy choices in other sectors. The chapter then describes how safety nets fit within the broader policy agendas for poverty reduction, risk management, and social sectors. It also describes the principal challenges to the acceptance and use of safety nets, especially in low-income countries, and provides cross-references to other parts of the book where the details of overcoming these challenges are elaborated. The chapter emphasizes that safety nets are never the whole or sufficient answer to poverty reduction or risk management, but must be fitted appropriately into the existing policy context. Chapter 3: Financing of and Spending on Safety Nets. Financing safety net pro- grams is not theoretically different from financing any other government program and is therefore seldom discussed in the safety net literature, yet policy makers concerned with implementing or reforming safety nets face a constant stream of challenges regarding their finance. This chapter is targeted to this audience, providing a brief synopsis of some of the pertinent public finance literature with illustrations from safety net programs. The first section focuses on the theory of how much governments should spend on safety nets. This is followed by a review of special considerations about each of the possible sources of funds for safety nets--reallocations from other expenditures, increased tax revenues, grants, and loans. The chapter also discusses how to secure countercyclical finance, findings from the literature on whether expenditures on the welfare state impede economic growth, and new data on how much developing countries spend in order to allow benchmarking. The chap- ter concludes with a discussion of how to share finance among levels of government. Chapter 4: Enrolling the Client: Targeting, Eligibility, and Intake. This chapter is the first of three on the processes common to all safety net programs. It briefly reviews the benefits and costs of targeting and the choice of targeting method. It then details how to implement four important steps that determine who is actually in the beneficiary group: precisely defining the eligibility criteria, conducting outreach to ensure low errors of exclu- sion, screening to ensure low errors of inclusion, and rescreening or exit policies to ensure that people move out of the program as appropriate. It concludes with a discussion of the administrative requirements to carry out these tasks. The chapter is a summary of a much wider body of literature and refers the reader to several other pertinent comparative studies. Chapter 5: Benefit Levels and Delivery Mechanisms. This chapter brings together a dispersed body of knowledge on the conceptual and practical details of program ben- efits and payments. It covers how to determine what benefit levels might be and how to structure them, reviews issues of labor disincentives and how they can be handled via the benefit structure, describes program elements designed to move households toward inde- 8 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS pendence through required or optional linkages to actions or services that are likely to help in that effort, and discusses how to handle payment mechanisms. Chapter 6: Using Monitoring and Evaluation to Improve Programs. This chapter discusses the value added by and the know-how involved in developing and using moni- toring and evaluation systems for safety net programs. A monitoring system is an essential management tool that regularly supplies information about how well a program is work- ing so that program managers can take action to improve the program's implementation. Program evaluation refers to an external assessment of program effectiveness that uses spe- cialized methods to ascertain whether a program meets some standards, to estimate its net results or impact, and/or to identify whether the benefits the program generates outweigh its costs to society. The chapter focuses on the most frequent types of evaluation used for safety net programs: process evaluation, assessment of targeting accuracy, and impact evaluation. For each type, guidance is given regarding the value of such an evaluation and how it should be conducted. Chapter 7: Understanding Common Interventions. Because there is no single rec- ipe for a safety net, policy makers must clearly understand the range of options they face. This chapter presents that menu of options: cash transfers of various sorts, food transfers, general price subsidies, public works, conditional cash transfers, and fee waivers for access to critical services. For each option, it describes key design features, outcomes, advantages, disadvantages, and lessons. The chapter emphasizes that even within a given type of pro- gram, there are many variations in detail and that how these are handled can modify the program for different circumstances and determine its degree of success. The chapter sum- marizes a vast literature and provides references to it. Many of the basic messages are not new, but many examples are fresh and the synthesis mature and useful. Chapter 8: Assisting Traditionally Vulnerable Groups. This chapter outlines the issues concerned with assisting people with disabilities, the elderly, and orphans and vul- nerable children. While the specifics vary somewhat, there are common themes in think- ing about how to serve these groups via safety nets. A key issue is whether to have special programs for these groups or to serve them within the social assistance programs designed for the wider population. This is problematic, since the members of the groups are not all poor, yet the group as a whole is poorer than average and the members have some spe- cific vulnerabilities. If they are to be helped by general social assistance programs, these programs may need to be modified somewhat. Moreover, income support is not the only public action needed to support these groups, and often it is not even the most impor- tant. Thus, the coordination of policy--or, in some cases, the integration of transfers and services--is even more important for these groups than social policy broadly speaking. Chapter 9: Weaving the Safety Net. The objective of this chapter is to help policy makers and sector specialists choose the right mix of safety net policies and programs to meet national goals. The weaving of the safety net consists of two interrelated compo- nents: fitting individual programs into a congruent whole and ensuring that the safety net sector complements the country's other social policies. The chapter presents a four-step process to assess the safety net sector: (1) diagnosing the sources of poverty and vulner- ability, (2) evaluating the effectiveness and efficiency of individual safety net interventions, (3) determining how to improve the safety net program mix, and (4) devising a concrete plan to implement the chosen strategy. 1. INTRODUCTION 9 Chapter 10: Customizing Safety Nets for Different Contexts. This chapter dis- cusses different country contexts and what they may imply for sensible safety net design and implementation. Six settings are covered--low income, middle income, in or follow- ing an economic crisis, following natural disasters, using safety nets to facilitate reform, and for rising food prices. For each setting, the chapter discusses how the safety net might be composed in terms of both the mix of programs and their specific tailoring to the situ- ation at hand. Actual program examples are provided throughout. The book also features two appendixes and a glossary. Appendix A: Basic Concepts of Poverty and Social Risk Management reviews social policy concepts used throughout the book to ensure a common understanding of key terminology and ideas. Appendix B: Main Features of Selected Safety Net Programs describes salient features of many of the programs from around the world discussed and used as illustrations throughout the book. It is intended as a selective reference on individual safety net programs. With regard to these programs, a word of explanation is in order as to how we are referring to them throughout the book. In general, we here use the names by which the programs are most commonly known in the literature, regardless of whether that term is English, non-English, or an acronym. Thus, we refer to Argentina's Trabajar program, In- donesia's various JPS (Jaringan Pengaman Sosial, or Social Safety Net) programs, Jamaica's PATH initiative, and Mexico's PROGRESA, generally without translation or explication beyond an indication of their relevance to the discussion at hand. All of these programs are fully described, and their names and acronyms translated as necessary, in appendix B. CHAPTER 2 The Case for Safety Nets KEY MESSAGES Safety nets deserve a role in development policy in all countries. They mitigate extreme poverty through redistribution of resources; they help households invest in their future and manage risks; and they help governments make sound policy decisions in macro- economic, trade, labor, and many other sectors. Safety nets face--and create--challenges to the implementing government. They com- pete for fiscal resources, require competent administration, and can result in negative incentives. These challenges demand prudent choices by program designers about the role, design, and implementation of safety nets. Fortunately, there are many options available to help manage the challenges. Safety nets are never the whole or sufficient answer to poverty reduction or risk manage- ment. They must operate within the existing policy context and be balanced with existing or planned safety nets, social insurance, and other social or poverty alleviation policies. No single prescription fits all circumstances. The mix of support to the chronic poor, the transient poor, and vulnerable groups will be complex, and, until the safety net is ad- equate for all, the subject of difficult and controversial triage decisions. "Nobody likes welfare--not the taxpayers who foot the bill, not the politicians who represent them, and not the poor who find welfare inadequate and personally degrading." --David Ellwood, Poor Support: Poverty in the American Family S afety nets contribute to poverty reduction and social risk management.1 Yet their ap- propriate scope is a fraught subject, revealing deep ambivalence and controversy among policy makers, analysts, and the general public in many countries. The wide variation in attitudes toward safety nets can be seen in the following paraphrasings of commonly held views: "We must provide for our poor--we can't let our children starve or the elderly beg." "Transfers discourage work among recipients and among those taxed to support them." "We don't need to give people fish, we need to give them fishhooks." This chapter shows how to reconcile these apparently contradictory and yet par- tially accurate views. It outlines the various arguments for having safety nets, describes the complementary role safety nets play in the broader set of poverty reduction policies and in providing adequate risk management options for the poor and the vulnerable, and outlines some of the challenges to safety nets' being an integral and permanent part of social policy in developing countries and how these can be managed. 11 12 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS 2.1 Why Should Countries Have Safety Nets? Safety nets can help achieve four objectives that are in turn part of larger poverty reduc- tion and risk management goals. · Safety nets and transfers have an immediate impact on inequality and extreme poverty. · Safety nets enable households to make better investments in their future. · Safety nets help households manage risk. · Safety nets help governments make beneficial reforms. As shown in figure 2.1, safety nets fit into the wider array of policies involved in poverty reduction, social risk management, and social protection. Safety nets are part, but not the whole, of each, and poverty reduction and risk management strategies overlap substantially but not entirely. Safety nets are not the only or even the principal tool for achieving any of the ends they serve, yet they can make a significant contribution. When situations are dire, they can help save lives. When situations are less dire--and programs are especially good--they can save or help build livelihoods as well. FIGURE 2.1 Where Safety Nets Fit in Larger Development Policy EQUITY Such as land re- distribution, enforcement of contracts & prop- erty rights, universal education, safety nets Includes labor policy, contributory social insurance, and social care services as well as safety nets SOCIAL PROTECTION SAFETY NETS SOCIAL RISK POVERTY REDUCTION MANAGEMENT For example, for small farmers, includes irrigation, microfinance, Foment pro-poor growth weather insurance, safety nets while providing services that facilitate the poor's participation in growth; safety nets are part of providing security SOURCE: Authors. NOTE: See appendix A for further explication of these concepts. 2. THE CASE FOR SAFETY NETS 13 SAFETY NETS HAVE AN IMMEDIATE IMPACT ON REDUCING INEQUALITY AND EXTREME POVERTY Safety nets can make poverty survivable or more bearable; this is their minimal function, accomplished simply by getting the transfers to the poorest. Societies have understood and valued this function for centuries, often finding sup- port for it in major religious teachings. For example, many of the rules set out in Deuter- onomy concern social justice; chapter 15, verse 11, says, "There will always be poor people in the land. Therefore I command you to be openhanded toward your brothers and toward the poor and needy in your land." The New Testament contains similar teachings. Luke 3:11 says, "He who has two coats, let him share with him who has none; and he who has food, let him do likewise." The Koran enjoins (2:177) that " righteousness is that...and give away wealth out of love for Him to the near of kin and the orphans and the needy and the wayfarer and the beggars and for (the emancipation of) the captives, and keep up prayer and pay the poor-rate." More secular, and technical, versions of the arguments that destitution and/or in- equality are to be remedied are presented by modern liberal theories of economics, which posit a social welfare function that weights the welfare of the poorer more than the welfare of the less poor. These theories say, in essence, that society benefits more if a poor person receives an extra unit of income than if a rich person does. There are many variations on the theme and numerous ways of weighting among individuals and income levels (see Barr 2004 for a discussion), but the basic notion is the moral judgment that welfare gains for the poorer are more important to society than those for the less poor. Popular support for such views is shown in opinion polls of private citizens and in summit documents signed by their governments. For example, the 2001 Latinobarómetro public opinion survey found that in all but 1 of the 18 countries surveyed, over 80 percent of the population believes the current distribution of country income to be unfair or very unfair (figure 2.2) This public attitude is also manifest in the international decrees signed by national governments. The Universal Declaration of Human Rights (UN 1948) im- plies social protection and safety net policies in article 25, stating, "Everyone has the right to a standard of living adequate for the health and well-being of himself and of his family, including food, clothing, housing and medical care and necessary social services, and the right to security in the event of unemployment, sickness, disability, widowhood, old age or other lack of livelihood in circumstances beyond his control." Article 23.3 goes further, saying, "Everyone who works has the right to just and favourable remuneration ensuring for himself and his family an existence worthy of human dignity, and supplemented, if necessary, by other means of social protection." (For a good discussion on right-based ap- proaches to social protection, see Piron 2004.) SAFETY NETS ENABLE HOUSEHOLDS TO MAKE BETTER INVESTMENTS IN THEIR FUTURE Safety nets allow households to take up investment opportunities that they would other- wise miss--both with regard to the human capital of their children and the livelihoods of household earners--despite credit market failures. Specifically, safety net programs can contribute to capital accumulation among the poor by preventing the negative outcomes 14 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS FIGURE 2.2 Perceptions of Fairness of Country Income Distribution in Latin America Very fair Fair Unfair Very unfair Average Argentina Bolivia Brazil Chile Colombia Costa Rica Ecuador El Salvador Guatemala Honduras Mexico Nicaragua Panama Paraguay Peru Uruguay Venezuela, R.B. de 0 20 40 60 80 100 Percentage of respondents SOURCE: Lindert, Skoufias, and Shapiro 2006, p. 24. Responses to question: "Do you think that the income distribution is...?" posed in the 2001 Latinobarómetro. of malnutrition and underinvestment in education, and by enabling investment in pro- ductive assets. Preventing Malnutrition Many children in developing countries are affected by malnutrition. In low-income coun- tries, 43 percent of children aged 0­5 are underweight, compared to 11 percent of their peers in middle-income countries (World Bank 2006m). Within each country, the poor are disproportionately affected by malnutrition; in fact, the prevalence of malnutrition is often two to three times higher among the poorest income quintile than among the wealthiest (Wagstaff and Watanabe 2000). Malnutrition accounts for about half of the 10 million deaths each year among children aged 0­5 in developing countries (Wagstaff and Watanabe 2000). Moreover, there is strong scientific evidence that malnourished sur- vivors are sicker, more disabled, weaker, less educated, and have a lower cognitive ability than better nourished counterparts.2 These outcomes are individually unacceptable and, in aggregate, reduce economic growth. The Food and Agriculture Organization has estimated that at least US$120 billion per year of benefits would be generated through the longer, healthier, and more productive lives of the 400 million people freed from food insecurity if the first Millennium Develop- ment Goal of halving hunger by 2015 were met (FAO 2002). The same report cites studies on India, Pakistan, and Vietnam that show that the combined effect of stunting and iodine and iron deficiencies reduced gross domestic product (GDP) by 2 to 4 percent per year. Not surprisingly, investments in reducing malnutrition are cost-effective. Behrman and 2. THE CASE FOR SAFETY NETS 15 Rosenzweig (2001) estimate that every US$1 invested in an early childhood nutrition pro- gram in a developing country could potentially return at least US$3 worth of gains in aca- demic achievement alone, without even considering the other benefits that would accrue. Inadequate income can contribute to malnutrition by curtailing diets in quantity and quality; reducing access to services such as health, education, water, and sanitation; and affecting the knowledge and time available for and dedicated to adequate feeding of young children. The transfers inherent in safety net programs can allow households to increase the quantity and quality of food they consume. They can also help provide com- plementary inputs to good nutrition and alleviate constraints to adopting the behavioral changes promoted by nutritionists. They can, for example, enable households to purchase containers for transporting, storing, and treating water or to purchase water itself and soap so that promoted hygiene practices can be followed. In some cases, additional income may allow households to rearrange responsibilities so that children have more adequate caretak- ers, which would in turn facilitate the multiple active feeding sessions needed each day for very young and already malnourished children. Many safety net programs have spillover effects beyond the direct transfer. The con- struction or maintenance of roads, markets, irrigation, drainage, domestic water supplies, schools, clinics, and the like accomplished through labor-intensive public works jobs can help improve livelihoods, food availability, and/or services for the poor. Prepared meals, take-home food rations, food stamps, and cash can all be (and often have been) linked to the use of health services, which usually include prenatal care, nutrition education, and growth monitoring; or to schools where children are then more accessible for nutrition education, deworming, vitamin supplements, or fortified school feeding. Preventing Underinvestment in Education There is ample and systematic evidence that chronically poor families are less likely to obtain adequate schooling for their children (World Bank 2005n). Children who do not have the opportunity to attend, or who are withdrawn early from school, face a lifetime of lower earnings (Hoddinott and Quisumbing 2003). Basic transfers can help households bear the direct costs of schooling--tuition and fees, transport, school supplies, uniforms. These costs can be quite substantial; a study of Bangladesh, Kenya, Nepal, Sri Lanka, Uganda, and Zambia undertaken by the United Kingdom's Department for International Development found that education spending was second only to food expenditures (DFID 1999). School fees alone can range from 5 to 20 percent of total household consumption (World Bank 2006c). And of course on top of the direct fees, the opportunity cost of the child's time for paid and unpaid labor must be taken into account. Good safety net programs have been effective in helping households build human capital. The largest body of evaluation evidence comes from the new wave of conditional cash transfer (CCT) programs that condition receipt of benefits on meeting prescribed (often quite substantial) levels of service use. Evidence is very strong that these programs raise school enrollment rates (box 2.1), especially among the most disadvantaged groups, and can raise the use of health services. Social pension programs too, even though ex- plicitly designed to protect the elderly poor, have had demonstrated effects on increasing the human capital of both children and the elderly in households (Carvalho 2000a; Case 2001; Duflo 2003). 16 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 2.1 The Motivating Force of Educational Stipends: The Bangladesh Female Secondary School Assistance Program Selina is 18 years old. She hails from Daragaon Village under Chunarughat Upazila of Habiganj district. Selina comes from a poor family. There are seven brothers and sisters in her family, and Selina is the third oldest. Her father, Abdur Razzak, is a retired guard of the Daragaon Tea Estate. He has a piece of land that he cultivates, but this land is too small to support his family. Abdur Razzak has always been eager to educate his children, but he has not been able to af- ford to educate his first two children up to secondary school certificate level. The new Female Secondary School Assistance Program renewed his optimism that he will be able to provide Selina with such an education. Selina was enrolled in Mirpur Girls High School in 1995, which is about five kilometers from her home. She received a regular stipend while attending grades 6­10. She used a part of the stipend money to pay for her commute to school; she had to walk about 1.7 kilometers to the Kamaichauri police box, from where she would catch a bus to the school. The commute was hard for Selina, but she knew that she had to maintain 75 percent attendance in school to con- tinue getting the stipend, so she was seldom absent. Because of her regular attendance, she was able to improve her results every year; when she was first enrolled in grade 6 she was 72nd in her class, in grade 7 she was 6th, and in grade 8 she was 5th. In 2000, Selina passed the secondary school certificate exam in humanities with a B average. SOURCE: Ahmed 2004a. Investing in Productive Assets Growing evidence indicates that cash assistance can help households not only subsist but actually improve livelihoods by investing a portion of the transfers they receive. Gertler, Martinez, and Rubio-Codino (2006) report that beneficiaries of Mexico's Oportunidades CCT program invested about 12 percent of their transfers, allowing them to raise their consumption by about a third after five and half years in the program. Earlier, Bezuneh, Deaton, and Norton (1988) found that food-for-work programs in northern Kenya dur- ing the lean season allowed households to purchase additional agricultural inputs and increase net returns from their farms by 52 percent. Sadoulet, de Janvry, and Davis (2001) note similar findings for PROCAMPO, Mexico's Program for Direct Assistance in Agri- culture, which involves a transfer to farming households meant to compensate for income losses expected from the adoption of the North American Free Trade Agreement. House- holds invest the funds to purchase inputs and serve as collateral for borrowing, resulting in a positive multiplier effect in the range of 1.5 to 2.6 for the program. This investment effect was found in social pension schemes as well in Bolivia, Brazil, and South Africa. The schemes studied are pure transfers targeted to the elderly for con- sumption purposes--a classic case of programs geared exclusively at avoiding destitution. Yet they have had important welfare impacts well beyond that goal. They have reduced poverty (Case and Deaton 1998); improved access to credit, thanks to the regularity of pension payments (Ardington and Lund 1995); and resulted in higher levels of investment 2. THE CASE FOR SAFETY NETS 17 in households' physical capital (Delgado and Cardoso 2000). More recent evidence from Bolivia's Bono Solidario program suggests that, in rural areas, pension recipients are invest- ing their transfers in smallholder agriculture; as a result, their food consumption has gone up by twice the amount of the transfer received (Martinez 2005). To allow households to invest, programs should offer benefits for a reasonably long period, and not withdraw them if the income or assets of the households increase slightly. It is not quite clear whether all safety net programs stimulate investment among their typi- cally poor beneficiaries with the effect proportionate on the size of the benefit. Some (see Carter and Barrett 2006, for example) argue that there are "threshold effects"--that is, until households reach a certain minimum threshold of welfare, they cannot invest effec- tively. The literature on this is incipient, and the evidence is not clear (see Lentz and Bar- rett 2005). The idea of such a minimum is consonant with the fact that the programs for which clear investment impacts have been shown have benefit levels higher than average in developing countries. Of course this may also be, at least in part, because the measurement of such effects is difficult and effects are easier to observe if they are large. Public works programs can help achieve the longer run welfare of households or communities through a different investment channel--the construction or maintenance of infrastructure that yields services to households. Access to roads can help households get their products to market; journey to work outside their villages; or obtain health, educa- tion, or other government services. In Bangladesh, Khandker, Bakht, and Koolwal (2006) found that certain road improvement projects led to a 27 percent increase in agricultural wages and an 11 percent increase in per capita consumption. Road improvement also led to an increase in schooling of both boys and girls. SAFETY NETS HELP HOUSEHOLDS MANAGE RISKS When families, especially poor families, face reductions in income or assets, they may re- sort to costly coping strategies that perpetuate poverty, such as selling their most produc- tive assets. Moreover, when risk becomes too threatening, households may try to reduce it, thereby making livelihood choices that reduce their earnings. A good safety net can reduce the need for either of these strategies which can trap households in poverty. Reducing the Incidence of Negative Coping Strategies There is clear evidence that families that suffer from short-term shocks may be forced to cut back on the feeding or schooling of their children; deterioration in nutritional or health status is found more often than withdrawals from school. Thus, Peruvian children suffered higher infant mortality during the country's 1988­92 economic crisis (Paxson and Schady 2004). Enrollment rates dropped during the Indonesian financial crisis, es- pecially for the poor and those in rural areas (Frankenberg, Thomas, and Beegle 1999). Hoddinott and Kinsey (2001) trace a whole series of effects on Zimbabwean children who were 12­24 months old when affected by a drought. They show that stunted preschoolers will have lower height during adolescence, will delay school enrollment, and will reduce grade completion. The magnitudes of these impacts are quite large: the 1982­4 drought shock resulted in a loss of stature of 2.3 centimeters, 0.4 grades of schooling, and a school start delay of 3.7 months for this age group. Using estimates of the values for the returns to education and age/job experience in the Zimbabwean manufacturing sector, the re- 18 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS searchers calculate the shock impact as translating to a 7 percent loss in lifetime earnings for the affected children. Safety net programs can help prevent such losses. Intheabsenceofsafetynets,shocksmayforcepoorhouseholdswithlowcopingcapac- ity to sell their productive assets. Families that have to disinvest in their livelihoods--eating their seed grain; selling their draft animals or the tools of their small enterprise; or default- ing on rent or mortgage payments and consequently losing their homes, farms, or work- shops--will find it very difficult to rebuild their BOX 2.2 The Loss of an Ox earning capacities. The effect will be all the more The long-run effects of asset loss are powerfully illustrated marked if there are inad- in this case study: equate credit markets,3 and, as discussed in box- Ato Mohammed, 55 and illiterate, resides in the Bati dis- es 2.2 and 2.3, the assets trict of South Wollo Zone (Ethiopia) and heads a house- hold of nine. He has [sic] been chronically food insecure required to rebuild liveli- for more than 10 years when he lost his only oxen due hoods are relatively large to drought. He sold the animal to buy food at the time and has not been able to acquire another. Currently, or lumpy so that a family Mohammed holds one hectare of farm land and he has must make a big purchase no grazing land. Since he owns no oxen, he has been before it can return to leasing out the land for share-cropping on a 50/50 shar- ing arrangement. Mohammed and his family members its full earnings potential are engaged in various types of daily labor activities for (Carter and others 2004; cash and food, and the household is a regular recipient Fafchamps, Udry, and of food aid. Mohammed asserts "oxen are the crucial produc- Czukas 1998; Jalan and tive asset that would liberate me from this insecurity Ravallion 2002; Lokshin trap." On the other hand, however, he does not want to take credit from a regional credit organization to buy an and Ravallion 2000). In ox as he does not want to be indebted and fears that the the wake of Hurricane debt may be passed on to his children if he fails to repay. Mitch in Honduras, for He fears that the ox may die due to lack of adequate feed or animal diseases for which there is no depend- example, Carter and oth- able animal health service in the community. He also ers (2007) calculate that fears that he may not be able to pay back since crop a loss of 10 percent of a failure is frequent due to insects and droughts. (Carter and others 2007, pp. 835­36). poor household's assets would result in a rate of growth in household in- come over the following 2.5 years 18 percent lower than if the assets had not been lost; a similar loss of assets would lower the growth rate of richer households by only 9 percent. When available, safety nets have reduced the incidence of negative coping strategies. For example, in response to the financial crisis of 1998, the Indonesian government put in place a system of targeted fee waivers for public health care and scholarships for poor schoolchildren. Both programs have been evaluated to show that service use fell less among recipient households than they would have in the absence of the programs (Cameron 2002; Saadah, Pradhan, and Sparrow 2001). Children benefiting from the pilot cash transfer scheme in Kalomo District, Zambia, are eating better and are less underweight (MCDSS and GTZ 2007). CCT programs targeted to the chronic poor have helped beneficiaries affected by shocks withstand them (see de Janvry, Sadoulet, and others 2006 for a review). In Nicaragua and Honduras, beneficiary households hit by the coffee crisis were able to 2. THE CASE FOR SAFETY NETS 19 maintain their children's BOX 2.3 A Poverty Trap in Shinyanga schooling and not increase child labor. In Nicaragua's In Shinyanga, cattle are a high-return investment (25 to case, consumption did 30 percent annually). Cattle are also a liquid asset that can not fall as much as for be used for consumption smoothing, which makes cattle nonbeneficiary families; ownership attractive. But they are also a lumpy investment. in Honduras, adults were Wealthier rural households have been found to specialize able to increase their labor. in cattle rearing, while poorer households derive a larger In Mexico, beneficiary share of their income from off-farm activities. Differences in children in households comparative advantage do not offer a convincing explana- that were hit by shocks tion for this phenomenon. Households specializing in off- were able to maintain farm activities have much lower incomes but are unlikely their school enrollments, not to have the skills required because cattle rearing is a in contrast to similarly traditional activity in the area. affected nonbeneficiary The lack of credit markets and the indivisibility of cattle im- families. ply that households must be able to put up relatively large amounts of money to invest in cattle rearing. However, Managing Risks Ex Ante poor households with low initial endowments from which Families that are so poor only low incomes are earned find it hard to save enough they cannot afford a bad to invest in cattle. That problem is exacerbated by the fact year may minimize the that, because of low endowments, the poor have limited variance of their incomes ability to cope with shocks. Consequently, such households in ways that also lower enter into safe, lower-return activities, making saving even the means. They may harder. That combination of factors explains why poorer plant low-risk, low-return households specialize in off-farm activities (such as weed- crops; abstain from in- ing or casual labor) that require few skills or investments vestments in fertilizer; but are safe. That pattern effectively traps poor households diversify activities rather in poverty, despite the attractive investment opportunities than specialize in those that exist in the area. with highest return; and SOURCE: Dercon 1997. keep savings in liquid but low-return forms. The evidence of the cost of such ex ante risk management mechanisms is hard to compile, and mostly comes from very poor rain-fed smallholder agriculture or pastoralism; intuition suggests that the issue could apply equally well to poor urban households. The calculated impact in studies to date is quite substantial (see Dercon 2006 for a review), indicating that the poor sacrifice as much as a quarter of their income in return for greater security. The underlying problem faced by the poor in addressing risk management is a lack of insurance against risk. Some safety net programs can provide an insurance function to help households avoid taking ex ante risk management decisions that lower their incomes.4 If households could know that in the event of a bad year they would have reliable access to a safety net program, they could make their income and investment decisions based more on return and less on security. Walker, Singh, and Asokan (1986) quantified this effect to some extent when looking at the income streams for landless agricultural laborers in two 20 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Indian villages, one in which the statewide Maharashtra Employment Guarantee Scheme operated and one in which it did not. Households in program villages had 50 percent less variable income streams than did their nonprogram counterparts. Ravallion (1991) con- trasts the high correlation of distress land sales with famines in Bangladeshi villages and the lack of such correlation in India, where public works programs were operating. For safety net programs to deliver this insurance effect, they must provide a credible ex ante guarantee of quick assistance in time of need--something that few programs to date have done. For example, a program could, like the Maharashtra Employment Guar- antee Scheme, provide a minimum number of days of employment on a guaranteed public works initiative. India's new National Employment Guarantee Scheme is meant to deliver such an insurance benefit, but evidence of its impact is not yet available. Another ap- proach could be a cash transfer with a means test or other entry criterion sufficiently agile to permit new entrants to be assessed and granted benefits promptly when their incomes decline. A few Eastern European programs have managed this. Regardless of approach, the program must have an adequate budget, since the guarantee of an adequate income floor is critical to the change in risk-taking behavior sought. Because this guarantee has rarely been available to households in practice and over long enough periods for them to gain confidence in it, the role of safety nets in ex ante risk management has been largely unrealized to date. SAFETY NETS HELP GOVERNMENTS MAKE BENEFICIAL REFORMS Safety nets can support good social policy with varying degrees of directness. A strong social assistance program can directly replace inefficient redistributive elements in other programs. Less directly, safety nets play a role in helping governments adopt or sustain sound macroeconomic, trade, and other policies. Least directly, safety nets may reduce inequality over the short term, thus tempering the high inequality harmful to the devel- opment of sound institutions that underlie good policy and governance. Replacing Inefficient Redistributive Elements in Other Programs Many sectors and programs historically have intertwined equity and efficiency goals; the general thrust of recent policy has been to focus on efficiency. Recognition of the need to think explicitly about the distributional impact of these policy reforms has been manifest in the call in the development community for "poverty and social impact analysis" to be part of the policy decision process (see World Bank 2003i). The goal is to design reforms with fewer losses to the poor or that compensate them, sometimes through sector-specific compensatory mechanisms, but often through a more general safety net program as dis- cussed below. Indeed, Kanbur (2005) and the World Bank (2005n) suggest that the use of a specialized redistributive mechanism such as a permanent transfer program is preferable to designing specific compensatory packages for each reform option. Take the example of labor markets. The growing consensus is that for them to be efficient and play their role in an investment climate conducive to growth and poverty reduction, they must be relatively flexible. Further, they must aim to protect workers, who will likely change jobs at least once and possibly several times over the course of their lives, rather than protect jobs per se. This focus implies a lessening of the role of labor market regulation in worker protection and an improvement of income support to the unem- 2. THE CASE FOR SAFETY NETS 21 ployed. Safety nets, especially in the form of workfare schemes, may be a sensible means of providing such income support. They can be used to complement unemployment insur- ance, as they were in Argentina and the Republic of Korea; or as substitutes for it, which was done in Bolivia and Peru (Vodopivec 2004; World Bank 2004e). Eastern Europe and Central Asia's experience with utility pricing provides another example of how the provision of social assistance can enable efficient policies in other sec- tors (Lampietti 2004; Lovei and others 2000; Saghir 2005; Shopov forthcoming; World Bank 2000b). Gas, electricity, and heat prices were set very low during the socialist era. With the transition to the market, this practice became unsustainable and by the mid- 1990s could no longer go unaddressed. Utility companies could not continue to shoulder the losses, and governments lacked the resources to cover the costs of the price subsidies. Domestic utility prices had to rise, and countries experimented with ways to reform the sector for greater efficiency while protecting consumers at least partially. Some countries chose to ensure a minimum provision of utility services explicitly via social assistance rather than implicitly through utility pricing. This approach has much to recommend it. It makes both the subsidy budget and the efficiency of the utility company more explicit and transparent. It allows for greater targeting, if that is desired, and can take advantage of eligibility and payment systems common to other social assistance programs. The approach also provides appropriate economic incentives to consumers, as they will save by reducing usage. In Bulgaria, for example, the same staff administers the heat- ing allowances system and the Guaranteed Minimum Income Scheme, using broadly the same methods, instruments, and budget level--0.23 percent of GDP for each program. However, eligibility thresholds are higher for the heating allowance, and the benefit is paid only during the cold months of the year. Where a social assistance program exists or can feasibly be created, such as Brazil's Auxilio Gás (Cooking Gas Grant) program, it can take the burden of social guarantees off utility companies and allow them to focus on efficient service provision.5 Facilitating Changes in the Economy Aimed at Supporting Growth There will be less opposition to reform when there are mechanisms to compensate losers or to assist the poor who often become poorer during a downturn; less opposition to reform allows for better macroeconomic policy and growth. Rodrik (1998) provides supporting empirics for this premise, looking at the presence of mechanisms for societal conflict reso- lution (safety nets and social insurance among them), macroeconomic policies, and robust growth. See box 2.4 for how the U.S. Federal Reserve Board chairman recently explained the issue in a business context and box 2.5 for an account of how the British Poor Laws helped fuel the agricultural revolution which in turn fueled the industrial revolution. Following the Latin American debt crisis of the 1970s and 1980s, many governments in the region implemented structural adjustment policies--macroeconomic and sectoral policies designed to downsize ailing branches of the economies that generated losses or that were kept afloat via costly fiscal or quasi-fiscal subsidies. These policies would allow a more efficient use of resources over the medium term. However, over the short term, workers in the affected industries would lose their jobs, suppliers to these industries would lose business, and so on. Governments soon recognized that safety nets could facilitate these reforms with immediate costs and delayed benefits. Chile, for example, instituted a 22 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 2.4 A Policy Maker's Take on Growth, Equality, and Policy Steven Pearlstein (2007, p. D1) of the Washington Post discusses U.S. Federal Reserve Board Chairman Bernanke's speech to the Omaha Chamber of Commerce: Perhaps the best part of Bernanke's speech yesterday was the graceful way he framed the tradeoff between growth and equality. One reason the U.S. economy is the most productive, the most dynamic, the most innova- tive in the world, Bernanke explained, is that we offer the biggest rewards to skill, effort and ingenuity. We also have an economic framework that not only allows companies and individu- als the flexibility to adapt to changes in technology or consumer tastes or competition, but re- wards them handsomely when they do. Bernanke says the flip side of this dynamism has been to generate not only a higher level of inequality, but also a higher level of economic insecurity. Now, he says, the only way to make these politically acceptable is to "put some limits on the downside risks to individuals affected by economic change." One way to limit those risks, of course, would be to restrict trade, impose new regula- tions on labor and product markets, or use the tax code to massively redistribute incomes. For Bernanke, the costs in terms of slower growth and higher unemployment would be too high. The better alternative, he argued, is to preserve the political consensus for open and flexible markets by offering Americans a stronger economic safety net. large public employment program in 1975 which continued for several years to accom- pany deep reforms of its economy (Raczynski and Romaguera 1995). Bolivia instituted an Emergency Social Fund in 1987; the World Bank's loan in support of it was the first project by the agency specifically designed to protect the poor during a macroeconomic adjustment initiative (Jorgensen, Grosh, and Schacter 1992). Such support has become widely accepted over the years. For example, Chu and Gupta (1998b, p. v) summarize the International Monetary Fund's position on the issue, noting that "the only realistic alter- native is to proceed with the necessary adjustment policies but complement them with the adoption of social safety nets." This logic might suggest that a safety net is a temporary solution to be implemented in the wake of a crisis. However, experience has shown that it is difficult to start a program from scratch and get it up and running during a crisis. One of the features of safety net policy most agreed on after the East Asian financial crisis was that safety net programs should be built during good times and expanded during bad; the Asian Pacific Economic Consortium's lessons and guidelines paper notes: Social safety nets should be in place before a crisis occurs. Permanent, rather than ad hoc, social safety nets can more effectively protect the poor from the adverse effects of crises without compromising longer-term goals. During good economic times, social safety net instruments help to alleviate poverty among the chronically poor and those suffering from the effects of non-economic shocks (APEC 2001, p. 6). Fostering More Inclusive Growth Transfers play both direct and indirect roles in reducing inequality. Reducing inequality should help create a "virtuous circle," leading to more inclusive institutions and thus indi- rectly to better policy and higher growth. 2. THE CASE FOR SAFETY NETS 23 BOX 2.5 The Developmental Effects of the Elizabethan Poor Laws The famous British Poor Laws were written in 1598 and 1601.They provided a system of social security--relief in kind for the "helpless" poor (the ill, the elderly, children), workfare or wage subsidies to the able-bodied, apprenticeships to children, and foster care for orphans.The poor laws were centrally mandated, but wholly financed by local property taxes and implemented at the parish level. Of course, their application varied from place to place and over the centuries. The World Bank provides an interesting view of their impacts on economic development: The comprehensive social security system provided by the Poor Laws had a number of highly significant economic consequences. In combination with laws (dating from the thirteenth cen- tury) granting complete alienability of land, it encouraged labor mobility and reduced the at- tachment to land holding as the only form of security for peasants. Individuals had a relative certainty of being provided for, wherever they moved to work in the economy, no matter what their property-ownership status. Landlords and farmers could reap the economic gains to be had from increased farm sizes, from enclosure, and from laying off workers or changing their labor contracts to more efficient weekly or day labor, without provoking the same degree of peasant protest as occurred on the continent. But equally, employers in England had a strong incentive only to do this if it made economic sense because, through the Poor Law, they would also have to reckon with their liability to pay for the families of the laid-off workers (World Bank 2005n, p. 120). New evidence is emerging that high levels of inequality can be costly to growth and poverty reduction (see De Ferranti and others 2004 and World Bank 2005n for extensive literature reviews). High inequality slows economic growth and development itself. When political and economic inequalities are great, they can lead to the development of institu- tions and policy choices that favor the generation of profits to particular groups rather than a broader base of growth. This narrowing of benefits is in turn bad for the invest- ment, innovation, and risk taking that underpin long-term growth. As an example, Haber (2001) attributes the large economic gap that opened between the United States and Mexico during the 19th century to the difference in the competitive- ness of their banking industries, which reflected the differences in political institutions in the two countries at that time. In the United States, political institutions allocated power to a broad base of people who wanted access to credit and loans, which led governments to allow free entry into banking sectors--which in turn resulted in a highly competitive market, low interest rates, and high investment rates. In contrast, Mexican political insti- tutions granted banking monopoly rights to a cabal of political supporters, which resulted in an oligopolistic market structure and lower levels of credit and investment, but gener- ated high profits for the select few. Initial differences reinforced themselves, resolving in the United States into a virtuous circle and in Mexico, a vicious one. High levels of inequality can also hamper the ability to manage economic volatility and worsen the quality of macroeconomic response to shocks. With higher inequality, the institutions responsible for sharing the burdens of adjustment work less well, and defini- tive policies are harder to establish. The sluggish growth performance in Latin America following the oil shock of 1973 is explained, in a cross-country regression setting, by the higher income and land inequality and higher murder rate during that period (Rodrik 24 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS 1999). High levels of inequality may make violent crime more pervasive, and crime is bad for growth. In Jamaica, a 1 percent increase in youth violence was estimated to decrease tourism revenue by 4 percent (World Bank 2003e). The direct role of transfers in redistribution is obvious and can be effective. Brazil's Bolsa Familia (Family Grant) and Mexico's Oportunidades programs each reach about a quarter of the national population with benefits of about a quarter of household base income. They lower the respective countries' Gini coefficients by 2.7 points, or about 5 percent. These unusually large impacts are due to the size and relative generosity of the programs involved. In Latin America as a whole, the full tax and transfer system reduces the Gini for incomes by only 2 points; in Europe, the Gini for disposable incomes is 15 points lower than that for market incomes (Perry and others 2006). This finding sug- gests that there is more room for tax and transfer policies in reducing high inequality in Latin America, and likely in many countries outside the region with similarly high levels of inequality. Additional Empowerment Effects Indications--albeit tending more toward anecdotal than econometric evidence--that safety nets may have empowerment effects that go beyond the transfer of income are beginning to emerge. These effects seem to stem from the way programs can pull their participants into new roles. CCTs, for example, have been shown to increase enrollments, especially in secondary school. In the long run, the increase in education achieved should lower inequalities in education, thereby lowering inequality in autonomous incomes. It should also ensure nation-building benefits of shared education and, if schools offer a sense of connectedness, help youth steer clear of costly risky behaviors. The impacts on enrollment are greatest for the most disadvantaged beneficiaries--the poor, females, and ethnic minorities--which should imply that the empowerment effects will be greatest for them too (World Bank forthcoming). There are intriguing, if not yet well-documented, indications that some CCT pro- grams may have even farther-reaching subtle and qualitative effects. In Colombia, mothers have to go to their children's schools regularly to handle the associated paperwork. Pro- gram officials report that this increased frequency of contact is breaking down traditional status-based barriers to teacher-parent communication (Combariza 2006). Levy (2006) reports that in Mexico, communities are now putting greater pressure on teachers to reduce their absenteeism. Participation in community groups is giving poor women new experiences in leadership and community action. In Turkey, women are registering marriages and children who would otherwise have been undocumented and are thus gaining protection under family law; they are also going to government offices, banks, and town centers to handle program-related paperwork--a type of errand many of the Kurdish mothers have never performed (Ahmed and others 2007). Voter turnout among the poor was higher in Brazil's 2006 presidential run-off than expected, which some analysts attribute to the workings of Bolsa Familia (Hunter and Power 2007), indicating that political power is being shifted along with economic power. These indirect effects of safety net programs seem to work toward increasing the inclusion or voice of the poor in ways that complement and reinforce the increase in income and may contribute to the formation of a virtuous circle. 2. THE CASE FOR SAFETY NETS 25 SAFETY NETS FOR PROTECTION AND PROMOTION The role of safety nets and their objectives show that they can have a protection and pro- motion function. They protect the poor from the worst of destitution and from falling deeper into poverty when faced with an economic shock. They also promote indepen- dence, allowing households to invest and thereby improve their livelihoods, and allowing governments to choose more efficient policies, which result in stronger growth and pos- sibly higher levels of consumption for all households. To meet these various roles, objectives, and functions, safety nets must incorporate specific design features (table 2.1). For a safety net to be able to protect the extreme or chronic poor from the full burden of their poverty, the net transfer (excluding the taxes that finance the transfer) must be redistributive, and its effects will be greater for more marked degrees of redistribution. Individual households facing shocks can be protected from irreversible losses if they gain access to the program in time; thus, program entry cri- teria and processes must be sufficiently open. Benefits will be proportional to the number of needy served--even if some cannot be accommodated, protection will be effective for those who are. The requirements are more demanding for safety nets to help households manage risk ex ante, since households must have a credible guarantee that assistance will be available when needed. Thus, both the intake processes and budget must be flexible and sufficiently demonstrated for households to trust the program. Promotion via assis- tance to sound government policy choices will be best achieved through a safety net that is permanent, but that can expand or add elements as needed to deal with specific shocks or reforms. TABLE 2.1 Safety Nets for Protection and Promotion Nature of benefit Safety net role/objective Protection Promotion Design elements required to deliver benefit Provide transfers that can X Progressive redistribution at least, often accomplish redistribution narrow targeting is chosen Enable households to make X May be inherent; evidence unclear better investment in their future whether any size transfer is enough for a promotive effect or whether transfers must move a household above a certain threshold to realize this effect Help households manage risks · Avoid irreversible losses X · Easy access once need is felt · Allow higher risk/return X · Credible guarantee that help will be activities available when needed Help government make sound X X Base safety net would be permanent; choices may be supplemented in times of covariate shock or with temporary compensatory programs to accompany some reforms SOURCE: Authors. 26 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Various arguments suggest that good safety nets should contribute to growth. This is, however, a difficult claim to substantiate with robust empirical evidence. Where effects should be direct--for example, safety nets allowing households to overcome credit market deficiencies and make productive investments--there is evidence at the household level but not at the community or macroeconomic level. Even if household effects are large, the weight of these households in total production is small, so the macroeconomic effects may not be large in aggregate, although clearly important in poverty reduction. A more sub- stantial aggregate effect on efficiency or growth might be expected where social assistance enables sectoral, trade, or macroeconomic reforms. However, because the causal chain is rather long and indirect, it remains impossible to quantify the full benefits of safety net policy. It must also be acknowledged that many safety net programs and systems operated today do not meet all the requirements to deliver their potential benefits. Few programs operate with credible guarantees and so are unlikely to change household ex ante risk management decisions. Many offer insufficient coverage or benefits to yield large protec- tion or promotion effects. And a few are admittedly outright failures of administration. Nonetheless, the general trend is toward more substantial safety nets, more sophisticated understandings of how to run them, and more credible estimates of their impacts. In 10 years, both safety net practices and the evidence of their worth will likely be much stronger than is the case today. In the interim, it is probably safe to say that relatively few policy makers or voters fully perceive the efficiency-enhancing role of safety nets, and base their support primarily on the redistributive rationale. This undervaluation of safety nets probably leads to their underprovision. 2.2 How Do Safety Nets Fit in Wider Development Policy? Safety nets, while extremely useful, are never the only or wholly sufficient solution to pov- erty and risk. Rather, they are part of a country's development policy. To determine the parameters of their role and see how safety nets fit in with the other instruments a govern- ment uses to address poverty and risk, it is useful to examine which groups of households can and should benefit from safety net programs. Safety nets may serve one or a combination of the following groups: · Chronic poor. Members of this group lack the assets (broadly defined) to earn sufficient income, even in "good" years. The Chronic Poverty Research Centre (2004) estimates that between 300 and 420 million people are chronically poor. This is a substantial subset of the 1 billion people--18 percent of the world's population--who live on less than US$1 per day (Chen and Ravallion 2007).6 · Transitory poor. Members of this group earn sufficient income in good years but fall into poverty, at least temporarily, as a result of idiosyncratic or covariate shocks ranging from an illness in the household or the loss of a job to drought or macroeconomic crisis. Transient poverty is apparently very substantial. Baulch and Hoddinott (2000) review a number of studies that show that, in a typical year, anything up to half of the US$1 per day poor may be "transient" poor, 2. THE CASE FOR SAFETY NETS 27 meaning that their stay in poverty will be relatively short--less than one, two, or five years, depending on the study. · Vulnerable groups. Membership in these groups overlaps with the chronic and transient poor since these same individuals may also have low assets or face shocks. However, some individuals within these groups will not be poor, especially where the vulnerability is an individual one and the individual is part of a nonpoor fam- ily or community. Some large vulnerable groups commonly served by safety nets are listed below; there will be others that are locally pertinent, such as minority ethnic groups. ­ People with disabilities. Statistics are problematic, but 10 to 15 percent of the world's population may be disabled, with 2 to 3 percent with severe disabilities that put them in need of income support (Mont 2007; WHO 2008). ­ Elderly. People aged 60 and above account for about 10 percent of the global population at present; this proportion is projected to reach about 21 percent by 2050. About 12 percent of this elderly population is older than 80; this proportion is expected to increase to about 19 percent by 2050 (UN 2002). ­ Orphans. There are 143 million orphans (children who have lost one or both parents) in Asia, Sub-Saharan Africa, and Latin America and the Carib- bean; of these, 16 million are double orphans who have lost both parents. In some countries, as many as 15 percent of all children are orphans (UNAIDS/ UNICEF/USAID 2004) ­ Refugees. There are 20.5 million international and internal refugees, returned refugees, internally displaced people, asylum seekers, and stateless people (UNHCR 2003). · Losers in reforms. The number of losers, and the extent of their loss, is very re- form specific. Global numbers for this category are thus unavailable, and analysis must be conducted for each case. Policy makers can then determine how broadly or narrowly to focus compensation and how much is required. Because transitory poverty can be as high as half of total poverty, most societies will feel the need for safety nets both to help households cope with shocks and to provide some sort of assistance for the chronically poor. These two groups overlap incompletely with those with specific vulnerabilities, making the triage process yet more complex. Transfer policy also is often motivated by a desire to compensate losers in the reform of other schemes that have affected patterns of income or welfare. The losers may not be poor, but transfers may be called upon to compensate them. The challenge is to strike the right balance among groups to serve, the reasons for doing so, and the instruments to use. Table 2.2 carries through the logic of the chapter, showing different goals that safety nets can help to achieve, the groups that can be reached, the specific roles/objectives of safety nets and some of the complementary policies for that group. The role of safety nets within the overall development policy mix has grown over the last 20 or more years, for two reasons. First, the move to markets and liberalization--not only in the centrally planned economies of Eastern Europe, the former Soviet Union, and TABLE 2.2 Possible Target Groups, the Role of Safety Nets, and Complementary Policies Motivation/goal Group Role/objective of safety net Design element required Complementary policy Mitigation of Chronic poor/ Provide transfers and support Progressive, possibly Labor-intensive growth; access to adequate poverty extreme poor to reduce inequality and narrowly targeted, health, education, water, electricity and unacceptable deprivation redistribution transportation services, microfinance and agricultural extension, and so on Vulnerable groups, Interventions to encourage inclusion in including the elderly, society and work opportunities; family law orphans, disabled, may help protect widows, divorcees, and displaced, groups suffering orphans; social care services from discrimination Increase Chronic poor/ Foment investments in human Some level is automatic, Same as above; also health and education household human extreme poor capital and livelihoods possible threshold effects policies (such as community health project) capital and livelihoods Help households Those vulnerable to shocks Prevent losses to livelihoods or Timely entry required to Stable economies, well-functioning labor manage risks (often the poorer among human capital avoid losses after shock markets, and social insurance programs them) to mitigate risks of sickness, disability, unemployment, or retirement to reduce number and severity of episodes of transitory poverty, especially for workers in the formal sector Chronic poor or chronically Allow adoption of higher Guarantee required to For those engaged in agriculture, especially exposed to high risks risk­higher return livelihood promote ex ante changes smallholder or rain-fed agriculture, strategies irrigation, microfinance weather insurance, or well-developed markets and access to supplemental nonfarm income Help governments Lower quarter or half of Provide compensation for Targeting either to specific make sound income distribution reforms or provide alternative losers, or lower portion of choices vehicle for redistributional income distribution objectives SOURCE: Authors. 2. THE CASE FOR SAFETY NETS 29 China but also in Latin America and India--has meant that the basic distributional or protection role has been increasingly allocated to safety nets as prices are freed, employ- ment less protected, and services less guaranteed. Second, the understanding that safety nets assist in promotion as well as protection is increasing, if still not universal. Safety nets thus enter more in the discourse everywhere, perhaps most notably in lower-income countries and in Sub-Saharan Africa. Instead of grouping populations by degree of poverty, policy makers and the public may use other ways to classify those who may need support. A common alternative cat- egorization is to look at the population along the life cycle. Table 2.3 provides examples of programs serving segments of the population based on age groups from infancy to old age. Regardless of how the population is disaggregated, it remains evident that safety nets are only part of the policies needed to support each group. TABLE 2.3 Examples of Social Protection Programs by Life Cycle Social protection policy Complementary Group served policy or service Regulation Social insurance Safety net Nonworking · Health care · Child labor laws · Universal child · Means-tested child young allowances allowances · Education · Maternity · Transfers linked to · Family law benefits maternal and child health programs · School feeding · CCTs Working · Labor-intensive · Minimum wage · Unemployment · Transfers poor or growth laws insurance · Workfare unemployed · Economic · Job security · General subsidies stability regulations for food, utilities, or · Severance pay housing Nonworking · Financial · Retirement age · Contributory · Transfers elderly system to pensions · Social pensions facilitate savings Special · Health care and · Affirmative · Disability · Transfers groups traffic safety action or insurance for to prevent compensatory people with disability investments disabilities for minorities, · Education worker safety inclusive of laws to prevent minorities, the disability, family disabled, and law to protect so on assets of widows and orphans SOURCE: Authors. 30 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS 2.3 What Are the Challenges to Safety Nets? Despite the many arguments in favor of publicly provided safety nets, there are still some reservations in the development community about their feasibility and desirability. This section summarizes these concerns and briefly explains how various program design and implementation features can be used to help address these issues so that, on balance, good programs will be beneficial in most settings. The qualifiers in that sentence are important. Much safety net practice around the world over the last 20 years has not been particularly good, and what works in one setting may not succeed in another. Fortunately, a recent explosion of innovations in safety net programs makes for many highly promising options, and there are numerous successful programs from which future initiatives can learn. Managing the challenges is complex, often requiring actions on multiple dimensions of design and implementation. The following discussion offers only brief summaries of these issues, with more detailed information provided as indicated throughout the rest of the book. CAN DEVELOPING COUNTRIES AFFORD SAFETY NETS? The premier reason safety nets are not a headline social policy on the development agenda is concern over whether countries--especially poor countries--can afford to transfer mean- ingful resources to their poor. This is a complex issue, which involves many nuances and trade-offs; chapter 3 provides a comprehensive treatment of the financing of safety nets. In the poorest countries, the sheer magnitude of spending that would be required to provide an adequate safety net is quite daunting when viewed relative to the size of the economy as a whole. This discussion focuses on long-term safety net programs to aid the chronic poor and those households facing idiosyncratic shocks because these needs are per- manent, the expenditures ongoing, and the related budget constraints most apparent. But safety nets are also needed to handle large covariate shocks such as an economic recession or natural disaster. Because such requirements are temporary and consequently less onerous, national governments and international agencies readily agree that safety nets are needed in these cases; indeed, humanitarian assistance usually is offered to countries in times of need. For example, consider the extreme case of Ethiopia, where annual per capita income is about US$100 (World Bank 2004a). To provide adequate food for all the inhabitants whose consumption is below the food poverty line would require an annual expenditure of about US$810 million--12 percent of GDP, or about one-third of all public spending. This expenditure would obviously compete for resources against many other unmet needs, since only 52 percent of appropriately aged Ethiopian children are in primary school, infant mortality is 117 per 1,000 live births (one of the highest rates in the world), and water supply coverage is only 24 percent (the lowest in Sub-Saharan Africa). It would be difficult to say that safety nets should be funded, or fully funded, when the opportunity cost is primary education, primary health care, or water supply systems. High Cost of Inaction The flip side of the high cost of providing adequate safety nets in poor countries is the high cost of inaction. How much will an economy lose by not providing safety nets? Mal- 2. THE CASE FOR SAFETY NETS 31 nutrition, for example, can cost (based on lost productivity) at least 2 to 3 percent of GDP (Horton 1999 on low-income Asian countries), and lock affected children in a cycle of impaired cognitive development and physical growth, lower productivity, less education, lower earnings, and higher health care needs--in short, the intergenerational transmis- sion of poverty. In Ethiopia, stunting affects 46.5 percent of children under the age of five (World Bank 2007r). To not address the issue essentially condemns the nation as a whole to poverty for at least the next generation. Malnutrition per se may not be a concern in less poor countries, but there may be issues of low or late school enrollment, repetition, high dropout rates, and child labor or issues of social inclusion and cohesion in general. Each of these problems is more prevalent among the poor than the nonpoor, and to ignore them ensures impoverishment of the children over their lifetime and perpetuates the causes of social division. Trade-offs and Balances The stark apparent trade-off between, say, vaccinations or schools and safety nets may be something of an exaggeration, or at least a mislabeling of the choice. Safety nets are often (and can usually be) composed in ways that complement traditional development spend- ing for human capital or infrastructure. So the issue is not so much one of transfers versus human capital or productive investments, but of balancing the use of tools to achieve these. How much should be spent to get teachers and classrooms ready for students versus on getting students fit for and in school? Should some of infrastructure construction and maintenance be organized in ways that provide safety net services in addition to infra- structure? These issues of fitting safety nets into other antipoverty and social policy are discussed in chapter 9. It is also worth noting that the large welfare states in Europe do not fund social protection instead of education or infrastructure but in addition to them. They thus have larger government sectors and apparently with no big cost to growth, a theme taken up in chapter 3. In this regard, the stunning poverty of Ethiopia represents an extreme case. More typical is a country like Brazil. In Brazil, the income gap constitutes 1.6 percent of GDP, less than 5 percent of the income of the wealthiest 10 percent of Brazilians, and is small in comparison with total social spending in the country. Thus, theoretically and in the aggre- gate, Brazil has the resources necessary for solving its poverty problems through redistribu- tive transfers alone, without raising taxes (World Bank 2001b). Similar situations exist for other high-inequality countries--Argentina, the Philippines, the Russian Federation, and South Africa, among others--though of course such calculations abstract from the very real issues of targeting, administration, incentives, and political economy within safety nets, and from the issues of the opportunity costs of other uses of the funds. Recent years have seen a growing consensus among scholars regarding the produc- tive role that safety nets can play in low-income countries both on their own terms and in complementing other efforts to achieve growth and human capital formation (Devereux 2002a; Lipton 1997; Sinha and Lipton 1999; Smith and Subbarao 2003). For example: An astonishing feature of the developing consensus about poverty, given the strong tide of anti-State sentiment in the 1980s, has been the widespread agreement that even very low-income countries can and should "afford" some types of public provision for 32 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS poor people whose health or age prevents work, or who are made unemployed by the vagaries of climate or market demand (Lipton 1997, p. 1006). Safety Net Spending May Replace Other Less Effective Spending Safety nets may serve as more efficient ways of redistributing income than alternative policies. For example, when Sri Lanka began its Food Stamp Program in the 1980s, it was not additional to existing policy, but a more cost-efficient replacement of the general food price subsidies previously in place. Similarly, Indonesia's new cash transfer program is not an additional burden on the budget but a lower-cost substitute for energy subsidies. The discussion of whether countries can afford safety nets has been implicitly couched in terms of pitting one set of high priority, pro-poor or pro-growth expenditures against another. This orientation puts the matter in the harshest possible light, since few governments spend so efficiently. All governments make a certain number of idiosyncratic funding decisions, and all exhibit patterns of sometimes large and unproductive expendi- tures. Subsidies to manufacturing are not uncommon; these are of limited value in pro- ducing growth or jobs and can sometimes be costly. Brazil, for example, gave tax incentives to auto manufacturers that cost over US$200,000 per job created; India did the same with a cost of over US$400,000 per job created. In the Philippines, the effective corporate tax rate declines from the nominal 47 percent to 21 percent once fiscal incentives to firms are considered; in Thailand, the effect is even larger, a decline from 46 percent to 7 percent (World Bank 2004e). Military expenditures represent another large use of funds that is never linked to development. Vietnam spends 7.1 percent of its GDP on defense, Cape Verde 3.2 percent, and Mali 2.3 percent (Chamberlin 2004); yet each country is poor with underdeveloped safety nets. Redistribution to the Rich versus Redistribution to the Poor The idea that governments cannot afford to redistribute income to the poor must be con- trasted with the evidence that they regularly redistribute income to the nonpoor. Energy subsidies are highly regressive and often more costly than safety nets. The Arab Republic of Egypt spent 8 percent of its GDP on several energy subsidies in 2004 (World Bank 2005c), and Indonesia spent up to 4 percent of GDP between 2001 and 2005 on fuel sub- sidies (World Bank 2007r). Similarly, countries dedicate resources to bailouts of insolvent contributory pension funds by transferring general revenues to support them. The expan- sion of Brazil's well-targeted CCT program Bolsa Familia to cover the bottom quintile of the population is raising some questions as to whether the country can afford to redistrib- ute so much. The program cost 0.4 percent of GDP in 2006. In contrast, the deficit in the main federal pension program covered from general revenues is 3.7 percent of GDP and delivers over 50 percent of its benefits to the country's richest quintile (Lindert, Skoufias, and Shapiro 2006). This pattern is not unusual, at least in Latin America. Figure 2.3 shows the distribution of general revenue­financed transfers for several countries in ab- solute terms. Safety nets are progressive, but their cost is small compared to that of the general revenue used to finance the deficits in nominally contributory pension systems. Another example of where governments have found money to assist the rich but not the poor is the bailouts made to financial sectors. In the East Asian financial crisis, Indo- nesia's bank bailout cost 50 percent of GDP (Honohan and Klingebiel 2000); spending 2. THE CASE FOR SAFETY NETS 33 FIGURE 2.3 Distribution of General Revenue­Financed Transfers for Selected Countries by Population Quintile PPP (US$) ARGENTINA PPP (US$) BRAZIL 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 1 2 3 4 5 1 2 3 4 5 Quintile PPP (US$) CHILE PPP (US$) COLOMBIA 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 1 2 3 4 5 1 2 3 4 5 PPP (US$) GUATEMALA PPP (US$) MEXICO 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 1 2 3 4 5 1 2 3 4 5 PPP (US$) PERU 70 60 50 40 30 Social assistance 20 Social insurance net of contributions 10 (that is, amount corresponding to tax- 0 financed coverage of the deficit in the 1 2 3 4 5 social insurance systems) SOURCE: Lindert, Skoufias, and Shapiro 2006. NOTE: PPP = purchasing power parity. Quintile 1 is poorest; quintile 5 richest. 34 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS on the accompanying safety net for the poor was about 2.4 percent of GDP in 1998/99, including food subsidies, public works, targeted scholarships, and fee waivers for health care (APEC 2001). In Korea, the bank bailout cost 27 percent of GDP (Honohan and Klingebiel 2000), while the spending on the safety net tripled from about 0.6 percent in 1997 to nearly 2 percent in 1999 (APEC 2001). Bottom Line on Finding Budget for Safety Nets Even where safety nets have a place at the table when resources are allocated, they will face budget constraints so tight that policy makers will have to make difficult triage decisions about how to allocate money insufficient to meet reasonable needs. There are typically three approaches that may be taken in different combinations in response to the dilemma. · Keep the role of safety nets small relative to possible statements of need. Ben- efits may be limited to only a portion of the poor, either by defining specific subcategories of individuals (usually those in the traditional especially vulnerable groups); using an eligibility threshold well below the poverty line; or providing only seasonal benefits (during the hungry season in agricultural economies or the heating season in cold climates). · Insofar as possible, ensure complementarities with building physical and human capital. This approach will provide twice the "bang for the buck" by helping the poor survive today and by reducing causes of poverty in future years. Prime ex- amples of this type of approach are workfare and CCT programs. · In very low-income countries, international assistance may be used to finance social assistance. Indeed, there is increasing willingness on the part of donors and countries to use aid in such ways. CONCERN OVER REDUCING WORK EFFORT One of the most common stumbling blocks for political support of safety nets is concern over the labor disincentives of welfare dependency (box 2.6). The fear is that potential beneficiaries will either work less after receiving the benefits or, if eligibility is tied to earned incomes or unemployment, will reduce their work efforts in order to qualify for the transfer. Both arguments and the evidence differ substantially across groups of coun- tries, classes of programs, and types of beneficiaries. The theoretical arguments are intuitive.7 Transfers provide unearned income and thus inherently will lower the incentives for recipients to work, as beneficiaries may trade some of the extra income for more leisure. This outcome is sometimes referred to as the "income effect." Transfers may also change a recipient's effective wage rate if their size is based on the recipient's income. This situation arises for verified means-tested programs, where the benefit level is reduced by a fraction of a currency unit for each additional cur- rency unit in earnings; the implicit tax on earnings is called the marginal tax rate of the program. This outcome is sometimes referred to as the "price effect." In the hypothetical perfectly means-tested guaranteed minimum income program where the size of the benefit is adequate to a decent minimum living standard and is reduced as income rises, the recipi- ent whose initial income is below the guaranteed income has no incentive to work. 2. THE CASE FOR SAFETY NETS 35 BOX 2.6 Is Dependency Always Bad? Concerns over "dependency" have long been cited by skeptics of safety nets. But dependency is an emotionally charged term, so it is worth pausing to consider what lies behind it and how robust the empirical evidence about it is. Lentz, Barrett, and Hoddinott (2005) provide a model of how to do this, writing about food aid in response to emergencies and reviewing a largely African literature. Lentz, Barrett, and Hoddinott (p. 10) consider that "a household or community exhibits depen- dency when it cannot meet its immediate basic needs without external assistance." They point out that dependence is not always bad, since the alternative to external assistance may be des- titution. They use the term "positive dependency" to refer to such welfare-enhancing assistance and "negative dependency" to refer to situations in which external assistance helps meet current needs but is achieved at the cost of reducing recipients' capacity to meet their own basic needs in the future without external assistance. In reviewing the many avenues through which such negative dependency may occur at the household level, the authors find little evidence that food aid discourages the labor supply of households that receive aid, crowds out remittances, or encourages moral hazard (the assump- tion of unwarranted risk). They point to an alternative definition of dependency, when rather than households it is govern- ments that come to rely on relief resources. But they show evidence that because aid is quite irregular--multilateral aid flows via the World Food Programme respond only weakly but at least predictably to shocks, bilateral aid flows from the United States do not respond to shocks at all--governments cannot become dependent on it. The authors also find that food aid has no persistent negative effects on national food production. Concerns over dependency in other settings are usually less precisely defined, often concerning issues about the opportunity costs of funds devoted to transfers and one or more of the negative disincentives possible at the household level. In this book, we have therefore found it appropri- ate to discuss each issue separately rather than taking on "dependency" in the generic. Developed Countries Concerns about labor disincentives have traditionally been strongest in wealthy countries with generous safety nets and high unemployment rates. However, the evidence shows that participation in safety net programs has only small or moderate effects on employ- ment or hours worked. In Ontario, Canada, in the 1990s, beneficiaries of a rather gener- ous safety net program reduced their work effort by 3 to 5 percent when benefits tripled from Can$185 to Can$507 (Lemieux and Milligan 2008). The bulk of the evidence comes from only one country, the United States. Here, for the numerous initiatives pro- viding small benefits, such as food stamp, nutrition, and child care subsidy programs, most studies have found no evidence of reduced work effort. Similar results were obtained for in-kind programs, such as housing programs or Medicaid (health insurance coverage for the poor) (Moffitt 2003). More generous pilot programs, such as the negative income tax experiments, did moderately reduce the work effort of participants. Male earners ben- 36 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS efiting from the negative income tax experiments reduced employment and earnings by 7 percent on average; for wives, only 17 percent of whom were employed, employment and earnings dropped by 17 percent (Burtless 1986). Among nonexperimental U.S. programs, only one--Aid for Families with Dependent Children--was shown to be associated with large reductions in work effort. Moffitt (2002b) found evidence that single mothers ben- efiting from the program had reduced their work effort by 10 to 50 percent; however, he also found that the program contained specific features that would make it especially susceptible to labor disincentives. (See box 5.3.) Consequently, extrapolating results from the Aid for Families with Dependent Children program to other safety net programs in developed countries would not be advisable. Developing Countries The theoretical model predicts that reduction in work efforts will be proportional to the size of the benefit (income effect) and the implicit marginal tax rate on earnings of the program (price effect). The theory thus supports the view that the impact of safety net programs on work disincentives should be smaller in developing countries, for four reasons: · Programs are less generous in developing countries; most safety net programs com- plement, rather than substitute for, the earnings of able-bodied beneficiaries. · Very few programs in developing countries use and are able to enforce effectively benefit formulas with marginal tax rates and frequent recertification of household income. · Many developing countries target their programs only to households without able-bodied adults (for example, Zambia's Kalomo scheme and Ethiopia's Direct Support program) or require able-bodied beneficiaries to work in return for ben- efits (all workfare programs, but also some cash transfer programs). · The static model does not take into account the fact that transfers help house- holds make productive investments in their future. The evidence supports the view that, in developing countries, safety net programs do not often reduce labor effort substantially. In Mexico, adult earners benefiting from the CCT initiative PROGRESA (now known as Oportunidades) worked as much as those in a randomized control group; at the same time, the program achieved its objective of increasing schooling and reduced child labor by 15 percent (Parker and Skoufias 2000; Skoufias and di Maro 2006). In Brazil, Leite (2006a) simulated the potential impact of Bolsa Familia on adult work effort and found that the transfer amounts would have very little impact. In Armenia, the employment rate and hours worked by adults in the Family Poverty Benefits Program were similar to those for a matched sample of nonparticipants (Posarac, Tesliuc, and Urdinola forthcoming). In Romania, a qualitative review of the Guaranteed Minimum Income Program found little evidence of an adverse impact on labor force participation; conversely, "because there is a small bonus for employment, there may be a small positive impact on participation compared with more traditional systems of aid" (Birks Sinclair & Associates Ltd. 2004). On the other hand, in Sri Lanka, Sahn and Alderman (1995) studied a rice subsidy program that induces labor disincentives through income effects; they found labor reductions on the order of 10 percent. 2. THE CASE FOR SAFETY NETS 37 Measures to Foment Work Effort There is increasing evidence, especially from developed countries, that measures to coun- teract welfare dependency exist and can be effective. In the United States, a number of welfare-to-work experiments reviewed by Hamilton (2002) and Greenberg and others (2002) found that a combination of work requirements, financial incentives for work, and/or services supporting welfare recipients increased earnings and employment on av- erage from 6 to 10 percent. Grogger and Karoly (2006) showed that welfare recipients (single mothers) respond to financial incentives and workfare tests in the way predicted by the static labor supply decision model. In Canada, beneficiaries of the Self-Sufficiency Project increased their full-time employment and earnings by 15 percentage points. In the United Kingdom, the welfare-to-work measures introduced under the New Deal for adults reduced the unemployment rate by 6 to 10 percent. Policy makers and administrators have a variety of tools at their disposal to manage labor disincentives, including the following: · Limit programs to those who traditionally are not expected to work anyway--the very young, the very old, those with disabilities, and so on, often referred to as the "deserving" poor. Although this limitation is a fairly common one, it results in only a partial safety net. (See chapters 8 and 9.) · Choose a targeting mechanism not directly tied to earnings--this leaves the rewards to working intact. Few developing countries use a means test or mini- mum income guarantee, though many transition countries do. Infrequent re- certification will also mute the labor disincentives. Most programs outside of Europe recertify only once every two or three years; some even less frequently. (See chapter 4.) · Set benefit levels to maintain work incentives. Most programs in developing coun- tries have very low benefits, often equivalent in amount to only a few percentage points of the poverty line. Thus they inherently leave plenty of incentive to work. The low benefits proffered are usually due more to fiscal constraints than concern over work disincentives, but the result is the same. In countries with a full suite of social protection programs, social assistance payments should be less than un- employment insurance and the minimum pension provided by the contributory pension system. (See chapter 5, section 1.) · When benefit levels are customized to earnings, ensure that there is still incentive to work--set exit thresholds higher than entrance thresholds, use sliding with- drawals of benefits as incomes rise or earned income tax credits to help make work pay. Alternatively or additionally, provide lump sum graduation benefits, or pay for allied benefits such as child care or transportation allowances for a period after work starts. Admittedly, these options are administratively demanding and will result in those above the poverty line receiving some program benefits. (See chapter 5, section 1.) · Link transfers to program elements, such as job training or placement, educa- tion, microcredit, social support services, meant to help households move out of assistance and toward independence. Such links also may be administratively 38 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS demanding, but they are fully consistent with broad social policy objectives. (See chapter 5, section 2.) Concerns over reduction in work effort will be strongest when programs are most generous and eligibility or benefit levels depend more on recent or current earnings. Since generous programs with customized benefit levels are becoming more common, features to manage work effort may be needed more often as well. But it is usually the middle- and upper-middle-income countries where program generosity is sufficiently great to make work effort a concern, and these are the countries most likely to manage the sophisticated program elements to mitigate the problem. Concern over work effort should thus lead the policy maker to consider various features of program design and of the balance across so- cial assistance of last resort, unemployment insurance, and contributory pension programs carefully, but should rarely imply abandoning social assistance as a policy tool. CROWDING OUT PRIVATE TRANSFERS Private transfers are important to the informal safety nets that arise when official public action is limited or nonexistent. If public safety nets are put in place, these private trans- fers might be diminished--a consideration that must be weighed in determining how and whether to implement the public program. To make this determination, policy makers must understand the adequacy of private social protection systems in a particular setting. In some countries or among some groups, they can be quite substantial and appear to exhibit the same features as a good public social protection system--they go from richer to poorer households and to those facing shocks such as illness or unemployment. For example, Cox and Jimenez (1997) found that 40 percent of black South Africans reported either receiving or giving cash transfers. Though undoubtedly helpful, informal insurance and interhousehold transfers are not sufficient safety nets. Many people are left out of such networks, and even for those who receive some assistance, it may not be enough to avoid poverty traps and the intergenera- tional transmission of poverty. Moreover, the entire support network may be affected by widespread shocks and thus unable to provide support to all members when it is most needed. Finally, private support can sometimes be part of a larger set of patron-client re- lationships that are not conducive to the client's long-term income growth (Glewwe and Hall 1998; Morduch 1999; Skoufias 2003). Next, policy makers must understand the extent to which the introduction of public transfers might affect private ones. One of the most credible investigations of this ques- tion is Jensen's 1998 examination of the effects of the expansion of the South African old-age pension program to Africans. Though means tested, the eligibility threshold and benefit levels in this program are quite generous. Jensen estimates that for those house- holds receiving private transfers, every publicly provided rand led to a reduction of 0.2 to 0.4 rands in private transfers to the elderly. This reduction in the burden of private support indirectly raises the income of poor donors. The South African social pension program is quite unusual in the generosity of its benefit--more than twice the median per capita monthly household income of Africans. The Nicaraguan and Mexican CCT pro- grams are also relatively generous, though their benefits are only about 15 to 25 percent of household income. In the Nicaragua program, the probability that program households 2. THE CASE FOR SAFETY NETS 39 will receive interhousehold food transfers is lowered by about 10 percent. In Honduras, where the CCT program benefit is only about 4 percent of annual household expendi- tures, no such crowding out of private transfers is found. In none of the three countries did remittance to the households fall (Nielsen and Olinto 2006; Teruel and Davis 2000). Program generosity would appear to affect the degree of crowding out, but the results may be rather context specific. Lentz and Barrett (2005) find no evidence that food aid receipt crowds out private transfers in pastoralist households in southern Ethiopia and northern Kenya. Gibson, Olivia, and Rozzelle (2006) found little evidence of crowding-out effects in China, Indonesia, Papua New Guinea, and Vietnam, concluding that crowding-out problems are, in fact, not a significant policy concern. To address issues associated with the crowding out of private transfers, public trans- fer programs should be designed as follows. · Programs should be of sufficient scope to cover people missed in the private sys- tem. They should be permanent and reliable programs so as to not undermine private systems without providing a better alternative. · If eligibility is determined through a means test, the income from private transfers may be excluded in whole or part from the calculation of income, or the income of the full household or even of nonresident parents or children may be consid- ered in the calculation. Such adaptations help preserve incentives for the contin- ued transfer of income within families, but they can be administratively complex and lead to errors of inclusion. · If possible, consider how to use public systems to reinforce private systems. In Zimbabwe, the traditional chief sets aside community land to be farmed by com- munity/volunteer labor so the resultant crops can be distributed to the needy in the village. Public subsidies to nonlabor inputs for the scheme should improve the yield and assist the private safety net system. POSSIBLE EFFECTS ON FERTILITY Economic theory holds that the demand for children is a function of individual preference and the cost of children, under an income constraint. Social assistance alters the income constraint and, depending on how benefits are set, may lower the direct costs of raising children. This theory raises the possibility that social assistance programs might result in higher fertility. Empirical evidence that social assistance increases fertility is scant. Gauthier and Hatzius (1997) look at family benefits and fertility among Organisation for Economic Co-operation and Development (OECD) countries and conclude that, while there is an effect on fertility, it is of a small magnitude: a 25 percent increase in family benefits would increase fertility on the order of 0.07 children per woman. Stecklov and others (2006) provide estimates of impacts on fertility of the Honduran and Nicaraguan CCT programs. They find significant though small positive effects in the former, but not in the latter. In Turkey, Ahmed and others (2007) show that the CCT program reduces fertility by 2 to 3 percent. Box 2.7 reports the reactions of this program's beneficiaries when evaluators held discussion groups aimed at understanding the impact of the program's pregnancy benefit. 40 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 2.7 Women's Reactions to Questions about Transfers and Fertility The Turkish CCT program includes a small cash benefit for pregnant women. Both quantita- tive and qualitative studies were done to understand how the program might affect fertility. The quantitative study showed the small decrease already cited. A complementary survey found that 97 percent of respondents said that women would not get pregnant because of the benefit. In fact, many thought such a question to be strange, humorous, absurd, or offensive. One woman (IIlyaskoy, Sengul G.) said, "Allah, were there women that got pregnant just for this money, re- ally? Ha ha ha!...of course you have to ask these questions, this is your duty! But mothers have to think about their children's futures as well" (Ahmed and others 2007, p. 61). Adato and others (2007, p. 135) report that women understand that bearing children has many costs and that these go far beyond what a small cash transfer can alleviate. They also docu- ment the reasons that many children are desired, including powerful cultural factors in favor of large family size. One woman (Nafia S. Beyüzümü, Van) summed it up this way: "I don't think a woman can give birth to get money...If a woman gives birth it is because first God, second her husband, and third her husband's mother want her to." There are several actions programs can take to contain any possible side effects re- lated to fertility; some may have unintended side effects of their own: · Keep benefits reasonably low or temporary. Given the total psychic, time, and monetary investment implicit in child bearing and rearing, low-level benefits from safety nets are unlikely to have much impact on fertility decisions. Public opinion surveys in OECD countries, which feature relatively more stable and generous assistance programs than others worldwide, revealed that this support might help families achieve desired family size, but would not increase it (Gau- thier and Hatzius 1997). · Keep benefits flat per household, or, if given on a per capita or per child basis, cap the total benefit or number of children. While this approach will minimize the incentives for fertility, it will also reduce the poverty targeting of the benefit as larger families are often poorer (even with appropriate treatment of economies of scale). Also, where there are marked differences in family size across ethnic groups, such caps may carry a political dimension far more important than the possible effect on fertility. · Introduce elements in the program that would tend to reduce fertility.The welfare- to-work reforms in OECD countries increase women's labor force participation, which usually discourages fertility. CCT programs require women to get minimal preventive health care and health education, which usually includes opportunities to deliver messages about the health benefits of breastfeeding and birth spacing, as well as family planning services. The availability of this information may re- duce unwanted fertility among adults. Also, the increased educational level of the female children in these programs will likely serve to lower their family size when they become adults. 2. THE CASE FOR SAFETY NETS 41 DOUBTS ABOUT ADMINISTRATIVE FEASIBILITY AND PROGRAM MANAGEMENT While concerns over administrative capacity are not trivial, there are often ways to deliver some sort of safety net program if due creativity is brought to bear on the issue. Tak- ing advantage of existing public systems outside the welfare agency, contracting out to private firms, simplifying design, and prudently guarding against the perfect becoming the enemy of the adequate will usually result in a feasible program. Chapter 7 contains many examples of acceptably administered programs in a wide range of countries and cir- cumstances. Chapters 4 to 6 go into much more detail about different facets of program implementation and administration, outlining requirements and different ways of fulfill- ing them, again with examples from a broad range of countries. Two politically charged concerns may hide under the more neutral term "administra- tive feasibility." Both have to do with who really benefits from programs carried out in the name of the poor or vulnerable. The first concern has to do with targeting and doubts about whether it can really be accomplished well, especially in low-income or low-capacity settings. (This theme is taken up in detail in chapter 4.) On average, targeting results are better for middle-income than low-income developing countries, but there have been successful cases of the use of all sorts of targeting instruments in low-income settings (Coady, Grosh, and Hoddinott 2004). The overall poor results are at least partially due to the nature of the programs cho- sen (general food subsidies) and to the stop-and-go nature of many other interventions, which makes it difficult to develop good systems. A second concern has to do with the potential for misuse of funds. Some believe that safety net programs should not be funded on the grounds that resources will leak away from intended beneficiaries. A number of safety net programs have famously suffered from graft and corruption, and, in some countries, the track record is rather poor. But a blanket abandonment of safety net policy would be akin to forsaking infrastructure projects because of reports of occasional contractor kickbacks and rakeoffs. The solu- tion instead lies in determining how to minimize such problems, about which much is already known. In addition to the broad government-wide governance agenda of due process, trans- parency, and accountability which is being increasingly explored and applied, there are a number of design and administrative features specific to social assistance programs that can help prevent fraud, error, and corruption. Key measures are summarized below. · Use program design to minimize incentives and opportunities for misuse of funds. ­ Ensure that program budgets are consistent with eligibility criteria. If funding is adequate to serve all those eligible, applicants will have little reason to offer bribes to get into the program. If program slots are rationed so that only a small portion of eligible applicants can be admitted, there is ample motivation for bribery and kickbacks. ­ Consider carefully the eligibility criteria. The simpler they are, and the less discretion they offer to eligibility intake officers, the less opportunity there will be for corruption. Where complex criteria are needed, reinforce mechanisms of control. (See chapter 3, section 6, and chapter 4.) 42 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS ­ Consider the benefit level for a participant with respect to the salary of intake officers. Ineligible applicants may offer a cut of their benefits in return for en- try into the program. If such an offer is low relative to officers' base earnings, it will be less attractive. ­ Conditionalities such as requirements for recipients to obtain health care, at- tend school, or work may help guard against "ghost" beneficiaries. In programs with no conditionality, an intake worker can easily register ghost beneficiaries. With conditionality, another official in the health clinic, school, or worksite would have to collude with the officer by providing certification of attendance. Public posting of the list of beneficiaries is another means of exposing invented beneficiaries. ­ Use payment mechanisms that move benefits from the treasury to the indi- vidual recipient with as few intermediaries as possible, as each additional link in the chain increases the potential for diversion of funds. (See chapter 5, sec- tion 4.) · Set up adequate administrative procedures. ­ Ensure that administrative processes are clearly defined, and that staff and oth- er resources are adequate to carry them out. A culture of compliance can only be created where rules are clear and reasonable. (See chapters 4 and 5.) ­ Institute a range of quality control procedures to ensure that eligibility crite- ria are respected, payments are audited, information systems have appropriate safeguards, and so on. (See chapter 6.) ­ Establish sensible tolerances in quality control procedures. For example, given the difficulty in measuring income, an initial eligibility evaluation and subse- quent recheck might arrive at slightly varying estimates of income. Only if the difference is substantial and larger than the expected measurement error should this variation be considered fraud. Such an approach also facilitates respect for rules and makes efforts to enforce them more cost-effective. (See chapter 6.) ­ Set up adequate grievance, appeal, and "whistle-blowing" procedures for appli- cants who believe they are eligible but were denied entry, for beneficiaries who are receiving incorrect payments or are requested to pay kickbacks, for program workers who suspect fraud by their coworkers, and for the general public that suspects irregularities of any sort. (See chapter 4, section 4.) ­ Take action against miscreants with meaningful penalties. · Use transparency and communications well. ­ Ensure that the eligibility criteria, benefit levels, and rules are clear to both the public and beneficiaries. People can only seek redress when they understand what is due. Conversely, clarity can help eliminate unwarranted appeals or claims of malfeasance. ­ Publicize cases of detected fraud and the penalties imposed. Some of these techniques involve trade-offs with other desirable features of safety net policy. For example, although the stratagem of fully funding safety nets should eliminate a source of corruption, it is not always feasible to do so given budget constraints. Keeping eligibility criteria and payment structures simple will reduce the probabilities of fraud, er- ror, and corruption, but it will make programs less precise in their targeting and lower the 2. THE CASE FOR SAFETY NETS 43 impact on poverty per currency unit spent on legitimate beneficiaries. Policy makers may need to forego some of the design options for minimizing fraud, error, and corruption, but will need to develop correspondingly more sophisticated administrative procedures as an alternative means of keeping problems in check. Because such systems take time to develop, they are more feasible or effective in permanent programs. SUMMARY Table 2.4 summarizes how the various challenges posed by safety nets discussed in this section may be handled. TABLE 2.4 Summary of How to Handle Challenges to Safety Nets Challenge to safety net Management strategy Affordability · Consider the costs of inaction · Keep safety nets lean · Leverage improvements in physical or human capital if possible Reduction in work effort · Craft eligibility criteria, benefit levels and structures, countervailing conditionalities appropriately Crowding out of private · Some is inevitable and not necessarily bad transfers · Some mitigating measures may be feasible Incentives for fertility · Craft benefit structure to minimize · Build in elements to shift preference for family size Administrative feasibility and · Employ design elements that minimize opportunities for corruption accountability · Develop administrative systems · Use communications and transparency SOURCE: Authors. Notes 1. Social risk management refers to how society manages risks (not to how to manage social risks), a conceptual framework introduced by Holzmann and Jorgensen (2000). See appendix A for a brief exposition. 2. Low birthweight and malnutrition among young children are well established as being linked with higher child mortality and higher risk of illness (with attendant costs for medical care and time requirements for caregiving) and, later in the life cycle, with lower cognitive development, lower schooling, and lower physical productivity. Iodine deficiency hinders cognitive develop- ment and increases child mortality. Vitamin A deficiency can increase morbidity and mortality and, in severe cases, cause blindness. In children, iron deficiency can reduce cognitive capacity and affect schooling and future productivity; in adults, it can impede hard work. Zinc has an appreciable impact on growth in children (Behrman, Alderman, and Hoddinott 2004; Webb and Rogers 2003; World Bank 2005n). 3. Adequate access to credit can help families avoid negative coping strategies, but the poor of- ten lack access to credit or have access only on particularly onerous terms (for example, from 44 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS moneylenders). Where credit is available, it can lead to large indebtedness, which can have repercussions on family welfare for years to come. 4. The underlying problem is uninsured risk. One means of addressing it is to improve insurance against the risk being directly faced. For example, weather insurance is being considered as a new option for reducing the income risks to small farmers of too much or too little rainfall, thus freeing them to adopt income-maximizing rather than risk-minimizing choices of crops and inputs. Though theoretically attractive, weather insurance has not yet been implemented anywhere at a large scale or long enough to see how much of the problem it solves. 5. Other sector-specific policies have been tried in various places, beginning with lax collections or "no disconnection" policies; these provided little predictability to the consumer, greater benefits to the nonpoor than the poor, and erratic revenue flows to the utility. Across-the-board subsidies, life-line or block pricing, and burden limit programs were also introduced; these too had efficiency drawbacks and, except for the across-the-board subsidies, not inconsiderable administrative requirements. 6. The US$1/day rate was found by Chen and Ravallion to be representative of the poverty lines found among low-income countries in the first years such calculations of global poverty were done. These calculations attempt to express in a common currency the purchasing power of varied domestic currencies; such purchasing power parity comparisons, while useful, are inex- act. US$2/day is more representative of poverty in middle-income countries. 7. The predictions that increased transfers will reduce the labor supply of beneficiary households are based on the static labor supply model (Moffitt 2002b). CHAPTER 3 Financing of and Spending on Safety Nets KEY MESSAGES The economic theory underlying the question of how much to spend on safety nets is the same as for other forms of government expenditure--that is, the marginal benefits of different types of expenditure should be equal to each other and to the marginal costs of raising public funds. Increases in spending are more likely to be justified the more the following conditions hold: the proposed safety net programs are "good," the programs are small, alternative uses for the funds are of low priority, taxes can be raised efficiently, and the combined package of expenditure and financing is redistributive. If countries wish to increase their spending on safety nets, they can reallocate expendi- tures, raise taxes, obtain aid grants, or borrow. Reallocation of funds from less important items is preferable when possible. If taxes are to be raised, the government must pay attention to the economic and political costs. If international grant (donor) finance is to be used, the government and donors should try to ensure that funding flows are stable and that procedures are conducive to building long-term implementation capacity. Debt finance is appropriate for safety nets when they benefit future generations in ways that will raise their productivity, and consequently future tax revenues, or when temporarily increased expenditures are needed as during a recession. Safety nets should be financed in a countercyclical manner, yet few governments man- age this. Developing countries' prospects for solving this problem seem slim until safety nets are fully financed in stable times and volatility is lower than has recently been the case. Expenditure reallocation in favor of safety nets during economic downturns along with generally prudent fiscal policy will help and have been put into effect, but to a degree insufficient to yield countercyclical funding for safety nets. The literature on the costs of the welfare state from countries of the Organisation for Eco- nomic Co-operation and Development (OECD) suggests that they have spent substantial sums on their social protection systems, but have financed them prudently and reaped benefits from them such that they have not suffered the reduced growth that economists often predict will accompany such high redistributive expenditures. Most developing countries spend in the range of 1 to 2 percent of their gross domestic product (GDP) on safety nets. Analysis of new data shows that spending on safety nets as a percentage of GDP is weakly but positively correlated with income and democracy. The analysis does not find any relationship between spending on safety nets and several other plausible variables, including governance, ethnic fragmentation, and public attitudes about inequality. When spending is broadened to include all social spending (safety nets, social protection, and the social sectors), we find more of the expected relationships. 45 46 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Conventional wisdom in the public finance literature suggests that redistribution is a role most appropriate for higher levels of government because of interregional equality con- siderations, but that subnational jurisdictions may be well placed to administer safety nets, because they may have greater knowledge of or contact with programs' client base. Thus many programs will involve multiple levels of government and shared responsi- bilities. Consequently, designing appropriate systems of intergovernmental transfers, managing the incentives contained in them, and developing and implementing shared administrative responsibilities can present challenges. F inancing safety net programs is not theoretically different from financing any other government program. The safety net literature thus seldom provides information about this topic,1 yet policy makers concerned with implementing or reforming safety nets face a constant stream of demands competing for limited funds. This chapter is targeted to such policy makers. It provides a synopsis of some of the pertinent public finance literature with illustrations from safety net programs and new data on spending on safety nets in developing countries. 3.1 The Theory on Expenditure Allocation The economic theory to help address how much to spend on safety nets is the same as for other forms of government expenditure: the marginal benefits of different types of expen- diture should be equal to each other and to the marginal costs of raising public funds. While the specifics depend on the type of program, the benefits tend to include improved equity, increased household welfare via investments or improved risk manage- ment, and such economywide effects as higher growth associated with successful reforms facilitated by safety net programs (see chapter 2). Typically, improvements in equity are measured by changes in indexes of inequality such as the Gini coefficient or by the reduc- tion in the extent of poverty among beneficiaries. Welfare improvements are gauged by the increase in recipient earnings; the improvement in their children's welfare in terms of nutrition, schooling, or child labor; or by the level of recipient savings and investments. The marginal costs associated with safety net programs include the cost of the transfers, administrative costs, and efficiency costs. The latter are of two sorts: they are either due to behavioral change by beneficiaries, such as reduced work effort, or to the economic costs of collecting taxes to finance the program. Box 3.1 illustrates these costs. Compared with other public interventions, such as building roads and dams and providing education and health, quantifying both the benefits and costs of safety net pro- grams in monetary terms is more difficult for two reasons. First, other types of programs are primarily or exclusively judged using efficiency criteria--that is, whether they gener- ate a high economic rate of return--whereas safety net programs are primarily judged by their contribution to improved equity. However, translating improvements in equity into monetary terms is an academic exercise dependent on subjective assumptions that cannot be tested.2 Second, safety net programs tend to have a diverse set of impacts, and quantify- ing their benefits and costs is only partially possible, a complexity less apparent for some other public actions. For example, in the case of a direct public road between two cities, an analyst can estimate the project's economic rate of return as the funds private operators 3. FINANCING OF AND SPENDING ON SAFETY NETS 47 BOX 3.1 Okun's Leaky Bucket In his classic treatise, Okun (1975) provides an intuitive explanation of what he refers to as the leaky bucket used to transfer money from better-off taxpayers to poorer ones. He enumer- ates the leaks as administrative costs, reduced or misplaced work effort, distorted saving and investment behavior, and possible changes in socioeconomic attitudes. Okun's idea of a leaky bucket is often cited by those who characterize transfers as costly and appropriate for only a small policy role, but Okun's comments on the size of the leaks in the bucket suggest that their magnitude is fairly modest, which recent research on safety nets in developing countries largely confirms. Administrative costs. These are the costs to the government of tax administration and to taxpayers of such items as recordkeeping. Okun deems that these are easily measured, are subject to policy control, and amount to only a few percentage points of overall costs at most. Experience in developing countries confirms that safety net programs can be run well for mod- est administrative costs: a useful rule of thumb is roughly 10 percent of overall program costs (see chapter 9 for more on the topic). Work effort. Okun (1975, p. 99) notes that the literature shows "virtually no effects on the amount of work effort of the affluent," a limited effect of transfers on the work effort of secondary earners in low-income households, and virtually no effect on low-income households' primary earners. A much greater effect can be found in tax avoidance behavior by corporations shifting remuneration and benefit packages in ways that reduce their tax liabilities. As the synopsis of labor disincentive effects in chapter 5, section 2, indicates, recent experience with safety nets shows that well-designed programs have modest and manageable labor disincentives. Savings and investment. These are important for programs that tax one generation to support another, as in pay-as-you-go pension systems, and presumably less so for safety net spending. Okun notes that in the United States, savings rates were 16 percent of GDP in both 1929, when taxes were low and flat and the social protection system was small, and in 1973, when taxes were higher and more progressive and the social protection system was much larger. He infers that a massive increase in the tax and transfer system had not lowered savings. We note that some safety nets actually help recipient households invest in their livelihoods, if not in financial markets. Socioeconomic attitudes. These are a less tangible concern and relate to the effort to balance the benefits of social inclusion against possible harm to the work ethic. Socioeconomic attitudes are a recurring theme in social assistance policy for all countries. Modern aggregate estimates of the cost of funds. The economic literature on the cost of funds uses general equilibrium models that essentially try to measure the leaks in Okun's buck- et. Estimates of the cost of raising US$1.00 for developed countries mostly range from US$1.00 to about US$1.50; about US$1.25 is common, although there are a few much higher estimates. Fewer estimates are available for developing countries, but they fall in the same range, perhaps tending to be a bit lower (Devarajan, Theirfelder, and Suthiwart-Narueput 2001; Martin and Anderson 2005). Warlters and Auriol (2005) estimate the average in 38 African countries as US$1.17. 48 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS save when transporting goods and people compared with alternative but longer routes. By contrast, a child welfare safety net program will generate a host of impacts, some of which are hard to quantify: the children may benefit from a more nurturing climate at home, better education, and improved health and subsequently stay in school longer and be more likely to avoid risky or illegal behaviors in their youth; as adults, they may be more likely to be employed and earn good wages. Nonetheless, a partial estimation of the ratio of marginal benefits and cost can be un- dertaken. So far, such evaluations have been done for only a handful of programs, mostly in the United States, but also in Mexico for PROGRESA (now known as Oportunidades). The evaluation of the opportunity cost of a given safety net program is best determined through a cost-benefit analysis (see chapter 6, section 5), which compares the program's net impact on its ultimate outcomes (as determined, ideally, from an impact evaluation) with the extra costs associated with implementing the program. Consider the following three examples. Beecroft, Lee, and Long (2003) examine the effects of the 1996 U.S. welfare re- form on the first cohort of beneficiaries affected in Indiana. The welfare reform focused on a "welfare to work" transition. Support became conditional on recipients pursuing a job search or education activities, and support for these activities was provided along with child care and transportation subsidies. Time limits on the benefits were set, with each family allowed only five years of benefits over a lifetime. The analysts consider two perspectives: that of the families receiving the support and that of taxpayers as reflected in the costs to the U.S. and Indiana budgets not only of the Temporary Assistance for Needy Families (TANF) welfare program, but also for food stamps and medical assis- tance, which have linked application procedures. According to Beecroft, Lee, and Long (2003, p. 104): The economic benefits of welfare reform to families--resulting mainly from increased employment--slightly outweighed the losses in welfare payments and other income. While changes in income varied across families, the typical family's economic position was very modestly improved. Welfare reform benefited taxpayers because savings more than offset welfare reform expenditures. Savings occurred primarily because clients spent less time on cash assistance, reducing benefit payments for the TANF, Food Stamp, and Medicaid programs. These reductions more than offset increased spending on employment and training services and child care subsidies. The budget savings were shared by Indiana and the federal government. The impact evaluation of Ecuador's Bono de Desarrollo Humano (Human Develop- ment Bond) program shows that it increased school enrollment by 10 percent and reduced child labor by 17 percent, with the effects concentrated among older children (Schady and Araujo 2006). Nominally a conditional cash transfer (CCT) initiative, the program aims to ensure that beneficiary households have their children attend school and obtain certain preventive health care services. The cost-benefit analysis focused on the program's educa- tion benefits and effects on increased consumption, because evaluation data are available for these impacts; possible impacts on nutrition and empowerment had not yet been stud- ied.The cost-benefit analysis used a national household survey to compare earnings among adults with different levels of schooling, estimate the increased years of schooling each child would receive as a result of the program, predict future earnings for beneficiary and 3. FINANCING OF AND SPENDING ON SAFETY NETS 49 nonbeneficiary children, discount these earnings over their lifetimes, and compare them with estimates of program costs, employing sensitivity analysis for key parameters (World Bank 2006l). The result varies depending on the economy's growth rate and whether the program becomes more effective at implementing the conditions that children enroll in and attend school (its "conditionality"). As shown in table 3.1, the cost-benefit analysis suggests that, under most scenarios, the program is TABLE 3.1 Cost-Benefit Estimates of the Education Effects of justified on the grounds the Bono de Desarrollo Humano Program, Ecuador of its education benefits alone; it is expected to Item 1% growth 3% growth 5% growth produce parallel improve- Without conditionality 0.62 0.82 1.13 ments in health and nutri- tion as well. For example, With conditionality 1.15 1.50 2.05 with a growth rate of SOURCE: World Bank 2006l. 5 percent and effective conditionality, the program's education benefits would be more than twice its total costs. Education benefits would exceed total costs even with only a 1 percent growth rate and conditionality, or without effective conditions and a 5 percent growth rate. The economic analysis also simulates the impact of the program's transfer element on consumption poverty and predicts that poverty falls by about 2.5 percent using a pov- erty line of either US$1 or US$2 per day. This is an upper bound of the short-run effect. If households save and invest any of the transfer, or if they reduce the number of hours they work, the immediate poverty impact will be lower. If the investments yield a return, the long-run effects on poverty reduction may be higher (World Bank 2006a). Note that neither of the foregoing examples explicitly values the welfare of the poor more than that of the nonpoor, although economists often think this should be done and that redistribution is one of the fundamental motivations of transfer programs. Alderman and del Ninno (1999) provide a relatively rare example of such an analysis. They introduce explicit distributional weights into the analysis of potential reforms to the value added tax (VAT) in South Africa. They construct a cost-benefit measure of a tax change with losses in personal consumption in the numerator, with distributional weights in the aggregation, and with revenue gains in the denominator. The higher the ratio, the greater the social cost compared with the revenue gained. Alderman and del Ninno find that when the welfare of all households is weighted equally, the VAT on maize, beans, and sugar is the most socially costly--or, conversely, that these are the best candidates for exemptions from the VAT. When higher weights are given to the welfare of the poor, maize remains the commodity most important to exempt, but kerosene becomes a better choice for exemption than sugar or beans. The difficulties in quantifying all the impacts of safety net spending and the com- peting ways in which funds may be spent in different sectors and programs and putting them in a common metric are significant. Thus most economists recognize that, in prac- tice, rigorously quantifying the marginal benefits from different forms of spending with each other and with marginal costs is generally not feasible (Besley, Burgess, and Rasul 2003; Devarajan, Theirfelder, and Suthiwart-Narueput 2001; Gupta and others 2001). Nonetheless, budget planners are deeply imbued with the principal that they should be 50 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS equated and will do what they can to respect it, even if they can only base their decisions on qualitative judgments. We can consider in a qualitative way when the marginal benefits of additional safety net spending would be most likely to outweigh the alternative uses of funds or marginal costs (see Coady and Harris 2004 for the theory and for an application to the Mexican reform that replaced general food subsidies with the PROGRESA CCT program in 1997). Benefits from additional safety net spending are more likely to be justified under the fol- lowing circumstances: · When the program is "good." As chapter 2 shows, safety nets can alleviate some of the misery of destitution, may help households invest in their livelihoods or their children's futures and manage risks, and can facilitate reform of government policies. Some programs will do better at producing some of these effects than others depending on their scale, consistency, flexibility, features, and the like. Chapters 2 and 9 provide criteria by which to judge programs in more detail, and the remainder of the book provides advice on how to achieve good outcomes. Funding proposals can be judged against this information when rigorous impact evaluations are unavailable. · When the net expenditure and/or tax package is progressive. One of the goals of safety nets is redistribution; thus, expenditures will be more justified if they are distributed progressively. Moreover, the productive effects will be more likely to be greatest for the poorest. Redistribution per currency unit transferred will be heightened if the expenditures are concentrated at the bottom end of the welfare distribution and if taxes are progressive rather than regressive. · When base spending on safety nets is low. When expenditures are low, more spending may well be justified, but as the program achieves something approach- ing adequate financing, the value of additional spending will decline. Higher funding would allow the program to expand from the poorest to the less poor, which implies a lower redistributive impact, and/or the larger budget could in- crease the level of benefits given to each beneficiary. This again implies less social value, as an additional currency unit of transfer will be less important for a less poor household than for a poorer one. · When alternative uses of funds are less important. Increasing spending on safe- ty nets is certainly more socially valuable than buying another palace for a dictator or his mistress. It is extremely likely to be more socially valuable than spending on a regressive gasoline subsidy that encourages pollution, but may or may not be more valuable than financing vaccinations or bednets to prevent malaria for the poor. · When the extra taxes to finance the expenditures have the lowest efficiency costs. Different tax instruments vary in the extent of distortions they introduce. Broadly based tax instruments--especially a VAT, and occasionally an income tax--will be less distortionary than choices such as trade taxes. Achieving the first three conditions will increase a program's marginal benefits, while achieving the last two will result in lower marginal costs. 3. FINANCING OF AND SPENDING ON SAFETY NETS 51 3.2 Sources of Financing for Safety Nets Money is fungible from one use to another, so the issue is whether an expenditure is justified in relation to the alternatives. Governments basically have four choices for how to finance a specific expenditure: reallocate expenditures from something else, increase taxes, find international grant financing, or borrow.3 Each of the four financing sources has advantages and disadvantages (table 3.2). Which option or combination of options is preferable depends on a country's situation: some have no possibility of increasing aid financing, some are so heavily indebted that further debt financing is unwise, and some have tax rates well above the average; but at the same time, some do have some flexibility on one or more dimensions (World Bank 2007e). TABLE 3.2 Options for Increasing Safety Net Budgets: Advantages and Disadvantages Financing source Advantages Disadvantages Expenditure · Finances programs within budget · No additional funds relative to the reallocation constraints budget · Increases overall productivity of · Many countries have low levels of government outlays discretionary spending Increased · More sustainable than other options · Economic costs taxation · Politically unpopular International · Increases availability of funds · Inflexibility in use of funds grants · Instability of funding · Donor coordination issues · Government autonomy issues Borrowing · Finances investment in productive · Currency mismatch in balance sheets activities in countries with low public · High debt service burden savings · Debt overhang impact on growth · Finances temporary expansion of programs during crises · Vulnerability to a solvency crisis SOURCE: Authors. REALLOCATING EXPENDITURES It is often possible to reallocate expenditures from other programs to safety nets. The advantage of expenditure reallocation is, obviously, that it does not require new resources and thus leaves the spending envelope unchanged. If the resources are reallocated from less effective or important programs to good safety nets, the overall productivity of gov- ernment outlays may increase. A disadvantage of expenditure reallocation is that the room for it may be limited. Government spending may be low, may be confined to equally important activities, or may have limited flexibility. Reallocation can also be politically difficult. Even where economists may see areas where the redeployment of funds could in- crease efficiency, politicians may see no realistic possibility of accomplishing the change. Reallocations to finance safety nets are most visible when the government reduces or eliminates an across-the-board subsidy and replaces it with a targeted transfer program. 52 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Jamaica, for example, eliminated general food subsidies in 1984 and used some of the sav- ings to fund its Food Stamp Program (Grosh 1992). Brazil reduced the subsidy on cook- ing gas in 2001 and funded the Auxilio Gás (Cooking Gas Grant) cash transfer program as compensation (Lindert, Skoufias, and Shapiro 2006). In 2005, Indonesia instituted a massive reduction in petroleum subsidies and reallocated half the funds implicitly saved to spending on health, education, and a new cash transfer program (Indrawati 2005). More subtle reallocations can also occur, as when other sectors are made more efficient and the savings reallocated. If reallocation within a given budget is impossible, the budget may need to be in- creased to accommodate the desired spending on safety nets. INCREASING TAXES Governments never take the option of increasing taxes lightly because it can have real economic costs, as well as the obvious political ones (see box 3.2 for a discussion of tax instruments). However, taxes in some developing countries are sufficiently below those of comparable countries that increased revenue collection seems economically feasible, even wise. Uganda, for example, raises the equivalent of about 11 to 12 percent of GDP in domestic revenues; this is much less than the 20 percent raised in Ethiopia, a country with otherwise similar fiscal characteristics, or the 25 percent average for low-income countries. Uganda might consider revenue enhancement seriously, and such enhancement is even a feature of policy dialogue in Ethiopia. Higher up the income scale, Chile and South Africa, which raise the equivalent of 24 and 25 percent of GDP in taxes, respective- ly, have much more potential to consider revenue enhancement than Brazil (45 percent of GDP) or Turkey (31 percent of GDP) (World Bank 2007e). Note that Chile and South Africa have managed to provide comprehensive safety net systems despite their relatively low expenditures. If taxes are to be raised to finance safety nets, they must not take more away from those who will benefit from safety nets than they give back. At a minimum, they should be neutral in their incidence. Chu, Davoodi, and Gupta (2000) provide a comprehensive overview of the tax incidence literature in developing countries. They note that 36 studies of 19 countries find that their overall tax systems are progressive in 13 cases, neutral in 7, and regressive in 7. The other studies have mixed findings or show insignificant effects. This indicates that, in most cases, progressively targeted safety net spending will be redis- tributive, but that this will not be true for a number of cases--underscoring the need to be cautious about financing increased safety net expenditures via increased taxation. OBTAINING GRANT FINANCING Grant financing from donor agencies does not solve the trade-offs issue: the use of funds still has opportunity costs. Grant aid and tax revenues should be considered as if they were part of one big budget that is allocated according to the merits of different uses of funds, with the grants increasing the size of the total pie, but not necessarily earmarking the size of the slice that goes to safety nets. The consensus around agreements and meetings such as the Monterrey Accords, the Millennium Development Goals, the Gleneagles Summit of the Group of Eight, and the African Action Plan is for increased donor support for low-income countries, which may indirectly increase the volume of resources available for 3. FINANCING OF AND SPENDING ON SAFETY NETS 53 BOX 3.2 What Tax Instruments Should Governments Use to Support Safety Nets? Governments can chose from several different types of tax instruments (see table). Public finance experts regard general revenues, that is, the pool of all government revenues, as the most ap- propriate source of financing for safety nets. Because safety nets are noncontributory benefits targeted to the poor, financing them through the broadest available base would ensure the largest degree of redistribution through the tax system. Financing them with progressive taxes would enhance the redistribution. Tax instrument Advantages Disadvantages General revenues · Large and stable tax base · Distortions of labor supply, saving, and (income taxes, consumption behaviors · Progressive incidence VAT, sales taxes) (income tax) · Regressive (VAT) · Procyclical (income tax) Payroll taxes · Protected in the budget · Regressive incidence · Linked to benefits · Labor market segmentation · Procyclical Earmarked sin · Politically viable · Usually yields limited revenues taxes · Tax may be desirable in its own right Cross- · Redistributive effect within a · Incentives for overconsumption subsidization program · Lack of fiscal transparency · Potential for contingent fiscal liabilities SOURCE: Authors. Payroll taxes are the classic instrument used to finance social insurance programs, so the ques- tion often arises whether they should be used to finance safety nets. In general, the answer is no for the following reasons: · If the proposal implies that, as with social insurance, the benefit is limited to those who have contributed, such an arrangement would exclude those most in need of a safety net. With high levels of informal and self-employment (up to half the economy in Latin America and much higher shares in Africa and South Asia), those outside the formal sector would be ineligible, but usually have lower incomes and no access to the risk mitigation con- veyed by the social insurance. · Payroll taxes contribute to the segmentation of the labor market into formal and informal sectors. Keeping payroll taxes as low as is consistent with their use in affiliation-based social insurance is thus desirable. · Payroll taxes are often less progressive than income taxes. The net impact is higher the more progressive the financing that supports expenditure on safety nets. · Earmarking of payroll taxes introduces rigidities in the budget and can favor inefficiencies. Some safety net programs in Colombia receive earmarked funds; critics believe this has induced an inefficient mix of programs (World Bank 2002d). 54 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS safety nets.4 Some donors are increasingly supportive of aid for safety nets specifically and so may label their support as being intended for safety nets. For example, a U.K. white paper on international development (DFID 2006) pledges to increase such spending in at least 10 African or South Asian low-income countries and double to 16 million the num- ber of families moved from emergency relief to long-term social protection programs. Three major problems arise with the way grant financing is usually delivered (see World Bank 2005e, chapter 3, for a cogent analysis of the issues powerfully illustrated with evidence from the health and education sectors). First, grants are often inflexible, cover- ing only some parts of program costs. Second, aid flows can be volatile, and most donors operate on the basis of one- or two-year commitment cycles. At worst, this can result in programs that start up and disappear during short time horizons and thus never get past the initial start-up or troubleshooting phase. Even an attenuated version of the problem can result in a reluctance to undertake administrative improvements that require a number of years to accomplish or to achieve their payoff. Third, development assistance can carry high transaction costs. Many donors may be present in a single country, and frequently each donor funds separate, often similar, programs, which precludes the realization of economies of scale. Donors also often insist on their own sets of procedures for accounting or procurement, so a country must run parallel administrative systems and devote effort to managing donors rather than to service delivery. As grant aid is meant for lower-income countries, such inefficiencies are especially regrettable. Ethiopia's experience with support for its safety net program prior to its 2005 re- form illustrates some of the problems common to grant aid. The funding was significant, averaging about US$265 million per year from 1997 to 2002, but it was generated on a system of annual emergency appeals and thus was volatile in amount, varying from US$152 million in 1998 to US$449 million in 2000. Based on policy, 80 percent of the aid was dedicated to public works programs, but the programs encountered problems that were exacerbated by the aid arrangements. The World Bank (2004a, p. 135) summarizes these problems: Food comes too late, the amount of food distributed is so diluted that each household receives too little to materially affect their welfare. In the case of works programs, the stop-start nature of programs prevents them having a sustained impact on the incomes of the poor; and the absence of counterpart funds and integration with local capital plans means they often do not result in creation of lasting, productive assets. Ethiopia's 2005 reforms of the productive safety net are intended to overcome several of these problems by establishing a government-driven system for aid to feed into, re- questing multiyear pledges, and changing administrative arrangements to pave the way to greater impact. The reform effort required significant political will to begin and will need to be sustained over subsequent years to achieve its full effect. BORROWING OR USING DEFICIT FINANCING The general wisdom in public finance is that debt financing is advisable only when the extra spending financed by borrowing raises the country's ability to repay the debt in the future--for example, spending on infrastructure or education. Borrowing to finance current expenditure with no impact on future income-generation capacity and produc- 3. FINANCING OF AND SPENDING ON SAFETY NETS 55 tivity can lead to a debt overhang, a lack of fiscal sustainability, and a greater likelihood of financial crises (Ter Rele and Westerhout 2003). Safety nets traditionally have been considered unproductive and merely redistributive. As argued in chapter 2, this may be an unduly harsh view, at least when safety nets are effectively run and targeted. Debt financing may be particularly applicable for safety nets when they benefit fu- ture generations in ways that will raise their productivity, and consequently future tax revenues, or when a temporary increase in expenditures is needed, as after an economic crisis or a natural disaster. Both these conditions pertained in 2001 when Colombia and Turkey borrowed money from the International Bank from Reconstruction and Develop- ment to found CCT programs to ensure that the economic crises in their countries did not impair the poor's ability to build human capital for their children. At the same time, such programs do not always scale back automatically after a crisis. Governments may borrow from international development banks in part to obtain the technical assistance and oversight that is often bundled with international develop- ment lending or to signal a multiyear commitment of funds to stakeholders, indicate a likelihood of technical quality, or the like. Borrowing from international development banks can entail some of the same problems as grant financing, although sometimes to a lesser extent, as the funding is more often multiyear and integrated into the government budget and the programs are executed by government agencies. 3.3 In Search of Countercyclical Financing for Safety Nets A special concern in financing safety nets is how to guarantee adequate resources during macroeconomic crises or following disasters. This is particularly important in developing countries. Ferreira, Prennushi, and Ravallion (1999) point out that crises in which gross national product declines over a 12-month period and/or inflation doubles to a monthly rate above 40 percent per year are rare in OECD economies, but relatively common in the developing world. In such crises, the living standards of many people--almost invari- ably including the poorest--will fall for some period of time. If safety nets are to protect the poor in times of economic downturn, they obviously will need larger budgets than in times of growth to grant benefits to the increased number of poor and to grant higher benefits to those who were already poor and become poorer. When funding increases in such times of need, it is called countercyclical (see Alder- man and Haque 2006 for a thorough discussion of what it takes to provide a countercy- clical safety net, including not only financing, but adequate targeting rules and ability to scale up quickly). The term is perhaps a bit pallid for the developing country context. It originated in the public finance jargon of the industrial countries, where the ups and downs are mild and are associated with the business cycle rather than the enormously greater volatility in developing countries. Unfortunately, in practice, safety net spending has tended to be procyclical rather than countercyclical (Braun and Di Gresia 2003). De Ferranti and others (2000), for example, find that even though seven Latin American countries did a good job of main- taining the share of targeted and social spending in the budget during a crisis, for every 1 percent decline in GDP, spending per poor person fell by 2 percent. 56 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Governments often try to protect budgets for safety net programs by increasing their share of the budget and reducing the shares of other items, but such reallocation is un- likely to fully protect spending per poor person during crises. Even when governments give safety net spending high priority, spending itself will fall. Crises involve declines in real wages and employment, which lower revenues. Moreover, the stabilization packages adopted to try to pull the economy out of the crisis often involve reducing fiscal deficits, with expenditure reduction usually an important means of doing so. Simultaneously, the number of poor people rises, as does the level of need among the chronically poor. Thus just when needs are rising, means are falling. The apparent solution to this problem is to prefund safety net program budgets. Some unemployment insurance funds work on this principal, collecting contributions in a special fund while workers are employed and paying out during recessions. Such arrange- ments do have their limits. In the early 1990s, several Eastern European countries reduced the replacement rate and duration of unemployment benefits, at least in part because of fiscal pressures associated with the transition (Vodopivec, Wörgötter, and Raju 2003). Some countries have tried to hold reserve stocks of grain, with less than desired outcomes (World Bank 2005h). Some countries, such as India, Mexico, and the Philippines, hold reserve funds for relief programs (Gurenko and Lester 2004). The state of Maharashtra in India has ear- marked a specific tax to fund countercyclical public works. While such funding may be ad- equate for localized emergencies, the needs imposed by large covariate shocks, such as the flood in Bangladesh in 1998 or the 2004 Asian tsunami, cannot be met without external support and/or macroeconomic consequences (Alderman and Haque 2006). Prefunding is quite rare for social assistance programs. Instead, countercyclical finance may be achieved through prudently low overall spending during stable times and increased spending or borrowing in times of increased need. Chile and Colombia are among the countries that have taken this route. Pressures on governments to spend revenues as they are collected have hampered the accumulation of fiscal savings to be used in the case of need. A typical example is the pro- cyclical accumulation of oil revenues (Alesina and Tabellini 2005; Davis, Ossowsky, and Fedelino 2003). The tendency is general, but pertains fully in the case of safety nets. Governments have used fiscal responsibility laws as a way to enforce broad countercy- clical fiscal policy. These rules are intended to aid in maintaining fiscal discipline and, per- haps equally important, to signal to creditors and other concerned parties that deficits during times of recession are not to be seen as a sign of irresponsibility but as part of a planned countercyclical policy. The overall performance of fiscal rules has been mixed. They are ap- parently neither necessary nor sufficient for overall fiscal discipline, much less do the results pass through directly to a single rubric of spending, such as safety nets (box 3.3). The desire for countercyclical funding for safety nets stems from the aim of serving those who need help. But it is pertinent to note that even in good times very few safety net programs in developing countries are fully funded. In industrial countries, most safety net programs are operated as "entitlement" programs--that is, all households meeting the eli- gibility criteria are guaranteed entry, with administrative processes and budgets (or budget flexibility) to back that promise. In developing countries, the vast majority of programs have some sort of a rationing mechanism to ensure that budgets are kept to an allowed 3. FINANCING OF AND SPENDING ON SAFETY NETS 57 BOX 3.3 Fiscal Responsibility Laws Fiscal responsibility laws can be of two different types: · Laws that mostly establish quantitative fiscal targets (Kopits and Symansky 1998), such as the government's overall deficit or a ceiling on certain high-priority spending levels, for example, the fiscal responsibility legislation in India · Laws that focus on enhanced fiscal transparency and public expenditure management, for example, as in New Zealand Countries' experiences show that unless important preconditions in term of fiscal transparency, budget accounts, political consensus, and enforcement mechanisms are met, fiscal responsibil- ity legislation in itself is insufficient to ensure that fiscal policy is sound. Also, fiscal rules tend to be less effective if they only cover the central government, are too specific, and do not foster a reallocation of government spending among programs. Countries that have successful fiscal rules include Brazil, where the law has fostered the cred- ibility of the government's policies, and Chile, where fiscal rules had a role in protecting social spending in 2004. Automatic stabilizers tend to be most effective in industrial countries. Failures of fiscal rules in Argentina and elsewhere in Latin America are among the factors that contrib- uted to macroeconomic instability (Singh and others 2005). An example of a fiscal rule that has not been fully enforced is the European Union's Stability and Growth Pact. amount: sometimes beneficiary rosters are opened once only and then closed for years, or they are opened annually to a number of beneficiaries who exhaust the fixed budget and then closed. There are several programs in Eastern Europe that have entry rules and administrative processes sufficient for an entitlement program but that have sometimes lacked the fiscal support and thus had to run in arrears. It is sobering to note that even these upper-middle income European Union accession countries cannot deliver an entitle- ment system. The larger magnitudes and greater frequency of downturns in developing economies implies that the fiscal risk that would be incurred by moving to an entitlement design would be greater than borne in the OECD countries where entitlement is com- monplace. With entitlement programs, the number of beneficiaries and program budget natu- rally fall once the general economy recovers. People get jobs, their earnings increase and place them over the threshold for a means test, or they become too well-off to choose to participate in self-targeting programs. When programs are not operated as entitlements, this decline in expenditures will be less automatic. Some households will leave the pro- gram as a result of their greater prosperity, but other eligible households that have been rationed out will be waiting to take their place. Reducing program expenditures may mean introducing tighter rationing, which may be politically difficult to do. Program managers and advocates for the poor will point to unmet needs and available resources and wonder why they cannot be matched up. In this way, the desirable goal of assisting the needy in good times comes into partial conflict with achieving the countercyclical husbanding of resources to help them during hard times. 58 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Prospects for achieving countercyclical financing for safety nets in developing coun- tries seem slim until safety nets are fully financed in stable times and volatility is lower than has recently been the case. Expenditure reallocation in favor of safety nets during downturns and generally prudent fiscal policy will help and have been put into effect, but to a degree insufficient to yield countercyclical funding for safety nets. 3.4 The Cost of the Welfare State in Developed Countries One of the implications of the basic theory of optimal spending is that a country that spends too much on safety nets will pay a price: it will fail to invest in other, more im- portant things and/or it will struggle under unduly burdensome taxes. So another way to think about how much effort countries might be able to devote to safety nets comes from looking at the empirical literature on the cost of safety nets. The literature on the costs of the welfare state predominately concerns OECD coun- tries, because that is where such expenditures are dramatically the highest and the issues most pressing. Atkinson (1999), Barr (2004), and Lindert (2004) critique older literature in ways that largely discount the concern that high expenditures on well-designed social protection systems will slow growth. They put forward the following arguments for how countries can afford substantial transfer systems, which are traditionally viewed as unpro- ductive and are expected to lower growth through the burden of the taxes to support them and the labor disincentives effects among recipients: · The simplistic models used to describe the labor-reducing effects of tax and trans- fer policies are too simple and extreme, and are therefore misleading. · The assessments of the cost of the welfare state have focused on the costs in terms of growth but have not tried to calculate the benefits that derive from the pro- grams, and as such are erroneously specified. · The empirical evidence shows that in practice, growth and welfare state spending are weakly correlated. Lindert (2004, p. 234) notes: "Within the range of true historical experience, there is no clear net GDP cost of higher social transfers."5 All three authors provide examples of how real-life social protection policies have design features to limit their potential distortions. Blank (2002) joins the chorus. She categorizes the ways in which program design can minimize the leaks in Okun's bucket by supporting those unlikely to work (the elderly, children, people with disabilities); im- posing job search, work, or study requirements on those who can work (often labeled as activation or welfare-to-work reforms) in industrial countries, CCT programs in middle- income countries, or public works jobs in low-income countries; or by investing, as in many programs for young children, certainly including those linked to their health or education, and possibly even general child allowances. Though not precisely quantified, a substantial share of safety net spending actually goes to such programs. The implications of the OECD literature for developing countries are as follows: · The literature is concerned with the whole package of social protection or of social protection and health insurance, so that the transfers considered average 21 percent of GDP for the OECD countries and range up to 30 percent for the 3. FINANCING OF AND SPENDING ON SAFETY NETS 59 highest spenders, an order of magnitude more than the range of spending on safety nets in developing countries. This implies that developing countries spend- ing little on safety nets may be able to spend a bit more without unduly harming their economies. · The OECD countries have essentially added their social protection systems to the list of other social service and infrastructure duties of government. Social protection has not come directly as a trade-off between, for instance, establish- ing universal education or good road systems, but in addition to them. In the low-income country setting, debates on safety nets are often couched as transfers versus development. Perhaps the issue should be rephrased in terms of whether safety nets are an important (additional) component of development policy. · The highest-spending countries have chosen a relatively efficient pattern of taxa- tion, more so than some of the lower spenders (Lindert 2004). The impact of social protection on growth depends not only on the magnitude of spending, but on how the spending is financed. · Because the OECD literature covers long-standing systems, their benefits are per- haps being realized and captured in effects on growth as a counterbalance to their costs. This suggests that the findings could easily apply to developing countries that have well-designed safety nets and less so to those whose programs are ineffective. The issue of how much a country can "afford" to spend on safety nets is dictated by more than the technical issues of economics. It involves choices between things society val- ues, and is thus deeply political as well. The literature on the political economy of support for social spending or safety nets shows that public attitudes on the topic vary within any given country and that the predominant view also differs among countries. The World Values Survey and the similar Latinobarómetro provide evidence on the strength of beliefs that poverty is largely due to FIGURE 3.1 Societal Attitudes about Poverty and Spending on individual behavior, such Social Welfare as laziness, or to forces outside the individual's Social welfare spending as a percentage of GDP control, such as bad luck, 20 Fitted values lack of family connec- Sweden tions, or the fault of so- ciety.6 This evidence can Germany 15 be tied to facts about the Spain welfare programs in the Norway pertinent polities, as il- Finland 10 lustrated in figure 3.1 Japan typifies the finding. The United StatesAustralia figure shows that social 5 welfare spending is much higher as a percentage of 0.4 0.6 0.8 GDP in countries where Poverty is society's fault public attitudes reflect a SOURCE: Alesina and Glaeser 2004. 60 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS view that poverty is caused by forces outside of an individual's control. Alesina and Glaeser (2004) show that among U.S. states, a similar correlation is apparent between beliefs about the cause of poverty and maximum benefit levels in the Aid to Families with Dependent Children program, the former principal welfare program in the United States. Lindert, Skoufias, and Shapiro (2006) report that, in general, Latin American attitudes about the causes of poverty, redistribution, and intergenerational mobility follow the European pat- tern and thus provide substantial support for safety nets (table 3.3). Other political factors are important as well. Substantial work has been devoted to the role of ethnic homogeneity and its effects on support for safety nets. Most analysts agree that homogeneity is supportive of safety nets and heterogeneity is harmful (Ale- sina and Glaeser 2004; Lee and Roemer 2004; Lindert 2004). The extent of democracy (Lindert 2004) and the form of its institutions--for instance, proportional versus majori- tarian representation--seem to matter as well (Alesina and Glaeser 2004). Societal attitudes about the causes of poverty define who the "deserving" poor are. In turn, the view of deservedness influences the choice of transfer programs and their design. The deserving poor are usually a subset of the poor who are poor through no fault of their own. The view of who among the poor is perceived as deserving of public transfers differs from country to country and in the same country over time. TABLE 3.3 Perceptions of Poverty in the United States, Europe, and Latin America, 1995­7 (percentage of respondents) The poor are poor because The poor have little The government's Society is They are chance of escaping response to poverty is Region and country unjust lazy from poverty inadequate Continental Europe 63.3 17.1 60.2 64.5 Latin America and the 65.8 28.3 62.0 67.7 Caribbean (average) Argentina 74.0 26.0 74.5 81.7 Brazil 75.7 20.5 70.5 -- Chile 55.6 36.9 58.5 58.7 Colombia -- -- 55.8 -- Dominican Republic 68.6 24.5 61.2 89.0 Mexico 65.8 24.6 56.9 71.1 Peru 56.5 34.2 47.1 44.8 Uruguay 77.2 12.4 73.5 80.8 Venezuela, R. B. de 52.9 47.1 59.6 79.9 United States 38.8 61.2 29.5 41.8 SOURCE: Lindert, Skoufias, and Shapiro 2006, table 3. NOTE: -- = not available. 3. FINANCING OF AND SPENDING ON SAFETY NETS 61 Consider the example of Bulgaria. In 1996 and 1997, Bulgaria experienced a deep economic crisis. GDP fell by 15 percentage points over two years, and unemployment rose from 10 percent in 1995 to 14 percent in 1997. The view that the whole country had suffered from the transition to a market economy was pervasive. To arrest the nega- tive consequences of the crisis on the poorest, the government increased the threshold of its Guaranteed Minimum Income Program, a cash transfer program designed to support everyone whose income fell below a certain threshold, by topping up their actual incomes to bring them to a standard minimum. This topping-up procedure meant that every ben- eficiary household received only the minimum it needed and spread the scheme's resources as thinly as possible. Such a program would, at least in theory, provide huge labor disin- centives. Why would people earning below the threshold work at all if their income were raised to the threshold level regardless of whether they worked? Given the sharp rise in unem- ployment at that time, such concerns were not central to the policy debate. Unemployment did not carry the stigma of laziness; rather, people were sorry for those down on their luck. Five years after the crisis, the circumstances were quite different: Bulgaria had regis- tered its fifth consecutive year of robust growth of 5 to -6 percent per person per year, but at 18 percent, unemployment was both persistent and high. Concerns about the negative impact of the Guaranteed Minimum Income Program on work incentives took center stage in the policy debate. In 2002, the government implemented a public works pro- gram targeted to the long-term unemployed receiving transfers from the program, who accounted for half of all beneficiaries. The public works jobs paid the minimum wage plus benefits. The program's coverage and spending fell by one-third while substantially improving its targeting performance. The shift in political attitudes about who was con- sidered to be the deserving poor brought about the 2002 reforms, influenced the mix of safety net programs, and altered program features to reduce labor disincentives. 3.5 Levels and Patterns of Safety Net Spending in Developing and Transition Countries Because indicating precisely what countries should spend on safety nets based on theory is so difficult, analysts often seek guidance based on benchmarks. Benchmarking is, of course, imperfect, as there is nothing to indicate whether the countries used in the bench- marks are spending the "right" amount. In addition, choosing benchmarks is an art. Countries are often compared with their neighbors, which may share the same historical and institutional forces--and may therefore share the same tendencies to "wrong" expen- diture. More thoughtful selection of benchmarks considers a wider set of countries at the same economic level and with comparable demographic characteristics, or ones that the country in question hopes to emulate in the future. To assist in benchmarking, we have developed a new dataset on spending in devel- oping and transition countries and have tried to understand some of the factors that may be associated with relatively high or low spending on safety nets. These factors include level of income, extent of inequality, governance, democracy, presence of different ethnic groups, and variation in public attitudes. We present our findings and provide the raw data in Weigand and Grosh (2008) so that analysts may use them for their own benchmarking exercises. 62 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Quantifying spending on safety nets is difficult. The conceptual definition used in this book does not fit within a single ministry's mandate, so the most easily and regularly obtainable sets of numbers on government spending are not useful for tracking spending on safety nets. In Peru, for example, the main safety net programs fall under half a dozen ministries and three different levels of government. And Peru only has about 20 major safety net programs, many fewer than commonly found elsewhere: Bulgaria has 34 pro- grams; Mexico over 100, spread through dozens of agencies and three levels of govern- ment. The International Monetary Fund's (IMF's) Government Finance Statistics is acces- sible, published frequently, and takes care to establish comparability, but does not have a category that closely matches the concept of safety nets as used in this book (IMF 2001). It lumps much social assistance in with social insurance to come up with a single figure for "social security and welfare"; other social assistance may fall under the "transfers to households and other organizations" category;7 and more will be reported in the accounts of the ministries that house or serve as umbrella organizations for the various programs, especially if these are in-kind programs. Despite their shortcomings, the IMF numbers have been the basis of a literature on safety net spending summarized in box 3.4. To fill the gap in knowledge about safety net spending in developing countries, we rely here on information provided by Weigand and Grosh (2008) that more closely follows the conceptual definition of safety nets used in this book. We supplement this information with data for a handful of OECD countries from the OECD Social Expenditure Database (OECD 2004b) and with data from the World Bank (2007c). Weigand and Grosh (2008) compile data from World Bank public expenditure reviews and other similar analytical work. These studies, performed as one-time or periodic reviews of social policy, try to sort through countries' budgets and programmatic structures to assemble comprehensive numbers, an exercise inherently different from that usually carried out for a given country as part of its annual budgetary process. The following three caveats to these data are in order: · Incomplete coverage. Weigand and Grosh provide data for 87 countries between 1996 and 2006.8 Coverage varies by region. It is high for Europe and Central Asia, with 25 of the 29 countries covered (and 96 percent of the population). Coverage is much less for Sub-Saharan Africa, with 9 of the 47 countries covered (and 18 percent of the population). · Comparability. Because the expenditure numbers compiled by Weigand and Grosh were calculated by the various authors of the many country reports, the precise definition of what to include in the safety net or social protection sector as a whole varies. Weigand and Grosh report the composites largely as they occur in the reports, trusting to the judgments of the authors of the individual reports to include what was pertinent and available in a given country. For health and education expenditures, Weigand and Grosh use World Bank (2007c), which has less serious comparability issues. · Interpretation. What countries do spend is not necessarily what they should spend. The reports underlying the data reported here were undertaken because the level of spending was a policy issue at the time the individual country studies 3. FINANCING OF AND SPENDING ON SAFETY NETS 63 BOX 3.4 Literature on Safety Net Spending Levels Based on the IMF's Government Finance Statistics Safety net programs typically represent about 1 to 2 percent or less of GDP in developing countries. This compares with spending levels of 2 to 4 percent of GDP in industrial countries (Atkinson 1995). Average spending levels tend to be higher in middle-income countries than in low-income countries, reflecting the low revenue base in the latter countries, but variability is large (Fox 2003). Spending levels also vary by region, with South Asian and Sub-Saharan African countries spending less than Latin American and Caribbean countries and countries in Eastern and Central Europe and the Middle East spending more (Besley, Burgess, and Rasul 2003). Various authors have tested for and found different factors that may affect the level of safety net spending or of social spending more broadly. Higher per capita incomes tend to be associated with higher spending on social assistance programs, while the incidence of poverty and inequal- ity are not necessarily good predictors of the level of spending on safety nets.This is because in many regions--for example, Latin America and the Caribbean--the system of social protection is split between social insurance for the (wealthier) formal sector worker and meager social as- sistance for the (poorer) worker in the informal sector (Fiszbein 2004). Schwabish, Smeeding, and Osberg (2004) find that inequality between the middle class and the poor (as measured by the ratio of welfare between those at the 50th percentile and those at the 10th percentile) has a small, positive impact on social spending, but that inequality between the ends of the distribution and the middle class (as measured by the ratio of welfare between those at the 90th percentile and those at the 50th percentile) has a large and negative impact. Also, spend- ing levels tend to be higher for countries with better governance indicators (Baldacci, Hillman, and Kojo 2004), but are not necessarily different in decentralized and centralized economies (Ter-Minassian 1997). Spending on safety nets tends to be correlated with government size, but is generally negatively correlated with fiscal deficits and inflation.This is because countries with unstable macroeconomic conditions are more likely to have insufficient resources to finance the safety net (de Ferranti and others 2000). were done. This suggests that at least some parties thought the spending level was inappropriate. Spending on safety nets as a percentage of GDP provides a summary measure of a government's efforts to provide safety nets. We also broaden our view to consider wider concepts of spending. We define social protection as the sum of safety nets (social as- sistance) and social insurance (pensions, unemployment insurance). We define the social sectors as the sum of spending on social protection, health, and education. The data show the following: · Mean spending on safety nets is 1.9 percent of GDP, and median spending is 1.4 percent of GDP. For about half of the countries, spending falls between 1 and 2 percent of GDP (figure 3.2).9 Some variation is apparent. For example, Bosnia and Herzegovina, Pakistan, and Tajikistan spend considerably less than 1 percent of GDP. At the other extreme, spending on safety nets in Ethiopia and 64 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS FIGURE 3.2 Safety Net Expenditures as a Percentage of GDP, Selected Countries and Years Senegal Maldives Philippines Bosnia & Herzegovina Lebanon Paraguay Pakistan China Uruguay Tajikistan Colombia Venezuela, R.B. De Chile Kyrgyz Republic Mexico Bangladesh Peru Jamaica Madagascar El Salvador St. Kitts & Nevis Sri Lanka Yemen, Republic Of Poland Romania Nicaragua Ecuador Vietnam Guatemala Bulgaria Albania Indonesia Jordan Latvia Mongolia Serbia Brazil Macedonia, Fyr Argentina Costa Rica Georgia Azerbaijan Grenada Egypt, Arab Rep. Of Moldova Panama Dominican Republic Croatia Russian Federation Morocco Montenegro Uzbekistan Bolivia St. Lucia Turkey Armenia Dominica Kazakhstan India St. Vincent Tunisia Czech Republic Kosovo Honduras Iran, Islamic Rep. Of Ukraine South Africa Slovak Republic Algeria Malawi Ethiopia Djibouti Mauritius 0 1 2 3 4 5 6 7 8 Percentage of GDP SOURCE: Weigand and Grosh 2008. NOTE: Kosovo data are for 2003. 3. FINANCING OF AND SPENDING ON SAFETY NETS 65 Malawi is nearly 4.5 percent of GDP because international aid is counted; these shares would be closer to 0.5 percent if only domestically financed spending were counted. Other high-spending countries--Mauritius, South Africa, and the Slo- vak Republic--finance their safety nets domestically. · Regional patterns are about as might be expected, with the Middle East and North Africa spending the most (2.2 percent on average), followed by Europe and Central Asia (1.7 percent on average), and Latin America and the Caribbean (1.3 percent on average). The smaller number of observations makes the aver- ages less robust for the other regions. For instance, the average of 3.5 percent for Sub-Saharan Africa is based on only six observations and includes external financing. · Spending on safety nets is less variable than spending on social protection or the social sectors (figures 3.3 and 3.4). To try to understand the sources of variation in spending patterns, we look at these patterns and their relationship to the following variables typically discussed in the litera- ture on developed countries: · Country income as measured by GDP per capita with purchasing power parity adjustments. The hypothesis is that richer countries will spend more. · Inequality as measured by the Gini coefficient. The hypothesis varies with the model of power assumed. A one-person, one-vote economy with higher inequal- ity will face more pressure for redistribution, because the number of people with incomes below the mean will be higher. In a model with elite capture of govern- ment, the elite may use private providers of social services and give little support to public ones, so higher inequality may lead to lower spending. · Voice as measured by the Kaufmann, Kraay, and Mastruzzi (2005) index for voice. We hypothesize is that greater voice will be positively related to spending on safety nets, social protection, and/or the social sectors. · Ethnic fragmentation as measured by Alesina and others (2002). The hypothesis is that greater fragmentation will lead to lower spending on safety nets, social protection, and/or the social sectors. · Democracy as measured by the Polity IV Project (2008). The hypothesis is that greater democracy will lead to higher spending on safety nets, social protection, and/or the social sectors. · Attitudes about inequality as based on a question from the 1990­2004 ques- tionnaires of the World Values Survey, which asks respondents to score their at- titudes on a scale with "incomes should be made more equal" at one end and "we need larger income differences as incentives for individual efforts" at the other. We hypothesize that spending will be higher when more people believe in the need for greater equality. In simple correlations, most of the factors have the expected sign, but the strength of the correlation is generally higher the broader the concept of spending used (table 3.4). For spending on safety nets alone, none of the factors examined correlate significantly. 66 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS FIGURE 3.3 Social Assistance and Social Insurance as a Percentage of GDP by Region, Selected Years Percentage of GDP 14 Social assistance Social insurance 12 10 8 6 4 2 0 East Asia Eastern Latin Middle East South Asia Sub-Saharan OECD Pacific Europe & America & Northern (n = 5) Africa (n = 23) (n = 9) Central Asia & the Africa (n = 9) (n = 25) Caribbean (n = 13) (n = 25) SOURCES: Weigand and Grosh 2008; OECD 2004b. NOTE: Not all the reports Weigand and Grosh use offer data on all categories of spending. For the OECD, we used 23 countries, as such countries as Mexico and Poland are already accounted for in the regional averages. FIGURE 3.4 Social Assistance, Social Insurance, and Social Sector Spending by Region, Selected Years Percentage of GDP 20 Health 18 Education 16 Social insurance Social assistance 14 12 10 8 6 4 2 0 East Asia Eastern Latin Middle East South Asia Sub-Saharan OECD Pacific Europe & America & Northern (n = 5) Africa (n = 23) (n = 9) Central Asia & the Africa (n = 9) (n = 25) Caribbean (n = 13) (n = 25) SOURCES: Weigand and Grosh 2008; OECD 2004b. NOTE: See note to figure 3.3. 3. FINANCING OF AND SPENDING ON SAFETY NETS 67 However, spending on social protection and the social sectors is significantly higher where income or voice are higher and lower where inequality is higher. TABLE 3.4 Correlations between Spending on Social Sectors and Other Factors Safety net spending Social protection spending Social sector spending Factor as % of GDP as % of GDP as % of GDP Per capita GDP (PPP) 0.0768 0.5045** 0.5460** Gini coefficient -0.1104 -0.3410** -0.2686* Voice 0.0678 0.2294** 0.2607** Ethnic fragmentation 0.1628 -0.0204 -0.0972 Democracy 0.1733 -0.0533 0.1907 Attitudes about inequality 0.1234 -0.1694 -0.1559 SOURCE: Authors' calculations. NOTE: PPP = purchasing power parity. Factors are measured as described in the text. * indicates that coefficients are sig- nificant at the 10 percent level or better. ** indicates that coefficients are significant at the 5 percent level or better. The results on measured inequality are worth noting: the correlations are all negative, that is, higher Ginis associated with lower spending on safety nets, social protection, and the social sectors as a whole. In examining the data in detail, most of the low Gini countries are in Europe and Central Asia, which historically has large social protection sectors; the high Gini countries are in Latin America and the Caribbean, which has a history of truncated welfare states. Because these two regions dominate the dataset, the inequality variable used may be capturing a historical legacy more than the real workings of inequality in relation to decision making. Figure 3.5 shows the more robust correlations for broader concepts of spending. The relationship with GDP is much more marked for the social sectors than for safety nets alone. In looking at attitudes to inequality, the finding for the social sectors echoes Alesina and Glaeser's (2004) findings for OECD economies presented in section 3.4; in contrast, the pattern for safety nets is not statistically significant and is of the opposite slope as would be expected. We interpret the pattern of results--that the correlates of social spending viewed broadly are more definitive than the determinants of spending on safety nets--to mean that societies agree that a certain floor of safety nets is required, but that they also have reservations about making the safety net too large. Thus, when support for social policy is higher, it tends not to be expressed through more spending on safety nets, but through more spending on allied social policies pertaining to social insurance, health, and/or educa- tion. This interpretation is consistent with the patterns of spending shown in figure 3.5. In sum, safety net spending as a share of GDP is not too diverse, with most coun- tries concentrated in the 1 to 2 percent range. There may be a case for those much below this range to move into it and for higher spending in low-income countries, but clearly for many countries, the most pressing issue will not be changing the size of the budget 68 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS FIGURE 3.5 Spending, Income, and Public Attitudes Expenditures on social assistance Expenditures on social spending as share of GDP as share of GDP 8 25 Mauritius Mauritius Poland 6 20 Belarus Croatia Ukraine Brazil Slovak Rep. Djibouti Macedonia, FYR Ethiopia Bulgaria 4 15 Djibouti Romania Uruguay Malawi Turkey Algeria Slovak Rep. Russian Federation Bolivia Ukraine South Africa Mongolia Chile Iran, Islamic Rep. Malawi Nicaragua Colombia Honduras Kazakhstan Costa Rica 2 Tunisia 10 Armenia Uzbekistan Georgia St. Kitts & Nevis Croatia Russian Fed. Peru Grenada Mexico Mongolia Costa Rica Argentina Madagascar Ecuador Romania Poland Dominican Rep. Madagascar Jamaica Jamaica Mexico Tajikistan Senegal China Uruguay Thailand 0 Senegal 5 Cambodia Philippines 0 5,000 1,0000 15,000 0 5,000 10,000 15,000 GDP per capita (PPP in US$) GDP per capita (PPP in US$) Expenditures on social assistance Expenditures on social spending as share of GDP as share of GDP 4 25 Algeria Poland South Africa Ukraine Croatia 3 20 Belarus Iran, Islamic Rep. Of Czech Rep. Ukraine Brazil Czech Rep. Macedonia, Fyr Bulgaria India Armenia 2 Turkey Uruguay Montenegro 15 Romania Russian Federation Turkey Croatia Russian Federation Argentina Moldova Argentina Georgia Chile Albania Colombia Jordan Armenia El Salvador 1 Romania Azerbaijan Vietnam El Salvador 10 Mexico India Iran, Islamic Rep. Chile Bangladesh Mexico Bangladesh Uruguay China Pakistan Pakistan Bosnia & Herzegovina 0 Philippines 5 2 4 6 8 10 2 4 6 8 10 Incomes should be We need larger Incomes should be We need larger made equal income differences made more equal income differences as incentives for as incentives for individual efforts individual efforts Ranked attitudes Ranked attitudes SOURCE: Weigand and Grosh (2008). NOTE: PPP = purchasing power parity. In the interest of legibility, not all data points are labeled. Gray shading indicates the 95 percent confidence interval around the regression. envelope devoted to safety nets, but making the most of that spending. That is the theme of the remainder of this book. 3.6 Delivering Safety Nets in a Decentralized World Conventional wisdom in the public finance literature suggests that redistribution is a role most appropriately handled by central governments, because they are inherently in the best position to handle interregional inequalities and risk pooling, but that subnational jurisdic- tions may be well placed to administer safety nets, because they may have greater knowledge of or contact with the client base for the programs (Musgrave 1959; Oates 1972). The idea that people of similar circumstances should be treated alike underlies the call for national fi- nancing for safety net programs. It also implies a role for national standard setting, not only in relation to eligibility criteria and benefit levels, but to features of service delivery as well. 3. FINANCING OF AND SPENDING ON SAFETY NETS 69 Decentralization has swept the developing world. In the last 25 years, governments of more than 75 countries have shifted more responsibilities to lower tiers of government. Most of these lower-level governments are elected, so the decentralization is political as well as administrative. The motivations for this vary. In Eastern Europe and the former Soviet Union, decentralization was part of the reaction to the former system of central planning; in Latin America, it was part of the transition to democracy; in countries such as Indonesia and Sri Lanka, it was a way to deal with regional and ethnic conflicts; and elsewhere, such as in Chile and Uganda, the goals were more explicitly related to the qual- ity of service delivery. The reality today is that a large share of safety net programs involve multiple tiers of government, which invokes what economists call the principal-agent problem. The national government is the financier, or the principal, and wants the programs it funds to operate in certain ways. It must rely on local governments--the agents--to carry out program-related functions, but cannot observe and control every action that these agents take. The solution lies in creating a mix of incentives and oversight mechanisms that bring the agents' actions in line with the principal's goals at an acceptable cost. THE CENTRAL GOVERNMENT AS PRINCIPAL FINANCIER OF SAFETY NETS Interregional inequalities can be significant and can present a conundrum: The poorest regions will have the highest poverty rates and the greatest need for social assistance for the chronically poor, but, at the same time, have the least capacity to tax their residents to raise revenues for distribution. For example, poverty rates in the regions of the Russian Federation vary from 3 to 56 percent. People with the same education levels and fam- ily composition are three times more likely to be poor in Dagestan Oblast or the Tuva Republic than those in the rich Tumen Oblast or in the city of Moscow (World Bank 2005l). This is not surprising in such a large, federal country; the situation is much the same in Brazil, China, India, and Nigeria. But such disparity also exists in much smaller, unitary countries. In Latvia, poverty estimates in 2004 ranged from 4 percent in Riga to 12 percent in Latgale (World Bank 2007i). Because local governments both finance and implement Latvia's Guaranteed Minimum Income Program, the relatively richer regions have considerably more funds available to spend than the poorer regions. This results in a perverse outcome whereby more than 40 percent of total social assistance transfers in Latvia go to people living in Riga, even though they are, on average, considerably better off than those residing in other parts of the country (World Bank 2007o). Experience from Bosnia and Herzegovina, Bulgaria, and Romania demonstrates even more acutely the problem of assuming that local governments can reasonably finance safety nets. In 1995, Romania introduced Social Aid, a last-resort social assistance pro- gram. During its first year of implementation, the central government's budget financed the program, while implementation was decentralized to local governments. The program covered about 10 percent of the population, a figure close to the estimated number of ex- treme poor. During 1996­2001, the responsibility for program financing was transferred to local governments, with no extra resources transferred from the central government budget. The program's caseload plummeted to 6 percent of the population in 1996 and 2 percent in 2001, with the highest reductions in number of beneficiaries occurring in the poorest municipalities (World Bank 2003h). Two key factors behind the program's col- 70 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS lapse were the inability of the poorest municipalities to pay eligible applicants and unclear legal provisions about the nature of the program. The central government altered and clarified program rules, but the finance problem and poor performance persisted until the program was recentralized in 2003. In Bosnia and Herzegovina, decentralized financing mechanisms have resulted in substantial interregional disparities in coverage, with poorer localities providing the fewest services. In the face of resource constraints, eligibility crite- ria for most benefits are ad hoc, as local welfare offices use discretion when rationing avail- able resources. These sorts of regional inequalities in social assistance can be particularly troubling when they are linked, as they so often are, to ethnic or other social divisions. Programs to protect those who have suffered shocks should be based on the largest possible risk pool, and should therefore be nationally financed. Thus most pension and unemployment insurance programs are centrally funded. The central government should also finance transfers related to natural disasters. Mexico's Natural Disaster Fund provides funding to local governments in areas hit by natural disasters to provide for temporary employment programs to help recipients replace lost income, help poor households re- build their housing, and assist local governments in reestablishing public services and infrastructure (Government of Mexico 2003). The weakness of subnational risk pooling is seen in an example from the United States. Until 1988, the federal government paid for unemployment benefits for workers unemployed as the result of a natural disaster; the states have since been made responsible for this function. To manage the implied fiscal risk, many states have adopted systems in which if the reserve fund falls below a certain threshold, benefit cuts and tax increases are automatic. Louisiana hit this threshold fol- lowing Hurricanes Katrina and Rita in 2005, when the single-state risk pool was too small given the dimensions of the disaster. Consequently, just when workers and businesses needed assistance, the system became less generous (Washington Post 2005). A number of observers see welfare migration as another reason why national govern- ments should play a central role in safety net finance and standard setting, though the case is not nearly as persuasive as for interregional equity and risk pooling. The economics literature and much political discourse in countries with federal social assistance systems are concerned with the idea that poor people will migrate from one region to another based on differences in the generosity of the regions' welfare programs. If this happens, or is even perceived as happening by policy makers and voters, it can lead each region to legislate a less generous program than it would otherwise have done. In the presence of welfare migration, the cost of an increase in a benefit would be not only the amount of the benefit increase multiplied by the number of current beneficiaries, but also the cost of the full amount of the benefit multiplied by the number of welfare migrants who would be attracted to the region. Similarly, a reduction in a benefit would both lower the cost per welfare recipient and reduce their number by inducing some of them to move away. Voters or legislators trying to balance costs with perceived satisfaction in providing for the poor will face a biased calculation. This issue is often referred to as "the race to the bottom," a catchy, though possibly exaggerated, term. The issue has been little studied in developing countries. For the United States, Brueckner (2000) reviews studies of the Aid to Families with Dependent Children pro- gram, the main decentralized federal welfare program prior to the 1996 reforms. The studies reviewed provide mixed evidence on whether or not the system induced significant 3. FINANCING OF AND SPENDING ON SAFETY NETS 71 welfare migration, although the author shows that states tended to set their policy taking the policy of neighboring states into account, which may indicate that policy makers were concerned about the issue. Harrison (2006) reviews the issue in Canada and finds little evidence of a race to the bottom. Dahlberg and Eadmark (2004) study asylum seekers in Sweden and conclude that they migrate from the locations to which they are initially as- signed to the three largest cities, which have more generous social assistance. Of course, these cities will also have more jobs and a more cosmopolitan atmosphere, which may be desirable to these immigrants. As concerns implications for other countries, Canada and the United States are thor- oughly integrated economically, and for individuals to move several times over the course of a lifetime is completely normal, both economically and culturally. Even in such countries, evidence on welfare migration is mild. The Swedish study is interesting, but considers a rather specialized subset of social assistance beneficiaries. We suspect that in many develop- ing countries, the economic and social costs of moving will be much greater than in Canada and the United States, and thus welfare migration is less of a concern. Three options are available for reducing the downward bias in program generosity that may result from concern about welfare migration: (1) make funding a fully national responsibility; (2) set national standards for program generosity; and (3) use a matching grant system of joint finance, with the national share being sufficiently high to offset the downward bias in local decisions about program generosity. LOCAL GOVERNMENTS AS IMPLEMENTERS OF SAFETY NETS The public finance literature notes that subnational jurisdictions may be well placed to administer safety nets. In particular, local government officials are expected to have greater knowledge of or contact with programs' client base and thus be able to reach the needy and exclude the non-needy. Local governments can also adjust policies to reflect local preferences, such as by serving locally preferred foods in school lunch programs. Further, decentralizing social assistance may make it easier for municipalities to establish links between social services and social assistance programs or achieve desirable moves toward service integration. This will be more important for some programs than others. For CCTs, this can be important if local governments are responsible for education and/ or health service delivery. For public works, the municipality or district may be the most sensible agency to determine what works should be carried out. At the same time, the extent of administrative decentralization is limited because of the desirability of national standard setting, the existence of economies of scale, and the likelihood of low capacity at local levels. PRINCIPLES AND MECHANISMS FOR SHARING FINANCIAL SUPPORT Good practices for managing nationally financed but locally implemented safety net pro- grams are in the process of being devised. Experience with the decentralization of social assistance has intensified in the last decade concurrent with trends in program manage- ment emphasizing monitoring for performance management and results-based financing and technology changes that have increased the ability to generate and analyze data on needs, clients, and service providers at a reasonable cost. 72 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS As table 3.5 shows, the various mechanisms for providing funds raised at the national level to local governments to run their safety net programs have specific advantages and disadvantages.10 In most countries with multilevel safety net programs, the financing is passed from the national level to subnational units using some form of earmarking. This ensures that the subnational units provide a minimum safety net, which is consistent with the reasons for financing safety nets at the national level. A common mechanism used by central governments to provide financing is open- ended capitation grants that cover all expenditures at the local level for the specified pro- gram. In one popular form of the model, the reimbursement formula is calculated as the TABLE 3.5 Advantages and Disadvantages of Financing Sources for Subnational Governments Financing source Advantages Disadvantages Assigning unconditional resources to subnational level for general use Unfunded · Conveys full discretion to the local · May result in insufficient spending mandate level Rights to raise · Provides a link between taxpayers · Increases the tax burden revenues and beneficiaries · May lead to duplication of taxes between locally (for · Gives incentive for fiscal the central and subnational governments example, from responsibility property taxes, · May yield small revenues surcharges, and · Local tax revenues often do not grow as user fees) quickly as income overall Borrowing · Finances investment projects to · May induce fiscal vulnerability, high debt increase future productivity service, and debt overhang and thereby contribute to macroeconomic instability Shared taxes · Provide a predictable source of · Local governments lack control over tax income with technical criteria and compliance and the tax base transparency of allocation · Yields are procyclical · Do not incur local tax administration costs Transferring earmarked resources to subnational level for safety nets Capitation · Ensure that funding is fair · Local governments have incentives for grants lenience in determining eligibility Block grants · Can be redistributive · No incentive to raise own revenues · Ensure vertical balance · Formula may be complex and lack transparency · Provide funds based on a formula Matching grants · Provide an incentive to cofinance · Limited by available funds programs Specific grants · Help redistribute assets among · Identification of programs may suffer regions from political capture · Avoid the need for counterpart · No incentive to raise own revenues funds · Subject to political discretion SOURCE: de Neubourg 2002; authors. 3. FINANCING OF AND SPENDING ON SAFETY NETS 73 average benefit per beneficiary, which is determined by the central government, plus a sum for administration. These schemes result in local providers having an incentive to maximize the number of social assistance recipients without regard to the fiscal costs of the benefits. The outcome will depend on the level of discretion allowed with respect to eligibility and on countervailing supervisory tools. The art is to balance these appropriately (table 3.6). TABLE 3.6 Options for Managing Local Governments' Incentives to Use Lax Entry Criteria for Centrally Financed Safety Net Programs Eligibility criteria Management tool Example Simple, little role for local · Little supervision required · Universal child allowances discretion Complex, such as a · Extensive quality control · U.S. Food Stamp Program means test or certification procedures · Brazil's Bolsa Familia program of disability · Local performance incentives Locally defined procedures · Hard budget constraint or cap · Indonesian JPS Scholarship on the number of beneficiaries and Grant Program assigned to each jurisdiction Any · Require local contribution, with · Romania Guaranteed Minimum higher contribution required where Income Program more discretion is given SOURCE: Authors. The simplest case with regard to eligibility determination is a universal categorical benefit program, such as a child benefit that is disbursed to the parents or guardians of all children in the country. The benefit is allocated based on a single, simple criterion that is easy to administer and monitor. Local providers will have an incentive to ensure that all residents in their jurisdiction benefit, but they cannot inflate the number of beneficiaries unduly, because the eligibility criteria are simple and the benefit is universal. In the case of a means-tested program, benefit allocation decisions are based on relatively complex criteria that require providers to make judgments, and they may have a tendency to be lenient with respect to these. Benefit providers do not have a financial incentive to limit program expenses, and may therefore allocate benefits in borderline cases to increase output and case numbers. Individual social workers usually find it more gratify- ing to help than to deny assistance to applicants, and local communities will welcome the extra infusion of cash. This is not a case of corruption, but of providers reacting rationally to the incentives designed into the program. Box 3.5 provides an example. When local governments have significant discretion in determining eligibility, the central government will need to closely monitor the processes or control the outcomes. It has three ways to do this: extensive quality control, incentive-based performance mecha- nisms, and budget caps. The U.S. Food Stamp Program involves a complex means test and is locally implemented but federally funded through an open-ended capitation grant. It is 74 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 3.5 Financing Arrangements and Incentives in the Netherlands The Netherlands' Alegemene Bijstandswet (National Assistance), the cash transfer of last resort, is provided through social service departments in local municipalities. Prior to 2001, 90 percent of the funding for the activities of these departments came from an open-ended capitation grant from the national government. The remaining 10 percent came from a block grant from the National Fund for Municipalities. A working group that assessed the system determined that the financial incentives given to the municipalities did not adequately encourage them to lower the number of beneficiaries either by reducing the number of new enrollees or by encouraging recipients to leave the safety net.This was because municipalities did not bear the costs of providing services and were not rewarded if costs were contained and if program effectiveness were improved. A new funding arrangement was initiated in January 2001 that reduced the national govern- ment's contribution to 75 percent, still in the form of an open-ended capitation grant; the remain- ing 25 percent was still provided in the form of a block grant. However, the budgeting rules were changed so that a municipality that spent less than the full amount of the block grant could use the remainder for other local policy initiatives. Early indications were that municipalities were more actively pursuing policies to help beneficiaries leave welfare rolls and find employment. The funding arrangements were changed again in 2004 to give municipalities even more in- centives to reduce participation. Municipalities now receive an annual budget divided into two parts, one for paying cash assistance benefits and one for providing labor activation services. If municipalities exceed the assistance budget, they can only request additional funds in ex- ceptional circumstances and only for amounts in excess of 110 percent of the original budget. If they underspend the budget, they may use the savings for any purpose. This has had three effects on how municipalities run programs: they devote their activation efforts to those who are most likely to find jobs easily, in general, the young unemployed; they use a stronger "work first" policy, whereby the unemployed are obliged to take any generally acceptable job rather than one that matches their experience or training; and they pay more attention to their gatekeeping role (van Berkel 2006). SOURCES: de Neubourg 2002 and van Berkel 2006. thus a candidate for excessive expenditures. The federal government manages this risk with a heavy set of quality control measures (box 3.6). Brazil's Bolsa Familia (Family Grant) program faces similar risks which it manages by using performance-based incentives. The central government monitors municipalities' performance using an index based on four elements: the quality of data collected for de- termining eligibility, the timeliness of recertification, children's compliance with the use of health services, and children's compliance with the use of education services. The munici- palities have an incentive to perform well, because the share of the program's administra- tive costs paid for by the federal government depends on the municipalities' performance scores. 3. FINANCING OF AND SPENDING ON SAFETY NETS 75 BOX 3.6 Quality Assurance of Eligibility Determination in the U.S. Food Stamp Program As the benefits provided by the U.S. Food Stamp Program are entirely federally funded but eli- gibility determination is handled at the local level, the federal government mandates a thorough system of quality control that it has developed to monitor and reduce errors in eligibility. Quality control measures the accuracy of states' eligibility decisions and benefit calculations. Each month, the states randomly select a specified number of cases from two sample frames. The first is a sample of all those who were provided benefits in a given month ("active cases"). The second is a sample of those who were denied benefits or whose benefits were terminated in a given month ("negative cases"). The states review random samples of a total of some 50,000 active cases and 30,000 negative cases each year. The federal government conducts a random re-review of about 30 percent of the cases to verify that the states' quality control review was conducted appropriately. The active case review is to determine whether households were eligible for benefits and whether they received the correct amount of food stamps during the month. State quality con- trol reviewers conduct detailed examinations of case files and in-depth field reviews, including interviews with adult members of each sample household and with others familiar with the households' circumstances. If a reviewer determines that a household received an incorrect al- lotment, the case is cited as a payment error. Reviewers calculate two types of payment errors: (1) overpayment errors, which include benefits issued to ineligible households and benefits paid to eligible households in excess of the appropriate benefit level (which varies with household size and income); and (2) underpayment errors, which measure errors in which eligible house- holds received fewer benefits than they were eligible to receive (they do not include the value of benefits that should have been paid to households that were denied or terminated from the program). These errors are then added (not netted) to yield the combined error rate, cited as a share of total benefits paid that month. Negative case reviews determine the share of households wrongly classified as ineligible or wrongly terminated in the total caseload.These reviews are less rigorous and usually consist of desk reviews of eligibility caseworker records. The Food Stamp Program attaches financial incentives to targeting accuracy. States are sub- ject to financial sanctions if their combined error rate (overpayment + underpayment) is higher than the national average. Conversely, they can receive enhanced administrative funding if their combined error rate is less than 6 percent and they do not have a high negative case error rate. This means that even if all states make progress in reducing their error rates, roughly half of all states can expect financial penalties if their rates are higher than the national average. In 2002, the average national combined error rate was 9.9 percent, and the states paid federal penalties of about US$46 million. SOURCE: Lindert 2005a. The problem of excessive enrollment numbers would become acute if local jurisdic- tions were allowed to determine their own criteria for poverty or eligibility and receive funding for everyone who met those criteria, as this would give them incentives to set cri- 76 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS teria that many people would meet. Thus, where local jurisdictions have full autonomy over entry criteria, the total number of permissible beneficiaries should be rationed so that jurisdictions face a hard budget constraint. This rationing should be done based on estimations of the poverty rate in each jurisdiction. This is the design used, for example, in the Indonesian JPS Scholarship and Grant Program that was put in place as part of the safety net response to the Asian financial crisis. Each school was told how many scholarships it could award, and a local committee determined which students would receive them. Another way of giving local jurisdictions the incentive to be strict in their decisions on eligibility is to require them to contribute some financing to the program. A contribu- tion of 10 to 20 percent is most common, but local contributions sometimes go as high as 50 percent. Extremely low local contributions provide correspondingly weak discipline, and so may be appropriate for programs with little local discretion or with budgetary caps. Higher contributions are appropriate when local governments have more discretion or an open-ended commitment from the central government. Matching grants are only appli- cable when local governments have independent spending and revenue authority and do not address the issue of interregional inequalities.11 In sharing financial responsibilities among levels of government irrespective of which financing option is used, it is important that programs receive adequate funding. In prac- tice, most of the programs discussed in this book are nationally financed; this is true, for example, of all the CCT programs and general food subsidies. In other cases, local governments cover a minor share of programs, often about 20 percent; examples include Ethiopia's Productive Safety Net Program and Romania's Guaranteed Minimum Income Program. Programs funded largely by localities are found primarily in large federal coun- tries; exemplifying such programs are the Maharashtra Employment Guarantee Scheme in India and cash transfer programs in several Brazilian states. Large-scale municipal pro- grams that operate in addition to national programs are also found in some megacities, for example, Mexico City's social pension. Funds should be assigned fairly and predictably; this can be ensured through the use of an appropriate formula to determine the level of transfers from the national to subnational levels. In the best cases, the formula takes population size, level of poverty, or sometimes a measure of tax capacity into account. Colombia's constitution, for example, stipulates substantial transfers to municipalities based on a formula that gives 60 percent weight to the number of poor people and 40 percent to population, fiscal and administra- tive efficiency, and progress in improving the quality of life (Ahmad and Baer 1997). In less optimal cases, the amount of central financing is not so clearly assigned. In China's ur- ban di bao (minimum living guarantee) cash transfer program, for example, the allocation of central budgetary resources to provinces is neither transparent nor stable, but is deter- mined in an ad hoc manner each year on the basis of a negotiated mixture of factors whose relative importance varies over time. The allocations from provinces to municipalities are similarly characterized by a lack of transparency and predictability. To date, the results are roughly in line with equity considerations: a number of better-off provinces received no central allocation in recent years and poorer provinces received much more--as much as 88 percent of financing in Ningxia and 100 percent in Tibet. However, the uncertainty makes planning by service delivery units difficult (World Bank 2007q). 3. FINANCING OF AND SPENDING ON SAFETY NETS 77 Transfers from national to subnational governments may be provided to support a wide set of local activities for which these governments are accountable, in which case they are unconditional. Alternatively, transfers may be tied to specific programs, in which case they are referred to as conditional or earmarked. Some commentators view conditional central grants as constraining local autonomy; others welcome them as a way to ensure a minimum provision of public goods, especially of safety nets. Note that the effects of grants on local spending decisions may not be a one-for-one increase in spending on the item nominally financed. Where subnational governments complement funding for centrally subsidized programs with local revenues, an increase or decrease in central funding may not be fully passed through to the target program, but tempered by changes in the allocation of locally raised funds. Theoretically, giving an unconditional grant to a region would allow some fraction of the income to be spent on current goods and services and some on other items such as investment and tax relief. In practice, governments tend to spend a large fraction of grants on goods and services. For example, many U.S. studies of the actual effect of various types of federal grants on state and local government spending suggest that nearly all grant funds are spent on public goods and services (Hines and Thaler 1995). This phenomenon is known as the "flypaper effect," a concept introduced by the economist Arthur Okun and captured by the phrase "money sticks where it hits." Numerous hypotheses have been put forward for this finding, including the explanation that politicians gain more politically from higher spending than from offering minor tax cuts to citizens. As a result, grants from the central government may have a greater positive effect on local spending than theory predicts. ALLOCATION OF ADMINISTRATIVE FUNCTIONS BETWEEN CENTRAL AND SUBNATIONAL GOVERNMENTS Even though funding may be largely central, actual delivery may be done through central agencies, through subnational entities, or through a combination of the two. Practices vary widely, and two of the most positively evaluated programs in recent years have taken opposite approaches. Mexico's Oportunidades CCT program is highly centralized, is well targeted, and has improved the use of health and education services. In Argentina's Tra- bajar workfare program, municipalities played a large role, the program was even better targeted, and it delivered useful infrastructure. Decisions about which level of government should be responsible for what function are often not "all or none." This section therefore discusses various factors to consider in determining, on a case-by-case basis, which level of government should be responsible for how much of which function. Economies of Scale The idea of economies of scale is that costs may not be fully proportional to the number of beneficiaries. Evaluation is a classic example, because the costs of designing an evalu- ation, writing data collection instruments, and analyzing datasets do not vary by the size of the dataset; moreover, through sampling, data collection costs increase far less than proportionately with an increase in the size of the population being studied. Thus, evaluating a program with 10 million beneficiaries costs little more than evaluating one with 10,000. 78 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS An initial rule of thumb is to keep functions that are subject to significant economies of scale at the national level. Table 3.7 shows how the World Bank (2006f) allocates func- tions for wage employment programs in India. The example presented is an analysis of the decentralized Village Full Employment Program, which is one of India's largest wage TABLE 3.7 An Example of a Functional Analysis of a Program and Responsible Levels of Government: The Village Full Employment Program, India Responsibility Cen- Dis- Village Village Function Activity tral State trict Block council meeting Implementation rules X X Policy Targeting X X design, standards Budgeting X X Standards X X Activity prioritization, action plan X X Planning Activity selection X X Human capital Skill development X X Asset Social capital creation Information dissemination X Physical capital Public works X Beneficiary selection Identification of beneficiaries X X Awareness raising X X X Recurring activities Provision of wages, food grains X Supervision & quality control X Operations Personnel Hiring and firing X Maintenance Accounting and financial X management Repairs X X Assets Recording of assets X X X Physical verification of assets X X Monitoring created and evaluation Audits Financial audits X Social audits X X SOURCE: World Bank 2006e. 3. FINANCING OF AND SPENDING ON SAFETY NETS 79 employment programs. The table makes explicit what is often obscured in theoretical writ- ing on decentralization: (1) multiple tiers of government may be involved, not just two; and (2) any analysis of how functions should be or are being performed needs to include all pertinent levels. Functions should be broken down into specific actions and then allocated to a spe- cific level of government. Consider eligibility determination, which is typically used as an example of a function that can be performed locally and is the most commonly given rea- son for having local governments administer safety nets. Determining eligibility requires person-to-person contact for each beneficiary; thus, a large part of the cost is proportional to the size of the program. Some aspects of the development of household targeting sys- tems are subject to economies of scale, however, such as developing the formulas and sup- porting software or running cross-checks against national data registries to verify informa- tion. Thus, some aspects of eligibility determination may be carried out more effectively at the national level and some more effectively at the subnational level (table 3.8). TABLE 3.8 Advantages of Alternative Allocations of Institutional Responsibilities for Household Targeting Systems Responsibility Decentralized Centralized Design · More involvement of local · More transparent with federal guidelines for authorities in social policy eligibility criteria, other design factors · System can reflect · Common framework for monitoring and evaluation local preferences and · Common software facilitates national database circumstances · Standard questionnaires more efficient and transparent · Less costly (economies of scale) Data · Empowerment of local · Better quality control, consistency of data collection collection authorities practices · Can be more efficient · Lower risk of manipulation by local authorities · Interviewers familiar · Better when local capacities are limited with local cultures and languages · Infrastructure network with local offices more likely to be in place Database · Databases can be tailored · Facilitates assignment of single identification management for use with other local number programs · Better quality control and auditing of databases · Facilitates building a national, consolidated database · Lower costs (economies of scale) · Facilitates cross-checks with other automated systems · Lower risk of corruption at local levels · Better when local capacities are limited SOURCE: Castañeda and others 2005. 80 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Decentralizing to the point of losing economies of scale is possible. In the United States, for example, administration of the TANF program is almost wholly decentralized to states, and often to counties within the states. Because the supporting information systems and payment contracts are designed and purchased at the state rather than the national level, the program's administrative costs are higher and it suffers from substantive information deficiencies: it cannot easily verify that households are not receiving benefits in multiple states or enforce the five-year time limit for participating in the program. The much more centralized Food Stamp Program has fewer such problems and greater econo- mies of scale in payment systems. Administrative Capacity Local governments may have limited administrative capacity, either generally or with regard to a particular program. Examples of general administrative capacity are the presence of electricity and the full complement of office equipment, such as telephones, photocopiers, facsimile machines, and computers. Program-specific capacity refers to the availability of sufficient staff who are adequately trained with respect to their roles. Both of these may be built up over time if the decision is taken to decentralize an existing program or to create a new one using a decentralized structure. The usual teething problems will arise while capacity is developed to match responsibility. This process can be aided by the provision of adequate funding and training and proactive management, but may still be a multiyear effort. Sometimes, however, a program may face constraints that are not just transitional, but structural. In Mexico, for example, municipal civil service rules are weak and mayors are allowed to serve only a single term. Because each election is followed by a high turnover of municipal staff, the ability to train or build local capacity is limited--which in turn limits the tasks that can be assigned to municipal staff. Clarity and Consistency However responsibilities are assigned, they must be clearly specified and understood by all pertinent parties, otherwise some tasks will not be done or will be duplicated in wasteful and often contradictory ways. Roles should also be consistent with funding mechanisms and capacities. Diagnoses of poorly performing programs are replete with examples of lack of clarity as to who does what and the ensuing problems. Table 3.7 illustrates an analysis done in an effort to sort out issues of clear and inconsistent allocation of respon- sibilities across actors. Management of Heterogeneity in Program Administration Whenever a program is carried out in many different places, its implementation will vary. Managing this heterogeneity in program administration is inherent in all large programs. The issue is especially salient for decentralized programs, as heterogeneity is likely to be greater when the acting units are different municipalities with their particular preferences and capacities rather than local offices of a national agency. There are essentially two facets to managing heterogeneity. The first is to ensure a minimum standard of service delivery. This can be accomplished either through imposing 3. FINANCING OF AND SPENDING ON SAFETY NETS 81 strict rules and monitoring compliance with them (as in the U.S. Food Stamp Program), or by establishing incentives for programs to meet those standards (as with Brazil's Bolsa Familia municipal performance index and central payment of administrative costs based on the index). The other aspect of managing heterogeneity is to learn from the variability by ascer- taining who the good performers are and what makes what them do work better than av- erage. The United States accomplishes this by having a large evaluation industry involved in its decentralized programs. Brazil's Bolsa Familia program has instituted a competition for innovations or good practices whereby municipalities submit ideas that are then shared with other municipalities. The United Kingdom's Job Centre Plus program has a small team of roving staff members who both help poor performers troubleshoot their problems and identify high performers and share their approaches. A MULTIFACETED APPROACH TO MANAGING A SAFETY NET IN A DECENTRALIZED SETTING In closing, we present a thumbnail sketch of how Brazil manages the issues pertaining to the involvement of multiple layers of government in Bolsa Familia to show how multiple tools can be used simultaneously to address the various challenges.12 Bolsa Familia is a CCT program, and so is relatively complex. In addition, Brazil is a federal country with 5,564 autonomous municipalities with a great deal of heterogeneity in capacity levels among them. Moreover, a number of municipalities also operate their own cash transfer programs. The basic assignment of roles for the Bolsa Familia program is as follows. Eligibility is determined through self-reported incomes which are verified through a combination of multidimensional proxy indicators and internal and external cross-checks. Municipalities are responsible for data collection for registering potential beneficiaries; Caixa Econômica Federal (a federal savings and credit organization) consolidates the national registry data- base; and the Ministry of Social Development makes final eligibility decisions. Payments are channeled through the banking system under a contract with Caixa. Municipalities play the lead role in registering potential beneficiaries, monitoring compliance with health and education conditions, and establishing social oversight councils. Brazil's three supreme audit agencies provide additional oversight. Given the level of complexity of the program and actors, extra thought and well- developed tools are needed to manage coordination and address principal-agent problems. The main approaches used are summarized in table 3.9. Different countries might choose a different set of tools, but the idea of thinking critically about what is needed and formulating a comprehensive approach to managing the challenges of decentralization is transferable. This particular set of tools seems to be balanced and appropriate for Brazil's context. The quality of service delivery seems to have improved since the introduction of the index of decentralized management, although definitive studies have not yet been done (Lindert and others 2007). Challenges remain, especially with respect to aspects of service delivery not covered by the index, and hetero- geneity will never be eradicated in such a large and diverse country. 82 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS TABLE 3.9 Management Solutions to Implementation Challenges in a Decentralized Context, Bolsa Familia Program, Brazil Challenge Solution Principal-agent dilemma · Requiring municipalities to sign joint management agreements with inherent in executing the Ministry of Social Development (which specify the roles and federal programs responsibilities of each agency involved and establish minimum via autonomous service standards) before receiving subsidies to cover the program's municipalities administrative costs · Ensuring that oversight and control audits include the activities undertaken by municipalities Heterogeneity in the · Assessing the quality of implementation by municipalities via a quality of municipal quantifiable index of decentralized management based on four key implementation resulting aspects of quality from capacity differences · Providing performance-based financial incentives (administrative cost subsidies) based on municipalities' scores on the index · Targeting training and capacity building to municipalities with low scores on the index Principal-agent dilemma · Using a performance-based contract for Caixa with enforceable caused by contracting sanctions for inadequate quality standards out payments and the · Ensuring that oversight and control audits also cover Caixa activities registry database Potential duplication with · Providing for vertical integration of subnational programs with Bolsa subnational CCTs Familia via joint cooperation agreements Need for mechanisms · Introducing the Bolsa Familia Innovations Award in 2006, which is to promote the sharing intended to promote the sharing of municipal experiences and includes of experiences and field visits as part of the awards process innovations across · Publishing descriptions of innovative experiences in the form of case municipalities studies SOURCE: Lindert and others 2007. Notes 1. A few exceptions to the general lack of information are Bose, Holman, and Neanidis (2004) and Coady and Harris (2004). Chu and Gupta (1998b) report the example of safety net imple- mentation accompanied by tax reform to finance it. 2. Welfare economics formalizes a moral philosophy that values a currency unit transferred to the poor more than a currency unit transferred to the middle class and substantially more than a currency unit transferred to the rich. The benefits of programs that improve equity via redistri- bution are estimated empirically by assigning different distributional weights to currency units received by households with different welfare levels. Essentially, the weights represent a way to translate a value judgment (how much a society values equity, and thus redistribution, as a means to achieve equity) into mathematical terms. However, there is no clear-cut empirical way to estimate the set of distributional weights of a society. Absent this, some researchers as- sume a certain function form (Squire and van der Tak 1975) and run sensitivity analyses with different values for the propensity for redistribution. 3. FINANCING OF AND SPENDING ON SAFETY NETS 83 3. Technically, a fifth option is to create money to cover deficits. However, experiences like those of Latin America in the 1970s and early 1980s have shown printing money to be such an in- flationary, inefficient, and inequitable source of financing that all prudent governments have discarded it as an option. 4. The Monterrey Accords were agreed on at the International Conference for Financing and De- velopment held in Monterrey, Mexico, in March 2002; the Millennium Development Goals were agreed on at the United Nations Millennium Development Summit held in New York in September 2000; the Gleneagles Summit of the Group of Eight was held in Gleneagles, Scot- land, in 2005; and the African Action Plan was agreed on at the Kananaskis, Canada, Group of Eight Summit in 2002. 5. Lindert (2004) suggests a number of explanations for this "free-lunch puzzle," concluding that countries that spend more have more pro-growth tax packages; welfare states have minimized the work disincentives of young adults; early retirement subsidies have little effect on GDP, in part because they are skewed toward less productive workers; unemployment programs raise unemployment, but raise productivity among those employed, and so have little effect on GDP; and many social transfer programs raise GDP per person even after accounting for the effects of taxes to support the spending. 6. The World Values Survey is a worldwide investigation of sociocultural and political change conducted by a nonprofit association funded by scientific foundations from around the world. Interviews have been carried out with nationally representative samples of the pub- lics of more than 80 societies on all six inhabited continents. For more information, see www.worldvaluessurvey.org/. Latinobarómetro is an annual public opinion survey conducted by Latinobarómetro Corporation, a nonprofit nongovernmental organization based in Santia- go, Chile, that involves some 19,000 interviews in 18 Latin American countries, representing more than 400 million inhabitants. For more information, see www.latinobarometro.org/. 7. "Social assistance and welfare" includes transfer payments (including in kind) to compensate recipients for reduction or loss of income or for inadequate earning capacity; sickness, mater- nity, disability, old-age, and survivors' benefits; government employee pension schemes; unem- ployment compensation; family and child allowances; other social assistance for individuals; and payments to residential institutions for children and the elderly. "Transfers to households and other organizations" includes transfer payments to private social institutions such as hos- pitals and schools, learned societies, associations, and sports clubs that are not operated as enterprises and current payments in cash to households that add to their disposable income without any simultaneous, equivalent counterpart provided in exchange by the beneficiary and that does not generate or eliminate a financial claim, and is usually intended to cover charges incurred by households because of certain risks or needs. 8. For the analysis and discussion in this section, the dataset used excludes Iraq, an outlier that has, because of its unique circumstances, been spending 15 percent of GDP on social assis- tance. 9. Country expenditures are rounded to the nearest half percent for this calculation. 10. This section draws on de Neubourg (2002). 11. Matching grants can actually exacerbate horizontal differences. Consider a 90-10 matching scheme. The local government that can put up 10 units of finance will end up with a total bud- get for the program of 100. The local government that can afford 5 units of finance will end up with a total budget of 50. Thus an initial budget difference of 5 becomes a final difference of 50 in absolute terms, even though in relative terms the difference is still 2 to 1. 12. This section is drawn from Lindert and others (2007). CHAPTER 4 Enrolling the Client: Targeting, Eligibility, and Intake KEY MESSAGES Concentrating resources on the poor or vulnerable can increase the benefits that they can achieve within a given budget or can achieve a given impact at the lowest cost. The theoretical gain from targeting can appear to be large. For example, if all the benefits provided by a transfer program were targeted to the poorest quintile of the population rather than uniformly distributed across the whole population, the budget savings or the difference in impact for a fixed budget would be five to one. In practice, the full theoretical gain is not realized, because targeting is never completely accurate, and because costs are associated with targeting. These costs include administrative costs borne by the program, transaction and social costs borne by program applicants, incentive costs that may affect the overall benefit to society, and political costs that may affect support for the program. The size of targeting errors and costs will differ according to the setting and the types of targeting methods used and must be assessed carefully in any policy proposal. Good evidence indicates that, for the most part, programs can focus resources on the poor to a moderate or high degree without incurring unacceptably high errors of exclusion and administrative, private costs, and/or incentive costs, although not all do so. Factor- ing in judgments on social and political costs is harder, partly because their metrics are so different, and partly because discussions about them are often more polemical than quantitative, but the widespread and increasing interest in targeting from policy makers suggests that these costs are not preclusive. A few methods of targeting and types of programs go hand-in-hand, for example, self- selection and commodity price subsidies. However, several different methods can often be used for a particular type of program; for instance, cash and food transfers can be targeted by means tests, proxy means tests, nutritional status or risk factors, geographic area, demographic characteristic, or self-selection. For a single program to use a number of methods is common and usually yields better targeting than a single method. Means tests and proxy means tests have the highest costs, but tend to produce the lowest er- rors of inclusion and are often good investments. Self-selection via a low wage rate and geographic targeting are also powerful and proven targeting tools. The details and quality of implementation will have a significant effect on targeting out- comes. Programs need extensive outreach to keep errors of exclusion low. Reducing errors of inclusion requires a definition of eligibility that sorts the poor from the nonpoor well and can be implemented at a tolerable cost. Targeting systems should be dynamic, allowing new or newly poor households to access the program and moving out house- holds that are no longer eligible. The inputs to good targeting outcomes include adequate staffing; well-defined rules of the game; clearly assigned and sensible institutional roles; 85 86 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS and adequate information systems, material inputs, monitoring, and evaluation. Systems also need time and effort to develop. A good household targeting system may be complex to develop, but can be used for many programs, not only for direct transfers in cash or in kind, but for entry into programs that pro- vide free or subsidized health care, schooling, training, housing, utilities, and the like. The shared overhead is not only efficient, but can lead to a more coherent overall social policy. F unds for safety net programs are scarce, and competing demands on the public bud- get are many. Thus policy makers always face pressure to use funds as effectively as possible and usually have to make significant trade-offs. Should a program serve many people but limit the level or the duration of benefits? Alternatively, should a program be more selective but more generous to those it does serve? The answers to those questions are part of the larger diagnostic process described in chapter 9 and have links to targeting (discussed in this chapter), benefit levels (chapter 5, section 1), and repercussions in rela- tion to the choice of programs (chapters 7 and 8). 4.1 Basic Concepts of Targeting Targeting is a tool policy makers sometimes employ to make a program efficient. It is not an end in itself. How much to target and how to do so depend on program-specific answers as to whether the gains outweigh the costs. POSSIBLE GAINS FROM TARGETING Targeting can increase the benefits that the poor can realize with a given budget (maximiz- ing impact) or can achieve a given impact at least budgetary cost (minimizing costs). This is accomplished by channeling resources to a target group, typically the poor or a subset of the poor. Targeting is an attractive option for many kinds of poverty reduction programs and expenditures. It is particularly important for safety nets, because in contrast to, say, educa- tion, the transfers confer a benefit that is largely a private good for the recipient household and because there is no natural limit to the amount a household might like to receive. The basic case for targeting is simple. Consider a country with 100 million people of whom 20 million are poor. The country's budget for a transfer program is US$200 mil- lion. With no targeting, the program could give everyone US$2. If the program could be targeted only to the poor, it could give each poor person US$10 and spend the full budget, thereby maximizing the impact using the given budget. Alternatively, it could give each poor person US$2 for a budget of US$40 million, thereby minimizing costs for a given impact. The theoretical gain from targeting can appear to be large. In practice it is less than indicated here, because targeting is never completely accurate and because it has costs, including administrative, private, incentive or indirect, social, and political costs. COSTS OF TARGETING Accurately distinguishing between who is and who is not needy incurs costs. The differ- ent types of costs are defined here, while evidence on their magnitudes is reviewed in the next section. 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 87 · Administrative costs are the costs to the program of gathering information to help make the decision about who should be admitted to the program. · Private costs are the costs to an applicant of applying for a program, including the time or cash costs of gathering the necessary information, traveling to the registration site and lining up for registration, complying with any preconditions, and so on. Private costs always reduce a program's net benefit to the recipient. If they are sufficiently large, they may discourage eligible people from participating altogether. · Incentive (or indirect) costs arise when eligibility criteria induce households to change their behavior in an attempt to become beneficiaries. They can be nega- tive, for example, when a program open only to those below a minimum income causes some households to work less so that they fall below that minimum in- come. Sometimes they can be positive, for example, a food ration or daily meal supplied only to those children who attend school may encourage some families to enroll more of their children or ensure that they attend school every day. · Social costs may arise when participation in a program carries with it some sort of stigma. Stigmatization may affect households' decisions about participating. For those that do participate, stigmatization may lower households' psychological welfare, if not their incomes. · Political costs can arise if the degree of targeting negatively affects the program's budget. TARGETING ERRORS In practice, program officials do not have perfect information about who is poor, because collecting such information is time consuming and costly. When program eligibility is based on imperfect information, program officials or the targeting rules they use may mistakenly identify nonpoor people as poor, and therefore admit them to the program (re- ferred to as an error of inclusion), or do the opposite, that is, mistakenly identify poor peo- ple as nonpoor, and thus deny them access to the program (referred to as an error of exclu- sion). Consider the matrix in table 4.1. Of 100 households, 20 are classified as poor (eligible) based on the poverty line (eligibilitythreshold).Now TABLE 4.1 Errors of Inclusion and Exclusion consider a program that gives benefits to 20 house- Welfare status holds selected according to Households Poor Nonpoor Total imperfect targeting crite- ria. Of these, 15 are poor Included in 15 5 20 (have incomes below the program successful inclusion error targeting poverty line) and 5 are nonpoor (have incomes Excluded from 5 75 80 program exclusion error successful above the poverty line). targeting Both the 15 poor house- holds included in the pro- Total 20 80 100 gram and the 75 nonpoor SOURCE: Authors. 88 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS households excluded are successful targeting. The 5 poor households excluded are errors of exclusion, while the 5 nonpoor households are errors of inclusion. An easy and preferable extension of this approach is to present similar information over the entire welfare distribution, showing by means of graphs or tables how many people fall in each decile or quintile and how many in each receive benefits.1 This shows whether funds are going to the very poor or to the moderately poor, if leakage is to the near poor or to the wealthy, and so on. It also permits differentiating the size of the benefit received. Presenting information on the entire distribution facilitates comparative work and bench- marking, as different programs have different eligibility thresholds. Figures 4.1 and 4.2 illustrate the presentation of information across the full welfare distribution and how pro- grams can be compared to one another. FIGURE 4.1 Share of Population by Quintile That Received A further extension Benefits in Selected Safety Net Programs, India, Fiscal 2004/05 of the basic framework is to take differentiated benefits Percent into account. Most pro- 14 School stipends grams still offer a uniform 12 Annapurna Scheme (non- transfer to participants, 10 contributory pension 8 but a sizable number now program) 6 tailor the benefit to the de- SGRY & National Food 4 for Work Program gree of need or to the size 2 Disability pension or structure of the recipi- 0 1 2 3 4 5 ent household (for more (poorest) (richest) details, see chapter 5, sec- Quintile tion 1). Customized ben- SOURCE: Based on data from World Bank 2007a. efits somewhat complicate NOTE: SGRY = Sampoorna Grameen Rozgar Yojana (Village Full Employment the ex post evaluation of Program). targeting performance, but not intractably so. FIGURE 4.2 Share of Benefits Accruing to Each Quintile for In relation to the Selected Safety Net Programs, India, Fiscal 2004/05 measurement of target- ing errors, information on Percent targeting is usually shown 50 as the percentage of pro- School stipends 40 gram benefits reaching Annapurna Scheme (non- the poor or a given decile 30 contributory pension program) or group of deciles. This 20 SGRY & National Food information comes from for Work Program a household survey that 10 Disability pension has some sort of welfare 0 measure, some measure of 1 2 3 4 5 whether the household or (poorest) (richest) Quintile its members benefit from a program, and ideally an SOURCE: Based on data from World Bank 2007a. indication of the level of NOTE: SGRY = Sampoorna Grameen Rozgar Yojana (Village Full Employment Program). 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 89 the benefit (for methodological information see chapter 6, section 5). The welfare mea- sure used may not, however, correspond closely to the welfare measure used to define the program's eligibility criteria or the definition of the household in the survey to the defini- tion of the assistance unit for the program. Thus while the usual survey-based information will show whether the funds are going to the poorer groups in society, it will not reveal whether the program's criteria are being followed well. For that, an internal quality control procedure will be needed as described in box 3.6. If the survey uses the same definitions of welfare and the social assistance unit used for the program's eligibility criteria, a model- ing exercise can be undertaken to establish who would be eligible to examine whether the rules, if properly followed, are well designed. 4.2 Results of Targeting Good evidence indicates that, on balance, programs can focus resources on the poor to a moderate degree without incurring unacceptably high rates of exclusion or of administrative, private, and/or incentive costs, although not all do so. Factoring in judgments on social and political costs is harder, partly because their metrics are so different, and partly because dis- cussions about them are often more polemical than quantitative, but the widespread and in- creasing interest in targeting from policy makers suggests that these costs are not preclusive. ERRORS OF INCLUSION Coady, Grosh, and Hoddinott (2004) summarize targeting outcomes from 122 targeted social assistance programs in 48 countries. The median program in their sample provides a quarter more resources to the poor than would random allocations. The best programs were able to concentrate a high level of resources on poor individuals and households. Ar- gentina's Trabajar workfare program, the best program in this regard, was able to transfer 80 percent of program benefits to the poorest quintile, or four times the share they would have received in a random allocation. The 10 programs with the best incidence delivered two to four times the share of benefits to the poor that they would have got with random allocations. Progressive allocations were possible in all country settings, in countries at markedly different income levels, and in most types of programs. A more recent and selective comparison of evidence (figure 4.3) confirms that in middle-income countries at least, programs can, but are not guaranteed to, have errors of inclusion as low as those found in some programs in countries such as the United States. Box 4.1 discusses targeting goals in much poorer countries. Targeting does not always work. The state-of-the-art as practiced around the world is highly variable. According to Coady, Grosh, and Hoddinott (2004), while median performance was good, in a quarter of the cases targeting was regressive, and thus a ran- dom allocation of resources would have provided a greater share of benefits to the poor. For every method of targeting considered except targeting based on a work requirement, the sample of programs included at least one example of a regressive program. Given that good incidence is achievable in many settings and that poor performance is also found in many, where poor performance is found, it may not be inherent, but can be improved upon. 90 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS FIGURE 4.3 Errors of Inclusion, Selected Programs and ERRORS OF EXCLUSION Countries Participation rates as re- ported in household sur- United States veys are the most com- Food Stamp Program Temporary Assistance for mon source of informa- Needy Families tion on errors of exclu- Latin America & the Caribbean Brazil sion; however, they only Chile report the outcome, not Jamaica its cause, and errors of Mexico exclusion have multiple Argentina sources. Where social as- Europe and Central Asia sistance programs are not Romania fully funded, as is the Bulgaria case for the vast majority Lithuania of programs, errors of ex- Hungary Estonia clusion will occur because Poland of the caps put on enroll- Moldova ment to keep programs Kyrgyzstan Republic within budget allocations. Albania For example, Argentina's Belarus Trabajar program cov- Serbia ered only 7.5 percent of Armenia the unemployed. Even Russian Federation though it had the lowest Georgia errors of inclusion in the Bosnia & Herzegovina Uzbekistan global Coady, Grosh, and Macedonia, FYR Hoddinott (2004) review, Azerbaijan it had high errors of ex- Tajikistan clusion. 0 20 40 60 80 100 Errors of exclusion Percentage of beneficiaries include the influence of in poorest quintile self-targeting. This in- SOURCES: United States: Lindert 2005a; Latin America and the Caribbean: cludes both the desirable Lindert, Skoufias and Shapiro 2006; Europe and Central Asia: Tesliuc and oth- ers forthcoming. aspect of the better-off NOTE: For Europe and Central Asia, the programs referenced are each coun- choosing not to partici- try's means-tested or proxy means-tested social assistance of last resort. The pate and the less desirable programs in Latin America and the Caribbean refer to each country's condi- tional cash transfer program, except for Argentina, which references the Jefes aspect of the needy being de Hogar workfare program. discouraged as well. Uz- bekistan's social assistance program uses community-based targeting via traditional groups of neighborhood elders. Micklewright, Coudouel, and Marnie (2004) break down the final outcomes into dif- ferent factors: households' awareness of the program; of those households that are aware of the program, the decision on whether or not to apply for a benefit; and for those that apply for the benefit, the percentage that are awarded the benefit and the amount of that benefit. The scheme fared quite well with respect to knowledge, with 85 percent of the 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 91 BOX 4.1 How Narrow Should Targeting Be? The question often arises as to how narrow targeting should be, especially when poverty rates are high. Is ensuring that benefits go primarily to the large number of poor satisfactory, or should benefits go to the poorest subgroup, say the bottom 10 percent of the distribution? Trying to concentrate benefits as far down the distribution as possible, even when overall poverty is high, is advisable for two reasons. The first reason is cost. Especially where many people are poor, the budget derived from taxes will be limited, because the pool of people who can pay taxes will be small. International assistance may provide a good deal of the social assistance budget in such cases, but it is still always insufficient to provide for all the basic priorities in a country. Thus safety net budgets tend to be low relative to needs, and the more so the poorer the country. The second reason to try to concentrate benefits as far down the distribution as possible is that most moral philosophy values the welfare of the poorest more than the welfare of the less poor. We illustrate the difference in welfare among the poor empirically with an example from Burkina Faso.When looking at the full distribution of welfare as shown in figure a, poverty is high and the distribution looks fairly flat, but when we zoom in on the poor end of the spectrum as shown in figure b, we see that the bottom 10 percent of the distribution has consumption levels of half or less than those who are just poor. Thus this group is much needier and will benefit much more from a transfer. a. All households b. Poor households Annual per capita consumption Annual per capita consumption (thousands of CFA francs) (thousands of CFA francs) 125 2,000 100 1,500 75 1,000 50 500 25 0 0 0 10 20 30 46 60 70 80 90 100 0 10 20 30 40 46 Household welfare rank (percent) Household welfare rank (percent) SOURCE: Based on data from the 3rd Priority Survey, National Institute of Statistics and Economic Studies, Ouaga- dougou, Burkina Faso. NOTE: The dotted vertical line is the poverty line. The richest 0.5 percent of households were omitted. Targeting the poorest households and not all the poor is conceptually appropriate when poverty rates are high. However, poorer countries may face capacity constraints that make setting up systems to target narrowly relatively difficult. This practical issue, rather than the conceptual one, may be important in determining the targeting choices made. poorest quintile being aware of the program. Claims were self-targeted, with those in the poorest quintile applying for a benefit more than twice as often as those in the top three quintiles. Nevertheless, only half the households in the poorest quintile that knew about the program applied for it. 92 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Errors of exclusion are often measured only approximately, as many household sur- veys do not contain sufficiently detailed information to calculate exactly whether house- holds meet program eligibility criteria. What is most commonly reported is participation by members of the poorest quintile of the population, but many of these may not have met any criteria for the program not related to poverty. A study of housing allowances in the Russian Federation (Struyk, Petrova, and Lykova 2006) makes this distinction and finds that when coverage was estimated as a percentage of eligible households rather than of all households, coverage rates increased by 40 percent or more. For example, par- ticipation rates in the city of Omsk were 5.3 percent overall, but 9.4 percent of eligible households. Even given a measure of participation rates, what their causes are and what should be done about the rates is unclear. It is perhaps for these reasons that errors of exclusion are not reported as often as errors of inclusion. Two recent studies in Latin America (Lindert, Skoufias, and Shapiro 2006) and in Eastern Europe and Central Asia (Tesliuc and others forthcoming) report coverage rates for the lowest quintile. In the Latin American study of 40 targeted programs, the mean coverage rate of the poorest quintile is 19 percent. It is higher but far from complete even for the large, well-known programs that anchor countries' social assistance strategy--for example, 32 percent for Mexico's PROGRESA (now known as Oportunidades) initiative. In the second study, the mean coverage rate for the poorest quintile is 42 percent. A study that reviewed experience in a small number of countries in the Organisation for Economic Co-operation and Development (OECD) was able to model take-up by eligible individuals more closely and finds that take-up rates are typically between 40 and 80 percent for social assistance and housing programs (Her- nanz, Malherbet, and Pellizzari 2004). While few studies can disentangle the root causes of low participation robustly, we believe that the main and largest source of errors of exclusion is budgets that are insuf- ficient to serve all those meant to be served. The second largest cause may be insufficient policy attention paid to outreach and administrative budgets too small to permit sufficient outreach efforts. The third, and possibly smallest, cause of errors of exclusion may be in- correct classification of eligible applicants as ineligible by program administrators.2 ADMINISTRATIVE COSTS Data on the administrative costs of targeting are scarce and their interpretation is prob- lematic. Measurement issues abound, because staff and systems are usually shared among the various functions of a single program, and often among several programs. For ex- ample, a social worker might determine eligibility, provide case counseling advice, make payments, and handle appeals issues for one or more programs. Thus determining the administrative costs of targeting requires detailed budget data and often a lot of imputa- tions, assumptions, or special data collection exercises to decide how to allocate the costs of systems and staff who work on more than one function and/or program. Moreover, what is a cost of targeting is not well defined. Time spent verifying reported levels of in- come is clearly a targeting cost, but registration procedures and databases of participants will be needed even for universal programs. Intake interviews and supporting databases are more complicated and costly for programs with complex targeting criteria, but not all of the costs are due to targeting. 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 93 Care must be taken when interpreting administrative costs. A large program that puts little effort into targeting will have lower administrative costs as a share of total pro- gram expenditures than a more narrowly targeted, and therefore smaller, program, because the administrative effort devoted to targeting is low and the total program expenditures are high. That does not imply that the investment in targeting was not worthwhile for the overall efficiency of the program. Consider figure 4.4, which illustrates the cost structure for a universal child allowance and for a targeted child allowance. For the targeted program, the ratio of total administrative costs to total costs (BDEG/ADEH) is higher than for the universal program (BCIJ/ ACIK), but the extra FIGURE 4.4 Conceptualizing Administrative Costs administrative costs of targeting (CDEF) are D E Targeting costs small in relation to the C F I Other admin costs B G J savings in benefits paid to the nonpoor children (HGJK). This simple il- Poor Benefits All children children lustration probably un- derestimates the value A H K that extra administrative costs can have on targeting SOURCE: Authors. outcomes. First, it does not capture the value that extra effort of the appropriate sort can have in lowering errors of exclusion. Second, this il- lustration is for child allowances, a program that is often universal rather than targeted. The desire to limit benefits to the poorest is often stronger for other programs. For those programs for which they have been measured, the administrative costs of targeting are usually reasonably low. Tables 4.2 and 4.3 provide data from the handful of recent studies that try to document these for some well-established means-tested and proxy means-tested programs. In table 4.2, specific programs are the unit of observation. Targeting costs average about 4 percent of total program costs, range from about 25 to 75 percent of total administrative costs, and in absolute terms cost US$8 or less per ben- eficiary in all but one case. The share of total costs is slighter higher than Grosh (1994) indicates for some older and less well-targeted Latin American programs that use similar targeting mechanisms, but is still hardly prohibitive. In table 4.3, household targeting systems used for multiple programs in five Latin American countries are the unit of ob- servation. Costs per interview cluster in the range of about US$3 to US$8 per beneficiary and the share of targeting costs in benefits transferred is less than 1.5 percent. The costs cited are for the most complex mechanisms: means and proxy means tests. Costs for other targeting mechanisms are almost by definition substantially lower, but are usually not well measured. In thinking about whether household targeting systems such as those in table 4.3 would be affordable in a new setting, three factors should be taken into account. First, some adjustments in pricing may be sensible, as labor usually accounts for the largest share of costs and its price is quite variable. To solve this problem in extrapolating lessons from U.S. eligibility processes for Latin America, Castañeda and Lindert (2005) approximate 94 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS TABLE 4.2 Administrative Costs of Targeting for Selected Means-Tested and Proxy Means-Tested Programs, Various Years Targeting costs as share of total... Administrative Program US$/ Country, program, and year costs costs beneficiary Albania: Ndihme Ekonomika, 2004 88 6.3 7 Armenia: Family Poverty Benefits Program, 2005 26 0.6 3 Bulgaria: Guaranteed Minimum Income Program, 2004 64 6.3 7 Kyrgyz Republic: Unified Monthly Benefit Program, 2005 24 2.3 1 Lithuania: Social Benefit Program, 2004 41 2.7 8 Romania: Guaranteed Minimum Income Program, 2005 71 5.5 25 Colombia: Familias en Acción, 2004 34 3.6 -- Mexico: PROGRESA, 1997­2000 40 2.4 -- SOURCES: Colombia: Lindert, Skoufias, and Shapiro 2006; Mexico: Caldés, Coady, and Maluccio 2004; other countries: authors' calculations. NOTE: -- = not available. Targeting costs include those related to outreach to beneficiaries, determination of eligibility, home visits, verification of information, and maintenance of databases. the number of hours of staff time required, valued at US$86 in the United States, and then reprice them with salary levels more typical of middle-income Latin American countries, thereby deriving a value of US$25. Second, looking at the size of the transfer to be targeted is important. Administra- tive costs accrue in absolute terms, not as a proportion of benefits. For example, spending US$10 per beneficiary in targeting costs to target a benefit of US$10 would be inefficient, but spending US$10 to target a benefit worth US$40 might make sense; it would certainly make sense for one worth US$100. This explains the low percentage of administrative costs in total costs for Argentina's Jefes de Hogar (Heads of Household) program or the large conditional cash transfer (CCT) programs in Brazil and Mexico, where the transfers are large. Another way to achieve good cost-effectiveness is to use the same targeting sys- tem for multiple programs. Colombia, for example, first developed its proxy means test to target subsidized health insurance, and later used it for targeting hospital fee waivers and its CCT, public works, youth training, and social pension programs. Armenia, Chile, and Jamaica also use their proxy means test for several programs. This can not only yield economies of scale in the targeting system, but can also lead to a more integrated package of support for households that may provide better risk management and more effective assistance for moving them out of poverty. Third, overall administrative costs, especially as a percentage of a program's overall budget, tend to be higher during the program's start-up phase. During the first seven years of implementation of Mexico's PROGRESA program, its administrative costs fell from 51 percent of the program's total budget to 6 percent. This was because of large up-front expenditures for systems (purchase of equipment, design of systems, definitions of proce- 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 95 TABLE 4.3 Estimated Total and Annual Costs of Household Targeting Systems, Selected Countries, 2002 person targeted person per a per per (US$) cost (US$) y targeted benefits beneficiaries cost cost registered of view millions) per b Country, targeting system No. (millions) No. (millions) Inter (US$) Annual registered Annual beneficiar Benefits (US$ Cost (%) Brazil, Cadastro Único 29.0 19.1 3.90c 0.40 0.60 877 1.4 Chile, Ficha CAS 5.6 1.9 8.40d 1.20 3.60 526 1.3 Colombia, SISBEN 27.0 12.9 1.80 urban; 0.20 0.40 941 0.5 2.90 rurale Costa Rica, SIPO 1.0 0.21 4.20 urban; 1.00 4.80 116 0.9 7.00 rurale Mexico, Oportunidades registry 36.9 21.0 4.90 urban; 0.40 0.70 2,300 0.7 6.80 rurale United States, TANF registries -- -- 86.00f,g -- 86.00g -- -- SOURCE: Castañeda and Lindert 2005. NOTE: -- = not available; CAS = Comité de Acción Social (Social Action Committee); SISBEN = Sistema de Selección de Beneficiarios (System for Selecting Beneficiaries); SIPO = Sistema de Información de la Población Objetivo (Information System for the Targeted Population); TANF = Temporary Assistance for Needy Families. Data presented here exclude costs of equipment and information systems. Not all of those registered will receive benefits, as some will be above the beneficiary threshold. a. Total amount of benefits for all programs using this targeting system/registry to select beneficiaries. b. Cost of the targeting system as a share of total program(s) budget. c. Recertification period has not been established. d. Recertification every two years (new interview required). e. Recertification every three years (new interview required). f. Recertification is annual. g. This is approximately equal to US$25.00 when referenced to salary levels typical of middle-income Latin American countries. dures, and the like) that yield benefits for multiple years, coupled with a gradual rollout of the program with successively larger numbers of clients served by those systems (Lindert, Skoufias, and Shapiro 2006). PRIVATE COSTS Private costs always reduce a program's net benefit to the recipient. If they are sufficiently large, they may discourage eligible people from participating altogether. Evidence on pri- vate costs is rarely quantified systematically and reported in internationally comparable ways, but assessments of such costs often give program managers some sense of the size of the problem and the factors that contribute to it. A 1998 assessment in Armenia un- dertaken to lay the basis for some of the reforms then on the drawing board found that among the poorest people, the causes for not registering for benefits included having in- 96 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS sufficient information about the safety net system; being unable to pay the bus fares, fees, and sometimes the under-the-table payments required to get all the required documenta- tion in order; and having difficulty standing in lines for long periods because of disability, pregnancy, or child care needs (World Bank 1999d). The costs were particularly high for those who needed to get a medical certification of disability. In addition, people were con- fused about whether those who needed to update the documents testifying to their place of residence first had to pay in full any back taxes or utility bills. Such barriers are com- mon, indeed almost inherent in, social assistance programs, but the Armenian program subsequently took a number of steps to reduce them. The highest private costs are for programs with a work requirement. Many of those who participate in public works programs would earn at least something, usually from intermittent work or self-employment, if they were not working on the public works jobs. Thus the amount they are paid (the gross wage) is more than the additional income they gain from the public works job (the net wage). The cost of the income foregone through not being able to pick up odd jobs or work in self-employment while on the public work- fare job is the private cost of participation in this kind of program, and is often high, on the order of a quarter to half of the benefit (see chapter 7, section 4). INCENTIVE COSTS The basic issue of incentive costs is outlined in chapter 2, section 3, and evidence on the topic is treated more thoroughly in chapter 5, section 2. Note that incentive costs pertain most directly to only two targeting methods: means testing that is based on current in- come and self-targeting through public works, because the work requirement may reduce the time available for work outside the program. SOCIAL COSTS Receiving public assistance can generate a feeling of shame about receiving it or being publicly seen to be receiving it. How important the issue is appears to be highly variable. One factor seems to concern general public attitudes about the receipt of public programs. Rainwater (1982) shows that stigmatization seemed to be a larger factor in program non- participation in the United Kingdom than in Italy, for example. Programs can do much to foment or minimize stigmatization. One way they do this is through the public portrayal of the program. In Jamaica, the government used publicity to minimize stigmatization of participants in the maternal and child health portion of its former Food Stamp Program. The program was open to all pregnant or lactating women and to children under the age of five using public health clinics (the use of which was in itself a source of self-targeting). However, when the program was initiated, the stress was on its universality. Publicity spots were run on television showing the pregnant wife of a cabinet minister signing up for her food stamps with the explicit goal of removing stig- matization (Grosh 1992). In contrast, in Armenia during reforms to the Family Poverty Benefits Program, the publicity campaign focused on the targeted nature of the benefits, stressing that they were meant for, and only for, the poor. The hope was that this would encourage more self-selection among applicants, that is, that stigmatization would keep the nonpoor from applying. 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 97 Programs also adopt quite different strategies about whether or not nonparticipants can identify participants. The assumption is that when nonparticipants cannot identify participants, the latter will be less likely to be stigmatized. In some cases, the entire ben- eficiary roster is confidential or the information system may incorporate safeguards so that social workers or program administrators can only access the subset of cases under their own jurisdiction. In other cases, the beneficiary roster is made public as part of the mechanism to ensure transparency and fairness. In the latter case, the nonpoor may be embarrassed to be seen on the roster and therefore not apply. If they do apply, they may be subject to social pressure or denouncement to the authorities and removed from the roster. Program participants may be identifiable in other ways. In Bulgaria's public works pro- gram, participants wear reflective safety vests when working outdoors. Though intended for their safety, some say they are stigmatizing. The increasing trend of paying benefits in cash rather than in kind and via banking systems, post office accounts, or checks is prob- ably making benefit receipt less stigmatizing than for the previous generation of programs. The debit card used in the Brazilian Bolsa Familia (Family Grant) program, for example, looks similar to any other credit or debit card carried by the middle class. This approach stands in sharp contrast to what poor Bolivians participating in the 1980s food programs experienced when they lugged home sacks of grain and tins of oil bearing the U.S. PL480 food assistance logo. POLITICAL COSTS As noted earlier, the technical justification for budgeting is that, for a given budget, nar- row targeting can maximize the program's impact on the poor, or for a given impact, can minimize the budget. But this does not take into account how a budget is actually determined, which involves political processes. Targeting choices and outcomes will play into those processes. The evidence on the political response to the degree or method of targeting is scant and debate is ongoing (box 4.2). Most of the modeling work focuses on models in which voters determine the program budget, and voters' interest in funding a program relates to their likely direct benefits from the program (Gelbach and Pritchett 2002; Pritchett 2005). In such models, more universal programs are predicted to have larger sustaining budgets, and under many constructions narrowly targeted programs would end up with no alloca- tion at all. Some of the more interpretive literature (Esping-Anderson 1990) follows a similar chain of logic. Other interpretive literature points out that sources of political support may be more varied. Voters may support a program because they value social justice or perceive indirect benefits to assisting the poor, such as being hassled by fewer beggars, facing lower risks of property theft or political instability, or feeling that they have met their social obliga- tions without having to respond to myriad personal or nongovernmental solicitations for support. Such voters will appreciate deriving these indirect benefits at as low a tax burden as possible, which may call for some narrow targeting. To the extent that social assistance programs are externally financed, most development agencies also have this bias in seek- ing the maximum impact on social indicators for the minimum budget, and thus will also favor narrow targeting. 98 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 4.2 Universalism versus Targeting Targeting is a hugely controversial topic, considered anathema by some and panacea by others when, as with many divisive topics, the most sensible view is probably somewhere in between. In relation to social protection, the universalist approach proposes that all citizens of a nation receive the same state-provided benefits. Targeting proposes that state-provided benefits dif- fer depending on individuals' circumstances. Proponents of both approaches understand that in most developing countries, current budgets do not allow a meaningful provision of transfers to all citizens, and also that targeting experience is far from uniformly excellent. There are two glasses of milk, each of them half empty and half full; the "camps" differ about which they per- ceive can be filled. Universalists are optimistic that the social unity resulting from a uniform provision of benefits will garner a sufficient budget (nationally financed in middle-income countries and donor assisted in low-income countries) to provide meaningful protection. Universalists believe that experience with targeting as a way to increase the efficiency of redistributive spending has been unsatisfac- tory to date, uninspiring in relation to hope for the future, and detrimental to efforts to increase the budget. In contrast, targeters have a more optimistic assessment of targeting experience and are hope- ful that bad experiences can be replaced by good experiences and that perhaps the good ex- periences can be improved. Targeters' pessimism concerns budgets, seeing both political and technical obstacles to budgets becoming sufficient to provide meaningful universal benefits. In reality, the distinction between the approaches is not absolute. Even the European welfare states that have gone the furthest in universal provision of child allowances, education, and health insurance and have extensive minimum wage laws, labor market activation and the like have last resort needs-based programs that are tightly targeted. Thus even though they may choose wider or narrower ranges of programs to target or different mixes of programs, all coun- tries need to understand how to target. Political support may also come from interest groups that are suppliers to the program or advocates for its beneficiaries. Farmers and teachers' unions may support school lunch pro- grams on these grounds. Political support can also be garnered from adroit mixes of interest. The recent upswing in social assistance budgets in Latin America seems to have been greatly assisted by the advent of the CCT program. Because they tie social assistance to more broadly supported goals of universal health care and education, and because they demand "good behav- ior" on the part of recipients, CCT programs have garnered much greater political support than their predecessor programs, despite being much more effectively and often narrowly targeted. An intriguing piece of empirical evidence comes from an analysis of municipal elec- tions in northeastern Brazil. De Janvry and others (2005) find that mayors were more likely to be reelected when their implementation of the Bolsa Familia program had the following characteristics: good coverage, low errors of inclusion, functioning social control councils, and higher impacts. The voters valued several of the same features that contrib- ute to a program's technical efficiency. 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 99 4.3 Targeting Options A number of different targeting methods are available for directing resources to a particular group. Some demand some sort of assessment of eligibility for each applicant (individual or household). Others grant eligibility to broad categories of people, for instance, all those residing in certain areas or all those of a certain age. Others are designed to discourage the non-needy from entering the program, but do not actually prohibit them from doing so. This section defines a number of the common methods and summarizes some of the main advantages and disadvantages of each. A great deal of detailed implementation know-how is available for many of these methods, and box 4.3 provides references. BOX 4.3 Resources on Different Targeting Methods The World Bank has contributed to the large body of material available on targeting. Some key readings by methods are listed below. Overview. Coady, Grosh, and Hoddinott (2004) provide a comprehensive overview of target- ing issues and methods and international comparisons based on a review of 122 programs in 48 countries. Individual or Household Assessment. Castañeda and Lindert (2005) provide an in-depth treatment of the implementation details of household targeting systems based on case studies of Brazil's unverified means test; Chile's, Colombia's, Costa Rica's, and Mexico's proxy means tests; and the verified means tests used in the United States. It is especially rich in dealing with the details of data collection, database management, and the like. Tesliuc and others (forthcoming) provide an in-depth treatment of the implementation of means- tested systems based on case studies in Albania, Bulgaria, the Kyrgyz Republic, Lithuania, and Romania and on Armenia's proxy means test. Conning and Kevane (2001) summarize the little that is known about community-based target- ing systems. Geographic Targeting. Hentschel and others (2000) provide one of the early descriptions of how to construct poverty maps with small area estimation techniques with illustrations for Ec- uador. Information on the World Bank methodology and experience with small area estimation poverty maps is available at the World Bank poverty Web site: worldbank.org/poverty. Henninger and Snel (2002) review the policy uses of poverty mapping in 14 countries and how institutional details can affect its use. Self-Targeting. Given the relative simplicity of self-targeting, fewer implementation details need to be considered. Subbarao's (2003) primer paper on public works reviews design features and experiences pertaining to self-targeting through wage selection, and Alderman's (2002) paper on food subsidies reviews self-selection through the choice of commodities. 100 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS MEANS TESTS A verified means test is usually regarded as the gold standard of targeting. It seeks to col- lect (nearly) complete information on households' income and/or wealth and verifies the information collected against independent sources. Where suitable databases exist and interagency cooperation can be obtained, information may be verified by cross-linking the registries of, say, the welfare agency, property registrars, tax authorities, social security agencies, and the like. When this is not possible, households may be asked to submit cop- ies of records of transactions, such as pay stubs, utility bills, or tax payments. Simple means tests with no independent verification of income are not uncommon. Sometimes verification is completely nonexistent in that a program intake worker simply records what an applicant says. Sometimes a social worker will visit the household to verify in a qualitative way that visible standards of living (which reflect income or wealth) are more or less consistent with the figures reported. Alternatively, the social worker's assess- ment may be wholly qualitative, taking into account many factors about the household's needs and means, but not having to quantify them. These types of simple means tests are used for both direct transfer programs and for fee waiver programs, with or without household visits. In the best of cases, means testing can be extremely accurate. However, means tests work best in situations of high levels of literacy and documentation of economic transac- tions. They are administratively demanding when combined with meaningful attempts at verification. Means testing is also the form of targeting most like to discourage work effort, because eligibility is linked directly to current income. Means testing is most appropriate where declared income is verifiable, where some form of self-selection limits applications by nontarget groups, where administrative capacity is high, and/or where benefit levels are large enough to justify the costs of administering a means test. Some countries have started with rudimentary systems and refined them over time. PROXY MEANS TESTS Proxy means tests generate a score for applicant households based on fairly easy-to-observe household characteristics, such as the location and quality of the household's dwelling, its ownership of durable goods, its demographic structure, and the education and possibly the occupations of its adult members. The indicators used to calculate this score and their weights are derived from statistical analysis (usually regression analysis or principal com- ponents analysis) of data from detailed household surveys of a sort too costly to be carried out for all applicants to large programs. The information provided by the applicant is usu- ally partially verified either by a program official collecting information on a visit to the home or by having the applicant bring written verification of some of the information to the program office. Eligibility is determined by comparing the household's score against a predetermined cutoff. The advantage of proxy means testing is that it requires less information than true means testing, and yet is objective. Moreover, because it does not actually measure income, it may discourage work effort less than a means test would. Proxy means testing also has some drawbacks. Administering it requires a large body of literate and probably computer- trained staff and moderate to high levels of information and technology. It also implies an inherent inaccuracy at the household level, as the formula is only a prediction, although on 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 101 average, good results have been observed. The formulas used usually rely on indicators that are fairly stable, and may distinguish chronic poverty well, but can be insensitive to quick changes in household welfare or disposable income, which may be frequent and large when an economy is suffering from a large downturn. Moreover, the formula and results may seem mysterious or arbitrary to some households and communities. Proxy means tests are most appropriately used where a country has reasonably high administrative capacity, for programs meant to address chronic poverty in stable situa- tions, and where they are used to target a single program with large benefits or to target several programs so as to maximize the return for a fixed overhead. COMMUNITY-BASED TARGETING Community-based targeting uses a group of community members or leaders whose prin- cipal functions in the community are not related to the transfer program to decide who in the community should benefit. School officials or the parent-teacher association may determine entry to a school-linked program, a group of village elders may determine who receives grain provided for drought relief, or special committees composed of community members or a mix of community members and local officials may be specially formed to determine eligibility for a cash transfer program. The advantage of community-based targeting is that it relies on local information on individual circumstances, which may be more accurate and less costly to collect than us- ing other methods. In addition, it can permit local definitions of need and welfare. At the same time, community targeting may encounter several possible problems. Local actors may have other incentives besides good targeting of the program. For example, the grant- ing or denial of benefits to different members of the community may lower the authority or cohesion of local actors involved in the decision. Also such a system may continue or exacerbate any existing patterns of social exclusion. In addition, if local definitions of welfare are used, evaluating how well community-based targeting works becomes more difficult and ambiguous. Community-based targeting may be most appropriate where local communities are clearly defined and cohesive, for programs that plan to include just a small portion of the population, and for temporary or low-benefit programs that cannot support administra- tive structures of their own. GEOGRAPHIC TARGETING With geographic targeting, location determines eligibility for benefits: people who live in the designated areas are eligible and those who live elsewhere are not. Few programs target only on the basis of geography, but many programs use geographic targeting in con- junction with other targeting methods, especially when programs are not fully funded. In such cases poverty maps can be used to focus the program in only some areas of the country or to allocate spaces in the program among subnational jurisdictions. The advantage of geographic targeting is that it is administratively simple, requiring none of the machinery for individual assessment programs described earlier. It will have no direct labor disincentive and is unlikely to result in stigmatization, as poor and nonpoor neighbors alike will benefit. Geographic targeting will perform poorly when poverty is not spatially concentrated. It also depends on the accuracy of the poverty map, a concern that 102 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS is diminishing in importance as small area estimation techniques improve and are widely applied.3 Political compromises may be required, as politicians from each jurisdiction will lobby to have their districts included. This may mean that a fixed portion of districts within each province may benefit, rather than the poorest districts overall. The most appropriate circumstances for geographic targeting are when living stan- dards across regions vary significantly, when administrative capacity is too limited or trans- fer amounts too low to make individual assessment methods sensible, and/or when ad- ditional self-targeting can be induced through the use of some public service used mainly by the poor. DEMOGRAPHIC TARGETING The usual and simple forms of demographic targeting are based on age, with child allow- ances and social pensions being the most common. Part of the rationale is that individuals may be particularly vulnerable in childhood and old age. The logic is somewhat clouded by the reality that most individuals, certainly children, and in most countries the large majority of the elderly as well, live in households with several individuals and generally one or more income earners. Because households tend to pool their resources, at least in part, many children and elderly do not live in poverty even though they do not generate income for themselves. Demographic targeting is obviously administratively simple. Moreover, it carries the appeal of universality, and is thus often politically popular, plus those participating in pro- grams targeted in this way are not stigmatized. The limitation of demographic targeting is that age may be only weakly correlated with poverty. Current research shows that observed correlations are sensitive to assumptions made about economies of scale and equivalence used in constructing measures of welfare, an area where economists agree that some correc- tions are useful, but do not agree on exactly how to do them (see box 8.2 for a discussion).4 Demographic targeting is a low-cost targeting method and is particularly useful when age and welfare are highly correlated or for programs that include an element of self-targeting to complement the demographic targeting. For example, food supplements may be given to children who use public health services in locations where private health providers siphon off much of the demand from the upper part of the income distribution, or a social pension may serve those elderly who are excluded from the contributory pro- gram, which usually serves the top end of the wealth distribution. SELF-TARGETING Self-targeted programs are technically open to everyone, but are designed in such a way that take-up is expected to be much higher among the poor than the nonpoor or the level of benefits is expected to be higher among the poor. One of the most common applications of self-targeting in social assistance is the use of low wages in public works programs to induce participation only by the poor. The less poor will be able to command higher wages elsewhere. The administrative costs of the targeting are quite low, although administering public works programs is not simple. As, by definition, such programs put people to work, they may be politically supportable. However, the net benefits are usually substantially lower than the gross benefits. In addi- tion, the associated stigmatization can be considerable. 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 103 The other common application of self-targeting in social assistance is in the subsidi- zation of staple foods that are more heavily consumed by the poor than by the nonpoor. In practice, few goods may be consumed more in absolute terms by the poor, because the poor consume less in total, and thus the benefits are often regressive, or mildly progressive at best. Errors of exclusion and stigmatization are concomitantly low. GUIDANCE ON CHOICE OF METHOD Most targeting methods are applicable to most programs, but a few programs and meth- ods go hand-in-hand. As just noted, self-targeting through the use of a low wage is only applicable to public works programs and is used by most of them. Similarly, food sub- sidy programs require the choice of appropriate commodities. Aside from such examples, choices must be made. Cash transfers, for example, have been targeted using virtually all methods and many combinations of methods. Coady, Grosh, and Hoddinott (2004) assess which methods deliver the best results in re- lation to errors of inclusion. In the sample of programs they review, 80 percent of the variability in targeting performance as measured by errors of inclusion was due to differences within tar- geting methods, and only 20 percent was due to differences across methods. Interventions that used means testing, geographic targeting, and self-selection based on a work requirement were all associated with an increased share of benefits going to the bottom two quintiles compared with targeting that used self-selection based on con- FIGURE 4.5 Targeting Performance by Targeting Method sumption. Proxy means testing,5 community-based 75th percentile selection of individuals, Median Coady-Grosh-Hoddinott index and demographic targeting 25th percentile 2.5 of children showed good 2.0 results on average, but with considerable variation. De- 1.5 mographic targeting of the 1.0 elderly and self-selection 0.5 based on consumption 0.0 showed limited potential y y ly y ks An An An for good targeting. Fig- or Other method elder oungy w ure 4.5 provides simple aphicalr y lic the the comparisons of the range An assessment Means-tested means-tested Geog to to Pub Consumption Self-selection of benefits for each meth- xy unity od. The use of multiple Pro argetingT targeting methods within argetingT Comm a single program generally Individual Categorical Any selection produced better targeting assessment method than the use of a single SOURCE: Coady, Grosh, and Hoddinott 2004, table 3.4. method. NOTE: The Coady-Grosh-Hoddinott index reports the share of benefits accruing The ranking by to the group observed divided by the share of the group observed in the total population--for example, the share of benefits going to the poorest 40 percent Coady, Grosh, and Hod- of the population divided by 40. For most observations, it reflects the distribu- tion of benefits to the poorest two quintiles, or to the poorest quintile or fraction dinott (2004) cannot be of the poor. 104 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS taken as a blanket preference for one method over another. It does not consider errors of exclusion, different elements of cost, or feasibility constraints. We have less good informa- tion with which to differentiate between methods in relation to these elements, but the following observations are probably true: · Methods may be ranked differently when errors of exclusion are factored in. Self- targeting through wages in public works programs, for example, may have among the lowest errors of inclusion, but tends to have high errors of exclusion, because programs are small relative to needs and are inappropriate for labor-poor house- holds. Given a sufficient budget, errors of exclusion will, by definition, be lowest for demographically targeted programs or for those that subsidize a commodity or service used universally. · Means and proxy means tests seemingly demand the highest administrative costs, but these costs may not be a large marginal addition to the costs of registering beneficiaries, and systems have certainly been set up for large programs or for combinations of programs that incur low costs as a fraction of benefits. Grosh (1994) shows that the extra costs are generally balanced by the more accurate targeting achieved. Private transaction costs are probably also higher for such pro- grams, but again, need not be prohibitive. · All transfers could reduce work effort because they supply unearned income to households. Means-tested transfers will have a somewhat stronger influence on work, because they link eligibility, and sometimes benefit levels, directly to in- come. To date, little literature is available on developing countries that indicates that such effects are large, possibly because the transfers are small and are often perceived as uncertain. · Stigmatization can occur in any nonuniversal programs, but can be greatly influ- enced by program design and implementation. · Much of the literature voices the concern that nonpolitical costs may be higher for more narrowly targeted programs, but generalizing about such costs is difficult. As concerns costs, means and proxy means testing systems will clearly have many of the highest costs, certainly program administrative costs; transaction costs to beneficiaries; and may have relatively high incentive, social, and political costs as well. Yet for a large number of programs these have not been prohibitive, and indeed, they seem to have been fruitful investments in efficient social assistance programs, especially for those geared to the chronically poor (box 4.4). Everything we know confirms that implementation matters tremendously to out- comes. In Coady, Grosh, and Hoddinott's (2004) comparisons, country context explained some, but by no means all, the variability. Targeting performance improved with the coun- tries' income levels (the proxy for implementation capacity), the extent to which govern- ments were held accountable for their actions, and the degree of inequality. Unobserved factors, however, explained many of the differences in targeting success. Improvements in the design and implementation of targeting methods thus have great potential. If pro- grams with poor targeting success were brought up to the median level of success, the share of program benefits going to the poor would increase by 10 percentage points. 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 105 BOX 4.4 How Do the Reasons for Adopting a Program Affect Targeting Choices? Much of the targeting literature, and this chapter, is geared toward targeting the chronically poor. A good deal of experience with targeting this group has been accumulated, and several good options are usually available in most settings. When programs are designed to assist those who have experienced some sort of shock, the range of most applicable options and implementation issues for each change slightly. Proxy means testing, one of the workhorse options for targeting the chronically poor, is usually ruled out, as the proxies used usually change slowly. Means testing can theoretically be used, but would entail both open registration procedures and recertification as frequent as every few months. This is somewhat daunting. Thus self-targeting will be even more desirable than usual as the targeting method. Categorical methods may also be applicable. Hurricanes, tornados, and landslides may have defined geographical effects, and assisting all those in affected areas may be sensible. Even a more general phenomenon such as a drought or flood may have broad primary and secondary effects that geographic targeting alone, or geographic targeting plus some sort of simple take on long-term welfare or risk, may be sufficient. Widespread crop loss in a region, for example, would directly affect most farmers, and would also probably have significant second-round effects on the demand for labor by landless laborers and on products from small-scale services in the area. When the motive of a program is to compensate losers in a reform process, distinguishing how specifically identifiable the losers and the extent of their losses are is important. If fully identifying these is possible, specific compensation may be offered. For example, when public enterprises are scaled back, fired workers are easily identified and can be offered severance pay, training, access to credit, and so on. Targeting such specific compensation is not an issue, although designing the package of compensation may be. When reforms are more general, for example a reduction of food or utility subsidies, the goal of a safety net program is not to com- pensate each loser for his or her loss individually and separately, but more generally to assist the poor, and possibly the middle class, and to make the reforms politically sustainable. In such cases, the eligibility threshold may be higher than for programs meant for the chronically poor and/or the program's time horizon may be relatively short. Usually such reforms are aimed at moving from untargeted programs to more targeted ones, so some sort of household targeting mechanism may be devised. Sometimes these are a bit rough in the first instance so as to allow the reform to move ahead quickly and the targeting system is improved over time. 4.4 Implementation Matters for Targeting Because outcomes depend greatly on implementation, and because even a single method can be implemented in a variety of ways, discussing implementation in detail is important. This section summarizes some of the key lessons derived from the literature introduced in box 4.1 with a focus on issues common across methods. Program managers are advised to read the reference materials on the specific methods in which they are interested. Figure 4.6 presents a generic outline of the successive steps involved in targeting from the general population to program beneficiaries. 106 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS FIGURE 4.6 From Population to Beneficiary: The Stages of Targeting General population Policy choice: definition of eligibility criteria Targeted population Take-up rate: determined by self-targeting Applicants Eligibility rate: determined by mix of applicants and eligibility criteria Beneficiaries Exit rate: determined by attrition and recertification Former beneficiaries SOURCE: Adapted from de Neubourg, Castonguay, and Roelen 2007. FROM THE GENERAL POPULATION TO THE TARGETED POPULATION One of the first steps in social assistance policy making is to define what benefit a pro- gram will offer: how much, under what conditions, for how long, and to whom. This step is complex and based on many factors that are discussed in detail throughout the book and brought together in chapter 9. The next step is to define the specifics of eligibility, translating from intuitive notions to specifics amenable to actual administration. The details of this can be quite complex. Suppose the decision is to target the elderly. How old is elderly? Is it everyone above age 60? Age 70? Age 80? Should the same threshold apply for men and women? Is the age to be used determined only by the needs of this program or in coordination with retirement ages in contributory pension schemes? Similarly, for a program targeted to the needy, how poor is poor? Should the threshold vary by family size or composition or by cost of liv- ing differences for different areas of the country? Or consider a program to assist single working mothers. The intuitive understanding of who is to be served may be clear, but a precise definition can be technically difficult to come up with. How much work is enough to count as working? Does working seasonally or at home count? At what age does a child become an adult, and thus a mother become just an adult woman? When is a mother single? When she is unmarried? What if she gets formal alimony or substantial transfers from the child's father? What if she is legally married but her husband has abandoned her or migrated elsewhere? How would a program official verify the hours of seasonal or at home work or support from absent fathers? Because the range of different sorts of programs and of various country contexts are both great, providing a list of criteria that might be used for all such definitions is hard. Instead, the following subsections deal with some general factors that may help program managers arrive at specifics for their situation. 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 107 Budgetary Implications The more generous the definition of eligibility, the larger the applicant pool and the greater the cost of the program (of course the size and duration of the benefit are impor- tant cost variables as well). For its universal social pension, Nepal defines the elderly as those aged 75 and over. This is high relative to most definitions of old age, but allows all those who meet the criterion to be served within the available budget. Similarly, in coun- tries with a high poverty headcount, assisting all the poor is infeasible, and so a program eligibility threshold might be set that is significantly lower than the poverty line. The Kalomo District Pilot Social Cash Transfer Scheme in Zambia, for example, targets only the poorest 10 percent in each village even though a third of the Zambian population is chronically poor, and the proportion is probably even higher in Kalomo. Policy Coordination Especially in a country with a fairly full set of social protection policies, having some coordination among them can be helpful. Thus the definition of old age used for a social pension might be the same as that used for a contributory pension scheme. Similarly, the definition of unemployment in a social assistance program might be the same as for un- employment insurance or verified by registration with the labor bureau. Administrative Feasibility and Accuracy Simple criteria are likely to be the easiest to administer and may be appealing for this reason, but they can also be inaccurate. A balance must be found. Consider, for example, a national eligibility threshold for a means-tested program. This is a simple criterion, but if the cost of living is much higher in urban areas than in rural areas, the program would be inherently biased against the urban population. Thus having a spatially differentiated eligibility threshold would be fairer. The ability to accomplish this will depend on the country's general administrative capacity and on having at least periodic access to data on how costs of living differ across the country. Transparency and Political Feasibility For transparency, simplicity may again be valuable, but transparency is partly a way to al- low the public to perceive that a program is fair. If a program is simple to the point that it seems unfair, it may garner more criticism than approval. In the end, programs must pass the test of political supportability, and targeting criteria can be critical in this regard. FROM THE TARGETED POPULATION TO THE POOL OF APPLICANTS Bureaucrats may define who is in the target population, but households will determine who is in the applicant pool as they make their individual decisions about whether or not to apply. To be able to apply, households first need to know about and understand the program, and then they must make individual calculations about whether the program's potential benefits outweigh its transaction and social costs. Programs may wish to induce a significant level of self-targeting, and thus hope that the better-off among the eligible population will not apply, but at the same time, if errors of exclusion are to be kept accept- ably low, measures need to be taken to ensure that the neediest will apply. Such measures are needed no matter what targeting mechanism is used. 108 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS In determining whether to apply, participants estimate the entire range of costs and benefits to them of the program. They take into account the benefit level and the transac- tion costs of not just the application process, but also of collecting the benefit and com- plying with any requirements, such as a work requirement. Thus many parts of program design and implementation are intended to address errors of exclusion. Here we focus on those related to the process of applying to the program. Have a Sufficient Budget for Adequate Outreach and Intake One of the principal reasons why households may fail to apply for programs for which they are eligible is that the program has an inadequate administrative budget to employ the various strategies described in the following subsections. To do so requires sufficient staff; enough offices and travel funds to get the staff to the offices; and adequate spend- ing on literature, media campaigns, the targeted dissemination of information to other intermediaries who may help reach clients, and so on. Ensure Adequate Dissemination of Information about the Program All those potentially eligible need to know about the program and who might be eligible, what sort of benefit they might get, and how to apply to or find out more about the pro- gram. Reaching the candidate pool can take diligence and creativity. Social assistance programs target those who may face significant barriers to information: they are likely to be less educated, less likely to speak the official language, less likely to own the televisions or radios or read the newspapers that are used for mass media campaigns, more likely to live in areas that are remote or underserved by government services, and more likely to belong to socially excluded groups. Governments can use various avenues to get the messages out. Brochures and post- ers in appropriate languages that are simply worded or rely largely on illustrations are usually part of a good effort. These can be disseminated through multiple channels, not only through the social assistance offices, but also through other service providers such as schools, health clinics, post offices, and municipal offices; through other authority figures such as those at local places of worship, community structures, and nongov- ernmental organizations; or through commercial agents such as local shops, bars, and marketplaces. Mass media campaigns can help, and so can training and/or information sessions for workers in the places that distribute information or for community groups in the poor- est areas. Strong evidence from OECD countries indicates that information costs are a significant barrier to participation and that removing them results in higher take-up rates (Hernanz, Malherbet, and Pellizzari 2004). Ensure Low Transaction Costs for Beneficiaries To apply for a program, applicants must usually go to a specified place with some set of documentation. To keep costs for applicants low, getting to the point of entry into the program must be convenient. Usually that means putting registration offices within easy reach of households, which can be achieved in various ways. If the government's district offices are in market towns that most potential applicants visit regularly, that may be sufficient. Alternatively, the network of offices can be made more extensive. That is usu- 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 109 ally expensive, so an intermediate approach of using periodic service delivery points can be helpful. In this system program staff are available on a regular schedule, say once a week or once a month, at some temporary location, perhaps in the village office, school, or health clinic. The extreme of convenience can be a door-to-door sweep of poor areas to notify potential applicants of a new program and assess their eligibility. A number of large, new, proxy means-tested programs in Latin America have taken this approach, at least in poorer districts for initial registration. The different approaches have various advantages (table 4.4). Convenience requires that time costs be kept tolerable as well, so lines should be kept short and the offices open at hours that allow the working population access without loss of work time. Keeping the required documentation to the minimum necessary is also important, as getting each piece of paperwork is likely to involve a whole new round of logistical and cost issues for applicants. Programs can do various other things to help keep the burden for applicants low. They can allow more easily available substitutes for formal documents when these are not readily available. Nepal's social pension program, for example, accepts TABLE 4.4 Approaches to Household Registration Survey sweep approach Application approach Definition · All households in a particular area · Relies on households to come to a local are interviewed and registered in a welfare office or designated site to apply for nearly exhaustive system benefits Advan- · Better chance of reaching the · Lower total costs because of self-selection of tages poorest, who are likely to be less the nonpoor out of the registration process informed than others (fewer nonpoor households are interviewed) · Lower marginal unit registration costs · Dynamic, ongoing access because of economies of scale for · More democratic: anyone has the right to be travel costs interviewed at any time · Permanent process helps build and maintain institutional structures Best · High poverty areas (more than · Moderate or low poverty areas suited for 70 percent of the population is poor) · Heterogeneous areas · Homogeneous poverty areas (rural · The program is well known and publicized areas, urban slums) · New programs, when there is a need for speed Examples · Brazil: Cadastro Único · Chile: Ficha CAS since the early 1990s of · Chile: Ficha CAS until the 1990s · Colombia: SISBEN targeting systems · Colombia: SISBEN · Costa Rica: SIPO using · Costa Rica: SIPO in poor areas · Mexico: registry for Oportunidades in urban approach areas · Mexico: registry for Oportunidades in rural areas SOURCE: Castañeda and Lindert 2005. NOTE: CAS = Comité de Acción Social (Social Action Committee); SISBEN = Sistema de Selección de Beneficiarios (System for Selecting Beneficiaries); SIPO = Sistema de Información de la Población Objetivo (Information System for the Targeted Population) 110 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS horoscopes as proof of age when applicants do not have birth certificates. Programs can also help applicants obtain their formal documents. The CCT programs in the Dominican Republic and El Salvador have components that help families obtain birth certificates for their children. Programs can arrange to supply documents at no fee as Armenia's Family Poverty Benefits Program does, or they can get information directly from other govern- ment offices, a change made recently in Albania. Physical accessibility for people with disabilities, the elderly, and mothers with small children should be considered, especially as these groups are likely to be more numerous in the target population than in the general population. Allowing designated proxies to carry out transactions on behalf of individuals in these groups may help. Or intake workers may need to do home visits to reach the homebound. In Albania, the Ndihme Ekonomika (Economic Assistance) program can arrange for village leaders to collect payments for distribution to the homebound. Have an Open Application Process The best arrangement is to ensure that people can apply for a program at any time. The social assistance ideal is an entitlement program that admits anyone who meets the eligi- bility criteria as soon as they apply and grants benefits immediately. If limitations on the budget ration the number who may be served, the preferred option is to recalibrate the eligibility criteria, for example, by lowering the eligibility threshold for a means-tested or proxy means-tested benefit, adjusting the age of eligibility for a social pension upward so that fewer people are eligible, or limiting the program to only certain areas based on a poverty map. However, programs often do not take this route. Program rosters are kept open until the beneficiary cap imposed by the budget is reached and then closed. This will obviously create errors of exclusion. New households are formed, some households move, others fall newly into poverty, and still others find out about programs belatedly. All these groups will be excluded with a closed registry system. Allowing households to register, even if they cannot benefit immediately may be useful, especially for programs that expect a relatively large turnover. FROM APPLICANTS TO BENEFICIARIES In self-targeting programs, or those targeted only by demographic or geographic criteria, application and acceptance into the program are essentially synonymous. Targeting meth- ods that use individual assessment have a further stage of targeting that demands signifi- cant information. Twenty years ago such methods were rare in developing countries, but today they are common, especially in middle-income and transition countries. The first element of an individual assessment mechanism is a set of criteria that does well at distinguishing welfare, is feasible to assess for each applicant, and creates few or tolerable disincentive effects. The next elements of household targeting systems concern the means of gathering, and usually of verifying, information about applicants, and then reaching a decision. Individual assessment mechanisms will produce the most accurate targeting if the information on which the eligibility determination is made is accurate. There are various ways of trying to achieve this. 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 111 No Verification The simplest option is not to verify any information, but households have an obvious incentive to convey wrong information to become eligible. Moreover, concepts such as total income are complex. People with multiple or seasonal jobs and those with earnings from self-employment where household and business accounts are intertwined may hon- estly not know how much their income is, or at least not as a program official would have calculated it. Household surveys that make serious efforts to build comprehensive income aggregates can include hundreds of questions for multiple time periods, enterprises, jobs, and miscellaneous sources of income (McKay 2000a). The utility of verification is supported by experience in the state of Maryland and in Brazil. In Maryland, for a brief period in the 1970s, the state experimented with self- declared income and the threat of audit (as does the U.S. income tax system). Case er- ror rates increased rapidly to 53 percent and payment error rates to 23 percent (modern payment error rates are 13 percent) (Lindert 2005). In Brazil, the means testing system used for several programs, most notably Bolsa Familia and its predecessors, did not verify income.6 It reported more poor households, more very poor households, and more house- holds reporting zero income than general household surveys would predict or than a small survey of registered households showed (Castañeda and Lindert 2005). Nonetheless, no verification may be done in some situations. This will be most appropriate when the main information collected is not monetary, as it is less likely to be distorted; when the benefit is small or one-time; or when speed of response is of the essence, as in a hospital emergency room when determining how to charge a patient. Applicants Provide Paper Documentation Applicants may be asked to supply paperwork that testifies to their income, expen- ditures, assets, or at least to those parts of these that are amenable to documentation and reveal something about their economic welfare. They may be asked for pay stubs, records of utility bills, tax bills, certificates that confirm that they do not benefit from other social programs in the country, or other similar documents. The Unified Monthly Benefit Program in the Kyrgyz Republic may require up to 25 such documents, depend- ing on households' specific circumstances, but analysts calculate that on average, about 5 are required. Having applicants provide documentation may put a significant burden on house- holds. Thus the usefulness of verification in reducing errors of inclusion must be weighed against the increase in transaction costs for applicants and possible increase in errors of exclusion. Moreover, the provision of some of the documents may generate administrative costs for the bureaus that supply them. Because these costs do not usually accrue to the social assistance agency itself, they are not usually counted. Such extra costs will not ac- crue for documents issued regularly anyway, such as utility bills, but if a household needs certification that it does not own land or automobiles or does not receive a pension, the appropriate bureau must produce such documentation only for use in targeting the social assistance program. Albania's Ndihme Ekonomika program switched to electronic third party verification to eliminate the need for such extra documents. Applicant-provided documentation of welfare works best when a program's client base regularly participates in a substantial range of documented transactions. It may work 112 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS less well where sources of income are largely informal and undocumented; where assets are so small that they are not taxable; where the real and property tax systems are undeveloped; or where the poor may not have utilities, or even if they do have access, the utility services may not bill consistently. In middle-income and transition countries, by contrast, building a useful, if partial, picture of household welfare with such documentation may be possible. Program Intake Workers Make Home Visits Proxy means tests are built largely around variables that are easy to verify with a home vis- it, such as the location and quality of the dwelling, the presence of utility services, and the ownership of durable goods. The problem with home visits is that they require additional staff time, and possibly laptop computers or other portable equipment. The benefits in reducing errors of exclusion must be weighed against the extra administrative costs. The additional staff time required can be somewhat reduced with the survey sweep approach described in table 4.4, because home visits are grouped geographically, thereby reducing travel time. Unfortunately, little experience is available to verify whether systematic use of spot checks or random sample home visits induce sufficient accuracy in reporting to completely substitute for 100 percent home visits. A Third Party Verifies Welfare: The Community Option Community-based targeting is predicated on the notion that community members will know who is poor in their communities because of routine transactions. They see on a daily basis who spends what, who owns what, and what livelihood strategies households resort to. Thus placing the eligibility decision in the hands of a suitably constituted com- munity group or agent implies verification at no cost either to the program or the appli- cant. However, the community group may incur costs that are not counted. Sometimes such groups carry out interviews or home visits and incur substantial unrecorded and un- reimbursed administrative costs. Certainly they spend some time in the decision-making process. Whether they bear any social costs in terms of resentment by others for decisions made or with respect to their main social roles is not well studied and presumably var- ies by situation, but may be significant for the accuracy, costs, and sustainability of such systems. A Third Party Verifies Welfare: The Electronic Option An alternative to having households amass documentation verifying their economic con- ditions and bringing it to welfare offices is for the social assistance agency to obtain the information directly from other agencies' records. Third party verification is becoming more common in middle-income and transition economies where the data systems of in- dividual ministries are fairly well developed. To use them for third party verification, legal issues regarding the confidentiality of records must be addressed and a technical means of cross-checking databases must be established via personal identification numbers, ad- dresses, or names. The usual approach is electronic verification by merging databases, either periodically or continuously. The solution does not, however, have to be high-tech. In Albania, the social welfare office receives a quarterly printout of who is registered with the unemployment office and consults it as it processes applications for the Ndihme Ekonomika. 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 113 Irrespective of the means of verification, the quality of the application interview is important. The information should be gathered in an accurate, complete, efficient, and polite manner. During the interview, the eligibility worker should not only elicit informa- tion from applicants, but should also supply information to them (box 4.5). BOX 4.5 Communications and Transparency in Program Intake Adequate face-to-face communications with program applicants and beneficiaries are impor- tant for several reasons: they are respectful to the client and thus lower social costs, they can help achieve the outcomes for targeting and for the household behavior sought and thereby increase the program's efficiency, and they can help a program to be judged as fair and thereby increase its political sustainability. Program staff should · explain the confidentiality policy with respect to information relating to the household's application or benefits; · use respectful, culturally appropriate manners; · provide multilingual staff or translation services as needed; · give all applicants information on clients' rights and responsibilities in relation to recertifi- cation, continued eligibility, and so on, as well as on who to contact and how if applicants have questions and how to file an appeal; · allow clients to ask questions. SOURCE: Castañeda and Lindert 2005. FROM BENEFICIARIES TO FORMER BENEFICIARIES Safety net programs are rarely designed to provide permanent support for individuals or households (although some social insurance programs may be). Thus programs need to have a way to move beneficiaries off the rosters. Note that this is a somewhat differ- ent issue than helping households achieve economic independence, which is discussed in chapter 5, section 3. Beneficiaries may move out of a program through natural attrition. Each year a cohort of children will exceed the age limit set for the child allowance and some of the elderly receiving social pensions will die. This attrition leaves space in the program for the next group of children born or those elderly just reaching the age threshold to be admitted. While the concept is simple and parallel in these two cases, the administrative implication is not. The child allowance program needs only to establish the child's age on entry and then stop payment an appropriate number of years later. This can be built directly into a data system or noticed easily in pencil and paper systems. The removal of deceased elderly from the roles of a social pension program is harder, as it requires that the program ad- ministrator learn of the death of the beneficiary. Even though families are usually required to make such notification, unless there is a significant lump sum death benefit, they have little incentive to notify the program administrator of the death and may not do so, or may 114 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS not do so promptly. Moreover, some elderly will not have families and neighbors may not wish to take on the public service of notifying the program. Social pension programs may try to obtain information from official death registries, but these are often incomplete. As a third strategy, social pension programs may build in a system to suspend benefits for persons who do not collect them for two or three consecutive months, hoping in this way to learn of deaths. Of course, the payments might go uncollected for other reasons, such as a nonfatal illness or disability or higher than average transaction costs for individual participants, and there are circumstances when aid is most needed. Beneficiaries may move out of a program on their own initiative as their circum- stances improve. One of the attractions of a public works program targeted through low wages is that as the economy improves, either seasonally or after a severe downturn, the number of workers willing to work for low wages declines and the program shrinks of its own accord. Similarly, some beneficiaries of other programs may withdraw or cease to col- lect benefits regularly if the benefit is small relative to the transaction costs of obtaining it. Again, given little incentive to formally notify a welfare office and withdraw, programs often build in the suspension of benefits to those who do not collect them regularly. This allows them to recognize an available slot and enroll another needy family. Again, however, noncollection may occur for reasons other than withdrawal. Individual assessment mechanisms need to build in recertification requirements. Usually these come in two forms. First, households are required to notify the social assis- tance program of any material change in their welfare, such as changes in income, house- hold composition, or address, whenever such changes occur. Second, households are re- quired to go through a rescreening process periodically. This rescreening process has been a particularly weak point of many household targeting systems. Often the systems are set up for new programs, and often these programs are implemented in a hurry because of economic or political imperatives and the focus is on getting people in. Only a few years later once the system is running does attention turn to how long people have received benefits. Ideally, rules pertinent to recertification are clearly defined from the beginning of a program, rules are explained to clients as they enroll, and systems are built to handle recertification interviews as a routine part of the workload. How often recertification ought to take place should be informed by an empirical look at how rapidly households move in and out of poverty, how sensitive the targeting systems are to that, and the costs of recertification. In practice, such studies have not gener- ally fed into policy decisions about recertification, but the requirements are fairly sensible. In Europe and Central Asia's means tests, recertification is quite frequent, at least annually, and frequently more often. In Latin American countries' proxy means tests, recertification tends to occur every two or three years. Recertifying more often for means tests than proxy means tests makes sense, as the variables measured in means tests are more sensitive to changes in short-term welfare. Proxy means test measure longer-term correlates, so what is measured is likely to change little even if short-run welfare has changed. To reduce the administrative burden and transaction costs of recertification for cli- ents, programs may vary the frequency depending on the type of household. In the Kyrgyz Republic's Unified Monthly Benefit Program, for example, families in urban areas are usu- ally granted benefits for only three months, whereas in rural areas benefits are granted for a year in the expectation that rural residents have less access to new jobs, and that if their 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 115 welfare were to change, in small villages the change would be noticed and the social worker would be able to detect any lapse in the required notification of change in circumstances. In Albania, the main provision is for monthly recertification, but when frequent changes in welfare are not expected, recertification can be required only annually, for instance, where the household head is disabled, over 70, or a single mother with multiple children. In both countries, social assistance workers have some discretion in assigning the recertifi- cation period for specific households. Programs may also build in explicit time limits. Time limits can serve two functions. Most obviously and importantly, time limits serve as a way to spread resources among more people. This will be especially important when program funding is much less than what is needed to support those eligible. This, for example, is the predominant motivation for the 18-month time limit for Bangladesh's Vulnerable Group Development Program. Even in more generously funded programs such as Mexico's Oportunidades, time limits are used in addition to the slow attrition resulting from the demographic criteria and the medium-term attrition that might happen through recertification via the proxy means test every three years. Families receive full benefits for four years in urban areas and six years in rural and semi-urban areas, with reduced benefits for a further three years. Even in fully funded programs where rationing is not an issue, time limits are sometimes used to ensure that the recipients of social assistance have incentives to become self-sufficient. That is the logic behind the five-year lifetime limit in the U.S. Temporary Assistance for Needy Fami- lies Program and the two-year limit for intensive psychosocial support in Chile Solidario. The administration of time limits has implications for the program's data manage- ment system. A national database will be required, or at least local databases will need to be organized in a way that makes cross-checking feasible. The Temporary Assistance for Needy Families Program, for example, has no national database, so the five-year time limit cannot realistically be enforced.7 Residents can move from one state to another and essen- tially restart their time in the program. Some cross-checking takes place, but it is far from complete. Where time limits are for life, this implies that databases include information not just on current recipients, but on past recipients. The administration of time limits also poses challenges in defining recipients. If a family unit consisting of a mother, a father, and one child receives assistance for two years and then the couple is divorced and each remarries, do the new family units each count as having been on welfare for two years? Does the one with the child? Do neither? What if a mother and her children are on welfare and then the mother dies or relinquishes the care of the children to a grandmother. Does the grandmother start a new entitlement allotment?The variations on the theme are nearly endless, but they are not only of academic interest, as family and household structures are fluid in many societies and often more so in low-income strata. MECHANISMS FOR HANDLING APPEALS AND GRIEVANCES In every program's transactions, mistakes occur. Even more often people may believe a mistake has occurred when it has not. Having mechanisms for handling these issues is important both for correcting the mistakes and for perceived fairness. A program without a way to address such issues runs the risk of wrecking its reputation. The concerns to be resolved via appeals can occur at any point in a program, but the largest share is usually concentrated around eligibility, as this is clients' first encounter 116 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS with the program and often the most complicated. Thus this chapter discusses this issue, though the same mechanisms will be useful for resolving complaints about payments, compliance with any conditions, and the like. Systematic, professional, rules-based procedures for handling grievances and appeals are a hallmark of modernity, accountability, and democracy in a social protection pro- gram. These are lacking in all too many safety net programs in developing countries, with the more common recourse being either that concerned clients grumble to friends and neighbors without taking any action to redress the problem with the program, or that they appeal in person or via a better connected intermediary to the highest level of official pos- sible. Such personal rather than rules-based appeals systems open the door to governance and accountability problems. They can also consume a great deal of time when managers are working on individual cases when they should be working on managing processes that affect everyone. Some programs are working diligently to modernize their appeals systems, but so far few case studies or overviews on the subject have been undertaken to help coun- tries learn from one another (for an example of a case study, see Planning and Develop- ment Collaborative International 2001). The information in this section is therefore much less well grounded than in the rest of the chapter. Mechanisms to handle appeals and grievances have the following three goals: · To resolve concerns according to the program's rules · To minimize costs to both clients and the program · To be, and to be perceived as being, accessible, simple, transparent, fair, and prompt Appeals and grievance processes involve multiple levels. Each higher level is more costly to both the client and the program and should be used only when the prior level has not resolved the issue. The Frontline Service Provider The bulk of complaints should be resolved easily. The file may contain a clerical error that can readily be corrected. Information may be missing that can be obtained from the client or from a government unit, for example, to cross-check information on eligibility or to verify compliance with a condition of the program. Misunderstandings about rules can be cleared up. A frontline service provider can often handle such issues promptly and easily in a face- to-face contact. Handling them at the lowest level of service delivery has several advantages, namely: it is most accessible to the client; it is often cheapest for the program; and it usually means that the social worker or service delivery unit that made an error or did a poor job of communications sees the consequences of that and can not only fix the specific case ap- pealed, but can understand what went wrong to prevent the same problem from arising in the future. Enacting this piece of the appeals system is quite simple: it requires only that the schedules of program intake workers be set in a way that allows a repeat interview with any applicant who wants one and that the workers have access to client files. In cases where more than one agency may be involved in providing frontline service, working out which will have the information and ability to solve which problems and making that known to participants is important. For example, in a CCT program the mu- nicipality might handle the determination of eligibility and complaints about that aspect of the program, but schools might verify attendance and contracted banks might make 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 117 payments. Applicants who did not get admitted to the program will clearly go first to the municipality to verify if that was correct, but to whom should participants go if they think the size of a payment was incorrect? They may discover the problem upon collecting the payment, but the bank teller would usually not be the person to whom to address com- plaints, as he or she would usually only have information on the amount to be paid, not on the reasons why. Thus where particular complaints are handled needs to be decided, adequate rules must be established for one agency correcting or acting on data from an- other, data systems must be built, and rules for who does what need to be communicated clearly to staff and to the public. The Higher or Independent Level of Appeal A second line of appeal needs to be available for parties not satisfied at the first level. This may be within the same agency at a higher level; for example, appeals may be handled by the district rather than the subdistrict office. It may be a specialized branch within the same agency, for example, an office of appeals (often for appeals across a range of pro- grams) within the same ministry or municipality. Alternatively, it may be an independent committee or an ombudsman's office. A second line of appeal will be particularly important when the cause of the problem is some sort of incompetence, negligence, or malfeasance in the frontline office, but will also be useful in other cases. If complaints are arising because of some sort of systematic flaw in the program's design or implementation, the authorities who handle the second line of appeal often have more clout with program management than frontline eligibility workers and may be able to help get troublesome policies or procedures changed. A second line of appeal may also give a sense of recourse to complainants who are not happy with the outcome of the first-line appeal, even if it was correct according to the program's rules. This can be particularly important for eligibility decisions, as the targeting rules are only approximations of the kind of justice a program seeks. Thus in some cases applicants may not be eligible according to the rules, but many people would agree that they should be. Administrative appeals should have clear service standards, that is, that cases will be resolved within a given time period and that complainants will receive a full explanation of the rationale for the decision. Judicial Appeals When all else fails, complainants may have access to the legal system. This is obviously the most expensive way to solve cases and therefore should not be used as a matter of course. Judicial appeals can be particularly powerful in setting or altering rules or their interpretation. In South Africa, for example, a constitutional court case has been brought against the social pensions program, claiming that granting the pension at age 60 for women but age 65 for men is unfair. The program is therefore lowering the eligibility age for men to match that for women.8 Examples of Promising Practice We close with some examples of promising practice. The cases have not been fully evalu- ated, but thumbnail sketches of them help illustrate the range of tools available for ad- dressing complaints. 118 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Communications and Clerical Accuracy. A great deal of the art of handling complaints is preventing them. Making clear what the program's rules are will help clients know whether they have been treated according to them and whether or not an appeal is needed. Having good pamphlets and posters available, holding group information sessions to ex- plain the rules, giving eligibility officers enough time to be thorough in their conversa- tions with applicants, having applicants review and sign their completed applications, and using double blind entry of data from applications can all help. Community Committees to Validate Eligibility Decisions. Several programs, for in- stance, Mexico's Oportunidades program, have formal consultation mechanisms where draft beneficiary lists are presented at formal community meetings. This gives commu- nities a chance to point out both errors of exclusion and of inclusion. In practice, few changes are made as a result and consist mostly of adding households. This is perhaps not surprising: helping a needy household missed in the original outreach campaign is a be- nevolent act that all in the community will value. In contrast, suggesting that a household is wrongly included is divisive and can carry social costs for the person who makes the observation. If the household is only a little too well-off to qualify for the program, prob- ably no one would want to take their benefit away, and if the household is glaringly among the well-off and powerful in the community, few may dare to suggest such a thing. To lessen that problem, in El Salvador's CCT program, nongovernmental organizations are allowed to challenge the inclusion of non-needy households and program officials then investigate, a technique designed to address the imbalance of power between intended participants and the powerful. Community Agents. Several CCT programs have so-called mother leaders. These are beneficiaries who are usually elected by groups of beneficiaries. They receive some train- ing in the program's rules and help convey information back and forth between clients and program officials. They do not have the power to make decisions on eligibility and payments, but can be useful in helping clients understand the rules and verifying that complete and correct information is being used. Call Centers. Providing information by phone (or even online) can be efficient when the telecommunications infrastructure is well developed and cheap. Clients may find placing a call much easier than traveling to an office and lining up. For a large program to run such a back office function may also be safer for its employees and cheaper. However, carefully determining access to information and the authority to make changes that call center staff should have is important. One variant allows them only to explain program rules. Another gives them full access to the files on all clients, which allows them to fix missing or incorrect information. Call center staff may even be able to seek information from other government agencies if necessary. Good call centers can be helpful, although they are never sufficient by themselves, as some part of the client base may find using them impossible or uncomfortable. Moreover, they have to be adequately staffed and equipped and well monitored. Calls that are never answered, or worse, answered incor- rectly, only harm a program's reputation. Community Appeals Committees. Armenia uses local social protection councils com- posed of five representatives of local government social sector offices and five representa- 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 119 tives of nongovernmental organizations. Among other functions, the councils hear ap- peals from those deemed ineligible by the proxy means test, but who consider themselves in need. The councils have the right to grant entry to up to 5 percent of the roster. This allows the Family Poverty Benefits Program to address cases where the rules and fairness do not quite match up, but in a transparent way. ADMINISTRATIVE CAPACITY TO SUPPORT TARGETING SYSTEMS Effective selection and intake of applicants into a safety net program requires sufficient administrative capacity. The specific requirements will depend on the targeting method selected and the myriad details of its implementation. Moreover, the systems for intake, payment, and provision of other services may be intertwined, so capacity is required for more than just targeting. This section therefore cannot be comprehensive, but it does il- lustrate some of the issues pertaining to administrative capacity with respect to eligibility for programs and how they influence costs and performance. Staffing Obviously adequate staff time and general skills are required. Surprisingly little docu- mentation about program staffing is available, therefore saying exactly what defines "ad- equate" in a quantitative sense is difficult. Documentation of even such basic facts as caseloads is uncommon, and moreover is hard to compare, because the range of tasks may not be comparable. In the city of Arzamas in Russia, each staff member in the one-stop shops that integrate the means tests and application procedures for several previously separate benefits can process 127 benefit claims per month (Institute for Urban Econom- ics, Independent Institute for Social Policy, and Urban Institute 2006). Castañeda and Lindert (2005) report that proxy means test interviews take about 15 to 20 minutes in Chile, Colombia, and Costa Rica. These interviews are done in applicants' homes. Staff can undertake interviews in about 15 homes per day (or about 300 per month) in urban areas using the survey sweep approach, where the outreach is scheduled ahead of time and concentrated in areas of high poverty. When interviews are on an on demand basis, staff can undertake only about 8 or 9 a day (160 to 180 per month) in urban areas because of the greater travel times between households. The Latin American systems can probably handle more applicants per month than elsewhere, partly because proxy means tests forms can be short, but probably also because, for example, the case workers in Arzamas also perform other tasks such as data entry. Where staff also provide significant counseling services, caseloads will be lower. The formal levels of education of program intake workers tend to vary with the country. In most countries, program workers have completed high school, or at least have had some secondary education. In higher-income countries, some workers may have an undergraduate degree, or at least have taken some university-level courses. In OECD countries, many workers have undergraduate degrees in social work. The minimum skills required relate to the targeting method used and how data collection is organized. For proxy means tests, much of the work required for data col- lection using massive survey drives is contracted out to professional survey teams that will understand the questionnaire and survey techniques, but may know little about the program or social policy. Their job is basically to collect the data that determine eligibility. 120 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS For community-based methods, the essential requirement is that the community members involved know the poverty situation of their fellow community members. They will need some understanding of the program to make effective decisions and to help get those se- lected signed up. Such knowledge is usually rudimentary, as the community members are not, by definition, full-time program workers. In the more classic case of eligibility work- ers conducting means or proxy means tests, they should fully understand all the ins and outs of the program's rules in at least a mechanical sense. A desirable, but less common, feature of programs is that program intake workers understand some of the goals and concepts behind the rules. When they do, they can make more effective decisions in borderline cases and they can be better ambassadors of the pro- gram to applicants. This is important. If intake workers make inappropriate decisions about eligibility, the program's effectiveness is directly impaired, and the lack of credibility that the program does what it says it does may impair its political sustainability. Clients' understand- ing of the criteria for eligibility is important in terms of their compliance with them, and again in terms of public perceptions about whether the program is operating fairly. A need commonly expressed, but not usually incorporated in hiring criteria or for- mal training for intake workers, is that they be trained in communications so that they can deal with conflict. An inherent part of their job is to say no to rejected applicants, which is never pleasant news to deliver or receive. Moreover, a subset of applicants may have mental illnesses or substance abuse problems, as these conditions often lead to poverty. Being able to handle such potentially difficult conversations skillfully is important for good commu- nity relations. It will also lower job stress, which can cause burnout and staff turnover. As the main knowledge that eligibility staff need is highly specific to the program, they acquire that knowledge through some sort of training provided by the program. Training is one of the areas of capacity building that often needs attention. Often on-the- job training consists only of watching workers with longer tenure do their jobs. Some pro- grams have formal training courses provided by the appropriate agency, but a high portion of these is likely externally financed, and thus may train specific cohorts of workers, but not be fully adequate for handling the ongoing training needed to deal with attrition and replacement of program workers. Rules of the Game Program rules and regulations must be clear and well defined. Certainly a trade-off exists between too much and too little elaboration of the rules. Programs with little elabora- tion of their rules may have problems making decisions that are consistent from one case to the next, especially for cases that are not clear-cut. At the same time, excessive detail may mean that workers are unlikely to know or understand them all or that they can be adequately conveyed to the public. Rules should also be defined in such a way that they incorporate reasonable, but not excessive, flexibility in legal terms. A few programs are defined in countries' constitutions, and are thus extremely inflexible, almost certainly too inflexible for good economic man- agement. Rules contained in laws can be changed periodically, but still require significant political will and time to adjust. Those rules contained in operational manuals issued by the executive branch of government are the most flexible. Some regional variation is ap- parent, with programs in Eastern Europe and Central Asia being supported by laws, and 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 121 programs in Latin American more commonly being supported by executive decrees and operational manuals.9 This partly reflects the relative roles of the branches of government in the different systems, but it may also contribute to the Latin American tendency for programs to come and go with new administrations. Information Systems and Technology Programs can be reasonably well targeted with little more than pencil and paper records kept in local offices as, for example, done by Cambodia's Japan Fund for Poverty Re- duction Girls Scholarship Program or, until fairly recently, the Unified Monthly Benefit Program in the Kyrgyz Republic. However, computerization can allow programs to in- corporate more sophisticated designs, services, and monitoring. A consolidated national database can help avoid duplication and track beneficiaries. Computerization is an es- sential prerequisite for cross-checking eligibility information with other databases, such as other program rosters, social security registries, and tax registries. Computerization of records can help track what services households have received and what other services they might be eligible to receive. Castañeda and Lindert (2005) consider experience with information systems for several Latin American means and proxy means testing systems. They identify the following lessons: · Proper identification of individuals is crucial. A unique social identification num- ber should be used, ideally one that is used on a countrywide basis to link registry information and beneficiaries with other systems and programs. Such cross-checks can reduce both errors of inclusion and fraud. · Software and coding systems need to be designed so that individuals are linked to their families or other assistance units. Such identification features have been stumbling blocks in many developing countries. They are not insurmountable, however. While countries would ideally assign individuals unique numbers at birth, in the absence of a single national identification number, registry question- naires often collect information on multiple identification numbers and then assign a new social identification number upon registration (and codes to link individuals to families). This is a feasible solution, provided that (1) data are consolidated and cross-checked in a single database system, and (2) the system has the capacity to be updated to reflect changes and can store and reference historical data. · Updates and recertification are important for tracking fraud and avoiding situa- tions such as "ghost" beneficiaries, which can emerge as registries become dated. They also allow for turnover in beneficiaries to make space for other poor families to gain entry into programs. · Database management should be designed so that it can respond flexibly to chang- ing policies and updates and can rely on common software (even if data entry is decentralized), with pretesting of systems, well-designed manuals, and adequate training for users. Material Inputs Just like schools and health clinics, social assistance programs need a suitable share of variable inputs to function effectively. In far too many cases, errors of exclusion are high 122 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS because of a lack of brochures, application forms, transport budgets for staff, and the like. Similarly, client services and monitoring are often deficient because of the absence of computerization. Institutional Roles As discussed in chapter 3, section 6, social assistance programs are often a joint effort by national and local governments. This leads to the question of who should perform which functions. Most important, roles should be clearly assigned. No clear blueprint of how that should be done is available. Table 3.8 shows the advantages of different arrange- ments for various aspects of household targeting systems. The case for local manage- ment is strongest for data collection, although it can certainly be organized centrally as well. The central level has a greater comparative advantage for database management, but that conclusion is partly based on the idea of optimal management of individual vertical programs. If local actors run many different programs, and especially if locally financed programs are an important part of the overall policy package, the overall coordination or integration of services for clients might be better achieved with more decentralized data- base management arrangements. Monitoring and Oversight Strong mechanisms for monitoring and oversight are crucial for all systems, but especially with decentralized data collection. While no system is completely immune to fraud or leakages, a variety of tools should be used to minimize them. A number of mechanisms are available, including supervising interviews, verifying information, comparing target- ing registries with other data, carrying out random sample quality control reviews, and encouraging citizen oversight (or social controls). Using multiple instruments strengthens the system. Time Details of design and implementation have a major impact on distributive outcomes. Too often, however, governments want to launch programs quickly, and they--and the con- sultants they hire to help them--do not pay enough attention to the necessary details that go into designing and implementing effective household targeting systems. Such systems take time to design, pilot, and implement on a large scale, usually at least 18 months. To get the best out of a system probably takes 5 to 10 years or successive rounds of critique (internal and external), adjustment, and critique. POLICY REFORM AND TARGETING SYSTEMS: AN ILLUSTRATION FROM ARMENIA As this chapter has shown, targeting outcomes depend on the big picture of policy choices about who the intended target group for a specific program is, the factors that will affect households' decisions about whether to apply for the benefit, and a host of details con- cerning how the selected targeting mechanism is implemented. This chapter concludes with a discussion of the reform of social assistance in Armenia, which shows how target- ing systems are built. In 1991, Armenia, a country of 3 million people located in the Caucasus, inherited a generous and regressive cash benefit system plus a system of parastatals obliged to provide 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 123 subsidized goods and services to "privileged" citizens such as veterans, members of the armed forces, mothers of many children, and persons with disabilities, through quasi-fiscal means from the former Soviet Union. The period following independence was difficult, with the disruption of trading patterns resulting from the dissolution of the Soviet Union, a major earthquake that wiped out a great deal of the nation's industrial capacity, and a con- flict with Azerbaijan which resulted in a trade blockade. By 1993, Armenia's gross domestic product had fallen to half of its pre-independence value. By 1998, the social assistance system consisted of 26 small, uncoordinated cash programs covering 15 percent of the population and providing small benefits to different categories of individuals (orphans, single mothers, large families, pensioners living alone, and the like). The entire social assistance system chan- neled fewer resources to beneficiaries than the regressive electricity subsidy scheme. In 1999, the government reformed the system by consolidating most programs into a sin- gle program, the Family Poverty Benefits Program.The program's design conformed to interna- tional best practice. It was implemented via a tightly run administration.The program targeted low-income households instead of categories of poor and not poor individuals. Eligibility was determined using a proxy means test tailored to Armenian conditions, where a large informal economy made income or means testing infeasible. Initially, the program covered 27 percent of the population, roughly similar to the share of the FIGURE 4.7 Fraction of the Social Assistance Budget Captured population living below by Each Quintile, Armenia, 1998 and 1999 the extreme poverty line, and provided more gener- Percentage of benefits received by quintile ous benefits than the pro- 35 grams it had replaced. At 30 the same time, the govern- 25 ment discontinued the in- efficient electricity subsidy. 20 The new design paid off. 15 An assessment of its target- 10 ing showed that the share of benefits going to the 5 poorest 20 percent of the 0 population had risen from 1 2 3 4 5 1 2 3 4 5 16 percent in 1998 under (poorest) (richest) (poorest) (richest) 1998 1999 the old system to 32 per- cent by 1999 (figure 4.7). SOURCES: Tesliuc and others forthcoming; World Bank 2002a. The early success of the program, quite singular in the Caucasus at the time, captured donors' interest, and the U.S. Agency for International Development funded a multiyear program of technical as- sistance supplied by Planning and Development Collaborative International. Since 1999, the program has received a large amount of technical assistance for capacity building, es- pecially for staff training and process evaluation. According to Planning and Development Collaborative International, the most effective assistance was (1) the recommendations for improved auditing in field offices; (2) the preparation of operational and staff manuals covering customer service, claims processing, human rights, social legislation, and office 124 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS administration; (3) the evaluation of the implementation of child-focused social policies; (4) the annual survey of the public's use, knowledge, and perceptions of social services; (5) the design and evaluation of two one-stop shops; and (6) the comprehensive outreach campaign using multiple media channels (press, television, radio, posters, and leaflets in- forming people of their rights). The institutional consolidation of the program (table 4.5) triggered better service and greater reductions in fraud and error rates. According to an Armenia Social Transi- tion Program survey, public awareness of the program increased from 50 percent in 2001 to 78 percent in 2004, and the service quality of the 55 frontline regional centers rose 25 percent between 2000 and 2004. Better trained staff and the introduction of an inte- grated database for determining eligibility helped reduce fraud and error rates. The share of benefits going to the poorest increased from 2004 to 2006. Thanks to economic growth, extreme poverty fell from 23 to 14 percent between 1999 and 2003, while 40 percent of the original caseload graduated out of the program. Thus the fiscal cost of the program went down by 30 percent, despite an increase in its benefits and the eligibility threshold. Had the prior system of categorical targeting re- mained in place, such a reduction in caseload would not have been possible. TABLE 4.5 Institutional Consolidation, Family Poverty Benefits Program, Armenia, 1999­2004 Improvement sought Action Lower exclusion errors · Run public information campaigns · Place points of service close to beneficiaries · Institute an appeals mechanism--here the Social Assistance Board, a community-based institution--with an important role in selecting beneficiaries Lower inclusion errors · Improve eligibility criteria (give more weight to children and the elderly) · Improve the benefit formula (give higher benefits to children and the elderly) · Cross-check databases with information about assets or expenditures Lower inclusion and · Reduce data entry mistakes through double entry of household records exclusion errors · Cross-check the eligibility determination process · Have coordinating unit staff undertake regular audits and other controls Lower private costs · Provide the documents required to determine eligibility free of charge · Locate the offices or points where applications are collected close to the beneficiaries · Implement a one-stop shop system Improved administration · Provide training · Provide adequate documentation for program staff · Implement a strong monitoring and evaluation system Greater economies of · Use the targeting instruments for other poverty-focused programs scale SOURCE: Authors. 4. ENROLLING THE CLIENT: TARGETING, ELIGIBILITY, AND INTAKE 125 Notes 1. A number of indexes are available that summarize the full distributional information into a single summary statistic and are sometimes useful. Ravallion (2007) reviews these. 2. In geographic targeting, when some areas, districts, or states are excluded from a program, obviously poor residents of these localities are excluded from the program. As few social as- sistance programs are targeted using only a geographic criterion (usually explicit elements of self-selection or individual assessment are used as well), geographic targeting and the exclusion errors it induces are usually symptoms of insufficient budgets. 3. Small area estimation techniques combine data from censuses and detailed household surveys to create poverty maps representative of small areas. An inherent challenge of poverty mapping is data. Censuses have data on every household, and thus are representative of small areas, but the details they contain are limited and are not sensitive predictors of household welfare. More detailed surveys, such as household expenditure surveys or living standard measurement study surveys, do a much better job of measuring household welfare, but have small samples repre- sentative of only very aggregated areas. In small area estimation, the welfare measure observed in the detailed dataset is regressed on a subset of variables that also appear in the census. Then using the formula so derived, household welfare is predicted for every household in the census and the predictions are aggregated into poverty maps with a finer level of detail than the use of survey data alone will permit and with greater accuracy than using census data alone will permit. 4. Economies of scale refer to the idea that two can live more cheaply than one, for example, heat- ing needs do not increase because an apartment has two residents rather than one. Similarly, some durable goods may be shared. Economies of equivalence refer to the fact that among goods that are less likely to be shared, requirements may vary from person to person. Chil- dren, for example, require fewer calories than adults. Deaton and Zaidi (2002) and Lanjouw, Milanovic, and Paternostro (1998) provide relatively simple technical explanations and ex- amples of the sensitivity of poverty profiles and policy conclusions based on how economies of scale and equivalence are defined. 5. When Coady, Grosh, and Hoddinott (2004) undertook their study, outcome data were only available for a few of the new proxy means tests. Since then data have become available for several more programs, all of which are quite well targeted. If these measurements had been part of the original study, proxy means tests would likely have joined the ranks of the methods that reliably produce progressive results. 6. Subsequently, some verification has been added. If reported consumption is 20 percent higher than reported income, a local program manager must investigate before forwarding informa- tion to the central level for decision. Information is cross-checked against the Ministry of Labor's income databases in some states against Department of Transport registries of automo- bile ownership (Lindert and others 2007). 7. The consensus around time limits was strong enough to pass the legislation mandating the time limit, but not strong enough to pass the relatively trivial budget appropriation that would have been required to implement a national database that would make the time limit more strictly enforceable. 8. Daniel Plaatjies, Executive Manager, Strategy and Business Development, South African Social Security Agency, conversation with Margaret Grosh, March 3, 2008. 9. For examples of some laws and operational manuals, see the "Implementation Matters" page at www.worldbank.org. Examples of operational manuals are in the country-specific subpages under CCT programs. CHAPTER 5 Benefit Levels and Delivery Mechanisms KEY MESSAGES To achieve its intended outcomes, a program's benefit level should be consistent with its objectives. However, budget constraints often make for hard trade-offs between cover- age and benefit level. Programs with benefits that are too small will have little impact on beneficiaries and administrative costs will be high relative to the level of benefits. Pro- grams with high benefits will have a larger impact on recipient households, but will have a higher fiscal burden, require more care in relation to design and targeting, and may induce greater work disincentives. In general in developing countries, programs with ben- efits that are too low are more frequent than programs with benefits that are too high. Benefits may be differentiated by household characteristics such as poverty level, size and composition, or specific needs or behaviors. Such customization will improve the poverty impact per unit of transfer, but will complicate administration and communication with the public and are thus more common in high-capacity settings. Participation in safety net programs has only small or moderate effects on employment or hours worked in developed countries and even smaller effects in developing countries. Moreover, policy makers have a variety of tools at their disposal to minimize work disin- centives, such as limiting the program to those who are not expected to work; adopting a targeting mechanism that is not tied directly to earnings; setting benefit levels to maintain work incentives; ensuring that incentives to work remain in place by customizing benefit levels in line with earnings; and/or linking transfers to such program elements as job training or placement, education, microcredit, and social support services. Experience is emerging with linking transfer programs to other services--voluntary or man- datory--that are designed to help households become independent. This is a promising field for experimentation. Tentative lessons suggest that mandatory links should be limited to cases where the supply of required services is ample, the services will be useful to all, or most transfer recipients already use the services. In a wide range of other cases, voluntary links through information, referrals, one-stop shops, and the like may be applicable. Payment mechanisms should be affordable, safe, reliable, and accessible to all benefi- ciaries. A number of different payment instruments are available, including cash, checks, vouchers, and in-kind benefits, that can be delivered using banks, automated teller ma- chines (ATMs), mobile pay points, private or public shops, and so on. The choice of ap- propriate delivery mechanisms depends on objectives, operational needs, administrative capabilities, and local infrastructure conditions. Investments in administrative systems and equipment related to payments can help increase service standards, reduce corrup- tion and leakage, and reduce costs in the long run. 127 128 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS 5.1 Determining Benefit Levels in Theory and Practice This chapter takes up the one of the basics of transfer programs: paying people. This sec- tion looks at the theory concerning the size of transfers, discusses the criteria for adjusting benefit levels to household circumstances, and reviews the evidence on the generosity of safety net programs in practice. SIZE OF TRANSFERS A basic question in any safety net program is how generous the program should be, that is, how much to pay. No clear-cut answer to this question is available. Ultimately, the level of the benefit is one of the products of the iterative process of designing a program, that is, the program designers select a benefit level such that the overall program will fit within its budgetary, administrative, and political constraints, while maximizing its outcomes for beneficiaries. However, this summary is too general to be a useful guide for practitioners. This section tries to break down the decisions and highlight the key elements and trade- offs that occur when selecting a program's benefit level. First and foremost, the benefit level depends on the objective of the program, and hence on the program type. The benefit level should be consistent with program theory, that is, the stylized model of how policy makers think the program's output will affect the outcomes they are trying to influence. A benefit level compatible with program theory will be the smallest transfer necessary to achieve the desired impact on intended outcomes (consumption, income, earnings, school enrollment, or use of nutritional or health ser- vices). Last resort programs aim to reduce poverty, hence the benefit level is set as a fraction of the income gap of expected beneficiaries. This is the case for programs that select benefi- ciaries using a proxy means test, such as in Armenia and Georgia. A number of variations on this principle are possible. In low-income countries, benefits are often set relative to the costs of an "adequate" food basket or the food poverty line. In guaranteed minimum income (GMI) programs, which are common in Europe and Central Asia and in countries of the Organisation for Economic Co-operation and Development (OECD), the level of benefit is the difference between the eligibility threshold and the income of each family. Programs that compensate poor consumers for one element of expenditure, so-called gap formulas, are used, for example, for family allowances that cover a portion of the cost of raising a child, heating allowances that cover the seasonal increase in heating costs during the winter, and food stamps that cover only the food poverty gap. Conditional cash transfer (CCT) programs encourage poor beneficiaries to invest in children's human capital by conditioning the benefit on the use of school, nutrition, and/ or health services. The level of benefit will thus reflect two objectives: reducing current poverty among beneficiaries and providing incentives for human capital accumulation. The principles for the first objective are similar to those for last resort programs. For the second objective, the level of benefits is set to compensate households for the opportunity cost of using the services. The total benefit to a household may include a few components. An education grant will compensate households for the opportunity cost of the time chil- dren spend in school and not working, plus for the direct costs of schooling. A health and/ or nutrition grant will compensate families for the cost of the time they spend taking their 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 129 children for health checks and/or attending nutritional education events. Some programs, like the Programa de Asignación Familiar II (Family Allowance Program) in Honduras and the Red de Protección Social (Social Protection Network) in Nicaragua, offered a supply grant to the service providers--schools and health posts--to cover the cost of improved service. The rationale of education, health, and nutrition grants is to increase the demand for education, health, and nutrition services as illustrated in box 5.1. Sometimes programs may offer unconditional grants to any poor household, with eligibility determined by the same principles as for last resort programs. In workfare programs, the benefit level is the wage rate. To ensure self-selection by the poor, the wage rate should be set somewhat below the wage level for unskilled workers. When other considerations, such as minimum wage laws, preclude setting such a low wage rate, programs have to ration demand by capping the total number of days of work to be BOX 5.1 The Value of a CCT Program's Education Grant: From Theory to Practice The Programa de Asignación Familiar II is a CCT program in Honduras that offers an education grant to poor children conditional on school attendance. To determine the value of the grant, the technical advice provided to the government by its consultant, the International Food Policy Research Institute, was based on both economic theory and microeconomic evidence. Economic theory suggests that each family demands a certain level of services, such as edu- cation, up to the point where the actual value of future educational benefits from sending a child to school is equal to the marginal cost of sending the child to school. The expected value of future benefits depends on, among other factors, the family's expectations about the child's future income and the relationship between education and income.The marginal costs of send- ing the child to school include the direct costs incurred when the child is sent to school as well as the opportunity cost of dedicating the child's time to learning instead of using it to generate income. Based on these expected costs and benefits, each family demands that level of service that will allow it to maximize its welfare over time.This maximization process leads to a demand curve that reflects the relationship between levels of service demand and price, assuming that consumers' preferences and incomes and the prices of other products remain constant. The sum of all services sought by each family produces an aggregate demand curve that can be interpreted as the relationship between service price and the number of families willing to pay this price for its use. The designers of the Honduran CCT program used household survey data to estimate that children provided about 3 percent of labor hours and 2 percent of household income, or about L 326 (about US$22) per year per child (about nine days of work during coffee harvest time). The direct costs of schooling were estimated to be L 6 (about US$0.40) per year for matricu- lation and fees; L 241 (about US$16) per year for books, uniforms, and supplies; and L 25.5 (about US$1.70) per month for 10 months for lunch and transportation money. Thus the total cost (adding up the lost income per child plus the direct costs of schooling) is about L 828 (about US$56) per child per year. SOURCE: IFPRI 2000. 130 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS provided to individual workers. In Argentina's Trabajar workfare program, for example, the wage payment was set slightly below the legal minimum wage. As the economic crisis that began in Argentina in 1996 became more severe, unemployment and wages worsened and program wages were adjusted downward. Participation was capped at 90 days per worker. In Ethiopia's Productive Safety Net Program, the labor market is so thin in some areas that fixing a wage rate that is both below the prevailing wage and delivers sufficient value is difficult. Thus the total benefit package is designed to fill the food gap during the three months of the hungry season. In this case, the wage may be too high to induce adequate self-targeting, so the number of days of work allowed is capped at five days per person per month. In-kind transfers have diverse objectives. If the program's objective is to provide a feeding supplement to schoolchildren, then the benefit level will be the cost of the food bundle. If the in-kind transfer is a vehicle for transferring income to poor households, then the same principles in determining the appropriate size of the transfer apply as for a cash transfer, though with some complications. If the in-kind transfer provides less of an item than the household would normally consume (is inframarginal), economic theory suggests that the subsidy is equivalent to a cash transfer of equivalent size, albeit with administra- tive costs that may be substantially higher. If the in-kind transfer is larger than what the household would normally consume, then the household may raise its consumption of the target good and/or may sell some portion of it, often at a discount, which will lower the real value of the transfer. A second element that is taken into account in setting the level of benefits is the pro- gram's overall budget constraints. As an example, consider a transfer program whose objec- tive is to reduce poverty. The process will likely start with an assessment of the poverty level in the country and then the selection of a subgroup of poor households that the program will serve (those who are "deserving" according to the values of the particular society). The first question to address is the affordability of bringing the consumption of the poor to the poverty line. Combining information on the number of poor and their income gap,1 policy makers can estimate the overall resource deficit among the poor. Knowing the magnitude of the overall resource deficit of the poor will inform policy makers whether measures to cover this deficit are affordable or not. Suppose that 16 percent of the popu- lation lives in poverty, the average consumption of the poor falls 25 percent short of the poverty line, and the poverty line represents 70 percent of gross domestic product (GDP) per capita. In this case, the overall resource deficit of the poor is 2.8 percent of GDP (the product of 16 percent 25 percent 70 percent), a crude, lower-bound estimate. The analysts should factor in that certain leakages of funds to the nonpoor are unavoidable. If the program's institutional capacity is low and income-based eligibility is not feasible, the use of more "leaky" mechanisms should be considered. For example, assuming that only 66 percent of the benefits will reach the poor, the overall financing required to bring the poor to the poverty line will rise to 4 percent of GDP (2.8 percent divided by 0.66). Estimating the financial effort required to eliminate poverty, measured either in ab- solute or relative terms, is rarely the end of the story. This initial estimate is often larger than the budget available for the program. Consider that the maximum budget that can be provided for a new program is 1 percent of GDP, not 4 percent. Dealing with this imbal- ance will typically involve an iterative process as illustrated in figure 5.1. 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 131 FIGURE 5.1 Reconciling Needs with Budget Constraints One option is to reconsider a program's generosity. Few safety net programs attempt to top Their average 16% of the up the consumption of consumption is population 25% below the their beneficiaries to the is poor poverty line poverty line. Many pro- grams provide benefits to Reconsider benefit bring beneficiaries up to level and a fraction of the poverty coverage line or to some arbitrary 4% of GDP level lower than the pov- The social safety net is needed to erty line, but the utility of budget envelope is cover this only 1% of GDP this approach is limited. consumption gap Extremely low benefits do not protect beneficia- ries from poverty--that is, they are not cost-effec- SOURCE: Authors. tive--and may not jus- tify their administrative costs--that is, they are inefficient. Peru used such a model in 2004, covering a large frac- tion of the population with a food-based transfer with low transfers per beneficiary unit. Lindert, Skoufias, and Shapiro (2006, p. 26), in their review of the redistributive power of social protection programs in Latin America, characterized the Peruvian safety net model as "giving peanuts to the masses." Not surprisingly, country-specific studies summarized in World Bank (2007m) find that such programs have had almost no impact on extreme poverty or nutritional status. Another option is to restrict coverage of the program or eligibility for the program. If program designers choose to restrict coverage, they may attempt to cover as many of the poorest beneficiaries that can be reached with an adequate level of the benefit given the budget constraint. If only 1 percent of GDP can be allocated for such a program, then the program may target only a quarter of the poor, that is, the poorest 4 percent of the popu- lation. Economic theory suggests that under the circumstances, directing the available resources toward the poorest is the best welfare-enhancing solution, as the marginal value of a monetary unit is higher for the poorest. A variant of this rationing process is to focus on specific vulnerable groups (see chapter 8), or on households deprived in a number of areas, for example, poor and living in substandard housing. If program designers choose to restrict eligibility, they may choose to restrict it to a subset of the poor considered to be deserving. As mentioned in chapter 2, section 4, the notion of deserving poor varies from society to society, but the most common definition is households or individuals who cannot support themselves through work. Thus, some programs may restrict eligibility to families with more than three children and/or elderly people, who together represent about 50 percent of the poor in many countries. Many programs would combine these two options, restricting both the generosity and the coverage of the program typically to the poorest and most destitute. 132 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS In the end, defining the appropriate benefit level is a balancing act: finding a level that is neither too high to generate dependency nor too low to lack impact. If the benefit is too small, the program fails to achieve its objective. If the program is too generous, it may have adverse consequences, such as reducing work incentives or crowding out private transfers, which would diminish or even outweigh its positive impact. Worldwide, pro- grams with benefits that are too low occur more frequently than program with benefits that are too high. OTHER ELEMENTS OF BENEFIT FORMULAS In addition to determining the average size of the benefit, program designers need to decide whether they would like to tailor the benefit level to the characteristics of the household, that is, to use a benefit formula. Deciding on Flat versus Variable Benefit Formulas Benefit formulas may be flat--that is, all beneficiaries receive the same benefit--or they may vary according to the characteristics of the beneficiary household in a number of ways. Some of the main variations include the following: · Benefits vary by family poverty level, with larger benefits for poorer families. · Benefits vary by family size or composition, with benefits determined by the total number of family members or of the number of family members not expected to work. · Benefits vary by the age of family members, for example, benefits tied to educa- tion may be larger for older children in recognition of the higher opportunity costs of their time or to cover the greater number of inputs they need such as textbooks. · Benefits vary by gender, for example, benefits tied to education may be higher for girls in countries with a marked gender gap in schooling. · Benefits vary over time, being higher during the hungry season or the heating season or at the beginning of the school year to cover enrollment fees, uniforms, and shoes. · Benefits vary by region to reflect differences in the cost of living in different areas. · Benefits vary with longevity in the program, tapering down after a certain period as a way to encourage families to leave the program. · Benefits differ in ways that promote certain behaviors even beyond a program's basic conditions. For instance, a CCT program might require school attendance all year to receive the base benefit, but provide a small bonus for good test scores at the end of the year. In general, variable benefit formulas will make a program more efficient, that is bet- ter able to deliver the level of transfer needed to raise most families toward the poverty line and/or induce the desired behavioral changes at minimum transfer cost. However, differ- entiating implies both obvious administrative costs and some less obvious costs because of the complexity involved. Much more effort will be needed to explain the formulas to cli- 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 133 ent families, to the public, and to program monitors and additional effort will be devoted to quality control procedures around the level of benefit determination. Private costs for applicants may also rise. Box 5.2 illustrates how Brazil's Bolsa Familia (Family Grant) pro- gram reconciled these conflicting objectives by choosing a relatively complex formula. Determining the Recipient of the Benefits Most programs around the world and over time have paid either the head of household or whichever member of the household carried out the transactions associated with register- ing for a program. Recently, however, program designers are putting much more thought into who within a household should receive payments. This reflects the growing recogni- tion in the economics literature that households contain members with different needs, preferences, and power and that various members may allocate the funds received differ- ently. The literature generally concurs that women will spend at least as much as men on children's welfare, in many cases more, and are less likely to favor boys over girls in doing so. The strength of this effect varies from place to place and study to study (Haddad, Hoddinott, and Alderman 1997; Quisumbing and Maluccio 2000), but the policy impli- cation is that transfers placed in the hands of women will help children's welfare at least as much, and sometimes much more, than transfers placed in the hands of men. Based on these findings, many new programs, especially CCT programs, explicitly deliver the benefit into the hands of the mother or a proxy for her.2 This is done in the belief that women are more likely to invest additional monies in the well-being of their children than men and the recognition that, on average, in most countries and households, women will be the ones bearing the implicit time costs of obtaining the required health and health education services and likely shouldering a large share of the household chores children would have done had they not been attending school. A smaller number of pro- grams, most often scholarship programs, transfer money directly to students. This is done to help motivate the students to study and ensure that they have as much influence over the money as if they had earned it themselves. Handling Inflation Benefits need to be increased from time to time to protect households from inflation. Many programs only do this in an ad hoc manner every few years and require special legislation or a decree each time. In such cases, the real value of the benefits and the pro- gram's impact usually plummet for a time before recovering. A more desirable procedure is to have a regular, perhaps annual, review of benefit levels as part of the budget cycle, or even an automatic indexing of benefits. In either case, program managers should consider not just how price levels have changed, but also how wages in low-skill occupations have changed. BENEFIT LEVELS IN PRACTICE The question of how generous safety net programs are in practice can be answered in many ways, and probably because of this, little comparative evidence is available. Some of the most common ways to express the generosity of a program are as follows: · By reporting the level of the benefit in local currency. This is not always simple, however, as programs often offer different benefits to individuals or households 134 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS in different circumstances. In such cases, information about the levels of benefits can be presented as a table as shown in box 5.2. · By reporting the level of the benefit in comparable purchasing power (in pur- chasing parity power dollars, for instance). The intent is to facilitate compari- sons across countries, but such information is difficult to compare, because the same type of benefit may be assigned to different assistance units (individuals, families, households) in different countries. Moreover, the adjustments for differ- ences in purchasing power may be insufficient to characterize a benefit as gener- ous or not across countries, as generosity is a relative concept. For instance, in the United States, a benefit of US$10 per person per month would be considered ungenerous, but in a poor country where a large fraction of the population lives on less than US$1 per day, it may be considered quite generous. · By reporting the level of the benefit as a share of the poverty line or other type of indicator, such as the minimum wage, the average wage, the minimum BOX 5.2 The Benefit Formula for Brazil's Bolsa Familia Program Policy makers debated many options for setting the level of benefits prior to launching the Bolsa Familia program in October 2003 (see, for instance, Camargo and Ferreira 2001; Ferreira and Lindert 2003; Lindert and others 2007). Some advocated higher benefit values for families with older children to reflect the higher opportunity costs of their staying in school and the greater risk of older children dropping out of school. Others supported benefits differentiated by gen- der. Still others suggested that benefit amounts vary by region to reflect regional differences in the cost of living. In the end, the program designers opted for a pragmatic set of benefits that (1) was simple to administer; (2) favored the extreme poor; (3) favored families with children, but with limits to avoid promoting fertility; and (4) prevented beneficiaries eligible to receive benefits from previous programs that were being replaced from losing as a result of the reform. This latter consideration was viewed as particularly important politically. Most families actually gained from the introduction of the new Bolsa Familia program range of benefits, as their aver- age value was significantly higher than under the prior programs. For those beneficiaries who received more under Bolsa Familia's predecessor programs because of multiple benefits, the excess amount was maintained under the new program as a so-called extraordinary variable benefit. This extraordinary benefit is to be maintained until those families who receive it no lon- ger qualify for the program benefit. No new beneficiaries will receive the extraordinary benefit. Only 411,579 out of a total of 11.1 million families received the extraordinary benefit. Bolsa Familia provides two types of benefits: a base benefit provided to all families in extreme poverty, regardless of their demographic composition, and a variable benefit that depends on family composition and income. For both extremely poor and moderately poor families, this variable benefit is set according to the number of children in the family (capped at three) and/or whether the mother is pregnant or breastfeeding. As the table shows, income transfers range from R$15 to R$95 (US$7 to US$45) per family per month. The average value of benefits paid during January­May 2006 was about R$62 (US$30). 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 135 pension, the social pension, or the level of unemployment benefits. For ex- ample, in the OECD, the generosity of safety net programs is expressed as a share of the wage of the average production worker in the manufacturing sector (see an- nex). Such comparisons are useful for comparing a program's generosity with the generosity of other programs or types of earnings within a country, for instance, to ascertain whether the benefit level is likely to create disincentives for work. It is less useful for cross-country comparisons. Our preference for comparing generosity is the ratio of benefits to the total consump- tion of beneficiary households. This measure can be estimated using household surveys that collect information on household consumption and the value of safety net benefits received during a certain period. This measure is preferable, as it takes into account many of the complexities of the provision of safety net benefits and transforms them into a single index comparable across households and countries: benefits are implicitly aggregated at the household level, the unit where they are shared and used to finance consumption.3 No. of children aged 0­15 and/or Quantity/type of Monthly Poverty level pregnant or breastfeeding women benefit Benefit Poor 1 1 variable R$15 (monthly per capita family income = 2 2 variable R$30 R$60­120) 3 or more 3 variable R$45 Extremely poor 0 Base benefit R$50 (monthly per capita family income > R$60) 1 Base + 1 variable R$65 2 Base + 2 variable R$80 3 or more Base + 3 variable R$95 SOURCES: Law 10.836 of January 2004 and Decree 5.749 of April 11, 2006. The average value of benefit transfers has fallen from its initial level of R$75 (US$25) at the end of 2003 because the program progressed from initially covering just the extremely poor and then gradually the more moderately poor. Unlike some other safety nets in Brazil, Bolsa Familia's benefits are not automatically indexed to inflation or minimum wage increases. The nominal benefit was held constant from 2003 until July 2007, despite a 16.7 percent increase in the cost of living. In July 2007, Decree 6.157 increased benefit amounts by 17 to 20 percent (depending on the category), thereby restoring their initial value. SOURCES: Lindert and others 2007; www.mds.gov.br/bolsafamilia/. 136 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS This measure, however, does not adjust for one element of program generosity: the dura- tion for which benefits are provided to eligible applicants. We used household- level information for FIGURE 5.2 Generosity of Selected Safety Net Programs, Europe and Central Asia and Latin America and the Caribbean, 55 cash transfer programs Selected Years, 2001­4 from 27 middle-income countries to illustrate how the generosity of these Family allowances programs varies by pro- (n = 15) gram type (figure 5.2). Last resort programs Figure 5.3 looks at 7 CCT (n = 20) programs in 7 countries in Latin America and the CCT programs (n = 8) Caribbean, and table 5.1 presents program-specific Social pensions information on 49 pro- (n = 14) grams in 20 countries in 0 20 40 60 Europe and Central Asia. Transfer as percentage of Figure 5.2 shows key pretransfer household consumption values of the distribution of SOURCES: Tesliuc and others forthcoming; World Bank forthcoming. generosity statistics as well NOTE: n = number of programs.The median value is the line inside the shaded as maximum and mini- rectangle, the 25th percentile is the lower value of the shaded rectangle, and the 75th percentile is the upper value of the shaded rectangle. Programs whose mum values. The median generosity is 1.5 times more than, or less than, the median were excluded. value of benefits as a share of the consumption of re- FIGURE 5.3 Generosity of Selected CCT Programs in Selected Latin American and Caribbean Countries, Various Years cipient households for the programs in Europe and Central Asia is 13.0 per- Nicaragua, Red de Protección Social (2000) cent for family allowances, 13.5 percent for last re- Mexico, Oportunidades (2004) sort programs, 9.0 percent for CCT programs, and Colombia, Familias en Acción (2002) 19.5 percent for social pensions. While the gen- Jamaica, PATH (2004) erosity of family allowance Honduras, Programa de and CCT programs is con- Asignación Familiar (2000) centrated in a narrow in- Ecuador, Bono de Desarrollo Humano (2006) terval, the values are more dispersed for other types of Brazil, Bolsa Familia (2004) programs. The higher gen- erosity of social pensions is 0 5 10 15 20 25 30 not surprising: these are the Transfer as percentage of household income programs meant to sustain households not expected to SOURCE: Based on World Bank forthcoming. work. NOTE: The unit used for Brazil is the transfer as a percentage of pretransfer household income. 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 137 TABLE 5.1 Generosity of Selected Cash Transfer Programs in Selected East European and Central Asian Countries, Selected Years 2001­4 Social pension Family allowance Last resort programs programs programs Other programs Poorest Poorest Poorest Poorest Country and year quintile Total quintile Total quintile Total quintile Total Transfer as a percentage of pretransfer household consumption Albania, 2002 9 16 n.a. n.a. 15 11 14 16 Armenia, 2003 n.a. n.a. 26 18 26 18 24 18 Azerbaijan, 2003 15 14 3 2 3 2 8 6 Belarus, 2002 45 26 20 13 21 14 15 9 Bosnia & Herzegovina, 2001 89 40 n.a. n.a. 27 16 58 30 Bulgaria, 2003 13 8 17 10 24 13 13 9 Estonia, 2004 n.a. n.a. 25 12 45 34 28 13 Georgia, 2002 40 23 n.a. n.a. 122 44 43 24 Hungary, 2002 12 12 40 21 24 19 41 22 Kazakhstan, 2003 n.a. n.a. n.a. n.a. 15 11 27 19 Kyrgyz Republic, 2003 n.a. n.a. n.a. n.a. 12 7 2 3 Lithuania, 2003 43 33 22 13 25 21 28 17 Macedonia, FYR, 2003 n.a. n.a. 53 51 53 51 111 77 Moldova, 2003 18 13 13 9 13 9 18 15 Poland, 2004 65 59 18 13 36 31 31 28 Romania, 2003 21 16 13 7 35 31 18 11 Russian Fed., 2002 32 23 6 3 5 3 16 14 Serbia & Montenegro, 2003 53 41 13 18 12 9 29 32 Tajikistan, 2003 3 7 1 1 2 1 5 7 Uzbekistan, 2003 n.a. n.a. 28 16 22 13 19 11 Median 27 20 18 13 23 14 22 16 SOURCE: Based on Tesliuc and others forthcoming. NOTE: n.a. = not applicable. Generosity is defined as the ratio of a transfer to household consumption. Consumption is cur- rent consumption (less expenditures on durables, housing, and health. 5.2 Managing Work Disincentives One of the most common stumbling blocks to receiving political support for transfer programs is concern about work disincentives or welfare dependency. Specifically, the 138 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS concerns are that beneficiaries will work less because transfers will reduce the pressure on them to work and will also reduce the rewards from working if benefits are reduced when household income increases. The theoretical arguments behind labor disincentives are intuitive (Ellwood 1988). First, any transfer provides unearned income, and thus inherently will reduce the pressure to work. The typical model assumes that beneficiaries will "trade" some of the extra in- come for more leisure.4 Beneficiaries will feel less urgency about taking a job or having all able-bodied household members working. As a result, people will not be as likely to work as they would be in the absence of the transfer program. This is sometimes referred to as the income effect. Second, when the amount of the benefit depends on the recipient's income (the poorer a household, the more money it gets), transfers may change the rewards from work- ing.This situation arises for verified means-tested programs where the benefit level is reduced by a fraction of a currency unit for each additional currency unit in earnings.The implicit tax on earnings is known as the program's marginal tax rate and is sometimes referred to as the price effect. In the hypothetical naïve GMI program where the size of the benefit tops house- hold earnings up to a minimum living standard and is reduced as income rises, the recipient whose initial income is below the guaranteed income has no incentive to work. Thus the theoretical model predicts that the reduction in work effort will be pro- portional to the size of the benefit (the income effect) and the implicit marginal tax rate on earnings (the price effect). The price effect will be manifest only for means-tested pro- grams with accurate and frequent verification of household income. The theoretical model outlined applies in particular to a class of programs and ben- eficiaries in developed countries: verified means-tested GMI programs serving able-bodied households. For these programs and country settings, beneficiaries face 100 percent mar- ginal tax rates, and households that earn less than the guaranteed income in the absence of the program have no rewards for working. To operate according to their theoretical design, GMI programs should be able to perform an accurate means test when someone applies to the program and then monitor the household's income continuously or at regular intervals to adjust the benefit level to changes in earnings. This is feasible only in countries where the informal economy is small and household incomes are monetized, documented, and verifiable. Few programs in developing countries operate under such conditions. The concerns about work disincentives should not be applied indiscriminately to all types of safety net programs and beneficiaries. First, the concern is less pertinent for programs targeted to beneficiaries who are not expected to work, such as social pensions, disability allowances, and, sometimes, allowances for single parents. Second, the model does not fit the typical beneficiary from a low-income country well, that is, a poor, credit- constrained entrepreneur.5 For this type of beneficiary, an injection of cash at zero interest rate may provide the additional liquidity required for a small investment, which may result in an increase in work effort. Furthermore, the model does not apply to a whole class of transfer programs such as workfare, where benefits accrue only if beneficiaries work. EVIDENCE FROM DEVELOPED COUNTRIES Concerns about work disincentives have traditionally been strongest in wealthy countries with generous safety nets and high unemployment rates. However, while the concerns are 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 139 common in most developed countries, most of the research and evidence is limited to the United States. Some evidence of labor disincentives is apparent for generous social assistance transfers with finely tuned eligibility based on verified means testing in the case of two pro- grams: the Aid for Families with Dependent Children (AFDC) and the negative income tax experiments. The AFDC program (box 5.3) was a relatively generous program that provided income replacement for single mothers. It contained specific features that made it especially susceptible to labor disincentives, such as reducing benefits dollar for dollar. BOX 5.3 Lack of Applicability of High Labor Disincentives in the AFDC to Safety Nets in Middle- and Low-Income Countries The AFDC program operated in the United States from 1935 to 1996. A number of nonex- perimental studies reviewed in Moffitt (2002a) suggest that single mothers benefiting from the AFDC reduced their work effort by 10 to 50 percent. The AFDC had a number of design fea- tures that encouraged reduced work effort, namely (1) it was an income-replacement program that provided relatively generous benefits; (2) it imposed a 100 percent marginal tax rate on beneficiaries' earnings; and (3) its target group was single mothers, a group that was expected, at least during the program's early days, not to work but to care for their children. In 1996, Tem- porary Assistance for Needy Families, a program that incorporated many design elements that encouraged beneficiaries to work, replaced the AFDC. Care should be exercised in extrapolating these results to other safety net programs in devel- oped or developing countries where strictly enforced means-tested programs with positive tax rates on extra earnings are rare and programs are less generous. SOURCES: Moffitt 1992, 2002a. The negative income tax experiments implemented between 1968 and 1979 supple- mented the incomes of poor working families by bringing them up to a fraction of the pov- erty line while allowing beneficiaries to keep a fraction of their extra earnings that took them above the poverty line.6 Evaluations found that the negative income tax approach moderately reduced participant work efforts. Male beneficiaries reduced their employment and earnings by 7 percent on average. Among female beneficiaries, only 17 percent of whom were em- ployed, employment and earnings dropped by 17 percent (Burtless 1986). Except for these two programs, the U.S. evidence shows that participation in safety net programs has only small or moderate effects on employment or hours worked. In the United States, most studies have found no evidence of reduced work effort for a host of programs with relatively smaller benefits, such as the Food Stamp Program, nutrition pro- grams, or child care subsidies (Blau 2003; Currie 2003). Studies obtained similar results for in-kind programs, such as housing programs or Medicaid, a program that provides health insurance coverage for the poor (Gruber 2003; Olsen 2003). Outside the United States, the evidence on disincentives to work resulting from safety net programs is extremely scarce. In continental Europe, for example, research has focused on the impact of generous unemployment programs on the work effort of the 140 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS unemployed (LaLonde 2003), not on safety net programs. This relative lack of interest is probably because GMI programs--the main culprit in terms of work disincentives--are small (in the European Union, most of them account for 0.2 to 0.4 percent of GDP and cover only 1 to 3 percent of the total population), while other programs use categorical criteria to cover population groups not expected to work (the elderly, the disabled, chil- dren) (OECD 2004a). Another robust evaluation of the work disincentives of income-tested cash transfers comes from Ontario, Canada. There the beneficiaries of a rather generous safety net pro- gram during the 1990s reduced their work effort by 3 to 5 percent when benefits tripled from Can$185 to Can$507 (Lemieux and Milligan 2008). EVIDENCE FROM DEVELOPING COUNTRIES As concerns developing countries, the theory supports the view that the impact of safety net programs on work disincentives should be smaller in developing countries than in developed countries for the following four reasons: · Many developing countries target their programs only to households without able-bodied adults, and in such cases the arguments are not relevant, for example, the direct support component of the Productive Safety Net Program in Ethiopia and the Kalomo District Pilot Social Cash Transfer Scheme in Zambia, or require able-bodied beneficiaries to work in return for benefits as is the case for all work- fare programs and some cash transfer programs. · Programs are less generous in developing countries. Most safety net programs complement rather than substitute for the earnings of able-bodied beneficiaries (box 5.4). · Verified means-tested programs are rare in developing countries. Few programs use and are able to effectively enforce benefit formulas with marginal tax rates and frequent recertification of household income. · The static model does not take into account that transfers help households make productive investments in their futures. Relatively less empirical research on potential labor-market disincentives associated with transfer programs is available for developing countries. However, as shown in the fol- lowing list, the few studies that have investigated the effect of safety net programs on adult work effort suggest limited labor disincentive impacts: · In Armenia, Posarac, Tesliuc, and Angel-Urdinola (forthcoming) do not find that the beneficiaries of the Family Poverty Benefits Program work less because of the program. The authors compare the employment rate and the hours worked by adults in two equivalent groups, applicants accepted versus applicants denied, with a proxy means test score close to the eligibility threshold using a regression discontinuity design. · In Brazil, Leite (2006b) simulates the potential impact of the Bolsa Escola (School Grant) program on adult work effort and finds that the transfer amounts have little impact. Even a 10-fold increase in the size of unit transfers under Bolsa Es- cola would result in negligible impacts on adult work effort. 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 141 BOX 5.4 Labor Disincentives in Very Low-Income Countries: The Kalomo District Pilot Social Cash Transfer Scheme, Zambia In the Kalomo district of Zambia, a cash transfer scheme supported by the U.K. Department for International Development and the German Agency for Technical Cooperation provides about US$10 per month to destitute households. The US$10 amount is based on the price of a 50-kilogram bag of maize, which enables beneficiary households to have a second daily meal. The assistance has to be meaningful, but not so large as to engender jealousy among those not receiving the support or compromise the scheme's financial sustainability once it is extended nationwide. Households with children get a bonus of US$2.50, reflecting the higher expenses of households with children. For simplicity, the amount is irrespective of the number of children. Evaluations of the scheme have not found that it created any disincentives for households. Ac- cording to one evaluation (GTZ 2008): The amount that households receive per month is only enough to permit them to have a sec- ond meal per day. It thus supplements the little that households can obtain on their own with their limited capacities and is certainly not an incentive to refrain from productive work. Most of the beneficiary households have invested part of their transfers in livestock or agricultural supplies at some point of time, showing that households have a strong interest in generating extra income and engaging in small productive activities. Social transfers of this type should therefore always complement other sources of income but help to avoid negative coping strate- gies. The level of transfers is a question needing careful consideration and testing. SOURCE: GTZ 2008. · In rural Ethiopia, Abdulai, Barrett, and Hoddinott (2005) find that the receipt of food aid is not associated with lower work effort in agriculture, wage work, or self- employment. The negative correlation between food aid and various measures of labor supply that may appear to suggest the existence of work disincentives because of food aid actually reflects the placement effect, that is, that food aid is targeted to the poorest communities and to the less able-bodied members of the community. Once these characteristics are controlled for, the data suggest that food aid leads to increases in the labor supply. · In Mexico, evidence from three studies suggests insignificant labor disincentives for the adults participating in PROGRESA. First, empirical evaluations show that the conditional transfers from PROGRESA did have a significant impact on re- ducing child labor, but had no measurable impact on the work efforts of adults. Specifically, Parker and Skoufias (2000) estimate that PROGRESA increased beneficiary families' average income by 22 percent and decreased children's labor force participation by 15 to 25 percent. They find no evidence of a reduction in labor force participation rates or work efforts by adults. Skoufias and di Maro (2006) find that PROGRESA did not affect the work incentives of adults from ineligible households in villages covered by the program. Finally, Freije, Bando, and Arce (2006) simulate behavioral responses and find that the Oportunidades program, the successor to PROGRESA, does not seem to affect adult labor sup- 142 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS ply. Their simulations show that such transfers would have to be far higher (more than double) before any labor disincentive effects would emerge. · In Romania, a qualitative review of the Guaranteed Minimum Income Program (Birks Sinclair & Associates 2004) finds little evidence of an adverse impact on labor force participation. Two design elements are suspected to have mitigated the work disincentives. One is the work requirements applied to all able-bodied beneficiaries. The other is an exit threshold that is set higher than the eligibility threshold. The evaluator considered that "because there is a small bonus for em- ployment, there may be a small positive impact on participation compared with more traditional systems of aid" (Birks Sinclair & Associates 2004, p. 27). · In Sri Lanka, Sahn and Alderman (1996) study a rice subsidy program that in- duces labor disincentives through income effects. They find labor reductions of approximately 10 percent. OPTIONS FOR MINIMIZING LABOR DISINCENTIVES Policy makers and administrators have a variety of tools at their disposal to minimize labor disincentives as discussed in chapter 2, section 3. We summarize five of them here, focusing on how to minimize labor disincentives generated by the transfer formula (the fourth and fifth options). However, reduced work effort is not unambiguously a good or bad outcome (box 5.5). One option is to limit programs to those who traditionally are not expected to work anyway, that is, the very young, the very old, the disabled, and so on, often referred to as the deserving poor. This is fairly common, but results in only a partial safety net (see chapters 8 and 9). One way to ensure the coverage of all the poor is to complement such programs with a workfare program. Ethiopia's Productive Safety Net Program provides an example. It covers all the extremely poor and combines a workfare program serving those who can work with cash transfers for households without labor resources. A second option is to choose a targeting mechanism that is not tied directly to earnings, which leaves the rewards to working intact. Indeed, few developing countries use a means tests or minimum income guarantee, although many transition countries do. Infrequent recertification will also minimize the labor disincentives. Most programs outside Europe recertify beneficiaries only once every two or three years or less often (see chapter 4, section 4). A third option is to condition benefits for able-bodied beneficiaries on a work test, which is, in essence, what a workfare program does. The application of this principle is not limited to a stand-alone workfare program. Most last resort cash transfer programs in transition economies require economically inactive applicants to register with the un- employment office and actively seek work. Some last resort programs, such as Albania's Ndihme Ekonomika (Economic Assistance) program and Romania's GMI Program, take this principle further by requiring able-bodied beneficiaries to work a certain number of days per month in exchange for benefits (one day per week in Albania and one week per month in Romania). Refusal to work results in stopped payments. In Bulgaria in 2003­5, when a large percentage of GMI Program beneficiaries were long-term participants in the program, the country implemented a temporary public works program, From Social As- 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 143 BOX 5.5 Is Reduced Work Effort Really So Bad? In most studies, the sensitivity of work effort to incentives is much greater for secondary work- ers--married women and children--than for men and is relatively low for men (Deaton and Muellbauer 1980). The reduction of work effort by secondary workers can be a tolerable, some- times even a valuable, outcome. A reduction in child labor is normally viewed as a desirable outcome, not a problem, especially if less work is associated with children being able to devote more attention to their schooling. This has been the outcome of CCTs for which the condition is related to schooling, which have reduced work effort by children and youth, but not by adults, in recipient families. Similarly for women, especially for the poorest women and the mothers of young children, fewer work hours may reduce their caloric expenditure and consequently improve their health and nutrition and/or give them time to provide better care for their children, thereby helping to reduce the intergenerational transmission of poverty (Engle, Menon, and Haddad 1999). Research on children's nutrition in poor countries shows that children may benefit from their mothers working if the child is older than one year, the substitute caretaker is an adult, the wage is good, and the woman controls the money (Smith and others 2003).This is a complex set of conditions that will not always obtain. Where the conditions are not met, a reduction in women's work effort may not always be a bad thing. Even in the United States, evaluations of the Temporary Assistance for Needy Families program reforms show that work effort by women did increase, which is generally viewed as positive, but their adolescent children showed some negative behavioral and achievement effects, apparently because of reduced supervision (Blank 2004). Moreover, the diets of children of working mothers tend to be of somewhat lower quality than those of nonworking women (Crepinsek and Burstein 2004). sistance to Employment, to move the able-bodied among the long-term GMI Program beneficiaries into workfare (de Koning, Kotzeva, and Tzvetkov 2007). By combining last resort programs with a work test, program designers can reap the benefits of both worlds: they gain the freedom to set benefit levels based on poverty reduction criteria without the threat of work disincentives. A fourth option is to set benefits at levels lower than adults can earn by working. Most programs in developing countries have extremely low benefits, often only a small fraction of the poverty line. Thus they inherently leave plenty of incentive to work. The low benefits are usually due more to fiscal constraints than concerns about work disincen- tives, but the result is the same. In countries with a full suite of social protection programs, this rule will translate into lower social assistance payments than unemployment insurance or the minimum pension provided by the contributory pension system. The most widely used mechanism to partly mitigate any work disincentives is to keep benefits substantially lower than the minimum wage as done in Bulgaria and Roma- nia (World Bank 2002b, 2003h) or the earning of low-skilled agricultural laborers as in the Kyrgyz Republic (World Bank 2003g). The conceptually desirable rule is to preserve an incentive structure that ensures that work is rewarded compared with welfare by setting social assistance benefits lower than unemployment benefits and the minimum wage. At 144 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS the same time, anything less than the minimum wage may be barely above a minimum survival standard. As noted earlier, social assistance benefits are rarely enough to ensure survival by themselves, but rather are a supplement that can help achieve a minimum standard of living. In thinking about benefit levels and possible reductions in work effort, the value of any noncash benefits must be taken into account. In a number of countries (for example, China, Jamaica, and Romania), recipients of needs-based social assistance also receive sub- sidized or free health insurance or waivers of user fees for health care. These may add substantially to the value of assistance, and thus to the possible incentive to reduce work effort or income in order to qualify for assistance. A fifth option for moderating labor disincentives is to use a benefit formula that reduces benefits on a sliding scale of less than one currency unit in benefit reduction for each cur- rency unit of increased earnings. This means that additional work effort will raise incomes. In practice, this is achieved by setting exit thresholds higher than entrance thresholds using the withdrawal of benefits as income rises, the provision of earned income tax credits to help make work pay, the provision of lump sum benefits on graduation from the program, and/or the initiation or continuation of payments for allied benefits such as child care or transporta- tion allowances for a period after work starts. These options are admittedly administratively demanding and will result in some of those above the poverty line receiving benefits. This approach is common in OECD countries but less so in developing countries, where benefit formulas tend to be much simpler. Romania's GMI Program adopts a sim- plified version of this approach (box 5.6). Households can also be encouraged to maintain their labor effort by having benefits with defined time limits or benefits that decrease over time. Both these features are incor- porated, for example, in Chile's Chile Solidario and Mexico's Oportunidades program. Finally, another option is to link transfers to program elements such as job training or placement, education, microcredit, or social support services intended to help house- BOX 5.6 Managing Work Disincentives in Romania's GMI Program The GMI Program is based on a simple idea. The GMI (often called the social minimum) is defined according to household size and composition. Households are then entitled to social assistance equal to the difference between the social minimum and their actual income from all other sources, including the imputed income from assets such as land and animals. This means that some household members could be employed in low-wage and/or part-time work and still be entitled to receive social assistance. Indeed, the existence of a working member increases the benefit entitlement by 15 percent (the entry threshold is the social minimum; the exit threshold is the social minimum plus 15 percent). Household members who are able to work are required to engage in community work if they are not otherwise employed. If they do not undertake the assigned community work, they lose their individual entitlement to assistance, although other household members will maintain their entitlements. Thus the GMI Program is a mixture of a negative income tax, social assistance, and workfare. SOURCE: Birks Sinclair & Associates 2004. 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 145 holds move out of assistance and toward independence. These may also be administra- tively demanding, but are fully consistent with broad social policy objectives. Concern about reduced work effort can lead policy makers to consider various fea- tures of program design and of the balance between social assistance of last resort, unem- ployment insurance, and contributory pension programs, but rarely implies abandoning social assistance as a policy tool. 5.3 Enhancing Safety Net Programs to Promote Household Independence Recent discussions about and some innovations in enhancing safety net programs have been aimed at not only providing a transfer, but also other assistance to help households increase their incomes in the near term. Terms such as graduation, emancipation, and pathways out of poverty are used as shorthand for this notion. Helping households increase their autonomous incomes is attractive because it · addresses the underlying problem of poverty rather than simply helping to mitigate it, · explicitly addresses concerns over welfare dependency, · bundles and customizes a variety of services for poor or vulnerable households that may generate positive synergistic effects. This section reviews some of the options and issues for program components other than basic transfers, grouping them loosely into two categories: enhancements that are explicitly part of the design of a safety net program and enhancements that link a safety net program with other interventions. The following three main approaches to making safety net programs more explicitly promotive and not exclusively protective are available: · Minimizing any incentives for dependency. As described earlier, the benefit formula can be adjusted in a variety of ways to encourage and smooth the exit from the safety net program. · Imposing conditions. Receipt of the transfer may be conditioned on behaviors that are expected to help households move toward independence. · Introducing nonconditional links to other services. Beneficiaries are assisted but not required to receive services meant to help them move into independence. TRANSFERS WITH REQUIREMENTS Imposing a condition that households do something that helps them establish an income independent of social assistance is an idea that has been sweeping the world, first in the form of labor activation policies in the United States and Europe, and then by means of the wave of CCT programs in the developing world. Labor Activation Programs Many OECD countries link social assistance benefits and active labor market programs that seek to increase the skills, employment, and long-run earning potential of partici- 146 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS pants through training, apprenticeships, job search assistance, subsidized job placements, and the like. In many cases participation in these is a requirement for the receipt of social assistance. In other cases participation in the labor activation programs is optional, but encouraged. Labor activation programs seek a more immediate increase in household independence than child-focused CCT programs, as they address today's earners rather than tomorrow's. Examples include Argentina's Seguro de Capacitación (Training Insur- ance) program and Bulgaria's From Social Assistance to Employment Program (box 5.7). Some of the most complex versions of such programs in OECD countries offer support for finding child care and provide transport subsidies and some customized assistance in helping identify and overcome individual or household-specific barriers to employment. For example, a caseworker might help potential workers who live too far from public transport or need to get to work at a time when public transport does not operate find car- pools or obtain bicycles or motor scooters. Caseworkers may be able to authorize special one-time payments to cover the costs of licensing or uniforms, or they may help arrange care for elderly or disabled household members. The concept of links to labor activation programs is appealing, but such programs can be complex and difficult to run, and the track record on increasing employment or income is mixed (box 5.8). Some proponents of such programs value the less tangible BOX 5.7 Bulgaria's From Social Assistance to Employment Program In 2002, Bulgaria started implementing its From Social Assistance to Employment Program in response to the increasingly high share of long-term unemployed relying on social assistance among the working-age population.The program's objective is to provide the able-bodied unem- ployed who are relying on social assistance an opportunity to work, earn their own incomes, and be reintegrated into the labor market.The program's main components are providing temporary employment through public works; having participants work in social services in nonprofit, so- cially beneficial activities, such as working in public recreational facilities or assisting people with disabilities; and providing adult literacy and qualification courses in parallel with employ- ment to increase participants' employability. The program provides employers with a subsidy to help cover beneficiaries' wages for up to nine months. The results of an interim evaluation were mixed (de Koning, Kotzeva, Tzvetkov, 2007). The program did indeed provide employment for a considerable proportion of the unemployed who would otherwise have stayed unemployed for much longer. On average, participation in the pro- gram halved the duration of unemployment. The program also generated substantial improve- ments in terms of participant self-confidence, social contacts, and job search motivation. Finally, the program clearly produced outputs and services that were useful to the local communities where projects were carried out. However, the program showed a low gross impact on employ- ment: only 8 percent of program beneficiaries found regular jobs following their participation in the program, compared with 16 percent of a comparable group of nonbeneficiaries. Further- more, some evidence indicated that about 14 percent of the program's projects were activities that would also have occurred in the absence of the program. A survey of employers revealed that the majority did not train the workers either before or during their employment. 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 147 outcome of keeping participants in the main channels of society (social inclusion) even if their earnings are not higher as a result of program participation. BOX 5.8 Global Experience with Active Labor Market Programs Active labor market programs are often targeted to the long-term unemployed; workers in poor families; and specific vulnerable groups with labor market disadvantages, such as young people, old people, and people with disabilities. Especially in industrial countries, the notion of activation of the working-age beneficiaries of transfer programs has received a good deal of attention in a renewed effort to reduce dependency on safety net transfers and counteract their possible labor disincentives. A review of 72 evaluations of such programs leads to the following general conclusions with respect to the most widely used interventions in industrial countries: · Employment services. These services include counseling, placement assistance, job matching, labor exchanges, and other related services. They generally have positive im- pacts on participants' postprogram employment and earnings. Costs are relatively low, so the cost-benefit ratio is often favorable. However, employment services, at least by themselves, are of limited use in situations where structural unemployment is high and demand for labor is lacking. The coverage and effectiveness of such services in develop- ing countries where many labor market transactions are informal are questionable. · Training for the unemployed. Participants often benefit from these programs in terms of higher employment rates, but not in terms of higher earnings. The few evaluations in developing countries paint a less favorable picture. Programs seem to work best with on- the-job training and active employer involvement. Results are more positive for women than for men. · Retraining for workers following mass layoffs. These programs generally have no positive impacts, although exceptions exist.The few successful examples typically include a comprehensive package of employment services to accompany the retraining, but these are generally expensive. · Training for youth. These programs are almost always unsuccessful in improving labor market outcomes, at least in developed countries. Investing earlier in the education system to reduce the number of dropouts and other schooling problems makes much more sense. While few studies for developing countries are available, evaluations in Latin America find positive impacts for programs that integrate training with remedial education, job search assistance, and social services. · Wage and/or employment subsidies. Most of these do not have a positive impact on workers and introduce substantial inefficiencies. Effective targeting may help, but at the cost of reducing take-up rates. · Public works programs. These can be an effective short-term safety net, but public works do not improve participants' future labor market prospects. · Microenterprise development and self-employment assistance. Some evidence sug- gests positive impacts for older and better educated workers; however, take-up is low. SOURCE: Betcherman, Olivas, and Dar 2004. 148 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Conditional Cash Transfers Worldwide, the use of CCT programs has surged in the last decade or so. The typical CCT program links the receipt of the transfer to such conditions as regular school at- tendance for households' children and/or regular health center visits for a defined subset of household members, generally children and pregnant and/or lactating women, though there has been some use of cash to give incentives to other behavioral change (box 5.9). The idea of the CCT is that while the cash transfer alleviates poverty today and compen- sates for the costs of children's education, the conditions help guarantee that the children in the household will have better human capital when they become earners in their own right, and thus be more likely to escape the intergenerational transmission of poverty. To date, the results of impact evaluations have been positive, although they have not yet determined whether the good results depend on the conditions to which the transfers are tied, the actual transfer amounts, or the common practice of designating mothers as ben- eficiaries. For more details on the design of and experience with these programs, as well as the political debate surrounding them, see chapter 7, section 5. Making a transfer payment conditional on certain behaviors by beneficiaries requires that the condition have the desired positive effect on household welfare in the short or long term for many households and negative effects on none or only a few households. The program will need sufficient administrative capacity to handle monitoring of compliance and penalties for noncompliance. Political support for the conditions is also necessary. NONCONDITIONAL LINKS BETWEEN TRANSFER PROGRAMS AND OTHER PROGRAMS AND SERVICES Some transfer programs are experimenting with providing links between their clients and other programs that may help them achieve greater independence. The nature of the link (provision of information, referrals, preferential treatment, or the like), as well as the types of programs that can be linked, will be largely dependent on the circumstances and char- acteristics of the target group. This section presents a nonexhaustive list of such programs and briefly discusses other program options. Documentation and Other Legal Services Lacking the right official identification documents is both a cause and a consequence of social exclusion. Participation in most public programs depends on proper documenta- tion, and transfer programs are no exception. As the poor and the vulnerable are most likely to be lacking documentation and at the same time are the target group for many services, this poses challenges to efficient program operation. Programs that require docu- mentation may well exclude the people who may be most in need, while programs that do not require proof of identity may end up confronting problems of inaccurate recordkeep- ing and possibly fraud. A number of CCT programs are trying to address this issue explicitly by organizing outreach campaigns to inform and help people obtain documentation, even hiring law- yers to help with the process as in the case of Nicaragua's Red de Protección Social. Even without explicit efforts by the program to encourage or require beneficiaries to get their identification documents in order, increased documentation can be a positive externality of transfer programs. The first qualitative evaluation of Turkey's Social Risk Mitigation 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 149 BOX 5.9 Cash Incentives Aimed at Behavioral Change Rather Than Income Support A number of programs are using the idea of giving cash or near cash as a positive incentive for changing individuals' behavior in areas other than children's health and/or education. These programs are not principally aimed at poverty, although the problems they are meant to solve may result in households being poor or facing a high likelihood of becoming poor. The following four main areas stand out. Gender Inequality. In South Asia, a number of secondary school scholarship programs are targeted to girls to address issues of gender inequality through increased education and are explicitly conditioned on the girls remaining unmarried. However, CCT-like programs that aim to improve women's status in society much more broadly, such as the Our Daughters, Our Wealth Program in Haryana, India, are also available.This program provides a series of cash payments to girls from low-income families from the time of their birth to adulthood. Within 15 days of giving birth, the mother receives Rs 500 (about US$12) in cash to meet the baby's immediate requirements for nutrition and medical care. Within three months of giving birth, the program invests Rs 2,500 (about US$60) in the Small Savings Scheme, an amount that will have in- creased to about Rs 25,000 (about US$600) that will be given to the girl when she reaches 18 years if she has not yet married. Finally, for every two additional years that girls delay cash- ing out the benefit and remain unmarried, they receive an additional Rs 5,000 (about US$60) (Population Council 1999). As the program was introduced in 1994, having a good picture of its full impact will not be pos- sible until 2012, when the first group of girls turns 18. However, a 2000 beneficiary assessment (Mode Research Private, Ltd. 2000) already showed promising results: more than 90 percent of community leaders felt that the scheme had helped to reduce female infanticide and gender dis- crimination and had promoted school enrollment, more than 97 percent of community leaders reported increased self-esteem and self-confidence on the part of mothers after getting benefits for their daughters, 97 percent of community leaders felt that the scheme would motivate girls to wait until they were older than 18 to get married, 34 percent of mothers believed that the money would solve the dowry problem, and 77 percent of mothers reported that they were accorded more respect in society. Seventy-two percent of the mothers also reported that the behavior of their husbands and mothers-in-law were better than expected in relation to postnatal care. HIV/AIDS and Other SexuallyTransmitted Infections. A handful of pilot programs in Burkina Faso, Malawi, and Tanzania are being designed to try to reduce HIV infections. The mecha- nism is to make cash payments to individuals who remain free of common and cheaply cur- able sexually transmitted infections. Participants are screened for these often, say once every three months. If they are diagnosed with a sexually transmitted infection other than HIV/AIDS, their payments stop and they are treated immediately and then readmitted to the program. The hypothesis underlying these programs is that they will work through two channels to decrease HIV infections. First, the incentive may lower risky sexual behavior. Second, because HIV trans- mission rates are higher among those with other sexually transmitted infections, treatment will lower the individuals' chances of HIV infection if they have sexual contact with a person infected with HIV. 150 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Project (Kudat 2006) indicated that because the provision of such documents was required for registration in the program, women who had not previously registered their marriages or obtained birth certificates for their children did so. Both administrators and the media recognized the increase in the registration of marriages and the number of birth certificates and citizenship cards as an important impact of the CCT program. Transfer programs can also encourage access to and use of other legal services, wheth- er by design or as a side effect. In the case of the Temporary Assistance for Needy Families (TANF) program in the United States, transfer recipients have free access to the services of the Child Support Enforcement Program, whereas others have to pay a fee. The lat- ter provides assistance with locating absent parents, establishing paternity, establishing a support obligation, and/or enforcing a support obligation. States that want to receive federal TANF funds are required by law to operate a child support enforcement program. This ensures that noncustodial parents contribute to their children's welfare and reduces the burden on the state (U.S. House of Representatives, Committee on Ways and Means 2000). Other Social Assistance Programs Families poor enough to be eligible for one social assistance program are also often poor enough to be eligible for others, and as many countries have many programs administered by many agencies, households may be able to build a substantial package of support by receiving all the assistance for which they qualify. Recipients of a transfer program may, for example, also be eligible for housing and/or utility allowances and/or fee waivers for health or education services. In a few cases, the eligibility determination process is uni- fied, but in many more cases, households must enroll separately with each agency admin- istering a benefit. One of the most common links is between countries' main social assistance benefit and access to health care via a fee waiver or subsidized health insurance payments. In 2001 in Armenia, for example, the government extended eligibility for the health services fee waiver program to all households that were receiving benefits under the Family Pov- erty Benefits Program (Angel-Urdinola and Jain 2006). In Jamaica, families in the PATH CCT initiative are similarly eligible for health care fee waivers, and in Colombia, families in the Familias en Acción program are eligible for subsidized health insurance (World Bank forthcoming). The link between access to health care and cash support is potentially important, as one of the most common reasons that households fall (or fall further) into poverty is because of a health shock that requires expenditures on health care and often implies missed work time and earnings. Social Care Services Vulnerable groups and the chronically poor may be especially at risk for having to deal with issues such as mental health problems, domestic violence, or substance abuse. While these issues are not limited to the poor or to vulnerable groups, they often face many more barriers in trying to overcome them than better-off groups, such as information and re- source constraints. To effectively help these groups address these issues and their implica- tions for overall household welfare, the receipt of income support alone through a transfer program will not suffice. They need adequate access to social care services. 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 151 Linking social assistance and social care can be achieved in a number of ways, rang- ing from providing clients with information on available services to providing them with a social worker who can refer them to particular social care services and coordinate their provision. While the type of intense and ongoing involvement with social workers on which Chile Solidario is based is rare in developing countries, a number of programs do try to use social workers at specific stages of programs' interaction with beneficiaries. In El Salvador, a social worker of a local nongovernmental organization (NGO) contracted for that purpose by the Red Solidaria, a CCT program, will visit households whose members are reported as not complying with the program's conditions. The social worker is respon- sible for determining the reasons for the noncompliance and for helping the household address them. In Brazil, the Bolsa Familia CCT program is starting a similar practice to refer households whose children do not comply with education or health conditions to municipal social assistance centers staffed with social workers to diagnose situations and establish family action plans that may include referrals to appropriate programs. Income Generation Interest in linking safety nets to microfinance is increasing. A number of pilots are cur- rently being run, such as the joint Consultative Group to Assist the Poorest and Ford Foundation programs in Haiti and India and programs planned for Ethiopia and Paki- stan. A number of the larger scale CCT programs, such as Bolsa Familia, Bono de Desarrollo Humano (Human Development Grant) in Ecuador, and PROGRESA in Mexico, are either considering or already starting to implement plans to link their ben- eficiaries to microfinance institutions. The idea is that the transfers from the safety net program are necessary for immediate poverty relief, but that access to vehicles for saving and credit, usually accompanied by some training in financial literacy or busi- ness development, can help beneficiaries raise their autonomous incomes and graduate out of social assistance. Experience to date offers some promising results in terms of the potential for such linked programs to graduate beneficiaries out of safety net programs and into microfinance. Two models have emerged of how transfer programs and microfinance institutions can work together to achieve such results. In the first model, the safety net program pro- vides a substantial part of the beneficiaries' preparation for accessing microfinance. In the case of the Rural Maintenance Program in Bangladesh, eligible women who are recruited to maintain earthen village roads are required to participate in a mandatory savings plan and receive training in numeracy, income-generating skills, and microenterprise manage- ment. They receive information about and are referred to local microfinance institutions. Their participation in the public works is limited to four years, but CARE, which manages the program, continues to provide business management advice for an additional year. Three years after graduating from the program, 79 percent of beneficiaries were still self- employed in microenterprise activities (Hashemi and Rosenberg 2006). The results of a similar public works and microfinance program in Malawi (Central Region Infrastructure Maintenance Programme) have been much more mixed. In this case, the successful link to microfinance institutions was missing, thus the remaining credit constraints did not allow the benefits from the program's savings mechanism and income-generating activities to lead to sustainable microenterprises (Hashemi and Rosenberg 2006). 152 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS In the second model, the microfinance institutions directly engage in the safety net program to provide savings plans and train beneficiaries to become their clients on graduation. In Bangladesh, the Income Generation for Vulnerable Groups Development Program, which distributes food grain to destitute women, organizes the beneficiaries in groups, administers a savings plan, and provides skills training, and thereafter gives them small, subsidized loans for income-generating activities. Nearly two-thirds of the 1.6 mil- lion women who have participated in the program have been able to graduate from pov- erty and become microfinance clients (Hashemi and Rosenberg 2006). Experience to date indicates that key factors for a successful link between a transfer program and a microfinance program are the existence of separate and well-functioning safety net and microfinance programs, clear messages to participants about the role of each, and usually separate administration of transfers and loans. The most promising se- quencing of activities is to give transfer recipients access first to a savings scheme and both numeracy and business training, as well as skills training for income-generating activities, and then follow that with small loans, possibly smaller than usual for microfinance, with eventual graduation into the microfinance institution's regular client pool. Links to other services might also help households achieve stable, independent in- comes or manage their funds and households more effectively, for example, adult educa- tion, financial literacy, and agricultural extension programs. However, little literature is available from developing countries on experiences with links of this kind to transfer pro- grams. Chile Solidario encourages adult education for those with an incomplete secondary education and has resulted in an increase of both the take-up rates of adult literacy and education programs by about 4 to 5 percentage points and of adult literacy rates by some 5 to 10 percentage points after two years of participation in the program. In El Salvador and Ethiopia, extension programs target the same geographic areas as the transfer pro- grams, but not necessarily the same households. OPTIONS FOR LINKING TRANSFER PROGRAMS TO OTHER PROGRAMS AND SERVICES Programs can operate along a continuum of closeness or integration. The following para- graphs provide brief descriptions of a number of these options and examples of programs using them. This area clearly offers scope for innovation, and program designers are cur- rently thinking creatively about further integration of services for the poor. Presumably the more tightly linked programs are, the more likely that individual households will actually gain benefits from both and realize any envisaged synergies. At the same time, this notion has its limits. Tighter links require much more effort to achieve. Moreover, the value of the link will vary from individual to individual; for example, a crop- based agricultural extension program will be of little help to the landless, and adult literacy classes will be fruitless for those who can already read. Thus efforts to create tighter links between transfer and other programs will generally be most worthwhile where beneficiary groups are relatively homogenous or where the services to be linked are expected to be use- ful to most, if not all, beneficiaries of the base transfer program. Moreover, tight linkages should be reserved for cases where the linked programs are functioning well. Experience with both CCT programs and linked social assistance and microfinance programs indi- cates that the transfer programs and the services to which beneficiaries are linked must 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 153 be well coordinated and operate efficiently and effectively independently of each other. Linking two programs that work well on their own will be more promising than linking ineffective programs or trying to make a single program offer different services that require widely different expertise and administrative capacity. Operating in the Same Geographic Area At a minimum, separate services can be delivered in the same geographic areas to serve the same or similar population groups. As this does not require any institutional links or referral systems or much in the way of exchange of information between programs, it is clearly a relatively easy and low-cost way of trying to reach a target group with various services simultaneously. Its disadvantages are the possibility of high errors of exclusion, duplication of certain program functions, and lost opportunities to realize possible syner- gies from an integrated services approach. However, it may nonetheless be a useful ap- proach in resource-constrained, low-capacity environments. Ethiopia's Productive Safety Net Program provides an example of this option. Through its design it is intended to link recipient households not only to transfers via public works or direct support for families with no adults who can work, but eventually to a wider set of initiatives designed to increase food security, including agricultural extension and microcredit. In practice, both the safety net programs and the food security programs are operating concurrently in the same areas, but each program is still finding its feet and tight linkages have not yet been pursued. Providing Information Information campaigns about each separate service can help households become aware of the full range of programs for which they might be eligible. The office of the transfer pro- gram where beneficiaries come to apply for a transfer or payment points where they collect their benefits can make fliers, brochures, and application forms available for a number of other social programs and services. Outreach workers from other programs could also attend gatherings of program beneficiaries. Many programs have done this in a limited or sporadic manner, and an improvement would be to make such information provision more thorough or targeted. An example of this approach is South Africa, where mobile teams from the South African Social Security Agency travel to remote areas to provide information and intake services for all of the agency's programs. These teams often arrange for similar mobile teams from the agencies involved in providing identification documents, health care, and the like to accompany them. Locating Different Services in the Same Office The program offices that beneficiaries need to visit to apply for and/or collect their ben- efits and receive services can be located in a one-stop shop office. This arrangement can help disseminate information and lower beneficiaries' transaction costs, as they will only have to visit one location rather than spending time and money traveling to separate offices for each program. One-stop shops may also increase beneficiaries' awareness of other services they might be eligible for, thereby further contributing to better outreach and higher intake rates. Ideally, locating different programs together does not only mean 154 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS the physical proximity of program offices, but also that staff working for the different programs have similar working hours and a basic knowledge of each others' services so they can better serve beneficiaries and refer them to appropriate programs. This arrange- ment does imply the need for certain structural and behavioral changes on the part of all the programs and services involved. In addition to the logistical requirements of possibly relocating offices to be in the same location and open at similar times, staff from the different programs must be able and willing to communicate with each other and learn about each others' services. Various U.S. states use this approach, and Jamaica is designing a pilot program. Integrating Intake Procedures Harmonized or unified application procedures can do still more. They could result in substantial cost savings, especially in the time staff and beneficiaries spend in dealing with applications, and could also result in better outreach and lower rates of exclusion, as eligibility for all programs is determined by one process. True service integration requires that staff from all the different programs represented agree on and follow the same intake procedures and share information on all cases. An interesting example of a one-stop shop with unified application procedures and shared information systems is the one-window experiment in the Russian city of Arzamas, which has a population of 110,000. Previously, four different agencies had separately ad- ministered 10 different benefits, most of them means tested. In 2002, the one-window pilot introduced a unified application form for all the major social assistance programs in the city. Applicants only had to visit one office and supply one set of documents verifying their eligibility for assistance, regardless of how many programs they applied for. Benefit processing was also consolidated. This reform achieved significant time savings for both staff and beneficiaries, cost savings for the programs' administration, and better outreach (box 5.10). Service integration can, however, have its pitfalls. When the main benefit program for poor families in the United States, TANF, was reformed in 1996, this led to a sig- nificant drop in application rates for two other antipoverty programs, the Food Stamp Program and Medicaid. This occurred because the application procedure for all three pro- grams was integrated, but the increased restrictiveness of the TANF program meant that many households did not apply anymore, and thereby did not have the chance to apply to the other two programs even though they may still have been eligible for them. With only about half of the eligible population receiving food stamps in 1998 compared with about 70 percent in the years prior to the reform, many states launched new outreach efforts, liberalized eligibility criteria for some programs, and simplified paperwork and intake procedures. Having Social Workers Provide Ongoing Support Vulnerable groups or the chronically poor who face multiple barriers to moving out of poverty may best be served by repeated and continued support from a social worker. This does not preclude any of the approaches already described. It does require additional processes for information sharing and referral between social workers and the programs and services their clients are beneficiaries of and/or eligible for. While the use of social 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 155 BOX 5.10 Introduction of a One-Window Approach in Arzamas, Russia This case illustrates the substantial improvements in program efficiency that can be achieved by evaluating its design and implementation. In 2002, the social protection administration of Arza- mas introduced a pilot one-window approach for providing social assistance benefits. The old system consisted of 10 benefits, 7 of them means tested, administered by 4 different agencies. The table shows what kinds of information needed to be verified for the three major benefits and clearly reveals the extent of duplication and the possibilities for harmonization. Housing Child School lunch Client information requiring verification allowance allowance allowance Income from employer X X X Pension income X X X Alimony income X X X Stipend income X X X Housing authority data X X X Payment records for rent and utilities for past 3 months X Employment center data X X X Passport data X X X Verification that child is in school X X Bank account number X X In the reorganized system, applicants can approach one specialist with all their requests for social assistance, fill in one application form, and provide one set of documents to confirm their eligibility for various benefits. Under the old system, clients had to visit four different locations to apply for, appeal denial of, or recertify for benefits, depending on the type of benefit. Under the new approach, all applications are received by small units in each neighborhood of the city and processed by a central unit. A process evaluation documented the savings resulting from the reorganization. First, staff time savings resulted from the unified benefit processing system: staff received 40 percent fewer re- quests for applications, and for every 100 eligible housing allowance applicants, they spent 31 per- cent less time for benefit processing. Second, staff efficiency improved. Under the old system, each staff member processed 85 benefits per month; under the one-window system, each staff member processes 127 benefits per month. Third, total administrative expenditures per benefit application dropped by 32 percent. Clients also experienced a reduction in transaction costs. Un- der the new scheme, the average client saves between 1.3 and 2.4 hours because of the reduced time required to collect the required documents, wait in line, fill out applications, collect benefits, and travel to the local one-window office instead of the former centrally located office. Finally, the experiment resulted in better outreach: 29 percent of clients claimed that they had learned about additional benefits during a recent visit to the one-window office. SOURCE: Gallagher, Struyk, and Nikonova 2003. 156 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS workers would appear to be a costly option, few evaluations are available of the relative cost-effectiveness of using social workers compared with other approaches. Chile Solidario (box 5.11) provides extensive social worker support to families to diagnose barriers to their independence and customize action plans to overcome them. The staff costs are higher than for many other cash transfer programs, but the cash benefit is low, which somewhat offsets the staff costs. The program's impacts are largely positive. Given these results, and perhaps also a growing recognition that the extremely poor and vulnerable face multiple barriers that no one program can address, other countries such as Brazil and Colombia are looking at the Chilean model and starting to emulate it. 5.4 Managing Payments The goal of a payment system is to successfully distribute the correct amount of ben- efits to the right people at the right time and with the right frequency while minimizing costs to both the program and the beneficiaries. Because circumstances differ across and within countries, no one universally applicable delivery system can be used everywhere. The challenge is to select the best delivery system that takes into account the program's needs, local circumstances, and beneficiaries' current and future needs. Box 5.12 outlines the main issues pertaining to the delivery of in-kind transfers, which call for additional expenses and logistical arrangements compared with the delivery of cash. PRINCIPLES AND GOALS Although the choice of delivery mechanism for cash benefits must be context specific, all delivery systems have the following common goals: · Ensuring reliability and regularity of payments. Reliability lowers transaction costs because people do not have to keep returning for the same payment. More- over, beneficiaries are better able to match their income and expenditure flows if they can depend on the benefit to be delivered at the specified time and place. Even small transfers are helpful if they are frequent and reliable. · Maintaining accountability. This means ensuring that all transactions are re- corded and that all the funds allocated are distributed to registered beneficiaries. Effective delivery systems must be able to prevent misappropriation of funds by program officials or by ineligible or fraudulent beneficiaries. If funds are lost along the way, the amount beneficiaries receive will be reduced and political support for the program will be undermined. · Reducing beneficiaries' costs. Beneficiaries face a variety of costs when they col- lect their benefits, including transport costs, fees to maintain and use bank ac- counts, identification card costs, opportunity costs of the time spent getting to the payment site and waiting, and possibly the costs of bribes and fees to receive the payment. Programs must endeavor to make the collection of benefits afford- able for beneficiaries. · Minimizing the cost of delivery (efficiency). This is one of the main goals of any delivery mechanism, as delivery costs can account for a substantial part of 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 157 BOX 5.11 A Transfer Program with Strong Psychosocial Support for Families: Chile Solidario Even though Chile experienced years of sustained income growth during the 1990s, which trans- lated into a reduction in the incidence of overall poverty from 33 percent in 1990 to around 15 per- cent in 2000, extreme poverty remained fairly steady at around 5.2 percent (World Bank 2005g). In 2002, the government introduced Chile Solidario, a program specifically aimed at households living in extreme poverty by using an approach that goes beyond improving the targeting perfor- mance of existing public programs or simply providing households with cash assistance. Program Design. The program is designed to address both the demand side and the supply side of public services serving the extremely poor and has two components. The first compo- nent targets households in extreme poverty using a proxy means test and provides them with psychosocial support through a local social worker for two years. The social workers work with households to assess their needs and help them devise a strategy to exit extreme poverty by identifying specific actions that household members then commit to undertake and which be- come the conditions of the benefit. These specific actions could be any of 53 different so-called minimum conditions for a family to move out of extreme poverty and are grouped in seven categories: identification and other legal documents, family dynamics, education, health care, housing, employment, and income. For the first six months, the degree of interaction between social workers and households is high, starting with weekly meetings for the first two months and slowly decreasing to meetings every other week and then monthly. After these initial six months, social workers meet with households every other month for another six months and finally once every three months for the second year of the program. During this two-year period, households receive a direct cash transfer, with the amount declining over time, and preferential access to a number of other social programs. After the two-year intensive period, households continue to receive a smaller direct cash transfer and preferential access to assistance pro- grams for an additional period of three years, but the social worker services are eliminated. The second component of the program addresses the supply side of public services by ensur- ing coordination among different programs. This is based on the recognition that an approach with isolated, sectoral programs does not address the multiple and interrelated material, as well as psychosocial, deprivation of the extremely poor. The long-term objective is to move toward a system of social protection that provides bundles of programs that are tailored to meet the specific needs of hard-to-reach households. Results. After the first two years of program implementation, significant gains were apparent along a number of different dimensions. In relation to education, preschool enrollment, enroll- ment for children between 6 and 15, take-up of adult literacy programs, and adult literacy all showed significant increases, ranging from 4 to 9 percentage points. In relation to health, en- rollment in the public health system, as well as preventive health visits for children under six and women, showed the most significant increases ranging from 3 to 7 percentage points. The results also showed a strong take-up of employment programs. No significant effects on house- hold income per capita were found. The evidence does indicate that, on average, participants increased their awareness of social services in the community and were more likely to be more optimistic about their future socioeconomic situation. SOURCE: Galasso 2006. 158 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 5.12 Delivering In-kind Transfers Several safety net programs deliver in-kind benefits with the objective of providing beneficiaries with take-home food rations or food that is ready to be consumed. Examples of these programs, discussed in more detail in chapter 7, are food rations, supplementary feeding, school feeding, and emergency food distribution programs. What Is Required? The distribution of take-home food rations presents many challenges, as food is bulky and therefore expensive to store and transport, and is also subject to theft and spoilage. In some countries, the distribution of rations relies mainly on government agencies, as in Bangla- desh and India, that have developed extensive systems of transport and storage facilities. In other cases, the private sector or NGOs may manage some of the distribution of rations. According to the logistics section of the World Food Programme's Web site (www.wfp.org/operations/), when emergency distribution programs are needed, United Nations organizations like the United Na- tions High Commissioner for Refugees and the World Food Programme have "turned the complex business of moving food into a fine art."The site describes the complex logistics needed to deliver the right amount of food to the right people at the right place at the right time. Preparing food for on-site consumption presents additional challenges.Food needs to be acquired, stored, and prepared. In some cases local workers, teachers, health workers, or volunteers are in charge of preparing and distributing the food. Efforts to relieve pressure on local institutions for food preparation include having the private sector deliver food to schools and clinics or using snack foods and products with quicker cooking times (Del Rosso 1999 provides more details on how to simplify implementation for school feeding programs). Where Does ItTake Place? Universal and targeted food distribution programs rely on a network of government or private stores to deliver in-kind transfers to beneficiaries. Special government ra- tion shops were common until the early 1990s in countries such as Bangladesh and Mozambique, but are now rarely used. Instead, private retail shops, which can make use of the existing retail system and marketing chain to reduce distribution costs, are being used more frequently, as in the Arab Republic of Egypt and India. Retail shops are usually paid a fee to carry out the transac- tions and manage the accounts. Programs must monitor the quality of the commodities that are used and ensure that the stores provide the right amount of food to the beneficiaries and do not exchange it for lower-quality products. In recent years, the use of electronic cards, which started in the U.S. Food Stamp Program and is now being incorporated in Mexico's Tortivales (Free Tortilla) program and tested in Egypt and the Indian states of Gujarat and Maharashtra, has facilitated the recording of transactions and increased accountability. Health clinics and schools are used to deliver food subsidies and supplements to women and children to improve their health, educational achievement, and/or school attendance. Thus par- ticipating clinics and schools have to be equipped with adequate storage and cooking facilities to handle the additional requirements of managing the food. Refugee camps or feeding centers are used to provide food to those who are either affected by natural disasters or have been displaced by conflicts. In these cases, the main challenge is to provide those affected with the proper nutritional requirements as they may not have access to any other food (Sphere Project 2004; UNHCR and WFP 1999). 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 159 overall program costs. As the program or its beneficiaries must pay delivery costs, any costs result in fewer benefits available for beneficiaries. To reduce such costs, program designers must look at context-specific factors to achieve the most effec- tive but least costly delivery system possible. While less technologically advanced distribution systems are often less costly in the short run, investing in more mod- ern delivery mechanisms may have long-run advantages. · Ensuring transparency. This entails making sure that both beneficiaries and nonbeneficiaries know the amount of the transfer. Transparency creates confi- dence in the program both for the beneficiaries and for the public at large. It can be enhanced by such simple means as providing public information and having an effective complaints procedure in place. · Ensuring security. The design of a benefit delivery system must address security. Security is an issue both for the program while it is transporting and distributing the benefit and for beneficiaries after they have received their benefits. Program designers must take steps to ensure that the program does not suffer losses as a result of thefts or attacks. Armed escorts may be required, and programs may have to purchase insurance to cover possible losses. This is particularly important in countries experiencing war or that are plagued by lawless groups. DELIVERY AGENCIES AND DELIVERY INSTRUMENT OPTIONS The methods of payment presented in this section cover the different types of delivery agencies and the means of delivering the benefits (table 5.2). TABLE 5.2 Distributing Agencies and Instruments Distributing agency Instruments used by beneficiaries · Banks, traveling banks, ATMs, branchless banking · Cash · Post offices · Bank accounts · Official ration stores, private retail stores · Checks and vouchers · Public agencies and offices, project offices, NGOs · Debit cards, smart cards, and cell phones · Payment centers SOURCE: Authors. Distributing Agencies and Locations As the table shows, a variety of agencies can distribute benefits from a number of locations. Banks, Traveling Banks, ATMs, and Branchless Banking. Banks, whether private or public, can be used to deliver safety net benefits in several ways. They can be used as a payment point where cash is issued to beneficiaries against a list of individuals or families. Banks can also cash checks and vouchers distributed to beneficiaries. Finally, they can maintain accounts in beneficiaries' names in which welfare agencies or programs can deposit cash. 160 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Regulations regarding the establishment of bank accounts vary considerably across institutions and countries. In general, one can distinguish between individual bank ac- counts that can be used for deposits, withdrawals, and so on and consolidated bank ac- counts that group some or all of the beneficiaries of a local branch under one general account. Individual bank accounts provide beneficiaries with greater freedom to make use of any additional features the bank offers; however, they might be too expensive to set up and operate. The additional benefit of a consolidated bank account is that it provides the agency implementing the program with the opportunity to retrieve funds that the intended beneficiaries have not collected after a certain time. Banks also offer a variety of delivery opportunities in addition to their offices to cater to people who live in areas without local branches. They can improve their geographical coverage through traveling banks, ATMs, other payment centers, branchless banking, or cell phone banking. Traveling banks feature bank employees traveling with the cash to be distributed to areas with no bank branches, thereby achieving greater coverage, which results in lower transportation costs to the beneficiaries who would otherwise have to travel to the nearest branch. Bangladesh uses mobile pay stations when beneficiaries of the Primary Education Stipend Program live more than five kilometers away from a local bank branch (Ahmed 2005). One of the disadvantages of this service is that it is more costly for the bank, which may then pass those costs on to either the beneficiaries or the program. Rent seeking and kickbacks are more likely to be an issue with mobile banks because con- trols might be less rigid outside regular offices.7 The potential for security problems during transport and at the payment site also needs to be evaluated. ATMs offer all the advantages of direct payment but minimize opportunities for dis- cretion and rent seeking. Other advantages of ATMs include accountability, automaticity, and potential for low operating costs, as well as the added feature of increased coverage and mobility. Box 5.13 describes possible uses of ATMs and their costs, together with point of service (or point of sale) (POS) machines. A combination of the state of infrastructure (especially electricity), security considerations, and costs will determine their suitability for a particular location. In a growing number of countries, branchless banking, often referred to as mobile banks, provides a new way to deliver money and other financial services to people without bank accounts through post offices and retail outlets such as gas stations. Branchless banks use information and communication technologies, such as debit, prepaid, and smart cards and cell phones, to transmit information between the agent and the customer or the bank. Branchless banking can be operated as an extension of the banking network as in Brazil, India, and South Africa or outside the banking network as in Kenya and the Philippines (Lyman, Ivatury, and Staschen 2006; Lyman, Pickens, and Porteous 2008; Porteous 2006). In Brazil, the state-owned Caixa Econômica Federal offers a simplified current account that can be opened at any branch or correspondent using only an identification card, tax file number, and proof of residence or an address declaration. Account holders have access to Caixa's entire branch and correspondent network (Ivatury 2006). Caixa has created an impressive network of banking correspondents that covers all 5,500 municipalities in Brazil. In 2004, Caixa had about 14,300 banking correspondents that included lottery houses, supermarkets, drugstores, and gas stations, compared with about 9,000 in 2001. Counting all branches and lottery houses, of Brazil's population of 170 million people, 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 161 BOX 5.13 ATMs and POS Machines: Conducting Transactions Remotely Using Electronic Cards An ATM is a computerized telecommunications device that provides a financial institution's cus- tomers with a secure method of performing financial transactions in a public space without the need for a human clerk or bank teller. ATMs range from portable stand-alone units weighing several hundred pounds to steel and concrete wall-mounted units weighing several thousand pounds to mobile units that have been incorporated into varying types of vehicles. The initial cost of an ATM ranges from US$10,000 to US$40,000. In addition, operating and maintenance costs can be as much as US$6,000 per year. Producing and distributing the cards required for accessing ATMs also incurs costs, although these are falling. Usually the program does not bear the costs of the ATM network, only of the cards for its beneficiaries. Sometimes the existence of a large program paying through ATMs will encourage the private sector to install devices in their stores as a way of attracting business. POS machines are communication devices that do not contain any money, but have the capabil- ity of authorizing transactions carried out in retail stores, restaurants, hotels, or mobile locations. Transactions can be performed in real time with connectivity to a central computer system via a telephone or the Internet or can be recorded on a smart card or computer. Shops and welfare delivery agencies can back up the records and then submit copies to banks, post offices, or welfare offices for subsequent reimbursement. POS machines, including hand-held models, can also connect to other devices such as global positioning system receivers, barcode scan- ners, smart card readers, cell phones, satellite phones, biometric fingerprint readers, portable printers, and audio-video devices. Costs range from US$300 to US$700 each, depending on the model. 160 million have ready access to Caixa. Those without ready access live, on average, 24 kilometers from a branch or lottery house. If other correspondent bank outlets, such as supermarkets and drugstores, are included, then everyone has ready access to Caixa (Ku- mar and others 2006). Bolsa Familia uses this impressive network to deliver benefits to its beneficiaries (Lindert and others 2007). Overall, banks have much to offer as a delivery agency. They have considerable ex- pertise in handling and accounting for cash, well-established systems of controls, and audit trails to help minimize fraud along with a management culture that actively discourages fraud. Banks are characterized by automaticity: the use of computers and software that provide internal checks on transfers and payments. In addition, the use of banks ensures that uncollected funds are not subject to theft, for example, Bangladesh's Rural Mainte- nance Program sends uncollected funds back to a central account (Ahmed 2005). Banks also have experience in dealing with security issues. Colombian banks, which are accus- tomed to operating in an insecure environment, took on complete responsibility for this aspect of the Familias en Acción program, thereby helping the program avoid additional insurance costs (Lafaurie and Velasquez Leiva 2004). A side benefit is that beneficiaries gain financial literacy and confidence in dealing with banks if they are exposed to banking by collecting a safety net benefit that is delivered through a bank. Economies of scale are 162 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS attainable with banks because of their cash management experience, and this may permit savings. Further technological developments and increased competition in the banking sector are likely to improve cost-benefit ratios even more in the future. At the same time, using banks as the main delivery agency might entail some dis- advantages. To begin with, banking facilities are sometimes not available or easy to reach, especially in poor rural areas. The cost of poor geographical coverage is illustrated in Mo- zambique, where demobilized soldiers were given vouchers or checks to present at a bank branch or post office. Not all districts had branches or post offices, thus one-third of the beneficiaries spent between US$$2 and US$4 of their US$14 transfer on transporta- tion (Hanlon 2004). Moreover, remote bank branches may not hold sufficient stores of cash and may be difficult to monitor (see Save the Children 2001). The poor may also have difficulties dealing with banks, understanding the transaction, or paying account fees (Ahmed 2005). Bank fees for opening accounts and transaction fees, whether borne by the beneficiary or the program, will reduce the amount of the transfer to the beneficiary. As such fees might constitute a high percentage of the transfer, the program should negotiate as low a fee as possible. The negotiation process itself is sometimes a long and difficult process that requires special skills (box 5.14). Branchless banking promises to offer several solutions to the banking sector's con- straints, but regulations pertaining to its use and the size of accounts and transactions are still being developed. In Brazil, for example, monthly transaction volumes (debits and credits) cannot exceed R$1,000 (approximately US$140) per customer. Clients are al- lowed four withdrawals and four account statements per month, and additional transac- tions are R$0.50 each. Deposits and balance inquiries cost nothing. Post Offices. In several countries, post offices offer financial services similar to those of small banks and are being used as places to pay for goods and services other than postage and may also allow people to make financial transfers and maintain deposits. When used to make payments to beneficiaries of safety net programs, post offices usually make pay- ments based on a list of beneficiaries provided by the program or by cashing beneficiaries' checks or vouchers. Thus in many ways post offices may function like banks and are a good alternative to banks, as they are accustomed to dealing with cash and already have systems for trans- porting, controlling, accounting for, and safeguarding cash. In some countries they offer wide geographical coverage and have established delivery routes and systems, and so they can deliver payments to beneficiaries. In addition, beneficiaries who are unfamiliar with or intimidated by more formal institutions such as banks may be more familiar with post offices, making them more accessible. This is the case in India, where the post offices are used to delivering old-age pensions (Farrington and others 2003). In some cases, despite their coverage, post offices may not be as efficient as banks in transferring resources from central or local institutions. In Lesotho, for example, post offices are used to distribute old-age pensions. To collect their money, pensioners present their pension books, which contain a photograph of the pensioner as identification, at a local post office branch. However, funds are not transferred electronically to local branch- es. Instead, post office officials withdraw the funds from central branches and physically distribute them to 291 pay points. This creates additional costs, including providing these officials with security escorts. Also, the delivery of funds is sometimes delayed and pen- 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 163 sioners find that after paying for transportation or spending hours walking, they have to return later if their payments are not yet available (Devereux and others 2005). Retail Stores. Several programs deliver in-kind transfers to beneficiaries using food stamps or vouchers, for example, the public distribution system in India and the Food Stamp Program in the United States. Beneficiaries receive their allotted amount of food upon presenting the proper documents, namely, vouchers, passbooks, identity cards, or electronic cards. When the program uses cards, official ration shops or private retail shops can authorize or record the transactions using POS machines or terminals (described in box 5.13). The stores are responsible for managing the stocks of food and the financial transac- tions. This entails providing records of transactions to government officials for commodi- ties received from the government or passing the vouchers on to banks for reimbursement of their payments for commodities they have procured. Public Agencies and Offices, Project Offices, and NGOs. Other types of program implementation agencies may be directly involved in distributing benefits. Appropriate public agencies would be those experienced in making cash payments and handling the accounting associated with payments, including local government and welfare offices, as long as they have enough staff to perform all the necessary tasks. In Ethiopia, for example, cashiers from municipal finance departments or local branches of the Department of Ag- riculture disbursed payments for the Cash-for-Relief Pilot Scheme. Municipal safe boxes were used to store the cash. While early concerns about corruption and mismanagement proved to be unfounded, the cashiers viewed having to make the payments an unwelcome addition to their workload, and local officials recommended that in future, cashiers should be hired specifically to work on such projects. Municipal officials also recommended the use of separate safe boxes for any future projects (Save the Children 2001). NGOs can also be used as cash payment sites if they have more extensive networks in a region than banks or government offices or are directly involved in managing a program. NGOs may not, however, handle cash as efficiently as the latter (Ahmed 2005). Other possible payment locations include worksites for public works projects, a logi- cal choice when cash is distributed directly to the beneficiaries. In this case, the workers are paid based on a list of the participants that also indicates the amount of work performed. This system is more convenient for beneficiaries, but poses several logistic and managerial challenges for program managers; for instance, they must make security and other arrange- ments for transporting cash safely to worksites and record all the transactions. Payment Centers. Service point pay stations set up by government agencies or contracted out to private agencies are an attempt to reduce beneficiaries' travel costs. In this situation, government or program officials or contractors distribute transfers using simple paper and pencil recording systems or electronic devices such as POS machines or cell phones. In Zambia, beneficiaries of the Kalomo District Pilot Social Cash Transfer Scheme who live more than 15 kilometers from the designated bank visit the closest local service point, which is usually at a school or health center. Pay point managers are responsible for collecting the money and distributing it to beneficiaries (Devereux and others 2005). This not only makes collecting the money easier for beneficiary households, but facilitates the administration of the scheme and reduces banks' workload. 164 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 5.14 Managing the Contracting Process The social assistance agency often contracts out the processing and delivery of payments rather than managing these directly. This makes a good deal of intuitive sense, as it allows pro- grams to take advantage of the financial sector's expertise rather than having to create parallel structures themselves. However, a few issues need to be managed in contracting out. Demonstration Effect. In countries where a large cash transfer program has not been oper- ated in the past, financial service providers may fear that the services will be extremely expen- sive, and thus may either not bid or submit bids with high prices. In the early stage of Colombia's Familias en Acción, for example, program officials knew that US$1.30 per transaction was a standard cost for an ATM charge and were expecting to pay around US$2 per transaction. The first responses from private banks quoted US$10 per transaction. As the program negotiators were knowledgeable about the banking industry, they understood the implications of offering to deposit the funds five days in advance of the start of the payment cycle. This resulted in a revised transaction fee quote of US$1.74 from the state bank.The program continued to negoti- ate with private banks able to provide coverage outside that of the state bank, and their quoted transaction fees were US$1.31 to US$1.52 (Lafaurie and Velasquez Leiva 2004). Other countries such as Kenya have employed similar strategies of using a demonstration effect, for example, the pilot of the Cash Transfers for Orphan and Vulnerable Children Program, to show that offering the service at a relatively low cost is feasible and then opened the process to bidding. Contract Features and Issues. Contracts need to specify a range of issues other than the financial terms, including ownership of the database of beneficiaries. Ideally the social assis- tance agency should own the database used to deliver payments and maintain transaction records. At a minimum the agency should have full access rights to the database and it should be fully subject to audit if it is managed by another institution, as is the case for Bolsa Familia in Brazil (Lindert and others 2007). Issues relating to the technical features of the database will also need to be specified. The contract should specify desirable but realistic service standards. It might, for example, · give program clients the right to choose which of a number of alternative payment points they wish to receive payments from or give them access to multiple payment points (if Sixty-five percent of the beneficiaries of the Bolsa Familia program in Brazil collected their benefits from lottery points in 2005. Caixa managed approximately 9,000 lottery outlets. These outlets use POS devices connected to the bank network to process transac- tions (Lindert and others 2007). The use of pay points is appropriate where banks are not available and when their use lowers beneficiaries' transaction costs. At the same time, their use requires excellent scheduling and control mechanisms. If the transactions are recorded using paper and pen- cil, the control system could be operated by local observers such as community elders or local government officials who could help safeguard against corruption by ensuring that the correct people receive the cash. The use of electronic systems such as POS machines and cell phones can improve the transparency and accuracy of transactions. 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 165 payments are made through POS machines in retail stores, for example, this will allow at least some beneficiaries to comparison shop and reduce any problems with price gouging); · require a 24-hour hotline for clients to call with questions about payments, lost or stolen cards, forgotten security codes, and the like; · specify maximum acceptable queuing times, which can be reduced, for example, through the use of ATMs rather than teller services, but also by spreading pay days over the month rather than paying all clients on the same day. Performance-based incentive contracts can also be used to improve the efficiency of the deliv- ery system of contractors and other agencies. In Brazil in 2006, Bolsa Familia renegotiated its contract with Caixa for maintaining the database and delivering benefits. The new contract es- tablished 17 performance indicators to measure the level and quality of the services Caixa was providing, such as an index of duplication of registry entries, hours of availability, and delays in delivering benefits cards (Lindert and others 2007). The biggest risk is entering into an agreement with a contractor for services that are not clearly specified or taken into account. For example, if the fees are disaggregated for each of the services provided, some costs might be overlooked. The resulting cost of the delivery system might then become much higher than expected. A 2 to 3 percent fee for each transaction seems reasonable, but if it is combined with other fees for using cards, maintaining the software and the database, and providing assistance to customers, then the fees might add up to be more than 10 percent per transaction. Length of Contract. On the one hand, a longer contract might permit payment agencies to amortize the costs of up-front investments in technology, processing systems, and client train- ing that will both allow low unit costs over time and good service standards. On the other hand, the fewer the overall number of contractors, merely holding a contract for a period begins to create a competitive advantage. The program does not want to end up beholden to a monopoly payment provider. Delivery Instruments Benefits in cash and in kind can be delivered to beneficiaries in a variety of ways, nor- mally using one of two main methods. The first method is to have one of the distributing agencies directly distribute cash or in-kind benefits. The second method is to give the beneficiaries access to the benefits via checks, vouchers, direct deposits into personal ac- counts, smart cards, cell phones, and the like that can be redeemed at one of the distribut- ing agencies. Cash. Direct distribution of benefits in the form of cash is common in low-income coun- tries, particularly in the case of cash-for-work programs or emergency programs. The only thing needed to process payments is a list of beneficiaries or a muster roll. The beneficia- 166 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS ries present some form of identification, a passbook or a checkbook for recording transac- tions, sign some paperwork, and receive the cash. Payments can take place in a variety of places including banks, public offices, and worksites. Direct distribution systems are straightforward to set up. They may be the only avail- able option in poor countries. In the Meket Project in Ethiopia, for example, beneficiaries gather along the road or at the market on the day of payment. Names are called out in groups of 10 and the first person named in the group is given the list that all 10 people sign, usually with a fingerprint, which solves the problem of illiteracy, plus the ink stains prevent duplicate collection. The first person on the list also collects the money for every- one in that group, each beneficiary is informed of his or her transfer amount, and the cash is then distributed (Devereux and others 2005). Direct distribution of benefits by designated agents presents two main security con- cerns. First, security measures are required during the transfer of the funds when banks are not used, and second, systems must be in place to verify the identity of recipients. Bank Transfers. Some programs have made arrangements to transfer cash directly into the bank accounts of individual beneficiaries. Beneficiaries can choose whether to keep the money in their accounts or withdraw it. Direct (electronic) transfers to beneficiaries' bank accounts are advantageous, as they eliminate intermediaries, discretion, delays in payment, and rent-seeking opportunities. For example, if the central government passes funds for a safety net to local governments, any local government facing a financial crisis of some kind may temporarily borrow these funds, thereby delaying their disbursement to beneficiaries; paying beneficiaries directly prevents this from happening. Direct transfers of cash may also be an effective way to deal with security issues. Small, more frequent, disbursements help make beneficiaries less of a target after they collect their benefits. Another advantage of this form of payment is that it introduces beneficiaries to the banking system: it provides them with the opportunity to open and operate a bank account, which they might not otherwise be able to afford. Of course the banks might charge service fees to open and maintain the accounts. To reduce such costs, the program may impose some restrictions on the banks or a number of beneficiaries may share a single account. For example, the Rural Maintenance Program in Bangladesh delivers one pay- ment to 10 people who collect it together. Checks and Vouchers. Instead of delivering cash or in-kind benefits directly, sometimes beneficiaries will receive a check or a voucher that entitles them to receive the benefits in cash or in kind at a later time. A voucher is typically a piece of paper that can be used as a check and exchanged for cash, or it can be exchanged for goods and/or services at designated business establishments. This method of delivering benefits is more common in middle-income countries with more established systems for printing and distributing checks and vouchers. These instruments also require a good system of banks and/or post offices to redeem checks or vouchers for cash or local stores to redeem food vouchers. The lack of an efficient system for redeeming vouchers and checks can undermine the success of a distribution system, as illustrated by a 2001 pilot project in Cundinamarca, Colom- bia. This pilot used vouchers that recipients could exchange for cash or use to buy goods at a designated shop. One of the problems that emerged was that retailers sometimes did not 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 167 have enough cash on hand to honor the vouchers, which resulted in the majority of the recipients being dissatisfied with the arrangement (Lafaurie and Velasquez Leiva 2004). Checks and vouchers share many of the benefits of cash: they are easily transport- able, will not spoil, and do not require a large storage space. At the same time, they have the same security problems as cash (although they are less fungible) and cannot be trans- ferred electronically. They are also more expensive than cash because of the printing costs associated with vouchers that the program must pay for. A risk is that a parallel market may emerge if beneficiaries resell their vouchers at a discounted value, which will also reduce the benefits accruing to beneficiaries. Another risk is that merchants might charge a fee to redeem vouchers or may overprice their goods, and program managers must take care not to work with such establishments. Finally, a nonfinancial cost of vouchers is that those who receive them might be stigmatized. Harvey (2005) notes that a U.K. voucher program for asylum seekers was abandoned because beneficiaries were subjected to abuse and harassment in the community. The success of any voucher program hinges on the logistical arrangements made with merchants for redeeming the value of the vouchers (Harvey 2005). A program's effi- ciency will be seriously affected if merchants are not reimbursed for the vouchers they have accepted or if they are not reimbursed in a timely fashion. The program will also have to address the issue of vouchers that are not completely spent when they are redeemed. Debit Cards, Smart Cards, and Cell Phones. A number of programs are introducing new transaction methods that seek to reduce transaction costs and the use of checks and vouchers. These methods include several types of debit cards and prepaid cards, smart cards, and cell phones. In 1993, the U.S. Food Stamp Program, for example, introduced a system based on smart cards, which is known as electronic benefit transfers, to replace paper vouchers. By December 2002, the system delivered benefits to 90 percent of the 19 million beneficiaries through 145,000 retail stores using POS terminals (O'Connor and Silbermann 2003). Debit cards have a magnetic strip that contains information about a beneficiary's account. They can be used to withdraw cash from ATMs or to process purchases from POS machines. Each time a card is used, the stipulated amount is deducted from the cardholder's bank account. As these are electronic transactions, they are less costly than over-the-counter transactions at local banks, but as a debit card must be linked to a bank account, the account fees may cancel all or some of these savings (Lafaurie and Velasquez Leiva 2004). As debit cards require the use of ATMs or POS machines connected to tele- phone lines, this system may not provide adequate geographic coverage, as the high capital costs mean that equipment will only be installed where it is used frequently enough to make it profitable; however, networks of ATMs and POS machines are expanding rapidly in many countries (Ivatury 2006). At the same time, beneficiaries may find them difficult to understand or use. Prepaid debit cards are similar to regular debit cards. The only difference is that they come with a preloaded value, and when they are used the amount is debited directly from a central account. Thus individual accounts are not required. These cards carry less risk for banks in relation to cash custody because they have a limited amount of money attached to them and can be blocked if lost or stolen. The payment settlement process is faster because the funds have already been allocated, and any ATM or POS machine within the network 168 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS can be used. Security is enhanced because a personal identification number or password is required, plus beneficiaries can choose when to withdraw their benefits, thereby avoiding lines and waiting time (Lafaurie and Velasquez Leiva 2004). The transaction costs for the bank, and therefore for the program, are significantly less for prepaid cards than for indi- vidual debit cards, as the bank does not need to maintain individual accounts. Smart cards contain an electronic chip that can hold, and sometimes also process, a large amount of information. Simple smart cards are disposable once they are used up (prepaid phone cards are an example), but some smart cards have multiple applications. More sophisticated types of smart cards can be used in POS terminals and ATMs and for storing records such as health information (Gallagher 2005). Transactions can take place at ATMs or at POS machines at remote locations. As the required information is embedded in the card, a bank account is not needed and the POS machine does not have to be connected to a bank. Transaction records can be updated later from the card. While the transaction costs of using smart cards are even lower than those of prepaid debit cards, their main disadvantage is the higher cost of individual cards. The current cost of a smart card is approximately US$3, compared with only US$0.20 for a standard debit card. Cell phones have also been used effectively to conduct financial transactions. Indeed, cell phones contain a smart card and can be easily connected over the network of branch- less banks or to other telephones or POS remote devices. The amount of money that cell phones are allowed to carry is restricted, even though many people may have access to their own cell phone that they could use to access a financial institution that manages transac- tions. In the Philippines, special banking regulations allow people to transfer and hold small amounts of cash through cell phones. In South Africa, Celtel allows people to have a cell phone account and conduct transactions within the Celtel network (Porteous 2006). Cell phones can also be used to record the delivery of cash or in-kind benefits by any authorized agent. In the Democratic Republic of Congo, cell phones are used to deliver payments to the beneficiaries of the demobilization project using any of 100 cash points, which are small booths, each with a person sitting inside with a cell phone and a cash box. Beneficiaries provide their government identification to the clerk at the cash point. The clerk enters his or her own identification number into the cell phone and sends a text message to the central financial database operated by Celtel. Ten seconds later, the clerk receives a text message confirming that the money has been credited to his or her account and containing the information about the entitlement and pays the beneficiary. SELECTION OF A DELIVERY MECHANISM The system selected for delivering cash and in-kind transfers to beneficiaries should consist of the appropriate combination of delivery agencies and means of payments and should be well integrated with other operational processes. In addition, the system will need to be feasible enough to be improved over time, while taking current implementa- tion constraints into account. Taking Appropriate Context and Political Economy Considerations into Account Delivery mechanisms that are based on the banking system and make use of ATMs and POS machines in combination with the latest technological innovations, like smart cards 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 169 or cell phones, have significant advantages where they guarantee good coverage and suf- ficiently low unit costs. While each instrument used for distribution has some advantages, not all of them can perform adequately in every circumstance. Therefore the choice of system must take into account country-specific constraints in relation to the financial and technological infrastructure necessary to support the proposed delivery system. An assess- ment of the appropriate context should include the following elements: · A review of the country's financial infrastructure to verify the geographical cover- age and efficiency of the public and private banking sectors and of the postal sys- tem, including investigation of the costs of setting up and maintaining individual and group accounts · A review of the communications infrastructure that includes the availability of electricity, the frequency of power failures, and the availability and reliability of telephone lines and cell phones as well as the costs of using them · A review of the capability of retail stores in relation to redeeming in-kind vouchers and stamps In addition, the system must be compatible and integrated with other operational processes, such as the selection of beneficiaries and the reconciliation of accounts. The information about payments processed, including the number of people who received benefits and the amount of funds disbursed, has to be verified (reconciled) against the list of the program's currently eligible beneficiaries using the program's monitoring system. Therefore the level of development of the overall program's monitoring system will influ- ence the system selected and its development. When a good, computerized management information system and communication infrastructure are in place, programs can make frequent payments using individual accounts and quickly reconcile payments. Box 5.15 shows how a CCT might be delivered using a smart card system. When the monitoring system is not particularly sophisticated, using consolidated accounting and group pay- ment systems to facilitate recordkeeping and the transmittal of information is an easier approach. The principles described here governing the selection of delivery mechanisms also apply to special circumstances such as the distribution of cash transfers under emergency situations--even though, in such cases as emergency situations following a natural disas- ter, the amount of time available for assessing options and designing alternative models may be very limited and the infrastructure disrupted. Box 5.16 describes the most com- mon options available to deliver cash under such circumstances. Assessing the cost of delivering transfers is complicated, and care is needed when comparing the costs of alternative delivery methods. Information on the actual costs in- curred to deliver transfers is not comparable across countries and programs. One reason for this is that total administrative costs are frequently not broken down into similar sub- categories of costs or they are aggregated in different ways. In other cases, the costs na- tional banks or post offices incur are not charged to the government agency responsible for the program or they are extremely low as in Albania and Lithuania. When the delivery of payments is outsourced to private companies and banks in- stead of using government institutions, the actual delivery costs charged to the program tend to be larger in nominal and percentage terms. Sometimes, to reduce both the admin- 170 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 5.15 Processing Payments Using a Smart Card System The flow diagram illustrates a possible way of organizing the activities related to the payment and reconciliation process. 1. The management information system (MIS) creates the list of beneficiaries. This con- tains details about the amount of benefits they are supposed to receive and the type of compliance required. 2. The list of beneficiaries is sent to the bank, mobile system, or service provider from which the beneficiaries will receive payment. 3. Payment takes place. Beneficiaries collect the benefits at a bank or from a mobile system or service provider. For this transaction, they may use a card containing iden- tification data or identifying biometrics. 4. Transaction confirmation is received from the bank, mobile system, or service pro- vider, which is then loaded into the MIS and verified by the program administrator. CCT Program Administrator List of participants 1 with compliance information 4 Authorized MIS Service 2 Provider 2 4 2 4 Can be equipped with · Card reader · Satellite phone Traditional · Biometric reader Banking Mobile ATM System System Can be equipped with · Card reader Eligible · Biometric reader Beneficiaries 3 Holding cards 3 3 This system can be fully automated online and can process the information via the Internet or satellite while transactions are made. Alternatively, transactions can be made offline and up- loaded to the main system on a daily or weekly basis by telephone, satellite, or the Internet. SOURCES: Adapted from Gallagher 2005 and Datta 2006. istrative and total delivery costs of transfers per beneficiary, subsidies are actually delivered every two or more months. Table 5.3 presents the costs of delivery for some countries and programs. When the actual costs of delivering benefits are low per transaction, as in Eastern Europe, payments can be made on a monthly basis. When they are higher, as in 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 171 BOX 5.16 Delivering Cash in an Emergency The delivery of cash--and any other assistance--in an emergency may be more difficult than usual, as financial institutions, basic infrastructure, transportation systems, and communication systems may have been destroyed. Creti and Jaspars (2006) set out the following three main options for transferring cash: · Using the local banking system. If a reliable, preexisting banking system is still in place and accessible to beneficiaries, payments into individual or group accounts have the ad- vantage of being safer for recipients and project staff, because they do not have to handle cash directly. For example, in Iran after the 2004 earthquake in Bam, the banking system was still functioning and the government set up bank accounts for the beneficiaries and transferred cash directly into the accounts. Following the floods in Mozambique in early 2000, the recipients of cash transfers were given checks at designated distribution sites, where a commercial bank provided tellers protected by security personnel who could cash the checks (Harvey 2005). Similarly, the British Red Cross used a bank that provided ATM services to distribute relief transfers in Banda Aceh, Indonesia, following the 2004 tsunami (Adams and Harvey 2006). · Using local money transfer companies. The use of traditional local systems for trans- ferring cash may be an option when financial institutions, especially banks, did not exist before the emergency or were destroyed as a result of it. For example, in response to unrest in Haiti in 2004, Oxfam used local shops to transfer cash to beneficiaries. Local systems for transferring cash were also used in Afghanistan for the Emergency Support for Drought- and Conflict-Affected Populations (November 1, 2001, through March 15, 2003) and in Somalia in 2004 for the Emergency Cash Program for Drought-Affected Households (Creti and Jaspars 2006). In addition, programs can use insurance to reduce the security risks of distributing cash; for example, in Ethiopia, Save the Children takes out insurance to cover the risk of loss when transporting cash in areas that do not have banks (Harvey 2005). · Having an implementing agency make direct payments. Direct disbursement may be an option when the use of banks or local systems is not feasible. It may often be the quick- est delivery method to put in place, although it entails high administrative and management workloads.Various donors working in areas of Sri Lanka affected by the 2004 tsunami used this method for most of their public works programs and delivered cash through direct distri- bution to beneficiaries at worksites (Aheeyar 2006). Latin America, payments are made every month or every other month. When they are even higher, as in Bangladesh, payments are made on a quarterly basis. Political economy considerations should not be ignored when selecting and setting up new delivery systems or modernizing existing delivery mechanisms. The latter might not be easy because of resistance to new methodologies and technologies because they might be perceived as complicated and because they may require a large investment before any benefit payments can be made. A clear explanation is needed of the gains that can 172 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS TABLE 5.3 Costs of Delivering Benefits, Selected Countries and Programs Average Costs per transaction monthly transfer As % of Frequency of Country, program, and year (US$) US$ transfer amount payment Albania, Ndihme Ekonomika, 2004 26.0 0.13 0.5 Monthly Bangladesh, Primary Education 1.80 0.15 8.3 Quarterly Stipend Program, FY2002/03 Brazil, Bolsa Familia, 2007 42.0 1.10 2.6 Monthly Bulgaria, Guaranteed Minimum 25.00 0.07 0.3 Monthly Income Program, 2004 Colombia, Familias en Acción, 2004 50.0 0.60­10 1.2­2.0 Every other month Ecuador, Bono de Desarrollo 15.0 0.45 3.0 Monthly Humano, 2004 Jamaica, PATH, 2004 45.0 0.30­0.60 0.7­1.3 Every other month SOURCES: Authors' calculations based on Ahmed 2005; del Ninno and Ayala 2006; Handa and Davis 2006; Kolpeja forth- coming; Schady and Araujo 2006; Shopov forthcoming; personal communication with Joana Mostafa, consultant to World Bank Brazil office, April 22, 2008. be achieved in terms of security and efficiency and the creation of strategic alliances with systems used by other government departments and the private sector. In Colombia, for example, the Department of Social Welfare is in the process of creating a sophisticated Internet-based database and payment system for the beneficiaries of all social programs. The department was able to reduce the initial setup costs of such as system by working with the private sector. The government helped to set up a payment system for all formal sector employees using a small number of banks or specialized companies that coordinate firms' payroll disbursements with individual employees' contributions to health and pen- sion funds, thereby significantly reducing the cost of processing payroll statements. If international or bilateral organizations finance programs, they might have ad- ditional reporting requirements that could affect the type of delivery mechanism and the frequency of payments. In the case of CCT programs in Colombia, Jamaica, and Kenya, for example, payments take place every other month to give the programs more time to undertake the reconciliation process and reporting requirements. Adapting to Local Conditions and Avoiding Unintended Effects The main challenge in setting up a secure and reliable delivery system for safety net pro- grams is achieving the right balance among ensuring feasibility, reducing program costs, and reducing beneficiaries' costs. In making decisions, program managers should bear the following points in mind: · Managers should, to the extent possible, make use of any preexisting delivery systems and local infrastructure, including bank accounts, databases, and national identity systems. In doing so, programs can take advantage of economies of scale 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 173 rather than increasing up-front capital costs and focus on improving a preexist- ing system rather than setting up a completely new one. In Somalia, for example, money transfer companies are widespread across the country and international agencies have now set up partnerships with the remittance companies for deliv- ering transfers for humanitarian programs (Ahmed 2006). Using or expanding a preexisting identification system can facilitate the verification of beneficiaries' identities and cross-checking of their records across databases. The use of existing systems is also preferred after a natural disaster or other emergency. · Managers should be aware of the trade-offs among alternative options and reach an appropriate compromise. The most obvious trade-offs are between costs and efficiency on the one hand and accountability and frequency of payments on the other. For example, a less expensive system might not be able to verify beneficia- ries' identities properly and give rise to fraud. Similarly, although beneficiaries might prefer more frequent payments to help them smooth their consumption, more frequent payment will result in higher administrative and transaction costs unless the marginal cost of each transaction is kept extremely low. · Managers should select a disbursement system that makes use of locations that are accessible to most beneficiaries, thereby reducing their travel time and costs. Reducing travel and transaction costs is particularly important in poor rural areas. Another alternative is to arrange payments to coincide with other activities, for example, by providing payments at worksites. Finally, managers might wish to consider using a combination of payment providers to reduce transaction costs. · Managers should experiment with new technologies. Introducing new systems such as smart cards and cell phones can improve the quality of delivery systems and significantly reduce costs in the long run and can also be used in low-income environments with poor infrastructure. For example, smart cards, POS machines, and cell phone pay point stations can be used in the absence of telephone lines and electricity. Moreover, even though the initial investment might be substan- tial, if individual transaction costs are kept low, the program's overall costs may decrease over time as occurred with the U.S. Food Stamp Program. · Managers should consider contracting out the payment system and/or involving other supply services. In this case the challenge is to involve as many institutions as possible and then negotiate the contract with them. The negotiation process will be more difficult when only a few institutions can provide the desired coverage. Notes 1. The income gap is the ratio between the average welfare level of the poor and the poverty line. If, as is common in the poverty and safety net literature, the welfare level is measured as per capita consumption, then an income gap of 25 percent means that the average per capita consumption of the poor is 25 percent below the poverty line. 2. An exception is the pilot program being developed in the Republic of Yemen, which will ex- plicitly test the effect of transfers paid to the mother versus the father and of putting a share of the benefits directly into the hands of youth. 174 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS 3. Cross-country comparability is enhanced if the information from different surveys uses the same recall period and the consumption aggregate is comparable, that is, it includes roughly the same components and is adjusted for differences in household size or purchasing power in the same manner. At the same time, analysts should be cautious in undertaking such compari- sons without a good understanding of program eligibility criteria and of possible differences between surveys. 4. The predictions that increased transfers will reduce the labor supply of beneficiary households are based on the static labor supply model (Moffitt 2002a). 5. Poor households in low-income countries face severe credit constraints. They cannot access bank loans at market interest rates because of the small size of their businesses (and thus high overheads for banks), their lack of collateral, the informality of their businesses, and their lack of knowledge about how to deal with banks. Thus they tend to operate on parallel markets served by moneylenders who charge substantially higher interest rates than banks. 6. The United States undertook four experiments: in urban areas of New Jersey and Pennsylvania from 1968­72 (1,300 families); in rural areas of Iowa and North Carolina from 1969­73 (800 families); in Gary, Indiana, from 1971­4 (1,800 families); and in Denver and Seattle from 1970­8 (4,800 families). The negative income tax was a mirror image of the regular tax sys- tem. Instead of tax liabilities increasing with income according to a tax rate schedule, benefits varied inversely with income according to a negative tax rate (or benefit reduction) schedule. If, for example, the threshold for positive tax liability for a family of four was US$10,000, a family with only US$8,000 of annual income would, given a negative tax rate of 25 percent, receive a check from the Treasury worth US$500 (25 percent of the US$2,000 difference be- tween its US$8,000 income and the US$10,000 threshold). A family with no income would receive US$2,500. 7. In economics, rent seeking refers to the misuse of government authority or resources. 5. BENEFIT LEVELS AND DELIVERY MECHANISMS 175 Annex: Generosity of Safety Net Programs of Last Resort in OECD Countries Table A5.1 presents the key design features for safety net programs of last resort in OECD countries (OECD 2007). These programs, commonly referred to as welfare or social as- sistance programs, are noncontributory income support schemes where eligibility does not depend on beneficiaries' employment record or previous earnings. The benefit level is usually set to reflect basic needs in a country, and safety net programs are one of the main instruments of antipoverty programs. Maximum benefit amounts for a single person vary from 5 percent of the average worker's wage in the United States to 34 percent in Iceland. Claimants usually receive additional payments for depen- dent spouses and children depending on the number of children and their ages. In some countries such as Finland, Japan, and the Slovak Republic, benefits may be increased to cover housing, health, or education costs. Comparing the amounts paid for the first person to those granted for additional household members is particularly interesting, because they imply a determination of the relative financial needs of different household members. For a second adult in the household, typically a partner or a spouse, additions to the maximum benefit amount range from zero in Poland to 100 percent of the rate for heads of house- hold in Denmark, Hungary, and Portugal. For children, the range is from zero in Hungary, Iceland, the Netherlands, Poland, and the United Kingdom to more than 70 percent in the Czech Republic, Finland, Sweden, and the United States. Other design elements reviewed in table A5.1 are the presence of income disregards (a portion of the total earned income of the household or other assistance unit not taken into account when assessing income), the benefit withdrawal rate (the rate by which the amount of the benefit falls when household income increases), and some of the safety net benefits not included in the means test. In most of the countries, social assistance benefits can complement (or top up) other incomes, whatever their source. Hence the relative generosity of countries' social assistance schemes cannot be assessed without considering the interaction of social assistance with other benefits and earnings from work. In several countries, recipients of unemployment benefits are explicitly excluded from receiving social assistance. TABLE A5.1 Social Assistance Benefits, Selected OECD Countries, 2005 Maximum amount (% of average wage) Design parameters of means test Possible Children Generosity to top up Determination Head of Spouse eligible for of other Benefit Benefits ex- unemploy- of benefit Categorical house- or part- additional Per Other last resort last resort withdraw- cluded from ment Country levela eligibility criteria hold ner payments child benefits received benefits Income disregards al rate means test benefits? Austria Nat'l average 15 7 4 Rent None 100 Family Yes Belgium Nat'l rates 20 7 Depends 4­9 310 net income per 100 Family Rare on age & year w/ children, 250 number w/o children Canada Regionally 16 12 Depends 4­5 Rent Depends on family size 75 Increases (Ontario)b determined on age & in nat'l child number benefit Czech Nat'l rates 23 16 Depends 13­17 Dependents 16 None Republicc on age & number Nat'l rates Age > 25 32 32 1st child 10 Rent DKr 25,896 if part of 100 No Denmark employment scheme Age < 25 21 Finland Nat'l rates 14 10 Depends 7­10 Rent, health 20% of net earnings 100 None Yes on age & care, work-relat- (max. 1,800) number ed expenses Nat'l rates Age > 25 17 8 Of a single 8 On 100% of earnings 100 Specific fam- parent for 6 months, then 50% ily & housing for 9 months benefits 1st child of a 5 Franced couple 2nd child of 5 a couple Add'l child of 7 a couple Hungary Nat'l guide- Unemployed, & 11 11 None 100 None No lines benefits exhaust- ed, age > 18 (continued) TABLE A5.1 (continued) Maximum amount (% of average wage) Design parameters of means test Possible Children Generosity to top up Determination Head of Spouse eligible for of other Benefit Benefits ex- unemploy- of benefit Categorical house- or part- additional Per Other last resort last resort withdraw- cluded from ment Country levela eligibility criteria hold ner payments child benefits received benefits Income disregards al rate means test benefits? Iceland Regionally Age > 17 34 21 Unemployed, 17 None 100 Child sup- (Reykja- determined age 18­24, liv- port, family & vik) ing w/ parents rent benefits Funeral costs, dental bills, and so on Ireland Nat'l guide- 27 18 3 Adult dependent 18 100 Family Rare lines Rent/mortgage interest payments Regionally Depends on age 20 11 Depends 6 Medical, long- Net earnings of at 100 Yes determined of family mem- on age & term care, occu- least ¥100,080 (up to Japan bers number pational, educa- ¥398,280 for those (Osaka, tion, maternity earning more) Tokyo)c aid, funeral costs Housing costs 3 Korea, Nat'l rates 14 10 Depends on 8­9 Medical care, 30% of income earned 100 Single par- No Rep. of number education, child- under specific pro- ent birth, funeral grams costs, housing costs, & self- support benefits Luxem- Nat'l rates Age > 25 30 15 3 Supplementary 9 30% of payment rate 100 Family bourg adult (not part of family), allow- ance Nether- Nat'l rates Age > 22 25 11 Supplement for 7 None 100 Family & lands single parents housing (continued) TABLE A5.1 (continued) Maximum amount (% of average wage) Design parameters of means test Possible Children Generosity to top up Determination Head of Spouse eligible for of other Benefit Benefits ex- unemploy- of benefit Categorical house- or part- additional Per Other last resort last resort withdraw- cluded from ment Country levela eligibility criteria hold ner payments child benefits received benefits Income disregards al rate means test benefits? Regionally 13 8 Depends on 2­6 Housing benefit 13­29 None 100 Family determined age depending on family situation Norway (Trond- Supplement for heim) heating expens- es & family ben- efit supplement in December Poland Nat'l rates, Permanent 19 0 0 Periodic assis- None 100 Rare social worker benefit tance, temporary discretion for benefit depend- periodic as- ing on family sistance situation Portugal Nat'l rates Age > 17 15 15 7 Adult 10 Upon taking up em- 100 Family & ployment 50% of earn- housing ings for 1 year Slovak Nat'l rates 8 6 1st child 5­12 Health care, 25% of net income 100 Family Yes Republic only + ad- housing, protec- dition if > 4 tive & activation children allowances Spain Regionally Age > 24 23 4 4 4th dependent 3 None 100 Family Rare (Madrid) determined person in house- hold Sweden Nat'l guide- 13 8 Depends 6­10 Medical costs, None 100 None Rare lines, social on age & transport, child worker dis- number care cretion for supplements Housing costs (continued) TABLE A5.1 (continued) Maximum amount (% of average wage) Design parameters of means test Possible Children Generosity to top up Determination Head of Spouse eligible for of other Benefit Benefits ex- unemploy- of benefit Categorical house- or part- additional Per Other last resort last resort withdraw- cluded from ment Country levela eligibility criteria hold ner payments child benefits received benefits Income disregards al rate means test benefits? Nat'l guide- 16 9 5 Supplement 5 100 lines, social from 3rd person worker dis- aged>16 Swit- cretion for zerland supplements (Zurich) 1st child of 9 Housing, basic single par- medical costs, ent, child care Nat'l rates, Age > 24 or 10 6 Family premium 3 £260 for single person, 100 Housing, Yes personal single parent £520 for a couple, council tax & amount + £1,040 for single par- family United family pre- ent Kingdom mium Unmarried; age 8 19­24 United Nat'l rates 5 5 4 Rent Occasional income up 100 Earned Statese to US$120 income tax credit SOURCE: OECD 2007. NOTE: All amounts are shown on an annualized basis. a. "National rates" indicates that rates are uniform throughout the country. "National guidelines" means that national rates are recommended without being strictly enforced, in which case these guidelines are adopted for the purpose of the comparison. Where there is regional variation in payment rates, two approaches may be followed when calculating benefit amounts: the national average is known and used or the comparison relates to one particular representative region, in which case the entry is "regionally determined." b. Basic allowance plus housing allowance. c. The benefit is made up of two parts: an individual amount depending on the age of the child (and sometimes the parent or guardian) and a household amount that depends on the size of the household. d. The benefit is also available for people under 25 with dependent children. e. Amounts shown are for food stamps only. CHAPTER 6 Using Monitoring and Evaluation to Improve Programs KEY MESSAGES Monitoring and evaluation (M&E) systems are the hallmark of good public management. Yet such systems are still rarely used for safety net programs in the developing world. However, a new wave of results-oriented programs, such as conditional cash transfer (CCT) programs in Latin America and the Caribbean and workfare programs in Argen- tina and Ethiopia, have developed and used integrated M&E systems, which in turn have generated robust evidence that the programs are well implemented and are achieving their intended results. These programs demonstrate that strong monitoring systems sup- port credible program evaluations and that both provide feedback for improvements in productivity, effectiveness, and impact. A monitoring system is an essential management tool that regularly supplies information about how well a program is working so that program managers can take action to im- prove the program's implementation. Monitoring is a continuous process that takes place throughout a program's life and should be an integral component of any program. A good monitoring system is comprehensive, actively used, and adapted to the country and pro- gram context. Effective monitoring systems require a strategic focus and political support more than they require costly investments in information technology. They require ad- equate skills, management attention, and funding and take time to develop and mature. Program evaluation refers to an external assessment of program effectiveness that uses specialized methods to ascertain whether a program meets some standards, estimate its net results or impact, and/or identify whether the benefits the program generates outweigh its costs to society. The most frequently used types of evaluation in safety net programs are process evaluation, assessment of targeting accuracy, and impact evaluation. The value added of program evaluation is substantial, but until recently, evaluations of safety net programs have been relatively scarce in developing countries. During the last 10 years, at least minimal assessments of targeting accuracy have become increasing- ly available, and assessments of program impacts have become frequent for CCT and workfare programs, although they are still rare for other types of programs. 6.1 The Value of Good Monitoring and Evaluation Program monitoring matters because it helps managers to adjust program implementa- tion. Colombia's Familias en Acción CCT program provides an illustration. Its moni- toring system uses sample-based site monitoring or spot checks. Every six months, the 181 182 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS program undertakes interviews in 20 municipalities using questionnaires for a sample of participants, program officials, and local government officials covering 400 indicators of various aspects of the program, including enrollment processes, verification of compliance with conditions, payment systems, appeals, and quality of the health education compo- nent. The results indicate which aspects of the program are working well; how much program management varies across locations; and where changes in procedures, training, staffing, and/or other inputs are needed. For example, the monitoring revealed problems with long lines for payment, including people waiting outdoors in the rain; consequently, the program worked with banks to find various ways to address the issue. The monitor- ing showed that some children were not being served continuously, but experienced a gap between the preschool and school portions of the program because of the dates of their birthdays relative to the school year; as a result, the program extended the age limit for the preschool portion to ensure continuous coverage. Also, the monitoring found and the program dealt with a number of areas where staff needed more training for efficient program implementation. Without a monitoring system, financiers, policy makers, and the public will not know if a program is operating effectively and efficiently, and public funds with high opportunity costs may go to waste. Evaluation complements the monitoring system. There are three main reasons for evaluating a safety net program: to improve the program, to inform stakeholders about the program's performance, and to draw lessons for other programs. In general, programs are evaluated in response to a request for information from management or from the supervis- ing authority, for example, a ministry or parliament; as part of a government-wide, results- based management agenda; or to respond to a public concern, for instance, an accusation of corruption or poor management. During the last decade, safety net programs, especially CCT programs, have been at the forefront of a new evaluation culture. Of 36 CCT programs active in 2007, 29 have ei- ther conducted or have plans to conduct impact evaluations with credible counterfactuals (World Bank forthcoming). This is a far greater percentage of programs than have nor- mally undertaken evaluations. Moreover, in several countries the evaluations have been neither simple nor one-time undertakings: dozens have looked at various aspects of Brazil's and Mexico's CCT programs. The new evaluations have been important in making the case for individual programs and for safety nets being part of social policy. In Colombia, Jamaica, and Mexico, for exam- ple, CCT programs have continued across changes in government and as flagship programs, something relatively rare in a region where social programs tend to rise and fall with indi- vidual chief executives or ministers. The evaluations of CCT programs in Brazil and Mexico helped spur interest in such programs elsewhere. When they were reinforced by positive evaluations from Colombia, Jamaica, and Nicaragua, the stage was set for many countries to think more about transfer policies and especially CCT programs. Indeed, the programs' success captured the attention of policy makers in the developed world as well, to the extent that Mayor Michael Bloomberg of New York has financed a pilot CCT (NYC Opportunity) since 2006, and the United Kingdom began a similar program, Opportunity Revolution, in early 2008. Evaluations of social pension programs in Bolivia, Brazil, and South Africa and of public works programs in Argentina, Bolivia, and India have provided evidence about other types of programs that strengthen the case that safety nets are beneficial. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 183 Beyond securing budgets for the programs, the evaluations of CCT programs have been important in refining some countries' programs. Evaluations in 2000 showed, for example, that in Mexico, anemia was not declining among PROGRESA (now known as Oportunidades) program beneficiaries, as expected (Behrman and Hoddinott 2000). This led to a series of investigations and the discovery that the fortification of the food supple- ment provided was such that the bio-availability of iron was less than intended. Moreover, family members tended to share the supplement, and thus the targeted children received less than the intended amount. As a result, the supplement was reformulated and the program's nutrition education component was strengthened (Neufeld 2006). In Jamaica, a 2007 evaluation revealed that the increase in secondary school enrollment promoted by the PATH CCT initiative was disappointing (Levy and Ohls 2007), and the government that took office shortly thereafter decided to raise the pertinent benefits and differentiate them by grade and gender. The demonstration effect of these evaluations has been important. Mexico paved the way for evaluations of CCT programs, and other countries have emulated its lead. Evalua- tions have also spread to other types of programs. The 2004 Mexican Social Development Law requires the evaluation of all new programs and established the National Council for the Evaluation of Social Development Policy. The separate Transparency Law mandates that the results be made public. External evaluations are now done, and the summaries of the results are supplied to program managers, who must inform Congress each year what they are doing in response to the evaluations (Hernandez 2006). Mexico's Ministry of Social Development, which is responsible for Oportunidades and many other programs, has adopted a system of results-based monitoring. It plans to conduct evaluations of five national programs a year and to have installed such a system in half of its subnational agencies within six years (Rubio 2007). Colombia has also been moving to a similarly systematic evaluation culture (Guerrero 1999; McKay 2007). Despite the value of M&E, effective M&E systems have been rare in the field of safe- ty nets until the recent explosion of evaluations of CCT programs. Often a vicious cycle arises whereby when programs are small and/or have low administrative budgets, building the systems needed for good M&E is difficult, but without good M&E, programs may perform badly, or if they perform well, they are unable to demonstrate this to their funders and thus may not garner support for improvements or expansion. By contrast, programs that have good M&E may perform better as a result and establish their case for funds more clearly. We hope that this chapter, aimed at program managers and financiers, will help expand an M&E culture to many safety net programs in the developing world. 6.2 Distinct, but Complementary, Tools Monitoring and evaluation are distinct tools used to assess whether programs are on track and are achieving their intended results. Table 6.1 summarizes the key differences be- tween monitoring and evaluation. Monitoring provides information on how much money is spent, how many ben- eficiaries the program is reaching, and how efficiently the program is serving them. The monitoring information consists of indicators that are compared with targets to assess whether the program is on track. Managers act upon this information to correct and 184 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS TABLE 6.1 Key Differences between Monitoring and Evaluation Item Monitoring Evaluation Frequency Is a continuous, routine activity that should Is an infrequent undertaking be an integral component of any program Coverage Covers most programs Covers a few programs Depth of Identifies whether a program is being Identifies the change in outcomes information implemented as expected or whether the resulting from a program or whether the outcomes of the program show progress or program's benefits are accruing to its not intended target group Cost Involves low annual costs Involves high costs for each study Utility Aims at continuous program improvement Provides information for major decisions and accountability such as starting, ceasing, expanding, or reducing a program SOURCE: Burt and Hatry 2005. improve the program. Unlike monitoring, program evaluations are often one-time efforts directed at answering a few key questions about program implementation, targeting ac- curacy, and/or impact. An implementation or process evaluation investigates whether the program is operating as planned. An assessment of targeting accuracy asks whether the program is successful in reaching the poor. An impact evaluation looks at how the program affects key outcomes--for example, for a workfare program it would estimate how much participants' incomes have increased because of the program. As shown in table 6.1, the findings generated by program evaluations influence key decisions about the program's fu- ture. To preserve the objectivity of the assessment and because program evaluations require specialized skills, external consultants or firms typically undertake program evaluations. Program monitoring and evaluation complement each other in two ways. First, eval- uation provides answers to key questions that a monitoring system cannot address, such as what the program's impact is or whether it reaches the poor. While a good monitoring sys- tem may collect and track information on a program's key outcomes, it cannot determine whether a change in outcomes was entirely due to the program or to other factors. For ex- ample, coffee farmers covered by the Red de Protección Social (Social Protection Network) CCT program in Nicaragua lost 2 percent of their consumption (the value of everything they consumed in a month) between 2001 and 2003, despite receiving a generous benefit equivalent to roughly 30 percent of their consumption. The question was whether this loss was due to lower work effort on their part or to external income shocks. An impact evalu- ation (World Bank 2005m) showed that other coffee farmers, similar in all respects except for their participation in Red de Protección Social, had experienced a 31 percent drop in their consumption because of a severe coffee crisis that affected all of Central America in 2001­3. Program beneficiaries would have experienced a similar drop in consumption in the absence of the program. Only an impact evaluation, which uses specialized techniques to separate the program's influence on outcomes from that of other factors, can estimate the program's true impact. Similarly, a good monitoring system may provide a good de- scription of program beneficiaries, but cannot determine how poor they are relative to 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 185 nonbeneficiaries, as it does not collect information about the latter. Only an assessment of targeting accuracy based on a representative household survey that gathers information on household welfare and program participation can provide a reliable answer. The integration of M&E generates substantial synergies. For example, impact evalu- ation requires good monitoring data for three reasons. First, the monitoring system tells the evaluator whether the necessary conditions for a program to have an impact have been met. For a CCT program, such necessary conditions include that benefits are reaching the targeted beneficiaries and that the program was implemented as designed. The monitor- ing system helps answer these questions by confirming who the beneficiaries are, when they joined the program, what benefits were actually delivered, and beneficiaries' com- pliance with conditions. Second, without a good monitoring system that covers the key parameters of the program, an impact evaluation would measure the impact of a program without knowing what the intervention really was, making any replication or scaling up of that program difficult. Third, having a monitoring system in place that collects out- come information reduces the costs and increases the quality of the impact evaluation. For example, if the monitoring system routinely collects information about the nutritional outcomes of children in a program, the only additional information needed to estimate the program's nutritional impact is information about the nutritional outcomes of an equivalent control group. Most safety net programs have many systems that collect and process information, such as a database of beneficiaries, a payment system; an appeals and complaints system; a quality control and/ or internal audit system; FIGURE 6.1 A Typical M&E System a management informa- tion system (MIS); and financial, accounting, and MIS: personnel systems (fig- includes all databases kept by different program units; ure 6.1). A substantial part generates reports on resources used, outputs achieved, of this information is used and productivity levels; compares indicators across program units, by client at the level it is collected, characteristics, and over time that is, by frontline staff or a specialized depart- ment, and is rarely trans- Process mitted either up or down evaluation the program hierarchy. targeting M&E Service system standards The monitoring system assessment, and impact aggregates a subset of this evaluation information that captures critical aspects of how the program operates and Internal audit, quality control, and how it affects beneficiaries spot check units: and passes it on to upper periodically review critical activities management, which uses related to program objectives it as the basis for making strategic decisions about SOURCE: Authors. 186 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS the program, and to external stakeholders (the finance or planning ministry, parliament, or the court of accounts--for example, the National Audit Office in the United Kingdom or the Government Accountability Office in the United States) for accountability reasons. In addition to the information the program routinely generates, both upper management and external stakeholders require information about how the program affects its beneficia- ries, which is typically obtained from specialized data collection procedures--interviews, focus groups, satisfaction surveys, multitopic household surveys--used to evaluate differ- ent aspects of the program. 6.3 Development of an M&E System The design of a good M&E system should start with a thorough understanding of what the program is trying to accomplish, that is, its goal or mission, and how it uses its inputs to generate its outputs and achieve its intended outcomes. Such an understanding shared by all key program stakeholders is essential to ensure that all the information required for decision making is collected, analyzed, and used and that extraneous information is not considered. This section discusses the development of an overall M&E system. UNDERSTANDING THE PROGRAM'S OBJECTIVES The first step in developing an M&E system is gaining a clear understanding of the program's objective and strategy, and the tool for this is developing the program's logical framework. The logical framework is a graphical representation of how the program is expected to help improve the condition of its beneficiaries alongside a results chain: the program uses certain inputs that are processed to generate outputs that influence certain outcomes among program participants. Inputs are resources a program uses to deliver cash or in-kind transfers, for example, employees and financial resources. Outputs are the amount of cash, goods, or social services a program delivers during the reporting period, such as the number of beneficiaries served or the number of workdays provided by a work- fare program. Outcomes are events, conditions, or behaviors that indicate progress toward achievement of a program's mission and objectives. A common outcome indicator for a safety net program is the increase in beneficiaries' incomes. The main difference between outputs and outcomes is that program outputs are fully under the program's control, while outcomes are something that the program is contributing to, but that are also influ- enced by factors external to the program. Annex 6.1 provides some concrete examples of information and indicators to be collected for different types of safety net programs. The logical framework will identify other external factors that determine these outcomes and the risks that may impede achievement of these outcomes. We illustrate the use of the logical framework for the development of an M&E plan using Ethiopia's Productive Safety Net Program (PSNP) as an example.1 The program finances public works for able-bodied individuals living in food-insecure districts and cash transfers for poor households living in those districts unable to work. The public works component, in turn, creates community assets to improve the livelihoods of the poor. A review of the program documents reveals the logical hierarchy of its objectives, from pro- gram outputs to intermediate and final outcomes, as illustrated in figure 6.2. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 187 FIGURE 6.2 Logical Framework for Ethiopia's Productive Safety Net Program Final outcomes · 1 million chronically food-insecure households graduate to food-secure status · 2 million more households have improved food security Intermediate outcomes · Food availability improved through increased on-farm production and productivity (food and cash crops and livestock) · Food access improved through increased income from sales of cash crops and livestock and nonfarm income generation · Food utilization improved through health and nutrition interventions (mostly outside the scope of the PSNP) · Vulnerability to shocks decreased through asset protection and promotion and timely safety net interventions PSNP Other programs Household assets protected and community assets built in · Resettlement Program 263 food-insecure districts out of 550 in the country through · Income Generation Program timely and consistent safety net activities · Other Selected outputs · Households with available labor participate in public works based on local priorities and economic opportunities · Food or cash is provided in a timely manner · Households without labor are supported with direct transfers of food or cash · Shocks are mitigated with the timely delivery of food or cash Inputs · Public works, cash transfers, administrative costs SOURCE: Food Security Coordination Bureau 2004. However, M&E planning requires a level of detail well beyond the summary form of figure 6.2. Ethiopia's Food Security Coordination Bureau further elaborated the presenta- tion of the PSNP's logical framework, particularly at the output, outcome, and impact levels as shown in table 6.2. The program logical framework should also specify the main assumptions and risks that may affect the program, and the realization of these assumptions and risk needs to be monitored as well, because the program's success is based on the accuracy of certain assump- tions and the absence of certain risks. For the PSNP, the key assumptions were as follows: · Infrastructure and services, such as health, education, and roads, provided by the government but not by means of the PSNP will be supplied to rural communities at sufficient levels to adequately support the food-security status of households living within the communities. 188 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS TABLE 6.2 Logical Framework and Selected Output and Outcome Indicators for Ethiopia's PSNP Objective Indicator Impact 1 million chronically food- · Percentage of households with no food gap, that is, they have insecure households and sufficient food to meet their needs for all 12 months of the year, 2 million vulnerable households including support provided by the program attain food security within three · Percentage of households in need of food assistance over a three- to five years year moving average Outcomes Chronically food-insecure · Percentage of program beneficiaries who report 12 months of food households have ensured access from all sources including the program food consumption during the · Average number of months of household food shortages covered program perioda by the program Household assets protected · Percentage of the average change in asset levels of chronically (households' short-term food-insecure households vulnerability to shocks · Percentage of households reporting distress sales of assets reduced) · Percentage of households reporting consumption of seed stocks Community assets used · Percentage of households reporting satisfaction or direct benefits productively and managed in a from the community assets developed sustainable manner · Percentage of households regularly using three or more community assets developed by the program · Percentage of public works for which an ongoing management mechanism has been established Markets stimulated through the · Percentage change in the number of traders/retailers in local shift from food to cash markets · Percentage change in the volume of grain trade · Diversity of goods available in local markets Outputs Public works Appropriate payments (food · Percentage of participants receiving food and/or cash resources and/or cash) delivered to per month versus the planned number supposed to receive food targeted beneficiaries in a and/or cash timely and predictable manner · Percentage of food and/or cash delivered per month versus the amount that was planned to have been delivered · Percentage of districts completing 70% of distributions by end July · PSNP staff will have appropriate capacities, materials, and financial support to carry out the program's activities in a timely and effective manner. · Graduation from the program for most households during its three- to five-year time frame assumes an absence of extraordinary food crises. A severe crisis would slow the progress of many households toward food security. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 189 Objective Indicator Targeting undertaken according Percentage of community members who understand targeting to established procedures criteria Community assets Appropriate and good quality · Number of public works constructed, including kilometers of roads public works constructed constructed or maintained per targeted district · Number of structures constructed per targeted district (health posts, classrooms, grain stores, market structures, latrines) · Percentage of public works that conform to established standards Soil conservation measures · Hectares of degraded cropland and rangeland rehabilitated promoted and/or installed; · Hectares covered by soil and water conservation measures degraded areas rehabilitated · Number of tree seedlings planted · Number of communities participating in training and/or environmental rehabilitation Small-scale irrigation and · Hectares of agricultural and pasture land reclaimed per targeted district water harvesting developed, · Number of irrigation and water harvesting schemes developed per district improved, or established · Amount of land cultivated by small-scale irrigation HIV/AIDS awareness campaign Number of households receiving HIV/AIDS awareness training Management systems for · Percentage of communities with guidelines or bylaws developed community assets established for the management and protection of community assets · Number of visits to sites by a technical task force team per district per year · Percentage of local, district, and regional monitoring reports on actual versus planned activities delivered on time · Percentage of districts where the M&E plan is fully understood and implemented Direct support Appropriate food and/or cash · Percentage of participants receiving food and/or cash resources assistance provided accurately per month versus the planned number supposed to receive food to targeted beneficiaries in a and/or cash timely manner · Percentage of food and/or cash delivered per month versus the amount that was planned to have been delivered · Percentage of districts completing 70% of distributions by end July SOURCE: Food Security Coordination Bureau 2004. a. Disaggregated by male- and female-headed households and by direct support versus public works beneficiaries. · Government decentralization will continue, and staff positions at all relevant ad- ministrative levels will be filled with capable people who will remain at their posts. At the same time, the success of the PSNP may be threatened by certain risks, such as environmental impacts resulting from public works or negative market effects resulting from food distribution. These risks should also be monitored. 190 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS DEVELOPING THE M&E PLAN Typically, the development of an M&E system requires drawing up an M&E plan, testing and fine-tuning the proposed system, and having internal or external experts undertake periodic reviews. The program's management can develop an M&E plan based on the logical framework. The plan will identify the information needs of the main stakeholders, the indicators that will be tracked to respond to their needs, the methods for analyzing the data, and the use made of the results and by whom. During this process, management will identify what information is available internally from the MIS, spot checks, and the like, and what information needs to be generated by means of special data collection and/ or analysis techniques such as specialized evaluations. The outcome of this process is a written document, the M&E plan. The plan sum- marizes the agreements reached among program staff and key stakeholders and describes succinctly how the activities will be carried out. The M&E plan should include a section describing the institutional setup of the M&E system, the composition of the M&E unit, the unit's overall tasks, and the unit's upward and downward links and links with other stakeholders, and a diagram showing the types of information that will be collected and to whom it will be provided. A good plan will cover the following aspects: · The information needs of each stakeholder about the program's key outcomes and outputs · A few key indicators for each objective that should follow the logical chain of service provision, utilization, coverage, impact, and efficiency · The information sources and data collection tools for each indicator, including fre- quency of collection, agency responsible, sampling system, and data storage and documentation system along with who needs the data and how they will be used · The data analysis plan describing how the data will be analyzed; how often; by whom; how the results will be presented, discussed, and used for internal program decision making; and what information will be shared with external groups · The resources that need to be dedicated to M&E, such as the staffing of an M&E unit and the training needs required for the personnel involved in collecting, re- porting, and analyzing the data A good M&E plan is the result of a participatory process that involves various pro- gram stakeholders, including experienced program staff; experts from the institutions that finance, regulate, and supervise the program; donors; nongovernmental organizations ac- tive in the field; and M&E experts. A process commonly used for achieving a mutual understanding among these parties is known as participatory planning. The program team will invite these stakeholders to a planning workshop to produce the M&E plan. To this end they work, ideally under the guidance of a facilitator who un- derstands M&E concepts, to develop the various elements of the plan as already described. The stakeholders' input and collaboration is important to the success of the M&E plan, measured by whether managers and stakeholders use the information the M&E system provides to improve the program's operations. Not involving all key users of information will result in an M&E system that will not provide feedback on some aspects of the pro- gram, that may allow potential issues in implementing the program to go undetected, and/ 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 191 or that may not produce the hard evidence needed to mitigate any negative perceptions about the program. Most experts recommend developing the M&E function from the beginning of the program. This means providing resources for an M&E unit; applying an M&E lens to program operations to ensure that the information generated by different departments and units feeds into the M&E system seamlessly and continuously; including in the job de- scription of relevant program staff the collection, processing, and reporting of information related to their functions; and training this staff. Having an M&E system operating from the outset has a number of advantages, including the collection of baseline indicators on the characteristics of beneficiaries before the intervention, especially on those intermediate or final outcomes indicators that the program is trying to improve, and ensures a complete time series of data on critical indicators to track program performance. Most M&E systems are, however, built incrementally. Some programs, such as Brazil's Bolsa Familia (Family Grant) program, started with an inventory of the informa- tion collected by its precursor programs and then improved or expanded the system. Pro- grams that did not implement an M&E system from the beginning may feel the need or pressure from stakeholders to introduce one later. Whenever management or stakeholders feel they need to do this, they will use the same principals as described in this chapter; however, failure to develop the M&E plan at the beginning of the program will result in the absence of baseline information, which is critical for estimating the program's impact. COLLECTING M&E DATA A results-oriented M&E system requires a combination of data collection techniques. Administrative data can provide most of the input, process, and output data, as well as performance and efficiency information. Information about program outcomes and coverage requires survey data. Information on client and staff perceptions and activities undertaken by the program may require the use of rapid appraisal methods or participa- tory techniques. In designing the M&E system, the team should undertake an inventory of available databases (administrative data) and surveys (for instance, household surveys and public expenditure tracking surveys) and identify what indicators are readily available or could easily be estimated from these sources and where additional investments in data collection are required. Administrative Records Most safety net programs routinely collect and process administrative data, especially on inputs, processes, and outputs. Human resource departments manage information about program staff and accounting, and financial departments track financial flows including administrative costs and the value of program benefits. Other systems track information about the different activities performed by the program, such as eligibility determination, recertification, payments made, and verification of compliance with program rules (for both staff and clients). Many safety net programs do not make maximum use of administrative records, which are by far the most available source of data for monitoring. Most programs do not fully exploit such data to monitor program performance and limit its use to tracking trends in inputs and outputs. Some programs do not disclose such information, and some 192 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS do not collect such information to avoid accountability issues. Many programs do not invest in developing a well-conceptualized monitoring system or in the necessary infra- structure (dedicated staff, procedures, and information technology) and operate systems that may gather information lacking in relevance and validity. Whenever possible, monitoring indicators should be derived from administrative data generated by the program, as this information has a comparatively low marginal cost. Such data are cheap whenever such records exist or can be readily modified. They are also more likely to be used by program personnel as they are familiar with the data collection system and with the procedures for transforming raw data into indicators. Such information can be combined to generate performance indicators to monitor the efficiency or productivity of different program units and can be estimated by combin- ing output and input information or by comparing process indicators over time or against a benchmark. Examples of such indicators are the number of beneficiaries served per pro- gram staff member, the level of administrative costs for every US$1,000 transferred, and the number of applications processed in less than a week. The use of administrative data has several general advantages and disadvantages (table 6.3). Agency records usually do not provide sufficient outcome data, thus tracking outcome information often requires special data collection efforts. This is why identifying agency records that provide outcome information is important. For instance, CCT pro- grams collect information about beneficiaries' compliance with conditions from service providers, such as the percentage of infants immunized and the share of school-age chil- dren with adequate attendance records or good grades. Checking the quality of administrative data is good practice, as such data often will be incomplete, old, or unreliable. Some form of quality control is required to assess if the data cover all relevant units such as households and that information is entered accurately. Formulating clear instructions and definitions of what data need to be collected and how the indicators should be computed is also good practice. The information generated by the administrative system is not always the information managers need, and thus data collection systems may need to be upgraded. Romania's Guaranteed Minimum Income Program, for instance, monitors the number of transactions that take place in a given TABLE 6.3 Advantages and Disadvantages of Using Administrative Data to Monitor and Evaluate Programs Advantages Disadvantages · Provide detailed information about clients, · Quality of administrative data and their potential program outputs (transfers or services), and utility vary considerably outcomes · Regular and systematic checking for data · Provide longitudinal data on beneficiaries for quality is seldom performed some programs · Standardized data collection procedures may · Generate data at low cost for multiple program not be followed across program sites years · Privacy and permission issues may delay data · Accommodate changes and additions to data access and transfer more readily than, say, a household survey SOURCE: Heinrich 2004. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 193 period, but not the number of beneficiaries (more than one transaction per month for a beneficiary is a possibility) (Pop, Florescu, and Tesliuc forthcoming). Thompson (2004) reports a similar issue in relation to the Russian Federation's Child Allowance Program. Beneficiary Surveys and Citizen Report Cards Beneficiary surveys and citizen report cards are relatively simple tools used to obtain timely feedback from clients. Beneficiary or client satisfaction surveys assess a program's performance based on client experience with it. Such surveys are an important source of information about service quality, outcomes, client constraints in accessing public ser- vices, and responsiveness of government officials. For instance, clients can rate specific service quality characteristics, such as hours of operation, waiting times, and helpfulness of personnel. The survey can also collect information about intermediate outcomes, such as actions beneficiaries have taken because of the program and overall satisfaction. Such surveys need to be specific in relation to the constraints clients face in using the program as well as in asking for suggestions for improving services. Collecting information on a few key client characteristics (income, employment, housing, gender, age group, and the like) to disaggregate the results by group is common practice. Questionnaires are typically tailored to a program's specific organizational setup. The program M&E unit designs the terms of reference for the survey and subcontracts the work to a survey firm. Nongovernmental organizations and think tanks in several countries have made use of citizen report cards. Similar to service delivery surveys, they have also investigated the extent of corruption encountered by ordinary citizens. A notable feature has been the widespread publication of the findings. Representative Household Surveys A survey of program beneficiaries cannot reveal whether a program covers its target group adequately or whether the benefits are adequately targeted toward the poor. To answer such questions, representative household surveys are needed. Two of the most common surveys found in developing countries are multitopic household surveys and core welfare indicator questionnaires. Representative household surveys have three key advantages: (1) they generate esti- mates that are representative of a particular population of interest, for instance, the benefi- ciaries of a given program or a program's target group, with known precision; (2) they cost significantly less than a census of that population; and (3) they provide data about a wide range of client and program characteristics. For analyzing the effectiveness of safety net interventions, these surveys should col- lect information about program participation, household conditions that determine eligi- bility for the program, and outcomes that the program seeks to influence. In some cases, such surveys will not adequately capture information about programs with low coverage. To obtain representative estimates about such programs, oversampling of the program target group is necessary. Household surveys can be used to analyze individual programs by revealing the char- acteristics of program beneficiaries; the incidence of program benefits across income dis- tribution or other groups; and the extent of take-up, for example, if it is incomplete and why. Such surveys often collect information on participation in a number of safety net 194 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS programs, generating economies of scale in data collection. In such cases, the surveys can be used to investigate the incidence, coverage, and benefit adequacy of multiple safety net programs; to estimate and profile population groups that are not covered by safety nets; and to determine those groups that benefit from multiple programs. Impact Evaluation Surveys Estimating a program's impact on beneficiaries often requires a special household survey that will differ from the other surveys described in a few respects. First, an impact evalu- ation survey will cover a representative sample of program beneficiaries--the treatment group--as well as a control group that ideally is similar in all respects to the treatment group except that its members are not program beneficiaries. Unlike representative house- hold surveys, impact evaluation surveys are not representative of the total population. Sec- ond, impact evaluation surveys collect information about a program's expected outcomes, its possible negative effects such as disincentives to work, and a fair number of household and individual characteristics to verify that the treatment and control groups are indeed similar in all respects. Third, the sample size of these surveys should be large enough to detect the minimum expected change in program outcomes that would indicate to policy makers that a program was successful. Finally, the credibility of the evaluation increases when data are collected via panel surveys, which track beneficiaries at baseline time, that is, before program implementation, and then follow up with them over time. The follow- up surveys should be spaced to provide time for the program to have had an impact on its beneficiaries. As data collection is the mostly costly aspect of an impact evaluation, whenever pos- sible, collecting information on a wide range of outcomes and outputs improves the sur- vey's cost-effectiveness. The marginal cost of adding a few questions on additional outputs will likely be small; however, when different outcomes refer to different subpopulations of beneficiaries, such as the nutritional versus the education impacts of a CCT program, there may be implications for sample size as well as for the length of the questionnaire, and thus the marginal costs will be higher. Qualitative Techniques Qualitative data collection techniques are often a useful complement to surveys and at times are the only method that can be used to provide answers to key questions about a program's operations and performance. The most widely used techniques are key infor- mant interviews, direct observation, and focus group (or community group) discussions. Qualitative information is collected through intensive, often repeated, interviews with in- dividuals that are typically more in-depth than precoded questionnaires and explore why and how things are done or why people think something or respond or behave in certain ways because of the program. When the information required for decision making is somewhat technical and re- quires specialized knowledge, the quality of the informant is what matters and not the representativeness of the information. The key informant interview uses a series of open- ended questions posed to individuals who are selected because of their knowledge and experience in the topic of interest. Interviews are qualitative, in-depth, and semistructured and rely on interview guides that list topics or questions. This technique is especially im- 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 195 portant when an external program evaluation is undertaken. Interviews with key staff are the fastest and cheapest way to learn about a program. Direct observation, also known as trained observer rating, is a technique that uses a detailed observation form to record what observers see and hear at a program site. The information may concern ongoing activities, processes, discussions, social interactions, and observable results. To generate comparable and reliable information, observers should be provided with a clear rating scale that sets out what level of performance should be as- sociated with each rating level. This technique may be suited for quality control reviews investigating whether certain procedures are being implemented as intended. Russia used this technique to investigate whether the application process for means-tested cash pro- grams was being undertaken as intended and whether different staff or different offices were implementing the program in the same way. Questions about program operations and performance cannot always be precoded in questionnaires or derived from administrative data. When an M&E team is at an early stage of developing an understanding about how a program influences its clients and is formulating key questions about a program's logical framework, discussions with a par- ticular group, referred to as a focus group, can be helpful. A focus group discussion is a facilitated discussion with 8 to 12 carefully selected participants with similar backgrounds, for example, beneficiaries or program staff. The facilitator uses a discussion guide and note takers record comments and observations. A community group interview is a variant of this technique whereby the discussion is open to all community members. While qualitative techniques lack representativeness and precision, they offer a richer understanding than surveys alone of how a program operates as well as of causality (how and why clients react to the program). Such techniques can be used before the design and deployment of quantitative data collection techniques to help formulate key hypotheses or questions to be investigated through quantitative techniques or after to investigate extreme cases or responses in depth. The general recommendation is to use these techniques in combination--a process called triangulation--to validate answers generated by different data collection processes. REVISITING THE PLAN AND THE M&E SYSTEM OVER TIME Once the M&E plan has been completed and the sources of data have been identified, the next step is to pilot test the M&E system and make any necessary revisions to the data collection tools. During implementation of the M&E system, the written M&E plan is expanded into an operations guide that provides practical direction on how to use the M&E tools and methods that have been developed for a specific program. A critical element of this stage is ensuring that program staff and the M&E team have sufficient training to manage the M&E system. A trial run period will help adapt the M&E system to local realities and constraints. Once an M&E system has been implemented, it should be fine-tuned periodically in response to changes in the environment in which the program operates, changes in the program's design, or following a review of the M&E system. For example, with time the characteristics of the program's clients may change, for instance, if a local industry such as mining suffers a decline and miners' households became poor and eligible for the safety net program. In such a case, both the program's design and the M&E system need to be 196 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS adapted to better serve program clients. Also, over time managers can assess which indica- tors are useful, which are missing, and which are useless and need to be dropped. Periodic reviews of the M&E system would detect these changes and suggest what corrections are needed. Such reviews draw on information produced by ongoing monitoring systems, but are more comprehensive and in-depth. Periodic reviews of an M&E system should address the following questions: · Is the program's logical framework, that is, its cause and effect reasoning, still sound? · Are the assumptions made at the program design stage still valid or have key con- ditions changed? · Has program progress been as planned or have significant discrepancies--positive or negative--become apparent? Are outputs being produced? Is the program on schedule and within budget? · Have important procurement or technical assistance issues not been resolved? · How well does the program team work as a unit and with its partners and stake- holders? · Is participation in the program as extensive as anticipated? · Are actions being taken that will help ensure the sustainability of program ben- efits? · Are performance data being collected as planned, that is, will the data that are needed to demonstrate program effectiveness and impact be available as the pro- gram comes to an end? MATCHING THE M&E SYSTEM TO PROGRAM REALITIES This subsection illustrates how to apply the general principles of an M&E system to dif- ferent settings, from high to low capacity, and from a single program to programs oper- ating in complicated, federal, and/or multi-institution settings, using three examples of programs with good M&E systems: Brazil's Bolsa Familia, Ethiopia's PSNP, and Mexico's Oportunidades. All three programs have a number of characteristics in common: they have implemented strong, results-oriented monitoring systems complemented by one or more elements of program evaluation (of processes, targeting, and impact). While all three programs are well known primarily because of their impact evaluations, their strong monitoring systems were an important element in running the programs well enough to deliver the impacts. A number of conditions facilitate the development of an M&E system. Developing a system for programs operated by one national agency or ministry is easier than for pro- grams operated jointly by a number of institutions. As an extension of this case, programs operating in federal environments are more complex: monitoring is needed in particular to ensure that implementation units pursue the objectives specified by the federal center, but such systems are harder to develop than for programs that involve a single agency. Programs operating in low-capacity environments typically face additional difficulties in relation to lower levels of skills, weaker MISs, lack of availability of information technol- ogy, or even lack of basic services such as electricity and telephones in frontline offices. The 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 197 development of an M&E system is further facilitated by its supervisory bodies' demand for accountability and the presence or absence of a culture of M&E in the public sector in general. Mexico's Oportunidades program has an excellent monitoring system that operates in a high-capacity environment. From the beginning, the program was designed with a monitoring system that would serve the multiple needs of its stakeholders (figure 6.3). An operational component informs frontline and mid-level managers whether planned tasks are fulfilled in terms of quantity, quality of service, and efficiency. A strategic, results- based monitoring component informs upper-level management and external stakeholders whether beneficiaries' outcomes are improving or not. The information different stakeholders need is supplied by different data collection and reporting tools, including the MIS, the appeals and control system, social accountabil- ity systems, and beneficiary assessments. Since the early stages of program development, Oportunidades has had three structures in place to monitor its operations and results. The first structure, the system of monitoring and management indicators, extracts information from the MIS and from the providers of health and education services to track the number of beneficiaries and their composition, the beneficiaries' compliance with conditions, the payment of program grants, the intermediate outcome indicators in relation to education and health, and the management indicators that monitor different phases of the program. FIGURE 6.3 M&E Strategy of Oportunidades TYPE OF INDICATOR AND SOURCE USE OF INDICATORS Outcomes (especially intermediate outcomes) School enrollment and attendance SIIOP Utilization of health care services Coverage of target group, targeting accuracy Measure impact and results Outputs (every two months) Incorporation of new beneficiaries SIIOP Provision of health and education services Strategic Payments to beneficiaries level Measure Performance performance Efficiency indicators for local offices operational SIIOP Quality of records on conditions sentinel monitoring and Active engagement of municipal links points identification of Extent to which rules of operation are followed problem areas Dropouts from list of beneficiaries (every two months Processing of education and health compliance information or twice a year) Operations and day-to-day management Short-term Management MIS Quantitative benchmarks indicators (daily, Number of families needing recertification weekly, or level Training of community representatives monthly) SOURCE: Alvarez 2004. NOTE: SIIOP = system of monitoring and management indicators. 198 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS The system of monitoring and management indicators has been in operation since 1998 and generates a set of 64 monitoring and management indicators every two months (see annex table A6.2.1 for selected indicators). The second structure, a survey of beneficiaries and program providers called sentinel points, has been implemented twice a year since 2000 and produces information on perceptions of service quality (see annex table A6.2.2). The third structure consists of regular assessments of program operations by external ex- perts using the monitoring and management data. The public has access to all data and assessments on the program's Web site (www.oportunidades.gob.mx). The example from Ethiopia exemplifies the difficulties of running an M&E system in a low-capacity, low-income country and the need for continuous adaptation and sim- plification. Ethiopia's PSNP, implemented in 2004 by the Food Security Coordination Bureau, offers another example of a program with a good M&E system. In 2004, a task force within the Food Security Coordination Bureau coordinated the development of an M&E plan for the newly created PSNP. The program monitoring system was designed to track progress for a range of inputs, activities, and outputs for accountability purposes as well as to allow prompt corrective action as bottlenecks were identified. The key indicators to be tracked were determined based on the program's logical framework. Government staff at the local level collect the monitoring data using standardized forms. The informa- tion is then compiled and summarized at the district, regional, and federal levels. Training was provided to federal and regional staff. At the local and district levels, the information is collected using paper and pencil and is then converted to electronic form at the regional level, processed, and transmitted to the Food Security Coordination Bureau. The system aimed for simplicity to account for the low capacity of the program's frontline units. However, implementation of the monitoring plan encountered numerous logisti- cal obstacles, with only 40 out of 232 districts reporting (with delays) during the first year of program operation and the remainder not reporting at all. The major stumbling blocks included the lack of local staff (25 percent of positions remained unfilled during the first year), the poor qualifications and high turnover of existing staff, and the poor infrastructure in some districts (for example, about 20 percent lacked electricity). To gen- erate a minimum amount of monitoring data, a number of additional systems were put in place. First, to assess the program with respect to the number of beneficiaries and actual disbursements, the program instituted a sample-based emergency response system, where information was collected via telephone from around 80 districts on a twice-weekly basis. Second, four- to six-person rapid response teams were formed to perform spot checks (four times a year from the federal level and eight times per year from the regional level). Third, the Food Security Coordination Bureau instituted a system of roving audits to investigate compliance with financial rules, disbursements and payments, and appeals and complaints to provide more timely information on compliance than the normal annual auditing system. Finally, some 80 public works projects of the PSNP were reviewed twice a year to investigate both the quality of planning and implementation. In the meantime, the program further simplified its monitoring system though such steps as shortening the M&E manual from about 160 to 80 pages and invested more in training the staff involved in M&E activities. The simplification of the M&E system and the development of a less ambitious emergency response system were appropriate responses to low capacity. Even though the formal monitoring system is now starting to show some improvement and provide more 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 199 reliable data on basic program operations, the additional monitoring instruments have been kept in place, as they provide more in-depth and often more qualitative information on overall program performance. The third example is the Bolsa Familia program, implemented by Brazil's Ministry for Social Development and the Fight against Hunger (MDS). The program developed a strong M&E system even though it had to confront two types of complications: multiple agencies coordinated the program (the MDS, the Ministry of Education, and the Minis- try of Health) and the program was implemented through 5,564 municipalities (Lindert and others 2007; Vaitsman, Rodrigues, and Paes-Sousa 2006). The program's success was ensured by a combination of three factors: strong political support, gradual and ongoing drive to expand and improve the M&E system, and capacity to innovate. From the beginning, the team that developed the system had to fight against a broader institutional culture within the government whereby the application and use of M&E tools was not widespread or mainstreamed as an integral part of public policy tools. However, the Bolsa Familia program was one of the major initiatives of President Luiz Inácio Lula da Silva's administration, and both the president and the minister for social development wanted to prove that the program was well implemented and improved its beneficiaries' welfare. To highlight the importance of M&E and ensure the presence of staff dedicated to this function, the MDS created the Secretariat for Evaluation and Infor- mation Management with the same hierarchical status as the other secretariats. Most of the information required to monitor the program had to be obtained from other secretariats, ministries, and agencies, few of which operated nationally and most of which were local, such as schools and health centers. The solution endorsed by the M&E team was a pragmatic one: to build the system based on information available in the partici- pating agencies, champion the use of a unique social identification number at the individual level, create a list of specifications and database of indicators, provide support for the suppli- ers of information (such as training and software), and simultaneously undertake a quality control function. Gradually the program has built up an impressive M&E system. A major innovation in Bolsa Familia's M&E system was the introduction of perfor- mance-based incentives for municipalities, which played important roles in implement- ing various aspects of the program and in supplying information. Initially, the program faced challenges in relation to the timeliness and quality of the information received from municipalities, whose capacity varied substantially. To address this problem, the program developed an index of decentralized management that captures the quality of program implementation by each municipality. Based on their performance scores on the index, the MDS pays the municipalities a subsidy that covers some of their administrative costs for implementing the program, with higher subsidies provided for those municipalities that scored better on the index. 6.4 Monitoring DETERMINING WHAT INFORMATION TO COLLECT After developing the logical framework as described in the previous section, the next step is to define more precisely the indictors to use. 200 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Define Indicators Each category of information considered essential for decision making needs to be "trans- lated" into an indicator. As indicated in table 6.4, indicators are numerical measurements of program inputs, processes, outputs, and outcomes typically expressed as levels (for ex- ample, the number of beneficiaries in the program as of a specific date), proportions (for instance, the percentage of beneficiaries paid on time), or ratios (such as the number of nutrition education sessions held per amount spent). Good indicators should be valid (should measure the aspect of the program the deci- sion maker is interested in) and reliable (conclusions based on the indicator should be the same when the indicator is measured by different people). Good indicators should also be sensitive enough to measure important changes in the situation being observed and timely (collecting and analyzing the data should be possible fairly quickly). Moreover, the process of generating the information should be cost-effective: the information gathered should be worth the time and money it costs to collect it and in line with local capabilities and resources. Effective monitoring systems start by building on information that already ex- ists drawing on existing local data collection activities or working with indicators used for other similar programs. For clarity, indicators should specify the reference population they refer to and the period they cover or the date when they are computed. Many programs use a template or indicator specification sheet similar to the one adapted from Brazil's Bolsa Familia program presented in table 6.4 to specify how the data will be collected, analyzed, and reported, namely, the formula for calculating the indicator, its expected (or threshold) value, the source of the information needed to compute the indicator, the frequency with which it will be calculated, the levels of disaggregation, and the main users and uses of the indicator. A comprehensive monitoring system will track a battery of monitoring indicators capturing inputs, processes, outputs, intermediate and final outputs, and performance. Input and Output Indicators. The bulk of inputs consists of the budget for transfers, which generally account for roughly 90 percent of program costs and are usually relatively easy to quantify at the program level using budget documents. However, the level of ac- curacy of the information depends on the institution that is implementing and managing the program. For example, programs run by donors and nongovernmental organizations may not be included in a government's budgetary processes; therefore information on them may be hard to access. Staff time and other administrative resources are key to program delivery, but these inputs are much more difficult to quantify, as staff tend to work on multiple programs within an agency and programs may depend upon multiple agencies for their execution. Moreover, operational costs are often not broken down by the type of activities staff engage in, such as targeting or making payments. Thus managers often lack the basic information that would help them make decisions about improvements to administrative systems. Output indicators track the number of beneficiaries and the transfers and other services provided to them. A good monitoring system should be able to document the number and types of beneficiaries reached and the services actually offered to them and compare the results with the program's intent. TABLE 6.4 Sample Indicator Specification Sheet Adapted from the Bolsa Familia Program, Brazil Category Indicator 1 Indicator 2 Indicator 3 Indicator 4 Name of · Coverage rate · Average value of cash · Cash transferred by · Percentage of families that rose above poverty indicator transfer program as a whole line Description · Percentage of families with a · Average monthly value per · Total cash transferred by · Percentage of beneficiary families whose monthly per capita income of family of cash transferred program in a particular monthly per capita income at time of up to R$100 that receive cash by program in a particular location and during a registration was > R$50 and < R$100 that transfers from program in a location and during a particular reference rose above poverty line (per capita income particular location and during a particular reference period period of < R$100 per month) as a result of cash particular reference period transfer (in a particular location and during a particular reference period) Type of · Intermediate outcome · Output · Output · Outcome indicator Formula for · Number of families receiving · Sum of cash transfers · Sum of cash transfers · Number of beneficiary families whose monthly calculation Bolsa Familia cash transfers provided to families provided to families by per capita income at time of registration was divided by estimated number of by program divided program > R$50 and < R$100 that rose above poverty families with a monthly per capita by number of families line as a result of program cash transfer income of < R$100 multiplied by receiving program benefits divided by number of beneficiary families 100 whose monthly per capita family income at the time of registration was > R$50 and < R$100 multiplied by 100a Source of · Summary of beneficiary payrolls · Beneficiary payroll · Beneficiary payrolls · Unified registry system for federal information by municipality (National (National Secretariat for by municipality government's social programs (Caixa Secretariat for Citizenship Income Citizenship Income and (National Secretariat for Econômica Federal, National Secretariat for and MDS) MDS) Citizenship Income and Citizenship Income, and MDS) MDS) · Estimated number of poor families (Institute of Applied Economic Research and MDS) Frequency · Monthly · Monthly · Monthly · Yearly (continued) TABLE 6.4 (continued) Category Indicator 1 Indicator 2 Indicator 3 Indicator 4 Possible · Brazil, major regions, states, · Brazil, major regions, · Brazil, major regions, · Brazil, major regions, states, districts, level of districts, municipalities states, districts, states, districts, municipalities disaggre- municipalities municipalities gation Limitations · Indicator is calculated using an · Variations in data and · Not available · Indicator is calculated based on self-reported estimate of the number of poor existence of extreme income information families, not a census values (far below or above · Methodology for calculating indicator assumes average) may compromise that income reported at time of registration indicator's ability to reflect remained static over the period and has reality only been modified by cash transfers, thus · Indicator does not fluctuations in income that may occur over time show exposure of each are not incorporated individual in the family to · Interpretation of this indicator as a the benefit, as information measurement of program impact warrants about family size is not caution, as at the time of indicator calculation incorporated family income may be different from the income reported at time of registration Interpreta- · Program coverage rate in 2001 · In March 2005, average · In March 2005, total · In March 2005, number of beneficiary families tion was estimated at 58.4% of poor monthly value of cash cash transfers amounted whose monthly per capita income at time of families with significant variation transfer received by each to > R$430 million registration was > R$50 and < R$100 was (ranging from 28.9% to 63.4%) beneficiary family was per month, with per 1.5 million, of whom 169,500 (11.2%) rose between different areas of the R$65.56 for country as month amount varying above the poverty line as a result of the benefit country a whole, with amount from < R$2 million to varying between regions R$58 million depending and states on the state SOURCE: MDS 2007. a. Decree 5,749 of April 11, 2006, altered the values characterizing poverty and extreme poverty for families to R$120 per month and R$60 per month, respectively. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 203 Monitoring input and output indicators guards against different forms of implemen- tation failure--that is, situations where services are not provided as intended. The most common implementation failures are failure to provide any services, provision of partial services, provision of services of uneven quality, or provision of the wrong services. Fac- tors that may lead to implementation failure include lack of accessibility to the program resulting from, for example, location of offices, office hours, and requirements or costs associated with applying to the program. Outcome Indicators. Safety net programs are implemented to improve beneficiaries' consumption, incomes, wages, investment in human capital, or the like. To the extent possible, the monitoring system should measure these outcomes, as they provide critical information for both stakeholders and management. Monitoring outcomes will indicate whether the social conditions the program is trying to address are being ameliorated, but will not actually reveal the net impact of the program, that is, the change in outcomes caused by the program. Measuring the impact of the program requires a comparison of outcomes of beneficiaries with and without the program, which is what impact evaluation, discussed later, does. Nevertheless, monitoring outcome information is useful when program managers have strong reasons to believe that the provision of services by the program will have a sub- stantial positive influence on the outcomes the program is seeking. This will happen when the influence of factors other than the program is small or insignificant. For instance, a workfare program that rehabilitates school infrastructure will directly influence the func- tionality and esthetics of the school. In the absence of a natural disaster that might damage the building, the outcomes can be safely attributed to the program. While the program will generate most of the data required to monitor program per- formance in relation to inputs and outputs and capture it by means of its MIS, gathering information about program outcomes will often require additional data collection efforts, such as surveys of program beneficiaries. Exceptions include, for instance, a CCT program that routinely tracks information on growth monitoring or administers surveys to track the test scores of students in the program. Table 6.5 presents guidelines for collecting outcome indicators. One difficult trade- off concerns the number of such indicators. The desire to keep data collection costs low TABLE 6.5 Guidelines for Collecting Outcome Indicators Good practice Bad practice · Collecting indicators for program beneficiaries · Focusing on one or a few indicators that do not capture fully the goals of the program · Developing indicators of preprogram to postprogram change whenever possible, but · Ignoring the possibility of corruptibility of bearing in mind that change may not be due indicators, which occurs when program staff has solely to the program discretion in interpreting them · Collecting information about participant · Misinterpreting outcome indicators: changes in satisfaction with the program program outcomes are not necessarily due to the program SOURCE: Rossi, Freeman, and Lipsey 1999. 204 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS may mean that the M&E system focuses on just a few outcome indicators, thereby failing to capture the entirety of what the program is seeking to influence. Despite its usefulness, safety net programs rarely use one class of intermediate out- come indicators: information about the quality of services. Such indicators deal with how well a service was delivered based on characteristics important to consumers of that service (Hatry 1999). The typical service quality characteristics that could usefully be tracked include wait times, staff helpfulness and knowledge, convenience of service (accessibility of location, hours of operation), awareness of program services, condition of facilities used by program beneficiaries, and overall customer satisfaction. Performance or Efficiency Indicators. From a managerial perspective, monitoring sys- tems that only track information on inputs, outputs, and outcomes are operating at well below their potential. The same information can be used to develop performance moni- toring indicators--indicators that capture the program's overall cost-effectiveness, or the efficiency of a subset of program operations. Performance indicators are referred to using different labels, such as efficiency, effec- tiveness, or productivity indicators. Sometimes these are used interchangeably, but in this book we assign a specific meaning to each. We organize and distinguish among different performance indicators based on the logical framework, or the program's results chain, as illustrated in figure 6.4. Here we distinguish between full cost-effectiveness or cost-benefit analysis, which compare the program benefits with FIGURE 6.4 Framework for Distinguishing among Different its costs, and indicators Types of Performance Indicators that capture individual dimensions that con- Outcomes tribute to the program's overall cost-effectiveness. Effectiveness Every program has a value Outputs chain of delivery, from Cost-effectiveness procuring the inputs, Efficiency of analysis service delivery through organizing them Cost-benefit Inputs analysis to deliver transfers or ser- Procurement vices, and finally realizing efficiency the desired impact. Every Expenditures step in this value chain is important for achieving SOURCE: Authors a good cost-benefit ratio, and failures along the line will have a negative effect on the final result. If a program is not cost-effective, it is important to know which part of the value is causing the problem. Moreover it is more often feasible to assess performance on one of the partial or proximate indicators than cost-effectiveness in total. The value chain may be broken down into three elements: · Procurement efficiency, which assesses whether the program achieved value for money in relation to purchases of inputs. Examples of procurement efficiency in- dicators might be the average cost of food procured for school feeding programs, 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 205 of capital goods and materials for a public works program, or of the costs of staff with a given qualification. · Efficiency of service delivery, which considers how efficiently inputs were em- ployed to produce service outputs. An example of an indicator of service delivery efficiency might be the applications processed per staff member or per US$1,000 of administrative costs. · Effectiveness, which examines the program's results (the change in outcomes) per unit of output. Examples of effectiveness indicators are the reduction in poverty gap per US$1,000 in transfers or the decrease in malnutrition resulting from a package or nutrition education session provided under a CCT program. Monitoring program performance is an ongoing activity whose purpose is to un- derstand and increase the efficiency, effectiveness, timeliness, and appropriateness of a program's activities with respect to its goals. Box 6.1 lists the key uses of performance monitoring information. BOX 6.1 Key Uses of Performance Indicators · Identify problem areas and modify practices accordingly. · Identify the root causes of problems, develop action plans, and track progress. · Identify technical assistance needs, supply technical assistance, and track progress. · Tighten funding procedures and standards and reduce or eliminate funding for poorly performing programs. · Identify the need for policy or legislative changes. · Identify underserved groups. · Identify and disseminate successful practices. · Motivate staff and recognize and reward high-performing agencies, offices, and individuals. · Allocate resources, set priorities, and develop plans and targets. SOURCE: Hatry 1999. Monitoring systems should also verify a program's compliance with rules and regula- tions and provide mechanisms to control the level of error, fraud, and corruption (EFC). A good monitoring system will support the objective of reducing EFC in a number of ways. Most fundamentally, good monitoring and proactive management will detect prob- lem areas and address them. Important methods for preventing, detecting, and deterring EFC include hot lines that collect tips from the public; data matching systems that verify identification documents and information on well-being as reported for targeting; and internal or external controls, such as audits. Monitoring can also directly track the level of EFC detected by internal or external audit systems. Monitoring for EFC is still relatively 206 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS rare (box 6.2) but useful for ensuring efficient use of public resources, guarding against political manipulation, and demonstrating program credibility, all of which are vital for maintaining public support. Track Indicators over Time Once the list of indicators has been finalized, they are tracked over time. The frequency with which indicators are reported will vary based on the ease with which information can BOX 6.2 Tracking Error, Fraud, and Corruption Because safety net programs channel large amounts of public resources, making sure that these reach the intended beneficiaries is important. EFC reduces the economic efficiency of safety net interventions by decreasing the amount of money that goes to the intended benefi- ciaries and erodes political support for the program. Definitions. Although most safety net Intentional programs strive to transfer all their resources intended for beneficiaries Fraud Corruption to the right beneficiaries in the right amount and at the right time, a fraction Claimant Staff is lost to EFC (see figure). Customer Official Error is an unintentional violation of error error program or benefit rules that results in the wrong benefit amount being Unintentional paid or in payment to an ineligible ap- SOURCE: National Audit Office 2006. plicant. Official errors are due to staff mistakes, and customer errors occur when customers inadvertently provide incorrect informa- tion. Intentional abuses by claimants are fraud and by staff are corruption. Fraud occurs when a claimant deliberately makes a false statement or conceals or distorts relevant information regarding program eligibility or level of benefits. Corruption commonly involves manipulation of beneficiary rosters, for example, registering ineligible beneficiaries to garner political support, staff accepting illegal payments from eligible or ineligible beneficiaries, or diversion of funds to ghost beneficiaries or other illegal channels. Losses. The evidence on the amount lost to EFC is limited, and most of it comes from devel- oped countries. A recent study (National Audit Office 2006) finds that even well-run programs in high-capacity countries suffer from fraud and error. In five countries of the Organisation for Economic Co-operation and Development--Canada, Ireland, New Zealand, the United King- dom, and the United States--fraud and error rates for the entire social protection system ranged between 2 and 5 percent of total social protection spending. Corruption was not an issue in these countries.Within the social protection system, means-tested safety net programs had the highest fraud and error rates (5 to 10 percent of spending), followed by unemployment benefits and disability pension programs (1 to 2 percent). Old-age pensions had the lowest rates of fraud and error (0.1 to 1.0 percent). These figures should be viewed as lower bounds for the extent 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 207 be gathered; the costs of data collection, the sensitivity of the indicators, and the dynamics of the processes being tracked. For example, the number of children repeating a grade can only be measured at the end of the school year, whereas school attendance can be tracked on a daily, weekly, or monthly basis. Indicators such as the number of beneficiaries and the benefits paid can also be tracked on a weekly or monthly basis. Other indicators, especially those estimated through special surveys such as coverage of the target group, are probably generated only on an annual basis. Input and output indicators are more of EFC, as they come from a small sample of countries and programs with high administrative capacity and adequate procedures for minimizing EFC. The information from developing countries is scarcer, as only a few programs and countries have tried to measure the incidence of EFC, and measures to control EFC are less uniformly developed. A review of accountability in CCT programs in Latin America and the Caribbean (World Bank 2007c) did not measure EFC as such, but did document widespread use of a range of effective control tools. Elsewhere, the results are not always so encouraging. In Bangladesh's Vulnerable Group Development Program, Ahmed and others (2004) estimate that beneficiaries as a whole receive 92 percent of their total wheat entitlement. Initially, the program distributed a certain quantity of wheat to beneficiaries from large containers. About 2.5 percent of the wheat was lost during storage and transportation, a process that is consid- ered normal and unavoidable in the wheat merchandising business. In contravention of program rules, officials sold yet another 2.5 percent to recover the costs of transporting the wheat from warehouses to beneficiaries. Another part was lost to fraud or ad hoc distribution of rations to non-needy recipients (3.0 percent). Once these results were made public, the World Food Pro- gramme tried to reduce the extent of leakage by switching from distributing bulk wheat to distribut- ing wheat flour in sealed 15- or 25-kilogram packages to prevent short-weight or divided rations and by increasing the amount of information provided to beneficiaries on their entitlements. In addition, the wheat flour was fortified, thereby increasing the nutritional impact of the transfer. In India, Dev and others (2004) show that a sizable percentage of the commodities sold through the public distribution system are diverted to the open market, with the amount of fraud vary- ing depending on the commodity and the area of the country. More than 30 percent of rice and wheat and 23 percent of sugar are diverted. Because richer people are less likely than poorer people to claim their rations of subsidized wheat and rice, shopkeepers are left with latitude for selling those commodities on the open market and doctoring records. Diversion is more extensive in the northern and eastern regions than in the southern and western regions. The government is currently thinking of introducing a smart card system (see chapter 5, section 4) to improve accountability and reduce fraud. The two South Asian cases underscore the benefits of measuring EFC. Once it was quantified and the sources understood, both programs found that the political will and technical knowledge to address the problem were easier to marshal. 208 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS sensitive than outcome indicators; among outcome indicators, indicators of intermediate outcomes tend to be more sensitive than final outcome indicators. Set Targets Programs often set explicit targets to help judge whether a given level and trend of an indi- cator is positive or negative. Several approaches are available for setting targets in safety net programs. Targets may be determined based on, for example, assumptions about the ex- pected relationship between inputs, outputs, and possibly outcomes or based on what should happen during each step along the program's logical framework. The performance of other similar programs may help to inform targets, for example, setting a target for the proportion of labor to nonlabor costs in a public works program or gauging whether a 5 percent change in enrollment in a CCT program is good or bad. For instance, in the case of Panama's Red de Oportunidades (Opportunity Network), the targets for the percentage of children com- plying with education-related conditions and showing a reduction in chronic malnutrition (table 6.6) were set based on what other CCT programs had found to be feasible. TABLE 6.6 Selected Targets for Panama's Red de Oportunidades Outcome indicator Baseline Year 1 Year 2 Year 3 Year 4 Year 5 Percentage of households living in 50 50 55 60 65 70 indigenous jurisdictions receiving transfers Percentage of children aged 4­17 who 50 60 70 80 85 90 comply with education-related conditions Percentage of children under 2 years old 0 0 20 40 65 90 registered in the program benefiting from the strengthened health care package Percentage point reduction in chronic -- -- -- 2 -- 5 nutrition among children under 2 registered in the program in indigenous jurisdictions SOURCE: World Bank 2007l, annex 3. NOTE: -- = not available. Targets are often set based on a program's current performance. For instance, the target may be to improve some aspect of performance by some fixed amount, say 10 per- cent. A more discriminating way to do this is to ascertain what is feasible by looking at the performance of an organization's better-performing units. Some targets derive from administrative, legal, ethical, or professional standards. For example, in some cases service standards regulate such things as access to program prem- ises (hours of operation, minimum size of the reception area, and/or presence of ramps to facilitate access by people with disabilities). MAKING MONITORING INFORMATION USEFUL As the implementation of a monitoring system entails costs, collecting only information that decision makers need or that is needed to ensure an adequate level of accountability 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 209 in relation to stakeholders is important. In other words, the monitoring system should not become a data bureaucracy that collects information for its own sake. To make monitoring information useful, the information should be disaggregated, compared with some appropriate benchmark, and reported in a user-friendly format. Outcome, output, and efficiency indicators should be disaggregated for different client subgroups and different service characteristics to investigate whether the service is ho- mogenous. Table 6.7 presents some of the most common subgroups used to disaggregate monitoring indicators. If program target groups consist of a high-priority and a lowest- priority target group, then the monitoring report should capture this breakdown. In the case of Peru's Vaso de Leche program, which provides a daily glass of milk for children up to 13 years old, the priority group was preschool children, but the program's monitoring system did not break down recipients appropriately (World Bank 2007m). Eventually, an investigation by the Audit Court detected that the share of preschoolers in the program was falling, thereby prompting a revision of the eligibility rules. Had this indicator been a performance target, the situation would have been detected and resolved much sooner. TABLE 6.7 Examples of Disaggregation Subgroups and Benchmarks Information disaggregated by Information disaggregated by Benchmarks for comparison client characteristics service characteristics purposes · Age group · Region served (urban or rural) · To previous performance · Gender · Office or facility that provided · To agency targets the service · Race or ethnicity · Among categories of · Amount of assistance customers · Household income provided to individual clients · Among geographical areas · Household size · Mode of service delivery · Among organizational units · Location (urban versus rural, (especially useful for testing district, city, and so on) different approaches) · By type and amount of service · Difficulty of improving the · Individual supervisor or · To the performance of similar situation of the beneficiary (for caseworker programs in other countries example, very, somewhat, or · To the performance of private not difficult) sector organizations SOURCE: Hatry 1999. Another way to increase the usefulness of monitoring information is to collect per- formance and outcome indicators. These measure results; hence they focus management's attention on the program's key objectives and the degree of progress toward them. Take the example of a simple cash transfer program whose mission is to protect households against income poverty and help beneficiaries achieve economic independence. A monitoring sys- tem oriented toward inputs and outputs would probably track the amount of benefits paid in a given period (the inputs) and the number of clients served (the outputs). A results- oriented system would track, in addition, the number of clients who graduated from the program. A system that focused on inputs and outputs would not signal whether the program was succeeding in helping clients achieve economic independence, nor would it highlight which welfare offices were more successful. By ignoring such issues, the program would tolerate poor performance practices or units. 210 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Using monitoring information in decision making requires judgment. First, as out- come indicators are only partially under a program's control, decisions cannot be made based only on these indicators. Second, focusing on a narrow set of outcome indicators may jeopardize achievement of the broader program mission, as program staff would likely devote most of their effort to attainting the narrow targets, at the expense of other out- comes of interest. Finally, a monitoring system focused mainly on output indicators that does not take outcome indicators into account may generate perverse incentives. For in- stance, a CCT program that measures only school attendance and not academic perfor- mance incorporates incentives to bring students to school, but not for improving their knowledge and skills. HELPING ENSURE THE SUCCESS OF A MONITORING SYSTEM The following subsections discuss a number of factors that contribute to the success of a monitoring system. The M&E Unit Must Be Independent As the M&E unit fulfills an oversight function, it needs to have sufficient authority and direct access to upper management. To guarantee that monitoring information is as objec- tive as possible, the M&E unit needs to be shielded from the influence of other program departments and placed directly under the program manager or minister. At this level, the unit will be responsible for all programs operated by the respective ministry or agency, which means that it can apply similar evaluation standards across programs and achieve economies of scale in the use of highly skilled and specialized staff. For example, in Brazil, the M&E unit of the MDS (the Secretariat for Evaluation and Information Management) was created at the same hierarchical level as other, much larger, departments that operate different transfer programs to emphasize its importance and guarantee its independence and objectivity (Vaitsman, Rodrigues, and Paes-Sousa 2006). This was an innovation in public policy in Brazil, and it is still the exception rather than the norm in most safety net programs and social protection ministries. While many program M&E units report directly to heads of agencies or ministers, as is the case for all programs operated by Ar- menia's Ministry of Labor and Social Protection and for Ethiopia's PSNP, typically units are headed by lower-level managers. Unfortunately, many M&E units report to program managers, who may censor any information that indicates poor performance, thereby impairing the M&E unit's objectivity and impartiality. Coordination and Communication Are Essential, Particularly for Complex Programs The development of a good M&E system can be a challenge for programs that involve multiple providers, such as CCTs, or that operate across different levels of government, such as programs designed and financed federally, but implemented by lower levels of gov- ernment. Different actors' willingness to share the information needed to track program performance is a critical component of any M&E system. Solutions also need to be found for any logistical constraints to data sharing. For instance, the exchange of information may be hampered by differences or incompatibilities in MISs, by participating institu- tions tracking different indicators or defining the same indicators differently, or by the use of information technology systems that cannot communicate with each other. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 211 Transparency Is Critical Another well-documented example of monitoring (and evaluation) comes from the Kal- omo District Pilot Social Cash Transfer Scheme in Zambia operated with technical as- sistance from Germany. The pilot included an external M&E system operated by the donors to measure whether the approach was cost-effective and feasible, and generated the expected impact. The monitoring focused on the quality of program management, the effectiveness of the targeting, the payment of the transfers, and the beneficiaries' use of the transfers. Thanks to the investment in the external M&E system, the program's strengths and weaknesses are well known, and policy makers are well equipped to make decisions about its future. At the same time, the scheme has helped set standards for other programs, especially in Sub-Saharan Africa, because of its emphasis on transparency, dissemination, and accountability: all M&E manuals, reports, and survey data are accessible to any third party via the program's Web site (www.socialcashtransfers-zambia). Even though it is just a pilot, the program is better known than other large-scale initiatives in Sub-Saharan Africa because of its extensive M&E system and its transparency. Expensive, High-Tech Systems Do Not Ensure Success While information technology reduces the time, error rate, and costs of data collection, it is not always indispensable. Good monitoring systems existed before the advent of com- puters, and in many low-income countries, monitoring systems will require some combi- nation of paper and pencil records and computerized systems. Particularly in low-income, low-capacity countries, the use of less sophisticated sys- tems may work better given the scarcity of qualified staff and of information technology in frontline units. Low-income countries may respond to their constraints by reducing the amount of information collected and reported, by collecting information less frequently, by collecting information from a subset of the frontline units, and/or by relying on paper and pencil systems. The earlier example from Ethiopia illustrates this point. For the PSNP, the simplification of the M&E system and the development of a less ambitious emergency response system were appropriate responses to low capacity. Donors might consider focus- ing their assistance on developing the M&E functions of such programs if they accompany this with sufficient training. A Good MIS Supports and Enhances the Monitoring System A program MIS's principal function is to help the program carry out the many trans- actions needed to run the program. It contains lists of people who have applied to the program, those applicants who are eligible to receive benefits, those who have complied with any conditions, those to be paid, those who have collected payments, and the like. Its fundamental purpose is to ensure that each function is carried out correctly for each client. When information from the MIS is aggregated in helpful ways or compared with information derived from other sources, the MIS produces information that is valuable for monitoring purposes. Thus while its main purpose is not monitoring, an MIS is one of the workhorse producers of information used for monitoring. 212 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS SCALING UP: A SECTORWIDE MONITORING SYSTEM A monitoring system across the entire safety net or social protection sector is based on program-level systems. The existence of strong M&E systems at the program level-- especially for large, key programs--facilitates assessment of the whole sector. The same framework and methods used at the program level apply to the sector level (Habicht, Victora, and Vaughan 1999). The systemwide system should show whether services are delivered, are used, and cover the target population adequately and whether outcomes are moving in the right direction. Compared with program-level monitoring systems, sectorwide systems put more emphasis on tracking outcomes indicators and less on inputs and processes. Examples of sectorwide monitoring systems that emphasize intermediate and final outcomes are increasingly common in developed (Canada, New Zealand, the United Kingdom, the United States) and middle-income countries (Brazil, Chile, Mexico, the Philippines). Two examples follow, plus box 6.3 presents an example of coordination among European Union states. The United Kingdom's Department for Work and Pensions has developed a com- prehensive performance monitoring system. The department coordinates the work of four agencies that implement programs for families, the disabled, and the elderly (contributory or social pensions). The government signs performance-based agreements known as public service agreements with the department and with line agencies that specify a minimum level of outputs and outcomes to be achieved for the budgetary resources provided. For example, one objective is to ensure the best start in life for all children and end child poverty by 2020. Progress toward this objective is measured via two performance targets: BOX 6.3 Example of Supranational Monitoring System: The European Union's Mutual Information System on Social Protection The European Union established the Mutual Information System on Social Protection (MISSOC) in1990topromotecontinuousexchangeofinformationonsocialprotectionamongmemberstates (for information on MISSOC, see http://ec.europa.eu/employment_social/spsi/missoc_en.htm). MISSOC provides basic information on the organization of social protection in each country, as well as about the financing of social protection, with highly structured and comparable informa- tion in more than 300 information categories. This information is now available in the MISSOC database, which includes information from 2004 on.The database allows users to choose what specific information categories for which countries to display for their particular uses. MISSOC is the outcome of cooperation between the European Commission and a network of official representatives from each member state. It has become a central information source on social protection legislation, benefits, and financing in the European Union countries. Citizens use it to get basic information about social protection in other countries and to compare it with the social protection in their country--for instance, when preparing for moving to another coun- try. Researchers and students also use it to compare social protection systems and solutions in detail and to study changes in social protection over time. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 213 (1) reducing the number of low-income households with children (a joint target with the Treasury Department), and (2) increasing the proportion of parents participating in income-tested programs who receive income support for their children. A second example is the monitoring system in the Philippines. The Micro Impacts of Macroeconomic and Adjustment Policies Project provides information on the welfare status of the population, particularly vulnerable groups, to policy makers, program imple- menters, and the general public. The system combines survey and administrative data from several levels: community volunteers collect and process selected indicators at the local level; municipal planning and development coordinators combine the data for their local governments; provincial planning and development coordinators combine the data for their municipalities; and the National Statistics Office combines data for provinces, supplemented with relevant household data derived from national surveys. At the sector level, the emphasis is on identifying whether the safety net sector is fragmented, patchy, or neatly woven. Some issues are better addressed at this level, such as identifying poor or vulnerable populations that existing safety net programs do not cover, inefficient overlap of programs, and low take-up of certain programs. Hence the monitoring system should be able to capture any synergies and complementarities among programs and identify what program should be responsible for which functions. For exam- ple, Mexico's Oportunidades program emerged as an integrator program to ensure better access to and utilization of health, nutrition, and education services by the poor. In Chile, the Puente program fulfilled the same function. 6.5 Evaluation Evaluation serves several important functions. First, by providing feedback during the life of a safety net program, evaluations can help improve their effectiveness. They can also help guide decisions about whether to expand, modify, or eliminate a particular pro- gram or policy. Second, evaluations permit making programs accountable to the public. Third, they can help inform government decisions about spending allocations as part of a broader, results-based management system. Despite the usefulness of evaluation, until relatively recently, few safety net programs in developing countries were rigorously evaluated. This is beginning to change. Sound evaluations of at least a handful of programs of every type are now available, and many more are available for CCT programs. Some medium-income countries have introduced government-wide management by results reforms that mandate or provide positive incen- tives for regular evaluations of programs and policies. In all developing countries, donors are requesting evidence that the programs they are cofinancing produce results, thereby increasing the demand for evaluation. Many types of program evaluation exist. For example, Rossi, Freeman, and Lipsey (1999) distinguish between the following types of program evaluations: needs assessment, process evaluation, impact evaluation, cost-benefit or cost-effectiveness analysis, and tar- geting accuracy evaluation. We focus on the three most common types of evaluations of safety net programs: process (or implementation) evaluation, assessment of targeting ac- curacy, and impact evaluation. Evaluations of cost-benefit and cost-effectiveness are also helpful, but are rare so not treated in depth here (see box 6.4). 214 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 6.4 Cost-Benefit and Cost-Effectiveness Analysis An important question regarding any program is whether the costs justify the benefits, or whether it provides good value for money. Program costs and benefits are compared using two main methods. Cost-benefit analysis estimates both inputs and outputs in monetary terms to determine whether a program has net benefits to participants and to society as a whole. Cost- effectiveness analysis is used in cases where the benefits cannot be quantified monetarily, as is true for many social programs. Cost-effectiveness analysis estimates inputs in monetary terms and outcomes in nonmonetary quantitative terms. By definition, it is a comparative exercise, examining the unit cost of one program versus the unit costs of other programs. In principle, cost-benefit or cost-effectiveness analysis involves straightforward calculations.All the costs of program operation are tallied and subtracted from the benefits of participation. Typically, the program's costs are estimated from administrative data on staff salaries, overhead, and oper- ating costs along with estimates of participants' foregone earnings or opportunity costs of partici- pating in the program. Benefits are taken from the impact assessment. In practice, such analyses must overcome many operational difficulties, data constraints, and measurement issues. Comprehensive, large-scale evaluations may include a cost-benefit analysis; however, this type of analysis is frequently omitted in smaller evaluations where evaluators lack access to re- sources or to adequate data, and is more common in developed than in developing countries. In developing countries, full cost-benefit analyses have rarely been undertaken. For a good example of full cost-benefit analysis, see Coady (2000) on Mexico's PROGRESA and Econo- metría Consultores, Institute for Fiscal Studies and Sistemas Especializados de Información (2006) on Colombia's Familias en Acción. In the case of the Colombian program, the program's benefits are the present-value of the life- time increase in earnings of the program beneficiaries due to lowered incidence of underweight infants, lowered incidence of malnutrition and child morbidity among children aged zero to six years old, and increased years of secondary schooling. The effects of Familias en Acción on these outcomes are derived from an impact evaluation study and are then monetized using evidence from a combination of sources (for example, a net additional year of secondary school education is assumed to increase future income by 8 percent based on estimates of Mincerian rates of return; an increase of 0.4 kilograms in birthweight is assumed to increase future income by 5 percent based on international evidence). Considered were (1) program costs for the trans- fers and their administration, (2) private costs incurred by the household for additional food and education expenditures, (3) private household costs of collecting transfers, (4) infrastructure and input costs of additional school and health center supply, and (5) the public cost generated to finance the CCT. Comparing the benefit and cost figures, the authors estimate a ratio of benefits to costs of 1.59, which is high by traditional benefit-cost ratio standards and suggests that the CCT is worth its cost. This ratio also means that, even if the assumptions used in this model are imperfect, costs would need to increase 59 percent relative to benefits in order to reach a point where the benefits do not justify the costs. It should be noted that this analysis does not consider other benefits, such as decrease in poverty or inequality that results from the transfer. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 215 All three types of evaluation answer distinct, but complementary, questions. A pro- cess evaluation asks whether a program is being implemented as intended, and if not, why not. An assessment of targeting accuracy investigates whether program beneficiaries are in- deed the poorest members of the population. Impact evaluation quantifies a program's net impact on the outcomes it is trying to influence, that is, it examines whether a program, as delivered, is achieving its goals. Many programs, especially CCT and workfare programs in developing countries, as well as welfare programs in the United States, have undergone all three types of evaluation, giving rise to the term comprehensive evaluations (Greenberg and Shroder 2004). Safety net programs are increasingly being evaluated using multiple types of evalua- tion. The United States was the first country to pioneer this practice. Of the 146 impact evaluations undertaken in the United States between 1962 and 1996 and reviewed by Greenberg and Shroder (2004), half were complemented by a process evaluation. Recent- ly, middle-income countries have increasingly been adopting the practice of undertaking multiple types of evaluation for a given program. In particular, CCT programs in Colom- bia, Jamaica, Mexico, Nicaragua, and Turkey used process evaluation to complement more focused evaluations of targeting accuracy or impact. PROCESS EVALUATION Process evaluation--also known as formative evaluation, implementation research, im- plementation analysis, or descriptive evaluation--is probably the most common type of program evaluation (Rossi, Freeman, and Lipsey 1999) and documents, assesses, and ex- plains how a program is implemented. It can be used throughout the life of a program, from start-up to maturity, and whenever the environment in which the program is operat- ing changes, thereby requiring adjustments of the program. This type of evaluation may cover some or all of three basic questions (Werner 2004): (1) What is happening? (2) Is it what the program's designers want or expect to happen? (3) Why is it happening as it is? In many instances, the decision to undertake a process evaluation follows from an operational problem signaled by the monitoring system. Program managers learn about operational problems from monitoring reports, for instance, reports about complaints and appeals, quality control reviews, or audit reports. The problem may be specific to a par- ticular client group, say low take-up among minority groups; specific to a particular service unit, for example, a high complaint rate in one office or region; or related to a program- level issue, such as a reduction in caseload below target levels. However, the monitoring system only reveals the problem. It does not indicate why the problem emerged or how to solve it. To find answers to the latter, managers can ask for an internal review or commis- sion an external process evaluation. In other instances, process evaluation is a substitute for a missing or a poorly per- forming monitoring and internal review system. Process evaluation should complement, not substitute for, internal monitoring systems. For example, Zambia's Kalomo District Pilot Social Cash Transfer Scheme included process evaluation carried out by external evaluators, but this was coupled with assistance provided by the external evaluator to the ministry to build in-house monitoring capacity. Unlike impact evaluation, process evaluation cannot establish causality with a known margin of error. Process evaluation can uncover plausible reasons why a program is work- 216 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS ing or not, and hence can build hypotheses and theories that can be tested using an impact evaluation. In developing hypotheses about causal connections, process evaluation will use one of the following methods: (1) using stakeholders' accounts of why they take or fail to take specific actions related to program activities and goals; (2) associating variations in management, policies, operations, services, or other factors with observed differences in results; and (3) comparing actual results with predictions or expectations based on the model or theory underlying the program. The findings from process evaluations are behind many of the improvements in ef- fectiveness and efficiency of safety net programs. For example, in Armenia, a number of discrete process evaluations contributed to sustained capacity-building efforts that led to substantial improvements in the targeting accuracy of the Family Poverty Benefits Pro- gram (chapter 4, section 4, provides more detail). In another case in the town of Arzamas in Russia, city authorities worked with the Urban Institute to study whether the many programs the authorities were implementing had overlapping functions or stages. Finding a substantial degree of overlap across programs, the city authorities decided to replace the delivery system with a one-window approach, resulting in fewer exclusion errors, lower administrative costs, and lower transaction costs for beneficiaries (chapter 5, section 3). During program start-up, process evaluation is used to monitor initial program imple- mentation so that bottlenecks can be addressed speedily and good practice can be document- ed for future applications. After the program stabilizes and matures, a process evaluation can be used to provide ongoing feedback to management, especially when the monitoring sys- tem detects operational issues or when the factors that affect program effectiveness change, raising doubts about the continued relevance of some of its components. More generally, process evaluation is used (1) on an ad hoc basis to analyze an operational problem; (2) to complement an impact evaluation; (3) to keep stakeholders informed, that is, for account- ability purposes; and (4) to substitute for an inadequate or missing monitoring system. Emerging programs may use process evaluation to keep interest in a program alive and/or to secure support for scaling it up. In countries where the government or parlia- ment has endorsed a management by results agenda, such evaluations can be routinely mandated to produce indicators of program performance that are tracked over time. Rou- tine process evaluation also has other uses. Jamaica's PATH initiative, a CCT jointly fi- nanced by the government and the World Bank, uses a third party evaluator to monitor compliance with conditions and estimate the error rate in assessing eligibility. The error rate, in turn, determines the portion of the budget cofinanced by the World Bank. If legislation requires routine process evaluations, programs are constantly exposed to public scrutiny, setting the right incentives for them to improve their effectiveness. However, evaluations mandated from above can run the risk of being irrelevant by not re- sponding to genuine operational problems. The evaluation's sponsors may lose interest in such evaluations, or worse, fear that the results of the mandated evaluation may negatively affect the program's status quo and derail the evaluation to noncontentious areas of little importance. Process evaluations based on routine compliance with a legislative require- ment to undertake them regularly have little or no value. The questions covered by process evaluation are as diverse as the problems program managers encounter. Table 6.8 presents some examples of questions a typical process eval- uation will address in the case of a safety net program. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 217 TABLE 6.8 Examples of Questions Addressed by Process Evaluations of Safety Net Programs Aspect evaluated Questions Program · Is the program well organized? organization · Does program implementation follow a clear organizational structure? · How well do different groups involved in delivery work together (in terms of different staff within delivery teams and different programs and agencies with which the program in question must interact)? Program · Are adequate resources being used to implement the program? resources · Is program staffing and funding sufficient to ensure appropriate standards? · Are program resources (inputs) being used effectively and efficiently? · Are costs (per beneficiary, per benefits transferred, and so on) reasonable? Program · How did people hear about the program? availability · What is the level of awareness among the eligible and potentially eligible population? and participation · Do those eligible and potentially eligible understand the program's key elements? · Do all those eligible participate in the program? · Who participates and why? Who does not participate and why? Do particular groups within the target population not receive the program and why? · Do some people who are not eligible participate, and if so, does this suggest that the target population is poorly defined or that program delivery is poorly controlled? Delivery of · Are participants receiving the proper amounts, types, and quality of benefits and services and services? benefits · Is delivery of benefits and services consistent with the program's intent? · Are all components of the program being delivered adequately and consistently? · How much change has occurred since program implementation? · Does the program comply with professional and legal standards, for example, are appropriate complaints procedures in place? Participant · What are participants' experiences of contact with the program, for instance, how experiences were they invited to participate? What kinds and how many contacts with the program did they have? What was the duration and content of their contacts? · Are participants satisfied with their interactions with staff delivering the program, with program procedures, and with the services they receive? · Do participants engage in anticipated or intended follow-up behavior? Program · Are benefits and services delivered according to different models or by different performance organizations, for example, do banks distribute cash in some areas while post offices issues or private contractors do so in other areas? If so, how do these compare? · Are program resources and/or program delivery consistent and appropriate across all geographical locations? · Do variations exist across locations or models that provide examples of best practice or that are important interventions that should be considered elsewhere? Are variations occurring over time or for different groups? · Is the program focused on those who are easier to reach at the expense of those who are harder to reach? If so, what is the impact on the nature of the services provided and the net outcomes of these services? What would be the effect of shifting the balance between the easier and harder to reach? SOURCES: Purdon and others 2001; Rossi, Freeman, and Lipsey 1999; Werner 2004. 218 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Process evaluation is often undertaken as a complement to an impact evaluation. A process evaluation can significantly enrich the findings of an impact evaluation for the fol- lowing reasons (Greenberg and Shroder 2004): · A descriptive process evaluation is important for the replication or scaling up of a successful intervention that produced the expected impact. This is particularly important for complex programs such as Chile Solidario. · The findings of a process evaluation are important for understanding the results of an impact evaluation. If an impact evaluation shows no impact, this could be due to poor program design or simply to poor implementation of a well-designed program. Programs implemented heterogeneously across sites, regions, or clients will generate a modest average impact despite their good impact in areas where the program was implemented well. The evaluator will only become aware of this problem by collecting information on the quality of implementation. · An understanding of how and how well any intervention is actually being imple- mented and delivered is important before decision makers decide whether to keep or modify a program. · The evaluator requires information from the process evaluation if the replication of a program produces different results to assess why and how the replication sites differ from the intervention that was evaluated. The most common data collection tools used for process evaluations are interviews; observation; focus groups; and examination of the records of the program, agency, and/ or ministry. Interviews are by far the dominant technique and take place at all levels with both individuals and groups. Individual interviews are advisable when the evaluator wants to inquire about aspects of the program that people will not want to discuss in front of others. Group interviews are recommended when the evaluator wants to learn about con- flicting views. Observation reveals how things are actually done in practice. Focus groups are facilitated discussions with clients, former clients, applicants, and eligible clients to obtain their perceptions about how the program worked and what it did for them. Admin- istrative records are an important source of information about the number of beneficiaries; the number of clients contacted by, accepted into, or removed from the program, and so on; timing; bottlenecks; and the like. ASSESSMENT OF TARGETING ACCURACY All safety net programs aim, explicitly or implicitly, to channel their benefits to the poor, or a subset of them, typically the poorest. An evaluation of targeting accuracy, also re- ferred to as a targeting assessment, helps reveal whether this aim has been realized, asking questions such as what share of the beneficiaries of a safety net program is indeed poor? What proportion of the poor is covered or served by a safety net program? Have changes in eligibility rules succeeded in reducing the share of nonpoor beneficiaries? How much did the coverage of the poor increase after expansion of the program? A targeting assessment is a cheaper but less precise alternative for evaluating a pro- gram's impact on poverty reduction than a full impact assessment. A reduction in poverty brought about by the program will be a function of the program's coverage, generosity, 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 219 and disincentive effects, while errors of inclusion and administrative costs will raise the costs of achieving an impact. Targeting assessments quantify the program's coverage and errors of inclusion, thereby providing useful insights into why a program may or may not be having a strong impact. Moreover, a targeting assessment is feasible in many more cases than a full impact assessment, and even a partial assessment of a program can help policy makers know whether they can hope for results or need to take corrective action. Using a medical analogy (Habicht, Victora, and Vaughan 1999), one does not need to reevaluate the impact of a vaccine once it has been established that it works in many settings. If a previous impact evaluation has demonstrated that the vaccine works, for subsequent in- terventions, we need only demonstrate that the vaccine has reached the target groups; this is what a targeting assessment does. Note that not everyone shares our relatively positive view of targeting assessments. Ravallion (2007), for example, considers directly measuring the outcome variable of interest, which is some measure of poverty given the objectives of these programs, to be preferable. While impact evaluation is desirable, it is not always possible or sufficient. This subsection deals with this type of evaluation at length for two reasons. First, because all safety net programs are targeted and have, at least implicitly, the objective of alleviating poverty, an assessment of targeting efficiency is critical for such programs.2 Second, unlike other types of program evaluation, textbooks and other sources tend not to cover this type of evaluation. Measuring Targeting Accuracy There are many indicators of targeting accuracy, but all start with knowing who benefits from a program and who does not. This information can be reported across the whole distribution of income graphically, summarized into various single-number indexes, or re- ported for various subgroups. Table 6.9 summarizes some of the common measures used. In the evaluation literature on safety net programs, the assessment is typically referred to as a targeting assessment. Outside safety nets, the technique is known as benefit incidence analysis. The rest of this subsection focuses on various features of targeting assessment. The initial question is who is the group of interest. Is it the poor or the extreme poor as defined by some explicitly delineated poverty line? Or is it the poorest x percent of the population? Both are of value, but we generally prefer the latter, because it provides information across the full spectrum of welfare. For example, knowing if the benefits that miss the poor go to the very nearly poor or to the very wealthy is helpful. Equally impor- tant, poverty lines have an element of arbitrariness about them and are rarely comparable across countries, are often not comparable within countries across time, and are often disputed. Policy makers who disagree with an analyst's definition of the poverty line will find drawing conclusions about whether a program is sufficiently targeted or not difficult if information is presented only in relation to that one disputed poverty line. In contrast, the presentation of results across the spectrum of welfare will be useful to policy makers no matter their opinions about the poverty line, will be useful to program analysts and policy makers 10 years later in the same country, and will be helpful to policy makers and analysts worldwide now and in 10 years as a comparator. Of course, to choose one does not exclude choosing the other. Both are easily computed from the same raw data and may be presented side by side. 220 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS TABLE 6.9 Common Targeting Measures Measure Definition Concentration curve Share of total transfers going to the poorest percentage of the population ranked by household income per person Share going to the poor Share of transfers going to those who are initially deemed poor (or other reference group based on income) Normalized share Share of transfers going to a the poorest x percent of the population divided by that share; for example, if 30 percent of the transfer goes to the poorest 20 percent of the population, the normalized share is 30/20 = 1.5 Concentration index Area between the concentration curve and the diagonal along which everyone receives the same amount Coverage rate Program participation rate for the poor Targeting differential Difference between the coverage rate and the participation rate for the nonpoor Proportion of type 1 errors Proportion of program beneficiaries who are not poor Proportion of type 2 errors Proportion of the poor who do not benefit from the program SOURCE: Ravallion 2007. Looking at the Average Incidence of Benefits of a Safety Net Program In its most basic form, a targeting assessment describes how public spending is distributed across population groups, whether defined as deciles or poor versus nonpoor. Data Requirements. To undertake a targeting assessment, the analyst will need a house- hold survey, representative for the entire population and for the program's target group, and with information on household welfare and program receipt. Not all countries have representative surveys that collect information on household welfare and the receipt of program benefits. Sometimes information about the receipt of program benefits is miss- ing; in other cases, while this information may be collected, it is not representative of the subpopulation of program beneficiaries. This is particularly likely if the coverage of the safety net program is small. When this information is not available from a national survey, options include oversampling program beneficiaries and combining a survey of beneficia- ries with the nationally representative household survey. Montenegro provides an example of oversampling program participants. For its 2004 household survey, the government wanted to keep the survey small while obtaining repre- sentative information about the Family Maternal Support Program, a safety net program with low coverage. Given Montenegro's total population of about 600,000, a national survey of 600 households was considered sufficient for other issues the survey covered, but would generate an inadequate sample of program beneficiaries. Given policy makers' interest in the program's performance, the survey was augmented with a booster sample of 200 beneficiary households extracted randomly from the program lists. The same survey questionnaire was administered to 800 households. The survey weights were adjusted to take into account the oversampling of program beneficiaries (Institute for Strategic Studies and Prognoses 2004). 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 221 Jamaica decided to combine a beneficiary survey with a national household survey. To assess the targeting accuracy of the newly implemented PATH initiative, in 2003, the Statistical Institute of Jamaica fielded a beneficiary survey of 936 households one year after the regular Survey of Living Conditions had surveyed 6,976 households (Levy and Ohls 2004). The 2003 survey used the same consumption module as the previous year's survey, employed the same methodology for constructing the welfare aggregate, and was fielded just before the applicants received their benefits to generate a comparable consumption aggregate that captured beneficiaries' welfare level before the program. To assess the pro- gram's distributional incidence, the consultants hired to carry out the survey compared the 2003 data with the 2002 quintile cutoffs adjusted for inflation. To assess the inclusion and exclusion errors of the new program, the 2003 data were compared with contemporaneous poverty lines (Levy and Ohls 2004). Sometimes programs serve only part of the population, for example, only urban or only rural areas or only areas known to be poor. Nonetheless, reference to the national welfare distribution is needed to understand program effectiveness. Using quantiles spe- cific to that subpopulation for the analysis may be misleading. Consider a country where all the poor live in rural areas that has an antipoverty program that operates mostly in urban areas, where it successfully captures the less well-off households. Evaluated against the national income distribution, the program will show a high leakage rate, but when quantiles are constructed separately for rural and urban areas, the program will appear to be progressive. These kinds of results are not straightforward to interpret (box 6.5). For programs of national scope or that are nationally financed, we recommend always present- ing the incidence of benefits based on national quantiles and qualifying the results based on quantiles estimated for subpopulations such as regions or urban and rural areas. The exception would be analyzing programs run by subnational jurisdictions, especially with their own funds; in federal countries, for example, analyses of state-level programs--such as in Brazil, India, and the United States--for state policy makers would be done for state populations. Methodology. An analyst will estimate the incidence of benefits (or of beneficiaries) in four steps. The steps are conceptually simple, but the results can be quite sensitive to how each is performed (Demery 2000; van de Walle 2003). Thus performing a sensitivity analysis on some of the main choices made and reporting how sensitive detailed results or policy conclusions are to those choices are good practices. Reporting in detail on the choices is also important to allow proper interpretation of results and benchmarking and to allow analysts to reproduce results. The first step is the construction of a welfare measure, which is required to correctly rank households according to their standard of living. The most typical welfare indicators are per capita consumption or income, sometimes per adult equivalent consumption or income. A good welfare measure should be comprehensive and comparable across space, time, and different types of households. To be comprehensive, a consumption indicator should capture all its components, such as food, nonfood, and services, as well as the value of goods produced and consumed by the household and the imputed value of durables or the rental value of an owner-occupied dwelling. Similarly, a comprehensive income indi- cator will cover the incomes earned by all household members from formal and informal sources and the value of goods produced and consumed by the household. All income 222 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 6.5 Misleading with Targeting Statistics Using household surveys that are not representative of the entire population to assess the tar- geting performance of a national safety net program is misleading. Two contrasting examples from Ethiopia and Russia illustrate this point. The example from Russia illustrates how a poorly targeted program can be made to appear well targeted by restricting the analysis to well-off urban areas. Since 1995, Russia has operated a cash program, the Housing Allowance Program, to compensate poor consumers for the high costs of heating. To receive the allowance, households need to submit documentary evidence of their income and their heating expenses. Because such documents are available only to urban residents living in apartment buildings connected to the urban heating grid, most rural households cannot access the program. As most of the poor live in rural areas, the program misses most of the poor. The World Bank (2006) documented the large exclusion errors and low targeting accuracy using evidence from a nationally representative survey. Struyk, Petrova, and Lykova (2006j) reached the opposite conclusion based on a household survey drawn from three upper-income towns. They find that the program identified the less well-off households from these three cities well and thus concluded that the program is well targeted. The example from Ethiopia illustrates how a program that is probably well targeted may appear to be poorly targeted. The PSNP operates in food-insecure areas where it provides food-for- work for able-bodied individuals and cash transfers for destitute households whose members cannot work. The program's targeting efficiency was a constant concern of the authorities. A number of qualitative evaluations documented that program participants are poorer than the rest of their communities in terms of land or livestock owned. A household survey was fielded in the program areas in 2005 that Devereux and others (2006) used to present standard benefit incidence tables. Because the sample was representative only of the poorest regions in the country, the incidence results were mediocre at best: program beneficiaries came evenly from all quintiles. An uninformed reader of these results can wrongly conclude that the program is badly targeted. No assessment of the program's targeting accuracy based on the national in- come distribution has been carried out to date. or consumption information will be totaled as a monetary value per household. Because prices differ across space and time, the purchasing power of a given level of nominal total income or consumption will differ. For comparability, the indicators will be deflated by an appropriate price index and expressed in real terms. To compare the welfare of different individuals, real per capita consumption or income is typically expressed in per capita or per adult equivalent. For guidance on constructing a consumption-based welfare mea- sure, see Deaton and Zaidi (2002). For guidance on constructing an income-based welfare measure, see Eurostat (2003) or U.S. Census Bureau (2005). For an explanation of adult equivalencies, see box 8.2. The second step is to construct quantiles. As Demery (2000, 2003) shows, the results of a targeting assessment can be quite sensitive to the types of quantiles the analyst uses, be they individuals or households. Constructing these so that they contain the same number of individuals, not households, is preferable except when a previous targeting assessment 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 223 based on household quantiles exists, as for comparison purposes, the same methods must be used. This simplifies interpretation of the results.3 In the third step, the analyst has to identify the benefit given to each beneficiary (or assistance) unit--individual, family, or household. How this information is collected will depend on the type of program. For cash, quasi-cash, and workfare programs, this infor- mation is straightforward to collect via a survey. Finding this information in a multitopic household survey where each household (or individuals within the household) reports the sums of money received from the program during the reporting period is quite common. For other types of programs, notably fee waiver, subsidy, and in-kind programs, most sur- veys will only collect information on receipt of the program (a yes or no answer), but not the value of the benefit, especially when beneficiaries cannot estimate the cash equivalent of the goods or services received (McKay 2000b). If the survey collects information on the value of benefits participants receive, the analyst can produce benefit incidence tables in terms of beneficiaries and benefits. If the value of the benefit is uniform across individuals, the incidence of beneficiaries will be the same as the incidence of benefits. If the benefit is customized by household characteristics, the difference in incidence by beneficiaries and benefits can differ significantly. Sometimes even if the survey asks only about participation in the program and not the level of benefit received, it may be possible to simulate these as follows: · Some surveys may deliberately omit to collect benefit information if the benefit formula is simple and there are no payment arrears. For example, if a child al- lowance program offers a flat benefit to all children aged newborn to 16 years old, collecting information on program participation is enough. The analyst can impute the amount to each household with children of eligible age. · Some surveys gather information about individual or household circumstances that determine the level of the benefit. For example, the value of a heating sub- sidy whose level depends only on the type of dwelling (apartment building versus individual house) and location (municipality) can be obtained if the survey col- lects information on who received the program, the type of dwelling, and the municipality. When the distinction between benefit and beneficiary incidence is important, as for price subsidy programs,4 the analyst can try to obtain an estimate of the value of the subsidy by multiplying the number of units of the subsidized good or service consumed (as observed in the survey) by an estimate of the unit value of the subsidy made separately from the survey. Alternative definitions of the beneficiary unit may significantly affect the results. Depending on the type of program and the target group, the direct beneficiary of a safety net program may be an individual, a family, or a household. However, in a broader sense, all household members benefit from the additional resources provided by the program, thus a strong economic rationale exists for assigning benefits to the whole household when assessing the incidence of a program. Consider a child allowance program in a country where children account for 25 percent of the population and families with children ac- count for 60 percent of the population. If only direct beneficiaries are taken into account, the coverage of the program will be 25 percent of the population; but if all beneficiaries, 224 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS direct and indirect, are counted, coverage will be 60 percent. Given the negative correla- tion between household size and welfare level, using households as beneficiary units for safety net programs where the assistance unit is an individual will improve both coverage and targeting accuracy statistics. Whenever possible, the analyst should report both results. If only one set of results is to be reported, we prefer those based on indirect beneficiaries, as this is the only way to compare programs that serve different types of assistance units. Checking the quality of survey information against administrative sources is good practice, and a number of simple tests are available. The analyst can check the size of the difference between the estimated number of beneficiaries or spending level obtained from the survey with the same figure from administrative data. Is this difference statistically significant? If not, this is a necessary, but insufficient, condition indicating that the survey data are of good quality. Is the level of benefits reported by beneficiaries the same as in the program's schedule of benefits? If so, this is another indication that the survey data are of good quality. When possible, the analyst should repeat these tests for those subpopulations for which the survey is representative, such as rural and urban areas or regions. The analyst should always be careful to use the same reporting period when comparing flow quanti- ties such as program spending or number of beneficiaries. For an example of comparing survey and administrative data to assess the representativeness of the former, see Galasso and Ravallion (2004). Small programs--that is, those that cover a small proportion of the total popula- tion--are hard to capture by means of nationally representative surveys. The estimated coverage of such programs will be imprecise, because the sample size of a typical household survey is not large enough to capture a sufficient number of beneficiaries. In such cases, analysts cannot determine whether a discrepancy between survey data and administrative data is due to measurement error or to leakages or fraud. Several solutions are available in this case. The first possibility is to add a booster sample to an existing household survey to have a sufficient number of beneficiaries as dis- cussed earlier. Another possibility is to conduct a small census of the families in a village or enumeration area before selecting the beneficiaries to interview. The resulting sample of beneficiaries will be large enough to provide robust information about the characteristics of the beneficiaries, and the small census can provide better estimates of beneficiary par- ticipation. If a discrepancy between the administrative data and the small census informa- tion persists, the analyst can design a small local survey of current beneficiaries (a tracking or tracing survey) to ascertain if all the beneficiaries in specific areas actually exist. For example, the results from the initial small census carried out as part of a study of the Food Support Program in Pakistan reported a coverage rate of 0.6 percent, while administra- tive records reported almost 4.0 percent (World Bank 2007k). A follow-up tracing survey was able to find more than 85 percent of the beneficiaries, bringing the reported program coverage up to 3.4 percent. The fourth and last step is the calculation, presentation, and interpretation of the results. Targeting measures are calculated using the data collected following one or more of the measures outlined in table 6.9. Once the basic calculations have been done, assess- ing an individual program against its stated objective and against other programs is useful. The comparators may be selected because they represent alternative uses of funds for the country or because they use a somewhat different targeting system and thus yield insights 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 225 into feasible options. The selection of targeting measures used also takes into account the measures available for the proposed comparators. As perfect targeting is not possible and programs will always fall short of their in- tended goal in that respect, including feasible comparators is important. For benchmarking safety net programs, Coady, Grosh, and Hoddinott (2004) provide a comprehensive com- pilation of the incidence of targeted transfers for 122 programs in 48 countries. Lindert, Skoufias, and Shapiro (2006) provide slightly more recent information for 56 programs in 8 countries in Latin America and the Caribbean. Caveats. Having the right expectations of the end result of a targeting assessment is im- portant. The targeting assessment is a descriptive analytic tool. It will not reveal why benefits are distributed as they are and what aspects of the program should be changed to close the gap between the actual and the desired allocation of program resources. It does not explain incidence outcomes, nor does it generate specific policy implications. In the end, stakeholders will learn whether the program's benefits are being distributed equitably or not and how far these results are from the intended distribution or outcome and bring understandings of the context and other dimensions of the program to bear in determin- ing appropriate responses. At the same time, the method gives an incomplete picture of welfare effects, for example, ignoring the impact of transfers on other dimensions of welfare such as health, literacy, and nutrition. Similarly, it does not take into account the long-term effects of safety net transfers. For some interventions with positive externalities, such as the health interventions typically found in CCT programs, it wrongly assumes that the cost of provision reflects the benefit to the user: for example, the cost of immu- nization is low, but the benefits are large. Undertaking More Elaborated Forms of Targeting Assessment The advantage of the basic targeting assessment lies in its simplicity: it can be done quick- ly if the right data are available, it does not require exceptional analytic skills, and it pro- duces results that are relatively easy to understand and interpret. However, this simplicity brings with it a number of limitations, some of which can be handled by more elaborated forms of targeting assessment (table 6.10). Accounting for the Behavioral Responses of Recipients. Ideally, a targeting assess- ment should rank households by their level of income or consumption in the absence of the program. Only then, when the analyst knows how poor program beneficiaries would have been without the program, can he or she estimate the true incidence of the benefits. The problem is, of course, that direct observation of what recipients' welfare would have been in the absence of the program is not possible, and so some approximation of it must be made. Two simple, or so-called naïve, estimators are possible. One method is to subtract the value of the transfer from post-transfer consumption. The assumption behind this ap- proximation is that the extra income does not affect the wages and/or remittances received by the household, and is entirely consumed, that is, not saved or invested. If the receipt of transfers reduces the level of wages and remittances earned by the household, this method will underestimate errors of inclusion. The extent of the error will depend on the size of the behavioral changes. A second simple approximation is to use post-transfer consump- 226 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS TABLE 6.10 Some of the Main Types of Targeting Assessments Type of targeting Policy or research question assessment Explanation Was a counterfactual Accounting Households are grouped into quantiles based on welfare distribution observed consumption following safety net transfers estimated? Behavioral Households are grouped into quantiles based on counterfactual consumption before (net of) safety net transfers What distribution of Average Describes what shares of benefits accrue to each benefits is of interest? quantile for an existing program Marginal Describes or estimates how the benefits corresponding to an increase or decrease in the program are or will be distributed Were households ranked Static Ranks households by quantiles based on their level based on their current of welfare during a given period welfare or based on losses and gains in welfare during Dynamic Ranks households by quantiles based on changes in a given period? their level of welfare between periods Was the assessment done Ex post Describes how program benefits are distributed before or after the program across population groups was implemented? Ex ante Estimates how program benefits will be distributed if certain program parameters change SOURCE: Authors. tion under the implicit assumption that all extra transfer income triggered a proportional reduction of other incomes from wages and/or remittances. If the transfer is large and reduces the level of wages and remittances earned by the household, this will overesti- mate errors of inclusion. Consider, for example, the case of a perfectly designed, perfectly implemented minimum guarantee program that gives each poor household a transfer suf- ficient to raise it just above the poverty line. Before the transfer, all program beneficiaries were poor; afterward, none were. A targeting assessment using post-transfer consumption would conclude that the program was failing to reach the poor, even though it was actu- ally a success. Though neither of the naïve estimates is accurate, they provide upper- and lower-bound estimates for targeting outcomes. To estimate a household's welfare in the absence of a program most accurately, the analyst must model changes in the household's labor supply, remittances, savings, and credit, or alternatively, obtain this information from a comparable counterfactual group, an undertaking more usually carried out under the label of impact evaluation. These changes provide an estimate of a household's welfare in the absence of a transfer and allow the calculation of correct welfare rankings. The possible sensitivity of the results to the estimator of pretransfer welfare is dramati- cally illustrated by van de Walle's calculations for the Republic of Yemen (figure 6.5). The distribution of transfers is nearly equal across deciles if the observed post-transfer per capita expenditure is used, but sharply progressive if the full transfer is subtracted from per capita ex- penditure. Van de Walle estimates that the marginal propensity to consume out of the trans- 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 227 fer (MPCT) is about 0.5, FIGURE 6.5 Incidence of Social Protection Transfers Depends which means that con- on the Assumed Pretransfer per Capita Consumption, Republic of Yemen, 1999 sumption rose by 50 per- cent of the value of the Percentage of population in transfer received. In this households that received transfers analysis, she combines all 65 transfers: social assistance, 60 pensions, and private 55 transfers including remit- 50 tances. Since the latter 45 are large, they effectively 40 raise those who receive 35 them out of poverty and 30 the change in ranking of 25 households is larger than 20 found for most social as- 1 2 3 4 5 6 7 8 9 10 sistance programs. (poorest) (richest) Population deciles The limited evi- dence available suggests Per capita expenditures net of transfers that for safety net pro- Per capita expenditures net of 50% of transfers grams with moderate Per capita expenditures, transfers fully included generosity, the increase in income or consumption is SOURCE: Van de Walle 2002a. close to the value of trans- fer. In the case of five of the six CCT programs reviewed in World Bank (forthcoming), household consumption increased by almost the whole value of the transfer, equivalent to an MPCT of 1. More generous programs with an income-replacement role, like workfare and social pension programs, but also social insurance programs, such as unemployment benefits or contribu- tory pensions, will likely increase household consumption substantially less than 100 per- cent. This is because, in the absence of the program, households had to earn some income in other ways in order to survive. For example, in the case of Argentina's Trabajar workfare program, the average direct gain for participants was about half the gross wage (Jalan and Ravallion 2003), implying an MPCT of 0.5. All methods used to account for households' behavioral responses to transfers are data intensive. Most of the CCT evaluations from World Bank (forthcoming) are random- ized evaluations, with information collected from treatment and control groups. Jalan and Ravallion (2003) use propensity score matching to evaluate the distributional outcomes of Argentina's Trabajar program. Ravallion, van de Walle, and Gautam (1995) and van de Walle (2003) use panel data and instrumental variable models to estimate a reduced form equation of household consumption on transfer incomes. In most cases, the data for a targeting assessment come from a single cross-sectional survey, and hence are insufficient to estimate the MPCT reliably. In such cases, analysts can carry out a sensitivity analysis, estimating the before transfer counterfactual consump- tion based on a range of possible MPCT values. E. Tesliuc (2004) provides an example of 228 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS this approach for the Kyrgyz Republic's social protection programs. As expected, he finds that the results are quite sensitive to the assumed MPCT for generous programs like pen- sions or unemployment benefits and less sensitive for safety net programs with moderate generosity, like the Unified Monthly Benefit. We interpret the limited available evidence as follows. For programs with high gen- erosity and strong behavioral responses, the static targeting assessment may misspecify the counterfactual consumption, which in turn may lead to erroneous conclusions about tar- geting and benefit incidence (see van de Walle 2002a for an illustration from the Republic of Yemen). Thus for these programs, estimating the MPCT is important. When available data do not allow a robust estimation, analysts should at least undertake a sensitivity analy- sis. For safety net programs with moderate generosity, the results of a targeting assessment will not change substantially if behavioral responses are ignored. In this case, the best ap- proximation for the counterfactual income or consumption is the observed level of income or consumption minus the value of the transfer. Understanding Dynamic Targeting Assessment. Dynamic incidence is a term used to describe a case where quantiles are based not on households' current welfare, but on how household welfare has changed over time. It can therefore be used to describe whether a program is reaching those most severely affected by an economic shock. Dynamic inci- dence requires panel data. Sumarto, Suryahadi, and Pritchett (2003) study the static and dynamic incidence of two Indonesian safety net programs: the JPS Operasi Pasar Khusus (Special Market Operations), which sells subsidized rice to targeted households, and an employment cre- ation program in place in 1997­8 during the South Asian financial crisis. The former used administrative targeting while the latter were self-targeted via the wage. The authors classi- fied households into static quintiles based on the consumption level in May 1997 (before the crisis) and dynamic quintiles based on the change in consumption from May 1997 to August 1998. The authors find that the employment creation scheme was much more responsive to household shocks than the sales of subsidized rice. For example, a household from the middle of the expenditure distribution before the crisis that suffered the worst shock was four times more likely to have participated in the employment creation pro- gram than a household with a positive shock, but only one-and-a-half times more likely to receive subsidized rice. The authors therefore conclude that self-targeted schemes perform better during crises. An analysis of Hungarian programs (Ravallion, van de Walle, and Gautam 1995) looks at changes in poverty and cash benefits during the transition. It shows that social as- sistance was helpful in reducing poverty during a period of change accompanied by a great deal of transient poverty, though much of the effect was due to an increase in spending rather than an improvement in targeting. An analysis for Vietnam for 1992 and 1997 (van de Walle 2002b) shows that during this period of rapid growth and significant poverty reduction, the safety net was ineffective because of a combination of low spending, low coverage, and poor targeting. A good deal of movement in and out of poverty was occur- ring, and the safety net did not do well at targeting those who suffered shocks, although in this case the targeting was no worse than the targeting of the chronically poor. These analyses underscore the need to use dynamic analysis to look at how well transfer programs actually protect their beneficiaries from shocks. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 229 Simulating How Changes in a Program Would Affect Incidence. Most targeting as- sessments describe the incidence actually observed in a program, that is, the average inci- dence, but policy makers often want to know how the benefits or losses occasioned by a change in a program will be distributed. The simplest case is when policy makers are considering a proportional increase or decrease in the benefit level of a program. In this case, the standard targeting assessment is a marginal method: it gives a first-order approximation of the distributional consequences of a change in the level of benefits (Younger 2003).5 Changes in the benefit formula may also leave the distribution of existing benefi- ciaries largely unchanged, hence the standard targeting assessment can still be used to estimate the incidence of beneficiaries. However, these changes will affect the volume of benefits accruing to different quantile groups. To estimate the benefit incidence properly, analysts should use more complex methods that model the decision to participate in the program and apply this to the new schedule of benefits (Sahn and Younger 2000; Younger 2003). A more complex case is a change in eligibility criteria. Analysts can estimate the re- sultant distribution with various degrees of sophistication. The most common method is a simple calculation of who would benefit without taking behavioral changes into account. Countries considering introducing a proxy means test typically simulate the distribution of benefits assuming complete take-up and without modeling changes in labor supply, savings, or the like, and countries may do the same for other program changes. Figure 6.6 shows a calculation under- taken when policy makers FIGURE 6.6 Ex Ante Estimation of the Average Net Impact of were thinking through Reforming Fuel Subsidies in Indonesia proposals to replace In- donesia's energy subsidies Percentage of average per capita expenditures with a cash transfer (see 20 chapter 10, section 5, for Scenario 1 15 more discussion). 10 Another case of topi- Scenario 2 cal interest is an expansion 5 or a contraction of a pro- 0 Scenario 3 gram. The marginal inci- -5 dence will differ from the -10 average incidence if those 1 2 3 4 5 6 7 8 9 10 brought into or removed (poorest) (richest) Population deciles from a program are, on average, more or less poor SOURCE: Arulpragasam 2006b. than those already in the NOTE: Scenario 1: perfect targeting of cash transfer to bottom 28 percent of the population; scenario 2: slight mistargeting, bottom 40 percent of the popula- program. Consider a so- tion receives random cash benefits; scenario 3: greater mistargeting, bottom 60 percent of the population receives random cash benefits. cial insurance program that covers about 70 per- cent of the population, all at the upper end of the distribution. An expansion of coverage to 80 percent could be expected to benefit only those in the bottom third of the income distribution, a much more pro-poor marginal incidence for the benefits of expansion than 230 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS for the average incidence of the existing program. For a safety net program that is tightly targeted to the poorest 10 percent of the population, an expansion to cover the poorest 20 percent would also show different marginal than average incidence, but in this case the marginal incidence would be less pro-poor than the average incidence. Lanjouw and Ravallion (1999b) develop a political economy model where the poor and the nonpoor have different amounts of political power for influencing the alloca- tion of a program and where a given public expenditure has different costs and benefits. Lanjouw and Ravallion apply this model to three poverty alleviation programs using data from India's National Sample Survey for 1993­4--a public works program; the Integrated Rural Development Programme, a means-tested credit scheme; and the public distribu- tion system, a food rationing scheme--and find that additional spending would be signifi- cantly more pro-poor than suggested by the average incidence of participation. Van de Walle (2003) provides a basic reference and primer on how to compute marginal incidence with a single cross-sectional, repeated cross-sectional, or panel dataset. Younger (2003) provides a comparative assessment of different marginal targeting assess- ment techniques using as an example participation in secondary school in rural Peru in 1994 and 1997 and discusses when using each of the techniques is appropriate and the precision of different targeting assessment estimates. Understanding the Value Added of a Targeting Assessment A targeting assessment looks at an important intermediate outcome of safety net pro- grams: do benefits reach the poor and vulnerable? If benefits do not reach them, then the program cannot have an impact on them. If benefits do reach the poor and vulnerable but many others as well, the program may operate at relatively high cost. A targeting assess- ment does not yield as rich information as an impact assessment, but helps policy makers understand if some of the necessary conditions for impact have been met. As targeting assessments can often be carried out with no extra data collection or with only booster samples to regularly scheduled surveys and as the analytic techniques are not excessively specialized, they can be carried out more frequently than full impact assessments. IMPACT EVALUATION Impact evaluation estimates a program's causal effect on the outcomes it seeks to influ- ence--that is, it measures the changes in participants' well-being that can be attributed to a particular program.6 The specific technique for estimating impacts varies accord- ing to the setting, but the fundamental conceptual exercise remains the same. Impacts are determined by comparing the outcomes of program participants with the outcomes those same individuals would have experienced in the absence of the program. Such an experiment is impossible in practice, of course, and all methodologies center on ways of constructing a plausible comparison group or counterfactual. For example, a CCT program that provides cash to households conditional on their children attending school and obtaining regular medical care may have three main objectives: to increase school attendance, reduce morbidity, and reduce poverty among participants. An impact evalu- ation would determine whether the program was actually achieving higher graduation rates, fewer sicknesses among children, and less poverty than would have been the case without the program. In this case, impact evaluation will measure the graduation rates 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 231 among program participants and compare them with the estimated rates that would have prevailed for the same participants in the absence of the program. The program's impact on enrollment is the difference between the two outcomes. Stakeholders may find impact evaluations useful in a number of respects. For exam- ple, an impact evaluation of a cash transfer program for low-income households may esti- mate the program's impact on beneficiaries' earnings, total income, savings, food security, and poverty status (the main objectives); on child welfare and the utilization of other safety net programs (the intermediate objectives); and on employment, remittances, fertility, and marriage (expected collateral or negative impacts). Evaluations of CCT programs in Latin America have estimated their contribution to higher school enrollment and improved academic results, lower levels of child labor and malnutrition, better access to and use of health services, and improved health status. An impact evaluation of a workfare program will typically calculate participants' net wage gains and the indirect benefits to the com- munity of the assets created by the program. Because a program may affect many outcomes over varying time horizons and for different subgroups of the population, impact evaluations are rarely one-time events. Of- ten a series of evaluation reports looks at different dimensions of a program. Each wave of data collection may generate several evaluation reports and different queries may prompt the gathering of new data. Thus while we refer to "an evaluation," various products are more likely to be the case, in some situations initially planned as a complementary bundle, sometimes not planned that way, but evolving in that direction over time. The initial eval- uation of PROGRESA, for example, started with a set of complementary reports looking at the program's impacts on consumption (Hoddinott, Skoufias, and Washburn 2000), health (Gertler 2000), nutrition (Behrman and Hoddinott 2005), and education (Schultz 2000, 2004). Subsequent rounds of data collection and subsequent reports have gone into more depth; they have explored other outcomes, such as adult labor supply (Skoufias and di Maro 2006), migration and fertility (Stecklov and others 2005, 2006), and household investment behavior (Gertler, Martinez, and Rubio-Codino 2006). Many other studies have sought to answer a series of questions posed by management, such as how to achieve greater reductions in anemia or the reasons for lower impacts among indigenous popula- tions. The results of an impact evaluation enable policy makers and program managers to answer the following questions: (1) Does the program achieve its intended goal or goals? (2) Can the changes in outcomes be explained by the program or are they the result of some other factors occurring simultaneously? (3) Do program impacts vary across differ- ent groups of intended beneficiaries, across regions, and over time? (4) Does the program have any unintended effects, either positive or negative? Armed with such evidence, policy makers can decide whether to expand, modify, or eliminate a particular program. Many examples of rigorous and well-planned impact evaluations that have proved extremely use- ful are available, two of which are highlighted in box 6.6. Many others form the basis of our understanding of the potential of different sorts of programs. However, evaluations cannot answer every question policy makers might ask in rela- tion to a particular program. In particular, they cannot address many "what if" questions. What if the program were made national? What if the proxy means-test formulas were changed? Such questions can be examined, even if not definitively, using a variety of ex 232 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 6.6 Ignorance Has High Costs PROGRESA, Mexico. In Mexico, a change in government is a serious threat for the survival of social programs. Programs closely associated with the public image of a president are espe- cially vulnerable, as are those perceived to have been abused by a specific political party for electoral purposes (Levy and Rodríguez 2005). To prevent such risks, the CCT program PROGRESA invested in a credible impact evaluation and a transparent dissemination strategy. To ensure the credibility of the results, PROGRESA opted for a randomized evaluation undertaken by an external evaluator.The result was a series of evalu- ation reports. The impact evaluation showed that the program was cost-effective; that it selected its target population appropriately; and that it had a positive impact on education, health, nutrition, and diet.The results of the evaluation, including the survey data collected, have been placed in the public domain via the Internet. The results have been broadly disseminated, and special efforts have been made to ensure that Congress has complete information on the program's objectives, methodology, and results. Special efforts have been made not to associate PROGRESA with a particular political party. In 2000, when the Zedillo administration left office, the incoming Fox administration could de- cide on the continuation of the program based on hard evidence. Given its demonstrated suc- cess, the program was continued and expanded from rural to urban areas. Only the name did not survive the change in government: the program is now known as Oportunidades. The program gained such credibility that in the 2006 elections, all major candidates supported it and it survived another electoral cycle. National Job Training Partnership Programs, United States. In the mid-1980s, the U.S. De- partment of Labor commissioned a study of the National Job Training Partnership Act to study the effectiveness of programs funded by the act. The evaluation is one of the largest and most comprehensive of its kind ever undertaken. Some 20,000 program applicants from across the country were included in the experimental design to estimate the impacts on earnings, employ- ment, and welfare receipt of individuals served by the programs. The ambiguous results pro- vided by earlier, nonexperimental evaluations and the large budget for the programs triggered the decision to undertake the evaluation. The results were released to the public in 1994. Among the findings of the study were that the act had very different effects for adults than for youth. For adults, the programs raised participants' earnings by 7 to 11 percent and provided benefits of about US$1.50 for every US$1.00 invested; however, the programs had no significant impact on earnings for youth, and the costs to society exceeded the benefits.Following the release of the eval- uation findings, Congress reduced the budget for the youth component by more than 80 percent (US$500 million annually) and increased the budget for the adult component by 11 percent. Even though the evaluation took eight years to complete and cost US$23 million, it succeeded in having money shifted from a component of the program with no impact to more effective programs. The study has also yielded longer-term benefits in terms of improved knowledge and basic re- search. Academic researchers and others have used its rich dataset to study a range of topics from different aspects of job training interventions to evaluation methodology itself. SOURCES: Greenberg and Shroder 2004; Levy and Rodríguez 2005; Orr 1998. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 233 ante simulation and modeling methods that are generally not part of standard evaluations. Impact evaluations focus on assessing the existing program as implemented. Establishing a Counterfactual The essential conundrum of evaluation is that evaluators want to know what would have happened to program beneficiaries in the absence of the program, but can only observe them in the context of the program. Finding a good counterfactual--a control group similar in all respects to the treatment group except for receipt of the program--is key to the reliability of an impact evaluation. Credible evaluation uses robust statistical techniques to construct the counterfactual. These techniques include randomization or experimental design; quasi-experiments such as matching, regression discontinuity design, or double differences; and nonexperimental or instrumental variable methods (annex 6.3). All these techniques try to guarantee the comparability of the treatment and control groups by avoiding what statisticians refer to as bias (box 6.7). As increasing numbers of safety net programs are evaluated using robust tech- niques to generate a counterfactual, a number of quick and easy methods used in the past are quickly falling out of favor, such as before and after program comparisons or comparisons of beneficiaries with nonbeneficiaries. The latter employed two variants: comparisons with those who chose not to enroll in the program or comparisons with those who were not offered the program. The selection of a counterfactual group is easier for before and after or with and without program counterfactuals, but the groups are rarely similar to the beneficiaries of a safety net program, as we cannot know why some people enrolled in the program while others did not. Before and after comparisons may be biased by events that occur during the life of the program and affect program outcomes. For example, a cash transfer program may increase the income of beneficia- ries compared with the absence of the program, but cannot entirely mitigate a recession that reduces the incomes of the entire population. This was the case of the Programa de Asignación Familiar II (Family Allowance Program II) in Honduras. Between 2000 and 2002, an economic recession hit the country, reducing the consumption of the poor by about 14 percent. Program beneficiaries received a transfer equivalent to 6.0 to 6.5 per- cent of their consumption, not enough to fully mitigate the impact of the economic crisis. The impact evaluation showed that consumption by participating households was 7 percent higher than that by a comparable control group selected via randomization. A before and after comparison showed that beneficiaries' consumption fell by 6.5 percent from 2000 to 2002, and may misleadingly have led to the conclusion that the program had a negative impact on consumption. One of the major drawbacks of the quick and easy types of evaluation is that they tend to generate different results over time. For example, two separate evaluations of Peru's social fund completed a year apart and using different methodologies and data arrived at opposite conclusions on key impacts. This and similar examples reinforce the point that evaluation design is critical. Estimating Program Impact Four main methods are available for estimating the impact of a program: 234 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 6.7 The Problem of Bias Determining program impacts is a matter of accounting for as many of the personal, social, economic, and other factors that influence the outcome of interest to isolate the effects of par- ticipation in the program. This is usually addressed by comparing the outcomes of the treat- ment group with those of a comparison group where the groups are similar to each other in all respects except program participation. The similarity of the two groups in the absence of the program is therefore crucial. Differences between the comparison and treatment groups can result in a biased estimate of program impacts in two ways: · Differences in observable characteristics. If the treatment and comparison groups are very different from one another on measurable factors such as age, education, or eco- nomic status, then disentangling the effects of these variables from participation in the program becomes difficult. · Differences in unobservable characteristics. The two groups may differ in ways that are not measurable but that are related to participation in the program. For example, in- dividuals who choose to participate in a program may be more highly motivated or more able than those who decide not to participate, making them more likely to show positive outcomes even without the program. Analysts will attribute the resulting differences in the outcome of interest to the program, while in reality they may be due to the unobservable differences between the groups. This is often called selection bias. The only way to eliminate both sources of bias is to randomly assign individuals or households that volunteer to participate in the program into treatment and control groups.This experimental design ensures that, with a large enough sample, the two groups are statistically similar in terms of observable and unobservable characteristics. Experimental evaluation designs are expensive, however, and require advance planning and cooperation from the authorities. Careful nonrandom selection of the comparison group can sig- nificantly reduce the bias from observable characteristics and adequate data can help reduce the selection bias under certain circumstances, but there is no way to ensure that selection bias has been eliminated and no way to determine in advance how big a problem this will pose. A trade-off therefore exists between the preferred methodology of experimental design and the less expensive and timelier application of comparison group strategies. SOURCES: Baker 2000; Orr 1998; Ravallion 1999, 2008. · Comparison of means is a method of estimating impact using an experimental design that involves comparing means of treatment and control groups. A ran- dom allocation of the intervention among eligible beneficiaries creates compa- rable treatment and control groups. The program's impact on the outcome being evaluated can be measured by the difference between the means of the samples of the treatment group and the control group. This method can be used only if the counterfactual has been built using experimental and quasi-experimental design. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 235 Note that this method only uses observations made at one point in time and therefore assumes that the outcomes of the treated and the counterfactual popula- tions evolve in a similar way over time. · Double difference or difference-in-differences method is another estima- tion method that can be used with experimental, quasi-experimental, and nonexperimental designs. Impact is estimated by comparing the outcomes for the treatment and the comparison groups (first difference) before and after the inter- vention (second difference). This method requires baseline and follow-up data from the same treatment and control groups, ideally as panel data. If the samples for the follow-up survey differ from the baseline survey, they should be from the same geographic clusters or strata in terms of some other variable. · Multivariate regression is used with nonexperimental designs to control for pos- sible observable characteristics that distinguish participants and nonparticipants. If controlling for all possible reasons why outcomes might differ is possible, then this method is valid for estimating the program's effects. The differences in the mean outcomes of the two groups, participants and nonparticipants, conditional on the set of variables that cause outcome and participation, constitute the pro- gram or treatment effect. · Instrumental variables are used with nonexperimental design to control for se- lection bias. These variables determine program participation, but do not affect outcomes. Evaluators can often use geographic variations in program availability and program characteristics as instruments, especially when endogenous program placement seems to be a source of bias. Implementing an Impact Evaluation An impact evaluation involves several steps: (1) establishing evaluation objectives, (2) de- termining appropriate evaluation methods, (3) collecting data, and (4) producing and disseminating findings. A main factor that influences each of these steps is their cost. To provide the highest value, an impact evaluation should include the following key design features: · Clear objectives. Evaluation questions should be determined early during the process and should be simple and measurable. · Credible evaluator. The evaluator should have the required specialized skills. An evaluator external to the agency or government is generally preferable to ensure objectivity and independence. · Rigorous methodology. Experimental estimates are the ideal, but a well-chosen, matched comparison group may suffice. · Adequate sample size. The sample should be large enough to detect program ef- fects of plausible size. In addition, the size should permit assessment of program impacts on key subgroups of the target population as appropriate. Minimum de- tectable effects should be determined prior to implementation of the evaluation. · Baseline data. These are needed to establish the appropriate comparison group and to control for observable program selection criteria. 236 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS · Sufficient follow-up. Follow-up data should be collected after enough time has passed to plausibly detect an impact and should measure the relevant outcome variables. The costs of impact evaluations of safety net and other social programs vary consid- erably, ranging anywhere from US$200,000 to more than US$1 million, with an average for a rigorous evaluation probably amounting to around US$300,000 to US$400,000. Features affecting total costs include the number and type of policy questions to be ad- dressed, the methodology, the extent of collection of new data, the size and scope of the program being evaluated, and the level of local capacity. The largest single cost item is usually data collection, which can vary widely depending on sample size, complexity of the survey effort, and the number of rounds of data collection. In reviewing 125 evaluations of World Bank health, education, and social protection projects, Fiszbein (2008) reports that data collection accounted for half or more of evaluation costs. Deciding When Impact Evaluations Should Be Conducted Impact evaluations demand a substantial amount of information, time, and resources; therefore the public actions that will be evaluated should be carefully selected. This can be done by asking three basic questions. If just one is answered affirmatively, this should support the need for a rigorous impact evaluation: · Is the program of strategic relevance for national public policy? · Can the evaluation results influence the design of the program? · Will the evaluation contribute to improving the state of knowledge about a par- ticular type of program or policy and does the information generated have poten- tial future research value? One important caveat is that fruitful evaluations require sufficiently mature pro- grams. Even though programs may be testing innovative approaches, before they can be evaluated they need clearly defined objectives, well-delineated activities, and a stable in- stitutional framework for implementation. Many programs are not ready for an impact evaluation, and some programs will never be worth a full evaluation. Before deciding to evaluate a program, assessing whether the program is "evaluable" is important. Such an undertaking can save substantial human and financial resources. A program must fulfill a number of criteria before being evaluated (table 6.11). An evaluable intervention is stable as opposed to still developing or likely to change in major ways, is clear about its goals, is reasonably homogenous across different program sites, is substantial enough to have impacts, has enough participants to demonstrate results, and can be documented. Politics and the political economy play an important role in the decision to con- duct a program evaluation. The issues stem from principal-agent problems, where stake- holders--including the government or funding agency, the implementing agency, and the beneficiaries--do not have consistent incentives to support an evaluation. Several factors, as follows, work against the decision to undertake an impact evaluation: · Managers of highly constrained or poorly performing programs may fear docu- menting their poor results (Pritchett 2002). Negative findings have the potential 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 237 TABLE 6.11 When to Conduct an Impact Evaluation: A Checklist Issue Criterion Question Well-defined Are the program's objectives clearly defined? objectives or not Short or long term Can final outcomes be assessed before the evaluation ends? Nature of outcomes Unique or multiple Is expressing the effects of the program using few outcomes reasonable? Quantitative or Are outcomes qualitative in nature or difficult to measure? qualitative Heterogeneity Is the program a mixed bag of different interventions? Nature Implementation Is implementation likely to vary significantly within the program? of the program Unit of analysis Is the program directed at areas or communities and/or is its sample too small? Scale effect Is the potential impact on outcomes likely to be small relative to the scale of the problem? External Is the potential impact on outcomes likely to be small relative to Context influences other external influences on outcomes? Active or passive Is the effectiveness of the program sensitive to how clients choose clients to respond to it? SOURCE: Authors. NOTE: To proceed with an impact evaluation, the first three questions should be answered "yes," and the remainder "no." to hinder social agendas and damage political careers, thus for policy makers not to present their detractors with a club to beat them may seem easier and safer. Even managers of better-performing programs can fear results that are ambiguous or difficult to translate into policy actions. · The time horizons of policy makers, program managers, and evaluation processes differ. National governments usually serve for only four or five years and local governments for only one or two years, but evaluations can take several years from planning to results. Managers with many competing priorities may tend to focus on actions with more immediate payoffs. · Evaluation is a public good, yielding not only lessons that may be important to the specific program being evaluated, but to similar programs (Duflo, Glenner- ster, and Kremer 2008). As individual policy makers and the government do not reap the all the benefits of the evaluation, they will not have an incentive to invest fully in it. At the same time, program officials may support evaluation of their programs either to be able to improve them or to document their successes and use that documentation for seeking funds; ministries of finance or donors may support evaluations to help them make allocation decisions; and civil society groups may support evaluations as a way to increase government accountability. 238 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Governments and their international partners can do a number of things to encour- age impact evaluation, or the broader M&E agenda, at the program level. Governments may offer to help program managers by covering the costs of M&E by subsidizing such budgets in cash or providing in-kind assistance. They may also reward managers of pro- grams with good M&E systems through increased budgets or managerial autonomy. Gov- ernments can mandate regular cycles of M&E for all programs by law or as part of bud- get reviews or legislative oversight. In addition, they can put in place public information practices that implicitly raise the pressure on programs to provide information. Finally, leadership of and support for M&E can be articulated in policy statements from the chief executive on down (McKay 2007). Notes 1. The PSNP is part of a larger safety net, the Food Security Program, deployed by the govern- ment of Ethiopia to protect chronically food-insecure households. The Food Security Program includes three initiatives: the PSNP, the Resettlement Program, and the Income Generation Program. The government agency coordinating the Food Security Program, the Food Security Coordination Bureau, developed an M&E framework for the entire Food Security Program; however, to simplify the presentation of M&E concepts, this section only presents the logical framework of the PSNP. 2. Assessing targeting efficiency is less critical for other public programs that may aim for univer- sal coverage, such as the provision of health care and education, although the issue of whether the poor are included in such programs still remains and a targeting assessment will provide the answer. 3. If the incidence results are presented using household quantiles, judging whether the distribu- tion of benefits is progressive or regressive will be difficult without additional information. For example, if we find that 40 percent of program benefits accrue to the poorest population quintile, then we immediately know that, on average, each individual recipient in that quintile got twice as much as they would have received if the money had been allocated randomly. If the findings refer to household quintiles, we do not know how successful the program target- ing was. If the households in the poorest quintile are twice as large as the average household, then they would represent 40 percent of the population. In this case, the program targeting is no better than a blind, random allocation. We favor the use of population quantiles. 4. For normal goods, whose consumption increases with the income level of the beneficiary, rich consumers will capture a larger amount (and share) of the total subsidy. Hence beneficiary incidence will underestimate benefit incidence. 5. We term this first-order approximation, because eligible beneficiaries may respond differently to an increase or decrease in the benefit level. For example, more generous benefits may induce marginally eligible households to apply for the program. 6. This subsection draws heavily on Baker (2000); Blomquist (2003); and Prennushi, Rubio, and Subbarao (2002). 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 239 Annex 6.1 Sample M&E Indicators for Typical Safety Net Interventions I. Cash transfers Input indicators · Budget allocation and expenditures · Number of program staff by level Output indicators Beneficiaries · Number of program beneficiaries (total and as a percentage of the estimated tar- get population) · Key characteristics of beneficiaries: gender, age, level of education, number of dependents, employment status (or employment history depending on eligibility requirements), household income, location Benefits and services · Amount of benefits paid (total per payment period by area) Intermediate outcome indicators Access and satisfaction · Targeting efficiency as measured by inclusion and exclusion errors · Average time of program participation · Beneficiary satisfaction with program access and delivery Outcome indicators Depending on program objective · Increase in beneficiary household consumption · Increase in beneficiary consumption of specific products (for example, food) · Decrease in poverty incidence or depth Process and efficiency indicators (to be compared to targets, past performance, or other measurement units) Entry (outreach, targeting, registration, and so on) · Average (and range) of time for processing applications to the program (calendar days following the application) · Number of benefits processed per month per staff member of the implementing agency Payment · Cost of processing payments per beneficiary · Beneficiaries experiencing payment delays as a percentage of total beneficiaries · Beneficiaries not collecting their payments as a percentage of total beneficiaries · Amount of assistance provided to individual clients; office or facility that provided the assistance; method of service delivery (useful for testing different approaches); individual supervisor or caseworker 240 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Exit · Average (and range of) time for cancellation of the benefit (calendar days follow- ing a finding of ineligibility, fraud, or the like) · Beneficiaries whose benefit is canceled as a percentage of total beneficiaries by reason for cancellation Administration · Average administrative cost of program (and range) per beneficiary · Average benefit (and range) per beneficiary (depending on the terms of the pro- gram) · Total amount of benefits paid as a percentage of total cost to the government of the program · Amount used as a percentage of the amount allocated II. Food-based transfers Input indicators · Budget expenditures · Number of program staff by level · Quantity of food available Output indicators Beneficiaries Of food for education, school feeding or take-home rations programs · Number of schools covered · Number of children who received a ration or meal (by gender) · Number of rations or meals distributed Of nutrition (child and maternal health) programs · Total number of participating health centers, community centers, village volunteers · Number of pregnant and lactating women who received a monthly take-home ration of fortified food · Number of children aged 6 to 59 months who received a monthly take-home ra- tion of fortified food · Number of pregnant and lactating women who participated in health care and child nutrition training sessions · Number of take-home rations of fortified food distributed Intermediate outcome indicators Food for education, school feeding, or take-home rations · Enrollment of poor children · Quality of food served in schools Nutrition (child and maternal health) · Prevalence of poor and mothers participating in the program · Quality of services provided (amount of waiting time) Outcome indicators Food for education, school feeding, or take-home rations · School enrollment rate (by gender and grade) 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 241 · School dropout rate (by gender and grade) · Nutritional status of children Nutrition (child and maternal health) · Decrease in prevalence of low body mass index among women 2 to 6 months postpartum · Decrease in prevalence of iron deficiency anemia among pregnant and lactating women and children aged 6 to 59 months · Decrease in prevalence of malnutrition (severe and moderate) in children Process and efficiency indicators Food for education · Total number of people involved in schools per beneficiary served · Average frequency of distribution (and range) of rations and meals · Average cost (and range) of rations and meals (disaggregated as appropriate by region and other characteristics) · Average program cost per beneficiary Nutrition (child and maternal health) · Average cost (and range) of take-home rations · Average share in costs of fortified, blended food · Average cost (and range) of health care and child nutrition training sessions · Average cost per beneficiary (per child and per pregnant or lactating woman) III. Public works programs Input indicators · Budget expenditures for salaries, intermediate inputs, and administration · Amount of food available in the budget (food-for-work projects) · Number of program staff by level Output indicators Projects · Number of workfare projects by type (for example, with and without financing of materials) and by province or region · Project specific: actual kilometers of water or sewer lines or roads maintained or built · Wages paid to workers (per day, per month, by province, and overall) · Amount of food distributed as wages (for food-for-work projects) Beneficiaries · Number of workers participating in the program · Total number of beneficiaries employed in each activity · Key characteristics of beneficiaries: gender, age, previous economic activity, educa- tion level, number of children, previous participation in an employment or training program, household income, confirmation of education and health certificates · Actual number of unemployed people who received the minimum wage Intermediate outcome indicators Projects 242 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS · Location of projects in poor areas (correlation of number of projects and total expenditures with the incidence of poverty, number of unemployed poor, and so on within the country and within provinces) · Quality of projects completed · Utilization by poor communities of infrastructure built, expanded, or rehabili- tated under the program Beneficiaries · Number of low-income workers employed in the project (total target, gender- specific target) · Beneficiaries experiencing payment delays as a percentage of total beneficiaries Outcome indicators · Increase in net annual earnings of the average individual beneficiary · Number of program beneficiaries who transitioned from workfare to formal sec- tor employment · If the objective is to fight seasonal hunger: percentage of beneficiaries whose diet improved · Increase in second-round effects resulting from projects, for example, the number of people accessing roads or other infrastructure built or maintained Process and efficiency indicators Projects · Average time taken to select viable projects (in calendar days) · Number of projects appraised and evaluated per month (overall and by province) · Number of projects evaluated as a percentage of total projects per month (overall and by province) · Number of projects supervised per supervisor per month · Number of supervision visits per project per month (overall and by province) · Average number of supervision visits per project during project execution (overall and by province) · Number of workfare activities executed by province (with and without financing of materials) · Number of supervision visits to training courses and basic education courses · Percentage of projects located in poor areas (quintiles 1 and 2) (target = 100 per- cent) · Wages paid as a percentage of the contract amount · Average cost (and range) per project category · Average share of labor cost (and range) per project category · Average share of the cost for wages in food (for food-for-work projects) Additional related objectives (such as community involvement) · Percentage of projects with participation by nongovernmental organizations, civil society organizations, and so on (overall and by province) · Percentage of projects sponsored by nongovernmental organizations, municipali- ties, and the like (overall and by province) Jobs · Jobs provided per estimated target population (overall and by province) 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 243 · Poor (bottom quintile) workers as a percentage of public works laborers Administration · Amount spent as a percentage of the amount allocated by province · Efficiency of employment program (value of salaries received by workers as a per- centage of total government cost of program) · Labor intensity of projects · Unit cost, for example, by kilometer of road built · Average cost per beneficiary by project type IV. Conditional cash transfers The indicators used to monitor cash transfers are also used for conditional cash transfers. In addition, the following aspects are monitored. Output indicators Benefits and services · School attendance rate by children · Health care utilization by children from birth through six years old and not en- rolled in school Intermediate outcome indicators Access and satisfaction · Beneficiary satisfaction with availability and access to schools · Beneficiary satisfaction with availability and access to health care facilities Outcome indicators Education-based requirements · Change in school attendance, primary and secondary school · Change in secondary school enrollment Health care­based requirements for children and adults · Change in the percentage of children brought to health centers for preventive care · Change in the number of children aged newborn through age six and not enrolled in school who have received all required immunizations on time · Change in the number of poor pregnant and lactating women visiting health centers for timely checkups · Change in the number of poor elderly, disabled, and other beneficiaries visiting health centers Process and efficiency indicators · Administrative costs for beneficiary selection, delivery of cash, and verification of compliance 244 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Annex 6.2 Monitoring of Oportunidades, Mexico TABLE A6.2.1 Selected Monitoring, Evaluation, and Performance Indicators Used by Oportunidades, Mexico, 1999­2003 1999 2000 2001 2002 2003 Jan.­ July­ Jan.­ July­ Jan.­ July­ Jan.­ Indicator Feb. Aug. Feb. Aug. Feb. Aug. Feb. % of beneficiary 95.95 97.44 97.98 97.95 97.75 97.97 97.62 families monitored (84.80­ (88.71­ (94.49­ (93.78­ (91.34­ (91.63­ (93.26­ 120.23) 100.00) 100.00) 100.00) 100.00) 100.00) 100.00) % of children under -- 73.94 90.85 92.53 93.08 94.27 95.08 2 under nutritional (47.19­ (74.50­ (69.09­ (65.61­ (83.04­ (81.25­ surveillance 100.00) 100.36) 100.00) 125.09) 100.00) 99.99) % of children aged -- 72.76 90.37 91.54 92.93 94.00 95.12 2­4 under nutritional (43.28­ (78.15­ (68.57­ (74.86­ (82.77­ (79.90­ surveillance 100.00) 100.00) 100.00) 122.70) 100.00) 100.00) % of children under 7.38 20.20 16.12 16.29 15.75 16.46 15.48 2 with malnutrition (0.00­ (3.61­ (2.22­ (4.40­ (4.73­ (5.95­ (5.14­ 16.52) 33.55) 31.89) 32.87) 32.66) 32.61) 31.63) % of children aged 10.87 35.82 28.03 27.16 25.33 25.96 24.52 2­4 with malnutrition (0.00­ (5.93­ (5.52­ (5.98­ (5.69­ (7.77­ (7.85­ 24.09) 52.94) 50.33) 50.48) 50.03) 49.53) 47.76) % of pregnant 215.04 87.42 92.82 93.76 95.76 96.78 97.38 mothers registered (129.18­ (65.28­ (67.98­ (70.47­ (76.58­ (84.36­ (83.44­ with prenatal care 498.11) 100.90) 100.00) 101.03) 125.37) 100.00) 100.00) facilities Avg. # of prenatal 5.18 1.78 1.62 1.81 1.73 1.81 1.79 visits per pregnant (4.02­ (1.30­ (0.85­ (1.12­ (0.89­ (0.98­ (1.28­ women 6.32) 2.24) 3.09) 2.49) 2.22) 2.45) 2.57) % of nursing mothers 78.85 89.26 93.43 92.70 94.66 96.16 97.13 under monitoring (49.37­ (73.33­ (70.02­ (63.19­ (72.61­ (68.62­ (80.89­ 239.23) 100.00) 100.00) 100.44) 122.12) 99.98) 100.00) % of children 289.40 87.61 51.87 76.34 76.19 76.56 77.55 under 2 who (34.04­ (41.76­ (20.54­ (42.40­ (35.32­ (44.48­ (53.41­ received nutritional 857.12) 127.09) 94.05) 95.92) 104.39) 101.12) 101.19) supplements % of children 132.83 130.97 80.06 121.53 125.67 117.79 114.81 aged 2­4 who (30.50­ (70.80­ (40.00­ (78.07­ (51.17­ (46.83­ (62.50­ received nutritional 386.00) 363.96) 373.35) 462.50) 862.35) 322.33) 363.60) supplements % of pregnant 308.45 94.88 64.65 83.60 83.96 83.85 84.57 women who received (133.16­ (69.66­ (25.12­ (44.23­ (59.15­ (55.21­ (62.48­ food supplements 2,083.90) 130.47) 98.91) 99.66) 104.54) 107.41) 109.95) (continued) 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 245 TABLE A6.2.1 (continued) 1999 2000 2001 2002 2003 Jan.­ July­ Jan.­ July­ Jan.­ July­ Jan.­ Indicator Feb. Aug. Feb. Aug. Feb. Aug. Feb. % of nursing 238.60 72.29 58.07 79.90 81.24 79.63 88.32 mothers who (106.07­ (50.25­ (33.46­ (30.17­ (49.71­ (47.26­ (63.93­ received nutritional 1,407.35) 98.46) 98.52) 99.56) 104.46) 109.14) 140.48) supplements % of children under 16.53 6.36 6.72 6.06 5.40 4.28 5.02 2 who recovered (2.17­ (2.35­ (2.10­ (2.14­ (0.00­ (0.00­ (1.05­ from malnutrition 542.86) 50.00) 274.42) 80.00) 64.39) 12.27) 15.08) % of children aged 19.74 6.52 5.97 6.05 4.96 4.42 4.76 2­4 who recovered (1.81­ (3.10­ (2.57­ (1.72­ (2.10­ (1.68­ (1.94­ from malnutrition 879.41) 77.9)3 111.34) 85.87) 31.77) 11.71) 14.09) % of children under 2 -- 15.41 12.54 12.47 12.17 12.53 11.85 with mild malnutrition (3.61­ (1.78­ (4.40­ (3.96­ (5.24­ (4.55­ 23.92) 23.51) 23.79) 23.82) 23.26) 22.59) % of children aged -- 28.46 22.66 22.02 20.69 21.01 19.87 2­4 with mild (4.78­ (4.14­ (5.56­ (5.40­ (7.25­ (7.10­ malnutrition 42.67) 41.22) 39.61) 38.05) 37.34) 35.98) % of children under -- 4.28 3.24 3.43 3.25 3.55 3.25 2 with moderate (0.00­ (0.43­ (0.00­ (0.35­ (0.63­ (0.55­ malnutrition 8.49) 7.30) 9.16) 7.88) 8.27) 7.94) % of children aged -- 6.79 4.95 4.77 4.31 4.60 4.29 2­4 with moderate (0.83­ (0.50­ (0.31­ (0.29­ (0.50­ (0.58­ malnutrition 13.99) 11.8)8 11.86) 10.89) 11.81) 10.75) % of children under -- 0.51 0.35 0.39 0.33 0.38 0.39 2 with severe (0.00­ (0.00­ (0.00­ (0.00­ (0.00­ (0.00­ malnutrition 1.13) 1.08) 1.47) 1.09) 1.33) 1.37) % of children aged -- 0.58 0.42 0.37 0.33 0.35 0.36 2­4 with severe (0.00­ (0.00­ (0.00­ (0.00­ (0.01­ (0.00­ malnutrition 1.38) 1.30) 1.13) 1.09) 1.13) 1.19) % of children with -- 4.02 3.27 3.93 3.31 3.04 2.61 low birthweight born (0.90­ (0.40­ (0.00­ (0.00­ (0.00­ (0.00­ to beneficiaries at 9.74) 33.33) 6.94) 15.79) 14.29) 5.63) attended births Avg. # of families -- 581 504 615 534 641 680 served by each delivery point % of total # of -- -- 2.04 2.02 2.82 0.83 2.01 participating families (0.75­ (0.00­ (1.71­ (0.00­ (0.003­ that did not pick up 5.48) 8.52) 14.70) 3.69) 4.48) their benefits (continued) 246 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS TABLE A6.2.1 (continued) 1999 2000 2001 2002 2003 Jan.­ July­ Jan.­ July­ Jan.­ July­ Jan.­ Indicator Feb. Aug. Feb. Aug. Feb. Aug. Feb. Education grant recipientsa # of recipients aged -- -- 38.46 61.42 -- 8­17 enrolled in (36.63­ 40.17) (58.35­ 66.62) basic ed. as % of all children 8­17 on roster as of beginning of school year # of recipients aged -- -- 5.17 13.27 -- 14­20 in upper-level (1.85­ 9.25) (7.13­ 23.46) secondary ed. as % of all children aged 14­20 on roster as of beginning of school year # of recipients who -- -- 95.72 94.80 -- finished primary, lower (92.04­ 97.41) (85.83­ 96.38) secondary, and upper secondary ed. as % of all recipients enrolled in last grade of each level at beginning of school year # of recipients who -- -- 71.68 85.05 -- remained enrolled (67.63­ 75.66) (80.77­ 88.54) in school as % of all recipients enrolled in previous year as of beginning of school yearb # of recipients who -- -- 96.58 95.12 -- finished school year (91.70­ 98.22) (87.99­ 96.74) as % of all recipients who began that yearb # of recipients -- -- 83.94 84.19 -- enrolled in next (76.21­ 92.22) (80.26­ 90.53) grade as % of all recipients who finished previous year as of beginning of school year SOURCE: Levy and Rodríguez 2005. NOTE: -- = not available. Data are national averages; ranges are indicated in parentheses and represent minimum and maximum values at the state level. Where the range extends beyond 100 percent, this reflects program expansion, when new or future beneficiaries are reported as being served by health or education facilities and this number is higher than the current number of beneficiaries. a. The data refer to September­October of each year. b. For 2001, the figure refers to primary education and for 2002 to lower and upper secondary education. TABLE A6.2.2 Indicators Collected from the Sample of Sentinel Points, by Area of Residence, in Percentages, Oportunidades, Mexico, 2000-02 2000 2001 2002 1st 2nd Aver- 1st 2nd Aver- 1st 2nd Aver- Item panel panel age panel panel age panel panel age Rural (Social Security Institute and Ministry of Health) Human Has a physician 96.01 94.70 95.35 96.05 98.50 97.28 98.15 99.20 98.68 resources Has a nurse 91.90 92.65 92.28 92.65 96.00 94.33 95.75 -- 95.75 Receives S2 84.09 86.77 85.43 88.70 86.60 87.65 92.15 95.80 93.98 Sufficiency of nutritional supplement for children 95.49 86.74 91.11 96.90 97.30 97.10 98.10 86.00 92.05 Sufficiency of nutritional supplement for women 95.03 87.82 91.42 96.70 96.30 96.50 98.20 -- 98.20 Supplies Availability of tools 87.80 84.97 86.39 87.00 91.10 89.05 90.25 -- 90.25 Has teaching materials 81.94 84.67 83.31 88.80 90.20 89.50 87.80 -- 87.80 Sufficiency of curative materials 85.52 75.69 80.61 79.90 83.30 81.60 81.95 -- 81.95 Sufficient drugs 72.11 73.41 72.76 71.15 75.35 73.25 72.55 76.80 74.68 Medical care 92.24 94.15 93.20 94.40 94.15 94.28 94.65 -- 94.65 Monitoring of delivery of nutritional supplement 91.61 93.06 92.33 95.05 96.70 95.88 96.80 -- 96.80 Attendance at educational sessions 92.94 95.50 94.22 94.70 98.05 96.38 97.75 94.40 96.08 Verification of health component activities by health information 95.01 95.35 95.18 92.30 96.45 94.38 96.40 -- 96.40 Records system for general population Bimonthly preparation of nutritional supplement requirements 89.99 90.37 90.18 90.60 92.95 91.78 92.35 -- 92.35 S1s updated 75.24 71.00 73.12 81.45 85.85 83.65 85.45 -- 85.45 Absences per 2-month period in S2 95.79 97.95 96.87 97.95 97.95 97.95 97.60 -- 97.60 Timely delivery of S2 97.22 97.50 97.36 98.75 99.45 99.10 99.45 -- 99.45 (continued) TABLE A6.2.2 (continued) 2000 2001 2002 1st 2nd Aver- 1st 2nd Aver- 1st 2nd Aver- Item panel panel age panel panel age panel panel age Monitoring of nutritional status for less than 5 years 90.33 91.87 91.10 92.90 94.45 93.68 95.30 -- 95.30 Food Training of mothers in preparation of supplement 95.11 94.85 94.98 94.95 98.40 96.68 97.95 -- 97.95 Waited 1 hour or more during a health clinic visit -- -- -- -- 65.00 65.00 58.60 62.10 60.35 Pregnant or nursing women who did not receive a supplement -- -- -- -- 8.00 8.00 4.70 4.94 4.82 Indicated not receiving a supplement for children who needed it -- -- -- -- 4.00 4.00 3.80 4.90 4.35 General Were charged for supplement (conditions imposed) -- -- -- -- 12.00 12.00 17.30 1.30 9.30 Indicated acute respiratory infections as common health -- -- -- -- 57.00 57.00 64.40 -- 64.40 problem among children in their community Indicated acute diarrheal diseases as common health problem -- -- -- -- 37.00 37.00 28.50 -- 28.50 among children Urban (Social Security Institute and Ministry of Health) Human Has a physician -- -- -- 96.00 94.70 95.35 98.90 99.47 99.19 resources Has a nurse -- -- -- 91.90 92.65 92.28 98.43 98.10 98.27 Receives S2 -- -- -- 83.75 86.75 85.25 95.57 96.04 95.81 Sufficiency of nutritional supplement for children -- -- -- 95.50 86.75 91.13 89.42 95.20 92.31 Sufficiency of nutritional supplement for women -- -- -- 95.00 87.80 91.40 88.22 95.71 91.97 Supplies Availability of tools -- -- -- 87.80 84.95 86.38 63.80 70.97 67.38 Has teaching materials -- -- -- 82.80 84.65 83.73 64.41 73.46 68.94 Sufficiency of curative materials -- -- -- 85.45 75.70 80.58 75.29 62.97 69.13 Sufficient drugs -- -- -- 73.35 73.40 73.38 49.68 56.33 53.00 (continued) TABLE A6.2.2 (continued) 2000 2001 2002 1st 2nd Aver- 1st 2nd Aver- 1st 2nd Aver- Item panel panel age panel panel age panel panel age Medical care -- -- -- 92.55 94.15 93.35 93.27 95.97 94.62 Monitoring of delivery of nutritional supplement -- -- -- 91.55 93.05 92.30 93.51 98.94 96.22 Attendance at educational sessions -- -- -- 92.90 95.50 94.20 96.15 99.10 97.63 Verification of health component activities by health information -- -- -- 94.30 91.90 93.10 84.54 97.68 91.11 Records system for general population Bimonthly preparation of nutritional supplement requirements -- -- -- 90.00 90.25 90.13 93.24 95.30 94.27 S1s updated -- -- -- 76.10 71.00 73.55 82.66 89.19 85.92 Absences per 2-month period in S2 -- -- -- 95.80 97.95 96.88 94.80 96.33 95.57 Timely delivery of S2 -- -- -- 97.20 97.50 97.35 99.33 99.77 99.55 Monitoring of nutritional status for less than 5 years -- -- -- 90.20 91.85 91.03 95.03 93.94 94.49 Food Training of mothers in preparation of supplement -- -- -- 95.50 49.85 72.68 98.31 98.11 98.21 Waited 1 hour or more during a health clinic visit -- -- -- -- -- -- 42.40 30.30 36.35 Pregnant or nursing mothers who did not receive supplement -- -- -- -- -- -- 9.90 5.70 7.80 Indicated not receiving a supplement for children who needed it -- -- -- -- -- -- 13.00 14.00 13.50 General Were charged for the supplement (conditions imposed) -- -- -- -- -- -- 13.20 16.70 14.95 Indicated acute respiratory infections as common health -- -- -- -- -- -- 62.50 64.30 63.40 problem among children in their community Indicated acute diarrheal diseases as common health problem -- -- -- -- -- -- 30.70 30.30 30.50 among children SOURCE: Levy and Rodríguez 2005. NOTE: -- = not available.The panels are for various two-month periods during the year. Upon registration at a health clinic, beneficiaries are given a booklet containing a schedule of appointments for each household member.This information is entered on an S1 form brought to the clinic by the beneficiary, ensuring that a record of attendance is kept at the clinic.The S2 is a form for register- ing household compliance/noncompliance with health conditions. It must be filled out by a nurse or doctor at the health unit every two months to certify that family members visited as required. 250 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Annex 6.3 A Summary of Experimental Methods The methods used to select the counterfactual or control group, known as evaluation designs, can be broadly classified into three categories: experimental, quasi-experimental, and nonexperimental. These three evaluation designs vary in terms of feasibility, cost, degree of clarity, validity of results, and extent of selection bias. This annex, adapted from Ravallion (2008), summarizes these designs. The experimental (randomized) design involves gathering a set of individuals (or other unit of analysis) equally eligible and willing to participate in the program and ran- domly dividing them into two groups: those who receive the intervention (the treatment group) and those from whom the intervention is withheld (the control group). Experi- mental designs are generally considered the most robust of the evaluation methodologies. By randomly allocating the intervention among eligible beneficiaries, the assignment pro- cess itself creates comparable treatment and control groups that are statistically equiva- lent to one another given appropriate sample sizes. This is a powerful outcome because, in theory, the control groups generated through random assignment serve as a perfect counterfactual, free from the selection bias that plagues other evaluations. The quasi-experimental design consists of constructing a comparison group using matching or reflexive comparisons. Matching involves identifying people who are not participating in the program who are comparable in terms of essential characteristics to participants. Matched comparison groups can be selected before project implementation (prospective studies) or afterwards (retrospective studies). Many methods are available for selecting a counterfactual group using matching techniques, some more robust, some less robust. The following three methods are likely to produce a robust counterfactual: · Propensity score matching. The most widely used type of matching is propen- sity score matching, in which the comparison group is matched to the treatment group by using the propensity score (predicted probability of participation given observed characteristics). This method allows the analyst to find a comparison group from a sample of nonparticipants closest in terms of observable charac- teristics to a sample of program participants. Score matching is a useful method when the analyst has to match many potential characteristics between a sample of program participants and a sample of nonparticipants. Instead of aiming to ensure that the matched control for each participant has exactly the same value of the control variable X, the same result can be achieved by matching on the predicted probability of program participation, P, given X, which is known as the propensity score of X. The range of propensity scores estimated for the treatment group should correspond closely to that for the retained sample of nonpartici- pants. The closer the propensity score, the better the match. A good comparison group comes from the same economic environment and is administered the same questionnaire as the treatment group by similarly trained interviewers. 6. USING MONITORING AND EVALUATION TO IMPROVE PROGRAMS 251 · Pipeline matching. This is another widespread type of matching in which groups of beneficiaries who have already received an intervention are matched against groups of beneficiaries selected to receive the program in the near future. · Regression discontinuity design. This is a combination of traditional random- ized experiments and quasi-experiments. In regression discontinuity designs, par- ticipants are assigned to either the program or comparison groups on the basis of a cutoff score that was assigned before the implementation of the program; for example, the score on a proxy means test for a needs-based program of last resort or age for a child allowance or social pension. Typically those scoring above or equal to a certain cutoff value will be allowed to participate in the program and those who score below the value will not. The main assumption is that those just below and above the cutoff point will have similar characteristics other than their participation in the program. Reflexive comparison is another type of quasi-experimental design. In a reflexive comparison, the counterfactual is constructed on the basis of the situation of program par- ticipants before the program. Thus program participants are compared before and after the intervention and function as both the treatment and the comparison group. This type of design is particularly useful in evaluations of full-coverage interventions, such as national policies and programs in which the entire population participates. There is, however, a major drawback with reflexive comparisons: the situation of program participants before and after the intervention may change for many reasons independent of the program; for example, participants in a training program may have improved employment prospects after the program. While this improvement may be due to the program, it may also be due to the fact that the economy is recovering from a past crisis and employment is growing again. Unless they are carefully done, reflexive comparisons may be unable to distinguish between the program and other external effects, thereby compromising the reliability of results. The nonexperimental evaluation design uses statistical econometric multivariate methods to account for differences between the two groups. In this case, instrumental variables is one of the econometric techniques that can be used to compare program par- ticipants and nonparticipants correcting for selection bias. It consists of using one or more variables (instruments) that matter to participation, but not to outcomes given participa- tion. This identifies the exogenous variation in outcomes attributable to the program, rec- ognizing that its placement may not be random, but purposive. The instrumental variables are first used to predict program participation, then the program's impact is estimated us- ing the predicted values from the first equation. As with quasi-experimental methods, this evaluation design is relatively cheap and easy to implement, as it can draw on existing data sources; however, it poses a number of difficulties. First, the reliability of results is often reduced, as the methodology is less robust statistically. Second, the methodology has some statistical complexities that may require expertise in the design of the evaluation and in the analysis and interpretation of results. Third, although partially correcting for selection bias is possible, full correction remains a challenge. CHAPTER 7 Understanding Common Interventions KEY MESSAGES Many different types of safety net programs exist: cash and in-kind transfers, general subsidies, public works programs, conditional cash transfers (CCTs), and fee waivers for health and education. This chapter describes the key design features, appropriate context, implementation challenges, and track record of such programs. These must be well understood if policy makers are to make appropriate choices about which programs to use to achieve their desired objectives and reach specific target groups. The common grouping or labeling of similar programs--for example, as cash transfers or public works programs--disguises differences whereby programs with the same label may be designed and implemented in a variety of ways to make them more suitable for particular contexts. The choices about customization will affect which population groups the program will serve and the types of impacts that it can achieve. Those implementing programs should therefore not just copy programs used elsewhere: they should under- stand the principles underlying each type of program and how to customize it appropri- ately to the particular need and context. The quality and care with which programs are designed and implemented, from the se- lection of beneficiaries to the provision and monitoring of benefits, have a large impact on the efficiency and effectiveness of a given program. No program is a guaranteed suc- cess, and few are guaranteed failures. The role of good systems and adroit managers in getting the most from a program cannot be overemphasized. Improving programs is always possible: new ideas can come from within a thoughtfully managed program, from observing how other programs are operating differently, from innovating, or from new technologies. New ideas for how to design programs and how to meet the perennial challenges of targeting, payment systems, monitoring, and so on surface constantly. Program managers should stay informed and maintain a criti- cal eye in assessing how innovations used elsewhere may be applicable to a specific program. D eveloping countries employ a large number of safety net instruments to reduce and mitigate the effects of poverty and other risks on vulnerable households. The type of programs used, their objectives, their design specifics, and their implementation vary depending on a country's level of development; the amount of resources the country al- locates to social programs; and the regional characteristics, including both the economic and political environments. 253 254 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS The types of safety net programs implemented in developing countries have been evolving. The last 20 years have seen a marked move away from generalized, universal food subsidies toward more targeted programs and from the use of food toward the use of cash. For example, universal food distribution programs were popular in North Africa, South Asia, and Sub-Saharan Africa until the early 1990s, when they were proven to be far too expensive and ineffective in reaching the poor, especially in rural areas (see Alderman and Lindert 1998 and Tuck and Lindert 1996 on the reform process in Africa and Dev and others 2004 and Mooij 1999b on India). Following the financial crises of the late 1990s, Argentina and the Republic of Korea adopted public works programs, which were previ- ously used on a large scale mainly in South Asia. Several other countries, such as Ethiopia, Malawi, and Uganda, are now also using them. Fee waiver systems for health and educa- tion started in the 1980s and 1990s to accompany cost recovery programs, but the current trend is away from cost recovery upon service use and toward free access to education and insurance for health. A recent innovation is conditional cash transfer programs, which pro- vide income support to families while requiring them to make the necessary investments in their children's health and education. Three factors emerge as crucial for the success of a given safety net program in achiev- ing the goals outlined in chapter 2. The first is selecting the right program to address the needs of the intended beneficiaries given the underlying political and administrative envi- ronment or, in other words, selecting the appropriate tool for the job. The second factor is customizing the design of the selected program. The third factor is paying adequate attention to the details in all aspects of program implementation, from the selection of beneficiaries to the distribution and monitoring of benefits. Chapter 2 shows how safety net programs in general can make a difference in pro- tecting the chronically poor and the transient poor and how they can promote household investment and facilitate other government policies that help reduce poverty. This chapter illustrates how different programs can achieve those goals and how well they are suited to addressing the specific issues of different population groups by severity and type of poverty and by vulnerability or other relevant categories. Similarly, chapters 4 to 6 describe key program design functions--targeting, pay- ment mechanisms, monitoring, and evaluation--that are common to all programs. This chapter reorients our perspective and takes a program-specific view to examine the unique design and implementation features of each type of program and how programs can be used to achieve their intended objectives for specific population groups. Furthermore, the chapter highlights the key advantages and disadvantages of each program and the chal- lenges inherent in running effective and efficient programs. Several criteria can be used to organize individual types of programs into meaningful groups. In this review, we have organized programs into three basic categories (box 7.1). · Transfer programs in cash and in kind include programs that help protect poor households by providing them with the resources they need to maintain a mini- mum level of consumption. Properly crafted, they can help ensure livelihoods for the very poor and assist them in case of shocks. They include programs that deliver unconditional transfers to households in the form of cash or near cash, which includes vouchers, coupons, and stamps that provide almost the same pur- chasing power as cash. These are the most flexible programs and can be shaped 7. UNDERSTANDING COMMON INTERVENTIONS 255 BOX 7.1 Classification of Types of Programs Covered Programs that provide unconditional transfers in cash and in kind · Cash transfers, including near cash (vouchers, coupons, and the like). Needs-based social assistance, noncontributory pensions and disability transfers, family allowances, food stamps. · In-kind food transfers. Targeted food transfers and rations, other food-based programs, supplements for mothers and children, school-based feeding programs and transfers. · General subsidies. Subsidies for food, energy, housing, and utilities. Income-generation programs · Workfare or public works programs. Public works programs in which the poor work for food or cash. Programs that protect and enhance human capital and access to basic services · Conditional transfers. Transfers in cash or in kind to poor households subject to compli- ance with specific conditions in relation to education and/or health. · Fee waivers for health and education. Mechanisms to ensure access to essential public services, such as fee waivers for health care services, school vouchers, or scholarships. to achieve any of the four goals of safety nets. Other programs provide access to food by allocating rationed and subsidized food commodities to targeted popula- tions via ration shops; take-home rations; or supplementary feeding programs, which provide direct feeding opportunities for mothers, children, students, and/ or displaced populations in crisis situations. Food transfer programs can provide the same protection as other transfer programs and may also help improve the nutritional status of mothers and children. Finally, some programs use universal price or tax subsidies for basic commodities to ensure or increase the consumption of food and other essential commodities by poor households. · The main focus of income-generation programs is to provide low-skill jobs for the poor during the course of building, repairing, or improving local infrastruc- ture. These programs provide low wage payments in cash or in kind to members of poor households willing to work at that pay. Thus they provide some form of protection to chronically poor and vulnerable households from loss of income resulting from shocks. While the focus is on public works, other labor market interventions, such as job training, job placement, and microcredit programs, can also facilitate access to income-earning activities. Income-generation programs can complement transfers and provide opportunities for graduation out of such programs. · Programs to protect human capital and provide access to basic services for poor households provide conditional transfers to encourage the use of education or health facilities or other incentive provisions to lower the cost of access to basic 256 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS health and education services for the poor. Thus they also play an important role in promoting investment in human capital and provide a viable alternative in case of government reforms. Some programs have characteristics that make them unique and therefore hard to classify, while others may fall into more than one category. Food stamps, for example, are transfer programs that provide coupons that can be treated like cash, but that may restrict purchases to certain food commodities, and can therefore also be classified as in-kind pro- grams. Another example is food ration programs that do not restrict access to a particular segment of the population and might therefore be regarded as food transfer programs or as general price subsidies. Similarly, the Female Secondary School Assistance Program in Bangladesh can be classified as a conditional transfer program or as a scholarship. In the end, it is not the artificial classification that matters, but the design and implementation characteristics of specific programs that will ensure their success and effectiveness with respect to their broad objectives. The rest of this chapter provides systematic coverage of individual programs based on the six main groups of interventions described in box 7.1. Appendix B briefly describes many programs organized along the same six groupings of intervention and by country and also indicates the programs' level of expenditure and coverage. 7.1 Cash and Near Cash Transfers Cash transfer programs include the provision of assistance in the form of cash and other instruments almost like cash that can be used to transfer resources to the poor or to those who, in the absence of the transfer, face a probable risk of falling into poverty.1 For a list of cash and near cash programs, see table B.1 in appendix B. PROGRAM DESCRIPTION The main objective of cash and near cash transfer programs is to increase poor and vulner- able households' real incomes. The main difference between cash, near cash--such as food stamps, coupons, or vouchers that may be used to purchase food--and in-kind transfers is the amount of choice given to beneficiaries in acquiring the types of commodities they want to consume. Cash transfers obviously allow recipients to purchase anything they wish; near cash transfers, such as food stamps, can restrict recipients' choices to certain types of commodities; while in-kind transfers limit the selection to the commodities received. Some programs deliver transfers that are partially in cash and partially in kind; and others provide vouchers, coupons, or stamps, which are something in between in cash and in kind. The origins of cash and in-kind transfers go back to at least Roman times (Brown 2002; Hands 1968). The Alimenta (Food) Program, originally started by Trajan's prede- cessor Nerva and expanded by Trajan, provided food to poor children. The first example of a food stamp program is the U.S. program (USDA 2008), which operated between 1939 and 1943 and was restarted in 1964. Developing countries such as Sri Lanka in 1979 (Edirisinghe 1987) and Jamaica in 1984 (Grosh 1992) introduced food stamps to alleviate the short-term economic hardships associated with the elimination of general subsidies on food commodities or food rations. 7. UNDERSTANDING COMMON INTERVENTIONS 257 TYPES OF PROGRAM This subsection discusses four main types of programs: the first three are pure cash trans- fers for poor populations based on need or for special vulnerable groups such as the elderly or families with children (see chapter 8 for a description of the difference between regular and special poor and vulnerable groups). The fourth program, food stamps, provides a means for increasing food consumption.2 Needs-Based Social Assistance Needs-based social assistance programs are mostly means-tested programs and are com- mon in countries of the Organisation for Economic Co-operation and Development (OECD), Eastern Europe, and the former Soviet Union. Some income transfer programs based on needs are also found elsewhere in poor countries, for instance, in Mozambique and Zambia in Sub-Saharan Africa (Devereux and others 2005; Schubert 2005) and in Pakistan in South Asia (ADB 2006). The level of benefits and program coverage depend greatly on the fiscal resources available. Benefits are usually quite low, often around 5 to 25 percent of the cost of obtaining the poverty line basket of commodities. Some programs provide a regular monthly transfer, like the Food Subsidy Program in Mozam- bique, while others provide only occasional transfers in response to a shock (see Harvey 2005 for a description of the use of cash transfers in emergency situations). In addition, transfers can either be flat--that is, the same for all recipients--as was initially the case in the Kalomo District Pilot Social Cash Transfer Scheme in Zambia;3 can vary depend- ing on household resources as in the case of Romania's Guaranteed Minimum Income Program (Pop, Florescu, and Tesliuc forthcoming); or can vary with respect to household size as in Mozambique, where transfers range from US$3 to US$6 per month depending on the number of children (Devereux and others 2005). Noncontributory Pensions Many countries provide noncontributory pensions for some or all of those who do not fall under the country's contributory pension scheme and, in some cases, all those above a fixed age. Under these noncontributory schemes, benefits are paid without regard to past participation in the labor market. These schemes, which are discussed in more detail in chapter 8, are almost always financed from general tax revenues and are usually targeted toward the poor. The level of benefits varies from Tk 165 (US$3) a month in Bangladesh, roughly 10 percent of the average per capita income in 2003, to R 370 (US$106) a month in South Africa, equal to half the country's average household income and more than twice the median per capita income in 1993 (Bertrand, Sendhil, and Miller 2003; Duflo 2003).4 Family Allowance Programs Family allowance programs are common in OECD European countries, Eastern Europe, and the former Soviet Union. Benefits are often small--a few U.S. dollars a month, repre- senting a fraction of the cost of the food basket--although in some middle-income transi- tion states, including the Czech Republic and Hungary, they provide a more substantial contribution to the cost of raising a child. Family allowances can take various forms, such as means-tested child benefits similar to needs-based transfers as used in the Czech Republic, Poland, and South Africa (box 7.2);5 birth grants or universal transfers for all 258 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 7.2 The South African Child Support Grant The South African child support grant is a means-tested monthly cash grant given to the primary caregivers of children living in poverty. The main objective of the program is to provide support for poor mothers and poor families to care for their children (Monson and others 2006; Samson, MacQuene, and van Niekerk 2006). The program was introduced in 1998 to replace the state maintenance grant. As of 2006, the program provides a monthly grant of R 190 (about US$28) to 7.4 million poor children younger than 14; initially, the program had covered poor children un- der 7. The age eligibility has been expanded gradually over the past few years, and the amount of the grant has increased with inflation, while the means test cutoffs have remained the same. Implementation. The South African Social Security Agency, a separate national government agency, implements and administers the grant. To be eligible to participate, caregivers must present documentation showing that they have primary responsibility for caring for children, proof of the age of the children, and official proof of the employment and income status of the applicant and the spouse. To be eligible for the grant, the income of the primary caregiver and the spouse has to be R 1,100 (about US$170) or less for rural households (or households liv- ing in informal housing settlements in urban areas or shanty towns, which include communities of self-constructed shelters of unclear land tenure) or R 800 (about US$120) or less for urban households living in formal housing. The two threshold levels of income are designed to take into account the higher mean household size in rural than urban areas and the larger number of dependent children in rural areas (Rosa, Leatt, and Hall 2005). In many areas, applicants can obtain the required certifications for affirming marriage certifi- cates or divorce decrees, as well as making affidavits declaring the earnings of applicants and their spouses, from police officers. In some Eastern Cape sites, however, program officials or children under a fixed age (often children under 2 or 3 years of age or those under 16 or 18), which are most common in Europe and the former Soviet Union; and programs for the employed population, often with a special system for public sector employees, which are popular in OECD countries and in some middle-income countries such as Argentina (Lindert, Skoufias, and Shapiro 2006). Moreover, the transfers can be either in cash or in kind, for example, in the form of subsidies on school uniforms or children's goods. Food Stamp, Voucher, and Coupon Programs Food stamps, vouchers, and coupons are near cash instruments targeted to poor house- holds (Castañeda 1998; Grosh 1992; Hoddinott 1999; Rogers and Coates 2002) that they can use to purchase food at authorized retail locations. Retailers who accept these instruments can redeem them for cash through the banking system. The value of the food stamp is backed by the government's commitment to pay. Such programs have been implemented in Colombia, Honduras, Jamaica, Mexico, Romania (where the program was intended to help farmers after the 1997 planting season), Sri Lanka, the United States, and a few other countries. The denomination of the stamps varies from program to program: in some cases it is in cash, in others it is in kind. The amount of the transfer is often based on the gap 7. UNDERSTANDING COMMON INTERVENTIONS 259 community leaders provide official confirmation of an applicant's situation. The program's ad- ministrative costs are low: registering a new applicant into the program costs only R 19 (about US$3), less than 1 percent of the annual payment to beneficiaries (Budlender, Rosa, and Hall 2005). Payments are managed and monitored at the national level, but disbursed by third party contractors at the provincial level. Impact. According to some estimates (Leatt 2006), 65 percent of all children in South Africa live in families that would qualify for the program and 80 percent of these actually participate in the program.Take-up rates were lower when the program first started because of the lack of capac- ity of local governments and the difficulty of getting documentation for children and caregivers. Even though many improvements have been made to reduce the percentage of excluded chil- dren, a lack of documentation might still represent a constraint for poor children. The impact of the program has generally been positive. It has been linked to reduced poverty, higher labor market participation, and increased school attainment levels. Children who receive the grant are significantly more likely to be enrolled in school in the years following grant receipt than equally poor children of the same age who did not receive the grant (Samson and others 2004). Lessons. The government made substantial efforts and was able to increase the participation of poor children in the program, but some work remains to be done to reach those poor families without proper documentation. More work also has to be done to reduce the rate of dropouts from the program, which are likely to occur when the primary caregiver changes, often because of death from AIDS. between the amount of resources spent on food and the amount needed to acquire a minimum basket of commodities. In practice, the benefits are often worth only a few U.S. dollars and represent a small share of the cost of the food basket. For example, in Jamaica, the value of food stamps is only 12 percent of the food budget of the lowest quintile of the population, compared with the United States, where food stamps are worth 56 to 70 per- cent of households' mean food expenditures (Castañeda 1998). Some programs restrict households to buying only a few specific foods, while others allow them to purchase any foods they wish. The foods authorized for purchase in the Jamaican Food Stamp Program include rice, cornmeal, skim milk, and wheat flour, which constitutes a basic local food basket (Ezemenari and Subbarao 1999; Grosh 1992). These were the same foods previ- ously covered by general price subsidies. KEY DESIGN FEATURES Key challenges in delivering cash and near cash programs include making sure that pro- grams reach their intended beneficiaries and that the funds do not disappear along the way. For efficiency, setup and delivery must be handled in a reliable and efficient manner using available technology. 260 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Beneficiary Selection Beneficiaries of cash transfer programs are selected using a variety of targeting mecha- nisms. Needs-based transfers are usually targeted using either income or means tests as in Bulgaria, Hungary, and Romania; proxy means tests as in Armenia; or a combination of proxy and means tests as in Mozambique and Zambia. In this regard, note that Mozam- bique also uses health status indicators, such as being chronically sick or malnourished (Devereux and others 2005). Some countries, such as some OECD nations and Mauri- tius, have special provisions for identifying individuals with disabilities (Tabor 2002). Either the central government can set the selection criteria, which are then applied locally, or they can be decided at the local level so as to take local conditions, preferences, and priorities into account. In Albania and Uzbekistan, communes use local information to achieve better poverty targeting than could be expected on the basis of proxy indicators alone (Alderman 2001, 2002a). Beneficiaries of noncontributory pensions and family allowance programs are often targeted by age group and can be based simply on age and place of residence, as for Bo- livia's and Lesotho's pension programs or Hungary's and Romania's family allowance pro- grams. Selection can also be based on a means test, as in South Africa for both the pension and family allowance programs and Bulgaria and Poland for family allowances, or can be based on a proxy means test as for the pension program in Chile. Finally, food stamp programs can include self-targeting elements. In Honduras, the Bono Escolar (Food Stamps for Schoolchildren) benefit was distributed through primary schools in selected areas and the Bono Materno Infantil (Food Stamps for Mothers and Young Children) benefit was distributed at health centers in poor areas; thus, targeting was achieved through school attendance by children or the use of health clinics by mothers. Alternatively, such programs can impose limitations on the types of commodities that can be purchased with the stamps, as in Jamaica. Disbursement Methods The mechanisms used to distribute cash and vouchers to beneficiaries include banking systems, post offices, local institutions such as schools, and mobile distribution units. Family allowance programs in Eastern Europe and in the Zakat ("almsgiving") Program in Pakistan, a cash transfer scheme managed by the Ministry of Religious Affairs, have made the greatest use of checks and banks. Post offices have often provided a reliable distribution mechanism for the old-age pension program in India (Farrington and others 2003), the Food Support Program in Pakistan, old-age pensions in Lesotho, and other programs elsewhere. Other delivery mechanisms include teachers in Zambia or places where other cash transactions take place, like lottery kiosks in Brazil and Western Union offices in Somalia. Namibia and South Africa use armored cars to deliver cash directly to beneficiaries. Compared with cash, more planning and preparation are needed for the distribution and reclamation of food stamps, including a reliable system for printing and distributing the stamps and a good banking system so that retailers can redeem them promptly. Recently, the use of electronic benefit transfer systems has become popular in the United States and Latin America and shows promise for reducing costs and corrup- tion. Such mechanisms, which are discussed in more detail in chapter 5, require an effec- tive administrative distribution process and an up-front initial capital investment. 7. UNDERSTANDING COMMON INTERVENTIONS 261 Several programs have demonstrated that distributing cash efficiently is feasible even in difficult situations and remote locations. For example, in the 1990s, Mozambique dis- tributed payments to demobilized soldiers and made single payments to flood victims in the form of checks, and rural residents had no difficulty in cashing the checks (Hanlon 2004). Scope and Coverage The amount of funds allocated and the number of people covered by cash programs vary with respect to the size of transfers and the type of program. With the exception of Hungary and South Africa, most developing countries typically allocate the equivalent of less than 2 percent of gross domestic product (GDP) to public cash transfers, while Western European countries average the equivalent of more than 2 percent of GDP on social assistance programs and a good deal more on all social programs. For the most part, the population covered varies between 1 percent of the total population (as in the Food Subsidy Program in Mozambique), to 8 percent of the total population (as in the Guaran- teed Minimum Income Program in Romania), to 16 percent (as in the Ndihme Ekono- mika [Economic Assistance] program in Albania), to 34 percent (in the Unconditional Cash Transfer Program in Indonesia). Participation in noncontributory old-age pension schemes in several high- and middle-income countries varies according to targeting cri- teria and beneficiary location. Brazil and South Africa have two of the largest programs. Brazil's rural old-age pension has a rural focus and covers 4.6 million beneficiaries, while its urban programs cover only 0.7 million elderly beneficiaries. In 1998, South Africa's program covered 1.8 people of all races nationwide. Bangladesh provides an example of a newly established program in a low-income country for poor rural people and covers 1.2 million beneficiaries annually (World Bank 2005a). The coverage of family allowance benefits is much larger in terms of the percentage of the total population if participation is universal and is based on children's ages as in Hungary and Romania. However, the trend in Europe in the 1990s was to reduce the level of benefits and the number of children covered (Rostgaard 2004). The number of benefi- ciaries and total expenses are reduced when enrollment is restricted using means testing, as for Bulgaria's family benefit or South Africa's child support grant; the latter, for example, covers about 7.1 million of 13.5 million children, even though 8.8 million would be eligi- ble based on income poverty criteria (Leatt 2006). Costs and coverage are further reduced if eligibility is restricted to families working in the public sector or in formal employment, as in Argentina and most OECD countries. The coverage of food stamps varies greatly according to the targeting criteria used and the program's budget. Coverage amounted to 3 percent of the population in Honduras in 1992, 11 percent of the population in Jamaica in 1998, and 48 percent of the population in Sri Lanka in 1989. The U.S. Food Stamp Program acts as an insurance mechanism, as it is set up as an entitlement and all those who apply and qualify for the program are ac- cepted. Therefore coverage varies from year to year, from 27.5 million people in 1994, to 17.2 million in 2000, and 26.5 million (about 9 percent of the total population) in 2007. The level of benefits per household varies depending on the country and the year. In Sri Lanka, for example, the level of transfers fell from 32 percent of food purchases in 1978 to 20 percent in 1982. In the United States, it is equivalent to 25 to 50 percent of the budget of a family with two children (Castañeda 1998). 262 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS Administrative Costs The administrative costs of pure cash transfers are lower than for any other transfer pro- gram and typically range from just over 2 percent of program costs in Armenia's Family Poverty Benefits Program to about 10 percent in Bulgaria's and Romania's Guaranteed Minimum Income Programs. The costs are lower than for distributing food stamps, be- cause of the cost of producing coupons and setting up the mechanism for retailers to re- claim cash from the government in exchange for the food stamps they accept, which may add 2 to 5 percent on top of the costs of cash transfers. The cost of delivering food stamps is 2.0 percent of the total budget in Sri Lanka (Castañeda 1998), 3.0 percent in Romania (Castañeda 1998), 10.0 percent in Jamaica (Grosh 1994), and 13.5 percent in the United States (Castañeda 1998). These costs can be kept down if the setup and delivery of food stamps relies more on markets to make food available so that governments do not have to get involved in costly marketing operations. The costs of food stamp programs are lower than for in-kind food distribution programs because transporting, storing, and distribut- ing food in bulk is more expensive than moving food stamps around. Cash programs are less expensive than public works programs as they do not require material and tools. Fi- nally, cash transfer programs do not require the certification of compliance that is needed for conditional transfer programs. Implementing Institutions Social welfare and social security ministries often administer needs-based and means-test- ed transfer programs, including old-age, disability, and family allowances. For example, Poland's Ministry of Social Policy administers family benefits, and national and provin- cial departments of social development administer pensions in South Africa. Sometimes family allowances are distributed directly through the workplace or, for those who do not work, through local agencies. In some instances, the agency that administers a coun- try's mandatory contributory old-age pension scheme will also administer complementary noncontributory programs, as in the case of Lesotho, where the Department of Pensions of the Ministry of Finance administers the program (Devereux and others 2005). Food stamp programs are often managed by the ministry of welfare as in Jamaica before 2002 or the ministry of agriculture as in the United States. Some countries use a combination of ministries, including the ministry of health as in the distribution of some of the benefits dis- tributed by the Programa de Asignación Familiar (Family Allowance Program) in Honduras. Many programs are funded centrally, but contain some elements of decentraliza- tion that may include implementation (staffing resources), financing, or design (criteria and objectives). Several Eastern European countries finance programs centrally but imple- ment them locally. The administration of old-age pensions is often decentralized to local offices that may be part of subnational (provincial) governments, as in South Africa and Sri Lanka (Barrientos and others 2003, appendix C). Similarly, in Bulgaria, the targeted social assistance system has staff at the local level (local social assistance directorates) in the municipalities (Shopov forthcoming). OUTCOMES, ADVANTAGES, AND DISADVANTAGES Several studies and evaluations of cash transfer programs have shown that, in general, such programs have been effective in reaching their intended beneficiaries and had a 7. UNDERSTANDING COMMON INTERVENTIONS 263 positive impact on beneficiary consumption. Most of the results of such studies can be generalized to other transfer programs and to targeting efficiency. Incidence Evaluation studies show that cash transfer programs can be effective in reaching the in- tended poor households. Evidence from Eastern Europe shows that 50 to 80 percent of the benefits of needs-based transfers go to the poorest 40 percent of households (Tesliuc and others forthcoming). Preliminary studies of South Africa's child support grant in- dicate that it appears to be well targeted at children in poorer households (Barrientos and DeJong 2004; Case, Hosegood, and Lund 2005; Samson and others 2004). Family allowance programs are usually slightly better than distribution neutral, because house- holds with children, especially those with large numbers of children, tend to have a higher than average incidence of poverty. In the case of food stamp programs, between 50 and 80 percent of benefits go to the poorest 40 percent of households. In Jamaica, the poorest 20 percent got 31 percent of the benefits; in Sri Lanka, the poorest 20 percent got 40 per- cent of the benefits (Castañeda 1998; Coady, Grosh, and Hoddinott 2003). Impact The results of impact evaluations of cash transfer programs have been broadly positive for the families of beneficiaries and have not confirmed many of the negative externalities often feared. Evidence from research on unconditional cash transfer programs in develop- ing countries shows a positive impact on consumption and on human capital of children. In Ethiopia, Lesotho, Mozambique, and Zambia, children benefit from transfers even though they are not the programs' primary targets (Devereux and others 2005). In South Africa, the pensions, when received by women, had a large impact on the anthropomet- ric status (weight for height and height for age) of girls, but little effect on that of boys; this was not the case when men received the allowances (Duflo 2003), suggesting that the efficiency of public transfer programs may depend on the gender of the recipient. In Bolivia, Martinez (2005) finds positive effects of the Bono Solidario program on house- hold consumption and children's human capital. A significant fraction of this increase in consumption is derived from the consumption of home-produced agricultural products such as meats and vegetables resulting from the transfers being invested in productive ac- tivities. Child and family allowances in transition countries have proved to be effective in ameliorating the impact of structural change on households with children and have been reformed to act as safety nets (Barrientos and DeJong 2004). Food stamp programs have been shown to be effective ways of transferring income, increasing household income by as much as 20 to 25 percent (Castañeda 1998). Without the Food Stamp Program in Jamaica, the poverty gap would have been much worse during the early 1990s, when the Jamaican dollar was being devalued: households with elderly members and young children benefited the most from the program (Ezemenari and Sub- barao 1999). Evidence also indicates that food stamp programs tend to increase food consumption more than cash transfers (Breunig and others 2001; Fraker 1990; Fraker, Martini, and Ohls 1995). One possible reason for this is that households do not treat food stamps in the same way as cash. Another is that the stamps may fall under the control of women, who disproportionately favor expenditures on food and other basic needs. The 264 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS impact of food stamps on nutritional status is hard to demonstrate. In a few cases, food stamp use has been associated with increased consumption of protein and micronutrients (Butler and Raymond 1996). The possible negative effects of cash transfers include the disincentive to work; the misuse of cash resources; the change in the number of desired children; and the consumption of nonnutritious food commodities, for example, alcohol and tobacco. However, while old-age pensions may have some negative effects on the decision to work (discussed in more detail in chapter 5), people rarely seem to use cash transfers for antisocial purposes such as cigarette and alcohol consumption, and women are not necessarily disadvantaged by the use of cash rather than in-kind approaches. Concerns of corruption and insecurity may be more frequent in conflict situations (Harvey 2005). The empirical evidence shows (and sometimes program rules require) that child allow- ance programs do not increase the probability of having children, but rather encourage prolonged school enrollment, which leads to higher educational attainment and lower family size. The fear of crowding out might also be overstated. In Zambia, most people agree that transfers are insufficient and therefore are still willing to provide assistance, but at a reduced level (Wietler 2007). Cash transfer programs have not generally resulted in sustained price rises, even when they have successfully stimulated the local economy. In the Kalomo District Pilot Social Cash Transfer Scheme in Zambia, Schubert and Goldberg (2004) find that additional purchases of food, soap, blankets, and agricultural inputs have not resulted in local price increases. Advantages Cash transfers are the most direct type of intervention designed to support the poor and have a number of advantages as follows: · Once the administrative infrastructure is in place, the cost of operating cash trans- fer programs is often small and far less than the cost of providing assistance in kind. · From the recipients' point of view, cash transfers provide them with greater free- dom of choice in how to use the benefit to enhance their welfare and results in a higher level of satisfaction at any given level of income than in-kind transfers. Program beneficiaries also feel that less stigma is attached to the receipt and use of cash than of in-kind benefits. · Targeted cash transfers do not directly distort prices. In isolated and thin food markets in rural areas, cash transfers can cause an increase in the price of food, but if food markets are functioning well, cash and food stamps can strengthen local retail establishments. · Food stamps can protect consumers from price increases and be self-targeting. If they can be redeemed easily, often, and in small quantities in local stores, they can help stimulate retail markets, in contrast to local delivery of food that might instead depress local market prices. If denominated in kind, food stamps are not subject to inflation to the same extent as food stamps denominated in value. Self- targeting can be greater than with cash transfers if the use of coupons is limited to inferior, less preferred foods. 7. UNDERSTANDING COMMON INTERVENTIONS 265 Disadvantages Although most of the disadvantages of cash transfer programs are common to transfer programs in general, some concerns arise in relation to possible uses of cash transfers and care is needed to limit the following disadvantages: · Some argue that it might be difficult for women to maintain control of the re- sources and use them for the benefit of their children instead of promoting anti- social behavior, like the consumption of cigarettes or alcohol by men. · Unconditional cash transfer programs may distort preferences. Where cash pro- grams are strictly targeted based on income, they may result in a greater disincen- tive to work than in-kind transfers or public works programs. · Changes in product prices affect the value of cash transfer programs. If the amount of the cash transfer is not adjusted because of unexpected surges in inflation or product prices, it can lose its value and effectiveness. · Distribution costs tend to be higher for food stamp programs than for cash pro- grams. In addition, food stamps are more likely to be subject to theft and fraud than food or cash, and because the use of food stamps is restricted, they are less desirable than cash. · Cash transfers are attractive to local elites and unintended beneficiaries. As a re- sult, they may be more difficult to target effectively, and good control mecha- nisms are needed (as outlined in chapter 5) to ensure that they reach the intended beneficiaries. LESSONS AND SUGGESTIONS Cash transfer programs are the most basic type of safety net program and can easily be adapted to different types of situations for a wide range of beneficiaries. Most Likely Beneficiaries Cash transfers have been used effectively to address many of the needs of poor people. The intended beneficiaries include those who are poor and have a low level of consump- tion for a variety of reasons. Some households might simply have too few people working at wages that are too low, and therefore require additional support that can be provided by family allowances, common in Eastern Europe, and food stamps as in the United States. Other intended beneficiaries include households that do not have anyone who, due to age or disability, can be expected to work. These households can usually be reached with pensions as in Bangladesh and South Africa, family allowances, and food stamps. Cash programs can also provide temporary cash transfers to those who have suffered losses of assets, income, and/or consumption as a result of an uninsured shock; for example, as has occurred in Indonesia, Mozambique, and Pakistan after natural disasters. See also box 7.3 on the effectiveness and flexibility of cash transfers in emergencies. Appropriate Context and Political Economy Considerations Cash transfers are the most obvious and simple instruments for addressing poverty in most circumstances as long as food is available in the marketplace, otherwise food prices 266 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 7.3 Use and Effectiveness of Cash Transfers in Emergencies The cost-effectiveness and flexibility of cash, as opposed to in-kind support, is extremely impor- tant in an emergency, where quick delivery of assistance and adaptability to specific recipients' needs is crucial. Cash can also have positive effects on local markets and trade; however, the same disadvantages that pertain to all cash transfer programs apply in emergency situations. In particular, the security risk of moving cash around might be more pronounced than under normal circumstances (Peppiatt, Mitchell, and Holzmann 2001). The design of a cash transfer program must take into account all these factors and be adapted to the specific emergency. In terms of program objectives, the use of cash provides the opportu- nity to link the relief response with longer-term development concerns, but the immediate goal in emergencies is to assist the most vulnerable and most badly affected individuals--commonly the displaced, widows, orphans, and the elderly. As the Tsunami Emergency Recovery Program in Sri Lanka in 2005 did, the targeting method can be a combination of geographical criteria (the most flood-affected areas) and categorical criteria (displaced households), with additional community identification of beneficiaries. An option Willibald (2006, p. 332) suggests in the case of cash transfers to former combatants following a conflict situation "would be to give priority access to those willing to disarm first, without excluding those who do not have a weapon." The size and timing of the transfer depend primarily on the objective and can include the value of lost assets and a more regular transfer supporting livelihoods. In the aftermath of natural disasters in Maldives, Pakistan, and Sri Lanka, cash transfers first addressed the asset restora- tion objective, but also attached transfers to the status of housing reconstruction or rehabilita- tion. Oxfam has developed a number of measures to create a more secure environment for the use of cash, including limiting local knowledge of cash movements; limiting access to bank transac- tions; having small cash transfers between banks; decentralizing responsibility for disbursement and involving a number of staff; disbursing cash on an ad hoc basis; having small, frequent cash disbursements; disseminating information to all stakeholders (community elders, committees, politicians, and nonrecipients); using long-standing staff who are local to the area and trusted by the head office and the team; and ensuring that community members choose safe locations for their cash disbursements (Creti and Jaspars 2006; Khogali and Takhar 2001). Impact. Based on experience of cash transfers in emergencies, Harvey (2005, p. 36) concludes that "People spend the money that they are given sensibly, cash projects have not generally resulted in sustained price rises and women have been able to participate and have a say in the benefits from cash and voucher responses." Lessons. Even in difficult situations, cash transfers can be delivered safely and provide a quick and effective means of support for vulnerable populations after a disaster.The Emergency Cash Relief Program (implemented by Horn Relief and Norwegian People's Aid and funded by Ox- fam Novib Netherlands), for example, was able to distribute a total of US$691,500 to 13,830 drought-affected households in the Sool Plateau in Somalia in 2003­4, making it the largest cash response ever mounted in Somalia (Ali, Toure, and Kiewied 2005). 7. UNDERSTANDING COMMON INTERVENTIONS 267 will increase. Thus the fact that cash transfers are the main type of safety net in the OECD countries--potentially available, in one form or another, to more than 80 percent of the population of the industrial nations according to the International Labour Office (2000)--is not surprising. Yet far fewer cash transfer programs are in effect in developing countries, and those that do exist tend to grant only small benefits mostly to those who cannot work and are equivalent to less than 2 percent of GDP. Fewer social protection programs are based on cash transfers in developing countries for a number of reasons. First, governments' cash resources in poor developing nations are limited, as they are more likely to receive resources in the form of food aid. Second, mobi- lizing support for pure cash transfers can be difficult because of the lack of experience with targeting. Finally, governments and donors may give priority to programs that can relieve structural constraints to growth rather than programs to augment consumer demand or transfer income. Despite these constraints, demand for safety nets that incorporate some form of cash transfer is growing in developing countries. This has been met by renewed interest on the part of several donor agencies in promoting the use of cash transfer programs as a response to chronic poverty, food insecurity, and AIDS in countries of eastern and southern Africa with a high prevalence of HIV infection such as Ethiopia, Malawi, and Zambia (DFID 2005; Devereux and others 2005). The intent is partly to respond to the growing unmet need for social protection with predictable cash transfers and partly to respond to the idea that regular, predictable grants in the form of cash transfers to identified vulnerable groups offer a cost-effective way to reduce poverty and realize basic human rights (Schubert and Beales 2006). They can also help reduce inequality and ensure that the benefits of growth reach those living in chronic poverty both during normal times and during emergencies.6 Cash and near cash programs can also be used effectively in times of crisis or for the transient poor if they have to face changing economic conditions. Coverage can be expanded either by setting the amount of the transfer without limiting the number of beneficiaries at any given point in time using current programs or by having provisions for quickly expanding the number of temporary or short-term beneficiaries. The first strategy is achieved in the U.S. Food Stamp Program, which does not limit the number of benefi- ciaries, although they have to be recertified often. Among available instruments, family allowances and food stamps appear to be more politically acceptable. Policy makers often see family allowances as an important tool for preventing the intergenerational transmission of poverty. Food stamps are often claimed to be a good compromise between cash transfers and in-kind transfers, because they are tied to the merit good of foods, sometimes of particular foods. Of course, public support is likely to be larger if fewer restrictions are placed on the commodities included in the program. Food stamps also provide a way to help eliminate general food subsidies as in Jamaica and Sri Lanka. In addition, the agriculture sector and the private sector food in- dustry often support food stamp programs because they expand the demand for food and can be supported by food aid. The use of cash and near cash programs requires functioning markets and adequate provision of basic services to ensure that supply is sufficient to respond to increased de- mand (Barrientos and DeJong 2004). Indeed, these programs are only effective where food is readily available on the private market, where the problem for the poor is adequate 268 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS purchasing power rather than lack of access to stocked markets, where consumers purchase most of their food at the market and the retail market operates adequately, and where the type of food offered reflects poor people's preferences. Adapting to Local Conditions and Avoiding Unintended Effects The following presents ways to make cash and near cash programs more effective in help- ing poor people given local conditions and to ensure that transfers reach the intended beneficiaries, that they are not captured by nondeserving people, and that they are not subject to corruption: · Specify clear program objectives and benefit levels that are widely understood and that most people agree are sound, fair, and effective to build a broadly based constituency in favor of a cash transfer program.7 Cash transfer pro- grams in developing countries need not be overly generous. Given that low-in- come households often already derive some earnings from informal sector activi- ties or private transfers, cash transfers can be used to partly close the gap between their current level of consumption and a minimum level of desired consumption, rather than to provide full replacement income. Furthermore, limiting transfers to those who cannot work more than they already do, whether temporarily or permanently, contains the cost of transfers and reduces the adverse labor supply effects that high benefits may foster. The amount of the food stamp transfer can be smaller or greater than the household's current expenditure on food depending on the program's nutritional objectives. However, if not denominated in quantity terms, the value of cash transfers and food stamps needs to be updated periodi- cally to prevent the erosion of benefits because of inflation. · Use the best possible targeting method, given administrative capacity, to reach intended beneficiaries. Categorical, geographic, and community-based screening approaches can be effective alternatives in circumstances in which formal income and means tests are impractical (see chapter 4 for more details on methods to improve targeting). · Use effective payment mechanisms. Appropriate payment mechanisms can vary depending on administrative capacity and the availability of financial channels, from the use of banking systems as in Brazil and Colombia, for example, to post offices in India and a combination of fixed and mobile banks in Bangladesh. Emerging technologies and disbursement mechanisms (treated in more detail in chapter 5) can facilitate disbursement and reduce costs for both beneficiaries and program administration. Corruption can threaten the very existence of a program (see Datt and others 1997 for the case of Mozambique); it can be reduced by ensuring good administration and monitoring and by keeping transfers small (Farrington and others 2003). 7.2 In-Kind Food Transfers and Other Food-Based Programs In-kind food transfers and other food-based programs provide additional resources to households by making food available when they need it the most in the form of food ra- 7. UNDERSTANDING COMMON INTERVENTIONS 269 tions, supplementary and school feeding programs, or emergency food distribution.8 The main difference between food programs and cash-based programs is that the former use food as a resource and give beneficiaries a limited choice in relation to the types of com- modities they want to consume. For a list of in-kind food transfer and other food-based programs, see table B.2 in appendix B. PROGRAM DESCRIPTION The main objective of food-based programs is to provide for adequate food consumption and thus help poor consumers achieve and maintain better nutritional status when, in the absence of the intervention, people would be likely to curtail their food consumption, re- sulting in malnutrition, morbidity, and possibly death. At the same time, food-based pro- grams also tend to improve vulnerable households' participation in social programs, such as primary health care (including prenatal, postnatal, and well-baby care) and education. The use of in-kind transfers goes back to ancient Egypt and to the Roman Empire. Examples of recent food-based transfers can be found in South Asia since the 1944 Bengal famine (see Sen 1981 for an analysis of the causes of famines in which he stresses the role of unequal distribution of income). TYPES OF PROGRAMS The food-based programs covered in this section include food rations and other in-kind food transfers and supplementary feeding, school feeding, and emergency food distribu- tion programs. Food Rations and Transfers Food rations and transfers are intended to provide access to food to vulnerable and food- insecure households.9 In most cases, targeted households collect rations at designated public or private distribution centers either for free or at a reduced price. In India, for example, certified poor consumers, that is, those with an income below the poverty line, can purchase wheat and other commodities at reduced prices through the public distribu- tion system (PDS). Take-home rations are a special case of rationing in which rationed quantities of food are delivered directly to beneficiary households. Many such programs are found in Africa, Latin America, and South Asia, and include the Vulnerable Group Development Program in Bangladesh (box 7.4), the Gratuitous Relief Program in Ethio- pia, and the Comodores Populares (Community Kitchens) program in Peru. The main difference between these programs and general price subsidies, which are discussed later, is that they restrict access to targeted beneficiaries.10 Several food ration programs have their genesis in reforms of prior general food price subsidy programs. This is what happened to the PDS in India when it started to pro- vide commodities at lower prices to households below the poverty line and to the market price stabilization interventions operated by Indonesia's National Food Logistics Agency (BULOG) when it introduced a new targeting program (box 7.5). Many people have strong ideas about the use of food-based versus cash transfers. Food distribution programs have played an important role in social policy and develop- ment, partly because of the availability of food aid from Australia, the United States, and other OECD countries (del Ninno, Dorosh, and Subbarao 2007). The debate on the use 270 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 7.4 The Vulnerable Group Development Program, Bangladesh The Vulnerable Group Development Program, which originated as a relief program in1975, is a collaborative food security intervention jointly managed and implemented by the Ministry of Women's and Children's Affairs and the World Food Programme that targets about 500,000 extremely poor rural women. The main objective is to integrate food security and nutrition with development and income generation. Implementation. Local selection committees composed of government officials, elected local government representatives, and representatives of nongovernmental organizations select the female beneficiaries using prescribed criteria. While the program operates nationwide, it focuses on food-insecure areas. The allocation of the beneficiaries is based on a food-inse- curity map, whereby more food-insecure subdistricts have more beneficiaries. Beneficiaries receive a monthly ration of 30 kilograms of wheat over a period of 24 months. Since 2002, beneficiaries in three subdistricts have been receiving 25 kilograms of fortified whole wheat flour (atta) instead of grain. Beneficiaries are also required to attend training in income-gener- ating activities, such as poultry rearing, livestock raising, fisheries, and sericulture; participate in awareness sessions on social, legal, health, and nutrition issues; receive training in basic literacy and numeracy; and obtain access to credit. In addition, beneficiaries are required to make a monthly savings deposit of Tk 25 (less than US$0.50), corresponding to roughly 10 percent of the transfer, into an interest-bearing account maintained by the nongovernmen- tal organizations providing services to the Vulnerable Group Development Program in those areas. Impact. Evaluations have shown that the program has been extremely successful in targeting hardcore-poor women aged 15 to 49. However, only about two-thirds of these women seem to have "graduated" from absolute poverty to becoming confident microfinance clients who have not slipped back to requiring government handouts. Lessons. The program has proven that a combination of in-kind transfers and training is an ef- fective way to alleviate poverty in the short run and reduce it in the future. SOURCES: Ahmed, del Ninno, and Chowdhury 2004; Ahmed 2005; Matin and Hulme 2003. of cash rather than food has been receiving renewed attention in recent years (see Gentilini 2007 for a comprehensive discussion of the debate), partly because of the issues surround- ing the use of food subsidies in Europe and the United States, which have been generating large food surpluses that are often distributed in the form of food aid. Thus the decrease in the availability of food aid resources for development has resulted in a shift away from the use of food transfers (Barrett and Maxwell 2005). The key issue, however, is being able to determine the circumstances in which food transfers are appropriate and how to maximize their impact. Box 7.6 describes four key considerations to keep in mind to help make this assessment--the functioning of food markets, the level of transaction costs, the type and size of the transfer, and the preferences of beneficiaries. 7. UNDERSTANDING COMMON INTERVENTIONS 271 BOX 7.5 From Universal to Targeted Distribution, India and Indonesia In June 1997 in India, the existing PDS was transformed into the targeted PDS in response to the findings of several studies (for instance, Radhakrishna and others 1997) that the program suffered from poor targeting and high unit costs for handling grain. The new program differenti- ates the quantities households are allowed to buy and prices depend on their poverty status. The PDS used to provide all consumers with access to rice, wheat, sugar, edible oils, kerosene, coal, and standard cloth at subsidized prices through a network of registered shops. Since 1997, only households below the state-defined poverty line are entitled to a ration card, which allows them to buy a larger quantity of rice and/or wheat than before (10 kilograms in 1997, 20 in 2000, 25 in 2001) at a subsidized price equal to about 50 percent of the economic cost. Since 2001, those above the poverty line may purchase food grains at a discount rate (equal to 70 percent of the economic cost). India also increased the allocation of state quotas of poverty cards to poorer states, shifting from an allocation formula that favored states with the largest food deficits regardless of whether they were relatively poor. In Indonesia, BULOG, a publicly owned corporation, maintained a floor price and a ceiling price in order to stabilize prices through its monopoly control over international trade in rice through 1997. In 1998, Indonesia abandoned this policy and replaced it with Operasi Pasar Khusus (Special Market Operations), renamed Beras untuk Keluarga Miskin (Rice for Poor Families) in 2001, a targeted rice subsidy program for poor consumers (Kitano, Ariga, and Shimato 1999; McCulloch 2004; Pritchett, Sumarto, and Suryahadi 2002; World Bank 2006f). The reason for the change was a shift in the exchange rate following the 1997 Asian financial crisis, which turned a policy geared toward producer subsidies into one that required massive and unsustain- able consumer subsidies Under the new program, BULOG sold rice to 3.4 million households at a subsidized price of Rp 1,000 (US$0.10) per kilogram, compared with a market price of Rp 3,000 (US$0.30) per kilogram, as of August 1998. The program reached 10.4 million families in 1999 and 12 million in 2003. Each family, identified by the National Family Plan Coordination Agency using geographical and categorical indicators, was entitled to receive 10 kilograms (later 20 kilograms) of rice per month. On the whole, the operation was well implemented. In a short time, rice was being distributed in a relatively well-controlled and accountable way. The main issues were that some needy households were excluded because they did not have identity documents or were not on the preexisting rosters used to target program beneficiaries; families had to make a small copayment for the entire monthly rice ration, which meant they had to find ways to finance a payment that was larger than their usual daily purchase; and some communities chose to share rations rather than let the intended targeting stand (SMERU Research Institute 1998). The experience in India and Indonesia shows that shifting the primary mode of intervention is possible; however, program improvements are still needed. For additional information on the reforms in India, see Ahluwalia (1993), Dev and others (2004), Government of India (2001, 2007b), Mooij (1999b), Radhakrishna and others (1997), and Tritah (2003); for Indonesia, see ADB (2006), Ahmad and Leruth (2000), Daly and Fane (2002), Perdana and Maxwell (2004), Tabor and Sawit (2001), Timmer (2004), World Bank (2006f), and Yonekura 2005. 272 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS BOX 7.6 Cash and In-Kind Transfers: Alternatives or Complements? When are food transfer programs appropriate? What are the criteria to keep in mind when deciding how much to distribute in the form of rations and how much as cash? Program design- ers should keep the following four key considerations in mind when deciding if food transfer programs are appropriate or necessary: · The functioning of food markets, including access, transport, and storage, and how this is reflected in the prices of staples. If markets are well integrated across regions, cash transfers have an advantage because of the private sector's superior ability to move food and other goods more efficiently than the public sector. Further- more, some argue that providing cash can have a positive impact on petty trade and other economic activities (Devereux 2000). However, if markets are thin, poorly in- tegrated across regions, or monopolistic, the provision of cash may increase prices, which reduces the value of the transfer and may cause additional hardship to those poor households that do not receive any transfers (Devereux, Mvula, and Solomon 2006). A close monitoring of prices, not of production, is needed to assess the situa- tion. · The level of transaction costs for the program and for beneficiaries. Most of the argument about transaction costs refers to the high cost of distributing food provided by donors compared with the relatively lower cost of distributing cash. Food distribu- tion takes time to organize, requires storage and transport, and is subject to losses and pilferage, and the public sector tends not to be efficient at keeping costs down. However, in some places where marketing and transport channels are not developed, only the public sector can provide adequate supplies in local markets. Beneficiary transaction costs also need to be taken into account. These costs include the time and expense of going to local markets, which might increase if places are far or un- safe. · The impact of the form and size of the transfer in determining the level of food consumption. Poor households are more likely to consume food and to eat good food if they receive a small transfer. Some claim that men might use cash transfers to purchase such commodities as cigarettes or alcohol, and the literature indicates that small food transfers result in higher food consumption than cash transfers (del Ninno and Dorosh 2003; Fraker 1990). Moreover, Hoddinott and Islam (2007) and Jacoby (2002) show that households are more likely to stick (the so-called flypaper effect) to consumption patterns and intrahousehold distributions that have a positive impact on the nutrition of children if they have access to small transfers of good food. · The preferences of the beneficiaries. Beneficiary preferences may vary depending on circumstances. Even though beneficiaries may prefer cash simply because it is more flexible, they still want to maximize the level of the transfer and their control over it. This is why women in certain circumstance might prefer food to cash (see Ahmed, Quisumbing, and Hoddinott 2007 on Bangladesh and Sharma 2006 on Sri Lanka). 7. UNDERSTANDING COMMON INTERVENTIONS 273 Supplementary Feeding Programs Supplementary feeding programs are intended to provide food specifically to mothers and young children. The food may be prepared and eaten on-site--for example, at child feeding centers as in Bangladesh, Indonesia, and Thailand--or provided as a so-called dry ration to take home as in Chile, where food supplements are distributed on a monthly basis through the primary health care system. Bolivia, Colombia, the Republic of Congo, Guatemala, Indonesia, Jamaica, Peru, Senegal, and Thailand also have supplementary feeding programs (Gillespie 1999). Table 7.1 compares the benefits and costs of the two systems. Foods provided for on-site meals are usually a low-cost blend of grains and pulses with added fat or oil: a typical diet might consist of 500 to 700 calories per day per child (UNHCR and WFP 1999). The consumption of on-site meals presents several advantages over take-home rations. In particular, the food is consumed by the intended beneficiaries and its preparation is supervised; however, the cost is much higher. TABLE 7.1 Comparison of Delivery Options for Supplementary Foods Item Take-home rations On-site meals Consumption There is no guarantee that only the All rations are eaten under supervision, and of food by intended recipient, whether a child or help can be given to ill and undernourished recipient a pregnant or lactating mother, eats children. However, recipients may be the ration. Usually the household given less food at home (referred to as shares it, sells it, gives it to their substitution). animals, or wastes it. Responsibility Families take responsibility for Responsibility for feeding may be taken and feeding recipients and there are away from the family, but in small feeding education of fewer opportunities for education. programs, caregivers may help prepare families Caregivers spend less time and effort food and feed recipients. Feeding problems by not having to go to feeding sites. can be identified and dealt with. Logistics, Large numbers of people can be Many resources, including well-trained staff, organization, covered using fewer resources and and extensive supervision are needed. and costs facilities. Costs are lower. Costs are higher, including caregivers' opportunity costs in regularly attending the feeding site. SOURCE: Gillespie 1999. School Feeding Programs School feeding programs provide meals for children at school to encourage their enroll- ment and improve their nutritional status and ability to pay attention in class. They can vary from the provision of breakfast, lunch as in the Thailand School Lunch Project, or a midmorning snack as in Bangladesh (Ahmed 2004b), to a combination of these as in Colombia, Costa Rica, Guatemala and Peru. School feeding programs are often integrat- ed with other interventions, such as health and nutrition education, parasite treatment, health screening, and provision of water and sanitation.11 A few school feeding programs also provide an income transfer in the form of food to take home. The Ethiopian World Food Programme (WFP) school feeding program provides 274 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS a daily nutritious meal and a take-home ration of vegetable oil to girls in pastoralist areas who are at school for at least 80 percent of school days (WFP 2005). The Food for Educa- tion Program in Bangladesh provided wheat or rice to families that sent their children to primary school until 2001, when it was changed into a cash stipend program (Ahmed and del Ninno 2002; Ahmed, del Ninno, and Chowdhury 2004). In such cases, the food does not necessarily benefit only the enrolled child, but may be shared among the family. Emergency Food Distribution Emergency distribution of food includes direct provision of food; supplementary feeding for vulnerable groups; and therapeutic feeding during crises, emergencies, and situations in which people are displaced (see UNHCR and WFP 1999 for guidelines on the size and type of on-site and take-home food rations). In some cases, schools might be used for the distribution of food rations. Emergency food programs provide a safety net of last resort whose objective is to save lives by preventing starvation, malnutrition, morbidity, and possible death when public and private institutions fail to protect individual entitlements to food. In many emergency situations, such as refugee camps or camps for internally displaced populations, food transfers are the only source of food and may be the only resources households receive. Currently, emergency operations account for a larger proportion of official develop- ment assistance than ever before. The percentage of total official development assistance devoted to humanitarian emergencies has risen from 4 percent at the end of the 1980s to 10 percent in recent years (Development Initiatives 2006; OECD Development As- sistance Committee 2001). Moreover, emergency interventions represent an important component of food-based programs in general, and of food aid in particular, as food usu- ally represents more than half of humanitarian aid. In 2005, emergency relief operations in the form of food aid accounted for 64 percent of total food aid, up from 26 percent in 1991 (Wahlberg 2008), and included interventions following conflicts in Afghanistan, Iraq, and Liberia; droughts in southern Africa in early 2000 and 2001; floods in Bangla- desh in 1998; and Hurricane Mitch in Central America in 1998. KEY DESIGN FEATURES The implementation of food-based programs poses several logistical and implementation challenges besides the usual challenges of any other transfer program. It involves procur- ing and storing food, including food aid and local and international purchases; transporting food to local areas and distributing it (Jaspars and Young 1995); and finding ways to reduce waste, spoilage, and pilferage. The selection of commodities and of the geographic areas to be covered is also crucial to avoid unintended effects and achieve intended objectives. Beneficiary Selection The same factors that determine the success of targeting in the case of income transfers in general determine the effectiveness of the targeting of in-kind programs (see Edirisinghe 1987 on Sri Lanka). Additional targeting mechanisms can be used to improve the nutri- tional status of mothers and children. The best targeting mechanisms for food transfer programs include the use of individ- ual targeting mechanisms, such as means tests and proxy means tests; self-targeting meth- 7. UNDERSTANDING COMMON INTERVENTIONS 275 ods through the use of inferior commodities (box 7.7); or methods based on nutritional risk criteria, for instance, age or pregnancy.12 Targeted households usually receive a ration card that entitles them to a certain amount of food at a subsidized price as in the Arab Republic of Egypt and in India's targeted PDS. The progressiveness of transfers depends BOX 7.7 Inferior Commodities and Inframarginal Consumption The economics literature refers to goods that the poor consume in greater amounts than other segments of the population as inferior goods. This designation pertains to the purchasing pat- tern (or negative income elasticity) and not to the physical attributes of the commodity. Coarse grains, for example, may be inferior goods in the sense that households with higher incomes are less likely to consume them, but from the standpoint of nutritional quality, such grains are actually superior to the more popular highly polished or refined grains. The economics literature defines the amount of a commodity transferred or made available at a subsidized price as inframarginal if it is smaller than the amount that consumers would have chosen to purchase at the regular market price. The impact of a subsidy of a commodity on the poor will depend largely on whether the commodity selected is inferior and whether the level of consumption is inframarginal. Because subsidies on inferior commodities are self-targeting, the benefits of the transfer will be larger for poor people than for nonpoor. As benefits are proportional to the amount of the good the household purchases, subsidies on commodities with low, and ideally negative, elasticities will be progressive (assuming such commodities are available). If the focus of the program is narrower and the purchases of the selected commodity represent a comparatively smaller share of a con- sumer's budget, the amount of income that can be transferred via a self-targeted commodity sub- sidy decreases. Moreover, even the most favorable self-targeted commodities will only distribute between a half to two-thirds of benefits to the poorest 40 percent of the population while the most successful means-tested transfer programs have the potential to deliver more than 80 percent of benefits to the poorest two quintiles (Coady, Grosh, and Hoddinott 2004; Grosh 1994). The magnitude of the impact of a transfer on the amount of consumption of a rationed com- modity is larger if the level of the transfer of the commodity is not inframarginal, that is, it is larger than the amount usually consumed. Indeed, for an inframarginal transfer, the amount of consumption will increase in line with the increase in household income, as consumers will base their decision on how much to buy on the price available in the market. For a transfer that is larger than the amount usually consumed, the increase in consumption will be much larger and will be based on the price response. A smaller price paid (taking the transfer into account) will induce a higher level of consumption of the commodity subsidized. The overall impact on total nutrient consumption, as measured by calorie intake, tends to in- crease with the level of income. While one cannot assume that all income increments are spent on increased food consumption, food consumption might increase more than proportionally (see the evidence from South Africa in Alderman and del Ninno 1999; from Bangladesh in del Ninno and Dorosh 2003; and from food stamp studies in Fraker, Martini, and Ohls 1995). SOURCES: Alderman 2002b; Alderman and Lindert 1998. 276 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS on how well they are targeted, as well as the impact of waiting time and possible stigma on participation by poor and nonpoor households. The typical target groups for supplemental feeding programs are pregnant and lac- tating women and children under the age of three or five. Many programs give more benefits to those who are malnourished (underweight or stunted) or those failing to grow according to norms. The use of public service facilities and the selection of poor and nu- tritionally insecure areas may introduce a significant degree of self-targeting in countries where the middle- and upper-income groups use private health care. The selection of beneficiaries for school feeding programs presents additional diffi- culties, as targeting children within schools is difficult and may stigmatize them, and may therefore limit program feasibility, efficiency, and desirability.13 Occasionally schools use differential cost recovery to target meals as in Jamaica and the United States or feed only some children outside of school hours as was previously done in Chile. The selection of schools (the institutional targets) participating in school feeding programs is therefore the main criterion that can be used to target school feeding programs as in Costa Rica, although examples of effective targeting of school feeding programs are available--for example, the U.S. School Lunch Program, which has been able to avoid stigmatizing recipients. Gender targeting can also be used if female enrollment and attendance are particularly low. Emergency feeding is usually carried out in refugee camps or in areas affected by natural disasters. The issues related to targeting mechanisms in those situations depend on the type of intervention. Supplementary feeding programs may benefit either all members of a particular age (for example, all children under the age of two) or gender, depending on the level of malnutrition, may be targeted using anthropometric criteria (for additional details see Sphere Project 2004; Taylor and Seaman 2004). Disbursement Methods In most cases, governments use their own distribution channels to procure, store, and transport the food needed for food-based programs. Not surprisingly, government in- volvement in food marketing may lead to inefficiencies. In India, for example, program grain is procured in the northwest of the country rather than closer to the places where it is distributed to beneficiaries. The use of private retailers who are authorized to sell both nonrationed and rationed commodities is common and occurs in, for instance, Egypt, India, and Iraq. Such retailers have replaced dedicated ration shops, thereby increasing availability and reducing costs. Because of the long distribution process and the number of transactions that take place, some of the food may not reach intended beneficiaries, resulting in leakages from fraud or spoilage. Some of the leakage may occur at warehouses or at retail shops. In the Indian targeted PDS, observers report gaps between official estimates of the amount of food provided and household consumption of between 20 and 35 percent depending on the location and the commodity (Ahluwalia 1993; Dev and others 2004; Mooij 1999a). Rao (2000) finds evidence that some retailers simply sell the subsidized grain at the open market price, thereby increasing their margins. This is not surprising, as such back-door sales are inherent in any two-tier price system. Community organization and information can help prevent leakage and fraud. An in-depth analysis of the problems related to leakages in food distribution programs in 7. UNDERSTANDING COMMON INTERVENTIONS 277 Bangladesh by Ahmed and others (2004) finds that leakages for the Vulnerable Group De- velopment Program were only 8 percent, compared with the higher rates more common for other programs in South Asia, partly because of monitoring and evaluation throughout the system and partly because of women's empowerment at the local level to hold program managers accountable. As noted earlier, supplementary feeding can be distributed in the form of take-home rations or can take place on-site. On-site feeding can be more effective in ensuring that the target population actually consumes the rations, although it may not translate into an additional supplement to the usual diet if the food intake at home is reduced. Although it is relatively expensive to set up, it is useful for therapeutic feeding of severely malnourished children and for feeding vulnerable mothers (Gillespie 1999). The main delivery mechanisms for school feeding programs include meals prepared on-site, meals prepared in advance, bulk food, and coupons. Each model is associated with a different set of implementation issues. Preparing meals on-site in developing countries presents several challenges: the long distances to fetch water and fuel for cooking, the slow cooking facilities, and the lack of adequate personnel which is often overcome by using volunteers. In many cases, the current emphasis on the timing of meals to maximize the program's impact on educational objectives involves additional challenges. New program approaches to improve efficiency include using snack foods and products that cook more quickly and contracting out to the private sector (Del Rosso 1999). In the case of emergency feeding, beneficiaries are totally reliant on the food sup- plied and must receive a complete diet. In general, the process of determining ration size and composition, frequency of distribution, and criteria for program entry and exit are subject to similar considerations as other maternal and child health supplementary feeding projects, with the additional challenge that when the diet is insufficient or inadequate, it can result in deficiency diseases, for example, pellagra if the diet is based mainly on maize (Taylor and Seaman 2004). Scope and Coverage Food ration programs are important in India, where the PDS distributes rationed amounts of basic food items to about 160 million families (approximately 70 percent of the popu- lation); in Egypt, which has more than 48 million beneficiaries (more than 80 percent of the population); and Indonesia, where about 12 million households receive rations (about 23 percent of the population). However, some ration programs have been discon- tinued, such as Mexico's Tortivales (Free Tortilla) program; others have been reorganized, such as the JPS Operasi Pasar Khusus (Special Market Operations) program in Indonesia (Yonekura 2005); and some have been replaced by other programs, such as the rice ration program in Sri Lanka, which has been replaced by a food stamp program (Edirisinghe 1987; Tabor 2002). The value of the transfer varies from extremely small percentages to 6 percent of total household expenditure for the bottom 20th percentile of the popula- tion in Indonesia (Pritchett, Sumarto, and Suryahadi 2002; Sumarto, Suryahadi, and Pritchett 2000). The overall coverage of feeding programs varies greatly, ranging from 1 to 2 percent of the population in Honduras to 6 percent of the total population in Chile and 15 per- cent of the population in Peru. The coverage of specific population groups can, however, 278 FOR PROTECTION AND PROMOTION: THE DESIGN AND IMPLEMENTATION OF EFFECTIVE SAFETY NETS be much higher. In Chile, 80 percent of children under two and 70 percent of preschool- ers and pregnant and lactating women are covered by feeding programs (Kain and Uauy 2001). Compared with the other types of feeding programs discussed in this section, the coverage of school feeding programs is usually more limited. Basically, there are a few large programs, as in Bangladesh, where the School Feeding Program covers 1.2 mil- lion primary schoolchildren