THE World Bank 44009 The Rural Investment Climate Analysis and findings rEPorT no. 44009-GlB Agriculture & Rural Development Department World Bank 1818 H Street, N.W. Washington, D.C. 20433 http://www.worldbank.org/rural The Rural Investment Climate Analysis and Findings THE WORLD BANK AGRICULTURE AND RURAL DEVELOPMENT DEPARTMENT © 2008 The International Bank for Reconstruction and Development / The World Bank 1818 H Street, NW Washington, DC 20433 Telephone 202-473-1000 Internet www.worldbank.org/rural E-mail ard@worldbank.org All rights reserved. This volume is a product of the staff of the International Bank for Reconstruction and Development/The World Bank. The findings, interpretations, and conclusions expressed in this paper do not necessarily reflect the views of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accu- racy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or accept- ance of such boundaries. The 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. The International Bank for Reconstruction and Development/ The 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, http://www.copyright.com/. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA, fax 202-522-2422, e-mail pubrights@worldbank.org. Cover photos by World Bank staff members: Household in Tanzania by Scott Wallace; Rural Enterprise in Sri Lanka by Yosef Hadar; Community in Nicaragua by Curt Carnemark Contents FOREWORD ix ACKNOWLEDGMENTS xi A B B R E V I AT I O N S A N D A C R O N Y M S xiii E X E C U T I V E S U M M A RY xv 1 Introduction 1 1.1 Assessing the Rural Investment Climate 1 1.2 Implications for the Rural Agenda 3 1.3 Study Objectives 4 1.4 Organization of the Report 4 2 Approach and Methodology 5 2.1 Laying the Groundwork 5 2.2 Contributions of RIC2 9 3 Enterprise Performance and Investment Climate Constraints 15 3.1 The Enterprises and Their Environments 15 3.2 Enterprises and Variables 18 3.3 Specification of Enterprise Performance Regression Models 18 3.4 Regression Results 19 3.5 General Discussion 24 4 Entrepreneurship Choice and Enterprise Start-Up 29 4.1 Households 30 4.2 Entrepreneurship Among Rural Households 33 4.3 Activities of Households 35 4.4 Enterprise Start-Up 36 4.5 General Discussion 37 iii iv Contents 5 Perceptions About Investment Climate Constraints 41 5.1 Entrepreneurs’ Investment Climate Concerns 42 5.2 Modeling Issues 45 5.3 Estimation Results: EICOs 47 5.4 Estimation Results: EICIs 52 5.5 Pushing the Analysis Forward: Next Steps 54 5.6 Implications for RIC Methodology 55 6 Differences Among Communities 57 6.1 Community Characteristics and the Community Environment 57 6.2 Cross-Country Comparisons of Benchmark Indicators 58 6.3 Benchmark Indicators and Prices 60 6.4 Benchmarks and Community Characteristics 61 6.5 Regressions at the Community Level for Sri Lanka 62 6.6 Performance of Benchmark Indicators as Descriptors for the Investment Climate 64 7 Conclusions and Recommendations 65 ANNEXES Annex A. Description of Data 75 Annex B. Data Used for Enterprise Performance Analyses 81 Annex C. Enterprise Performance Regressions; Notes and Tables 87 Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 117 Annex E. Specimen Indexes Derived from RICS Data 143 Annex F. Entrepreneurship: Notes and Tables 153 Annex G. Benchmark Indicators 165 Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 181 Annex I. Econometric Analysis of RIC Survey Data 205 Annex J. Employment and Income Estimates in the Surveyed Communities 245 Annex K. International Benchmarking of the Nonmetropolitan/Rural Investment Climate 249 NOTES 253 REFERENCES 259 Contents v Figures 5.1 Enterprise Investment Climate Outcomes 43 5.2 Investment Climate and Entrepreneurship 55 7.1 Level of Seasonal Activity Among Enterprises in Tanzania 73 D.1 Comparing Distribution of Total Assets Across Countries 132 D.2 Income Shares as a Percentage of Total Household Income, by quintile 135 D.3 Comparing Distributions of Income per Capita, Normalized at the Country-Specific Mean 137 Tables 2.1 International Comparison of RIC Benchmark Indicators 11 3.1 Basic Economic Characteristics of Selected Countries 16 3.2 Sector of Operations, Sales, Net Value Added, and Productivity of Enterprises 16 3.3 Characteristics of Surveyed Enterprises 17 3.4 Enterprise Size Distribution in Terms of Sales and Net Value Added 17 3.5 Median Sales and Net Value Added, by Size of Community (US $) 18 3.6 Enterprise Productivity in Terms of Gross Sales and Net Value Added 18 3.7 Economies of Scale and Production Elasticities for Variant (4) 20 3.8 Economies of Scale and Production Elasticities for Variant (5) 21 3.9 Significant Contributions to Productivity by Enterprise and Community Characteristics 22 3.10 Significant Contributions to Productivity of Benchmark Indicators and Components 23 3.11 Significant Contributions to Employment and Capital Generation by Enterprise and Community Characteristics 24 3.12 Significant Contributions to Employment and Capital Generation of Benchmark Indicators and Components 25 4.1 Economic Activities of Households and Household Members 31 4.2 Average Total Household Income and Its Components 31 4.3 Measures of Inequality in per Capita Total Income 31 4.4 Cross-Pilot Comparison of Household Assets by Type of Asset (US$) 32 4.5 Measures of Inequality in Total Assets 32 5.1 Ranking EICOs Across Countries 44 5.2 Investment Climate Evidenced in Government Efficiency 45 5.3 Investment Climate Evidenced in the Legal System 46 5.4 Significant Enterprise Characteristics in EICO Equations 48 5.5 Significant Enterprise Characteristics in EICO Equations 50 5.6 Analysis of Variance of Total Burden of the Investment Climate 52 6.1 Comparison of Community Descriptors (Community Averages) 58 6.2 Comparison of Benchmark Indicators by Country and Community Population Size 59 6.3 Correlation Coefficients Between Benchmark Indicators 60 A.1 Enterprises Included in the RIC Surveys and the Enterprise Performance Regressions 76 A.2 Definition of Variables 76 B.1 Nicaragua: Selected Variables of Enterprises and Communities; Value and Distribution 81 B.2 Nicaragua: Selected Variables of Enterprises and Communities; Value and Distribution 82 B.3 Sri Lanka: Variables Used in Regressions of Enterprise Performance; Value and Distribution 83 vi Contents B.4 Sri Lanka: Selected Variables of Enterprises and Communities; Value and Distribution 84 B.5 Tanzania: Variables Used in Regressions of Enterprise Performance; Value and Distribution 85 B.6 Tanzania: Selected Variables of Enterprises and Communities; Value and Distribution 86 C.1 Specification of Regression Variants 88 C.2 Percentage of Variation Explained in Enterprise Performance Regressions 90 C.3 Explanation of Employment and Capital Generation per Enterprise 95 C.4 Sales Regressions: Nicaragua 98 C.5 Net Value Added Regressions: Nicaragua 100 C.6 Sales Regressions: Sri Lanka 102 C.7 Net Value Added Regressions: Sri Lanka 104 C.8 Sales Regressions: Tanzania 106 C.9 Net Value Added Regressions: Tanzania 108 C.10 Labor Input Regressions: Nicaragua 110 C.11 Capital Input Regressions: Nicaragua 111 C.12 Labor Input Regressions: Sri Lanka 112 C.13 Capital Input Regressions: Sri Lanka 113 C.14 Labor Input Regressions: Tanzania 114 C.15 Capital Input Regressions: Tanzania 115 D.1 Sri Lanka: Availability of Sampling Weights by Province 117 D.2 Age Distribution (%) of Household Members: Sri Lanka, Tanzania, and Nicaragua 118 D.3 Distribution (%) of Households by Household Size and Gender 119 D.4 Distribution (%) of Household Members and Household Heads by Level of Eductation in Selected Countries 120 D.5 Coding of Schooling Level: Original and Converted 121 D.6 Distribution (%) of Households by Average Human Capital Stock 121 D.7 Distribution (%) of Households by Average Human Capital Index 122 D.8 Potential and Actual Workforce: Number of Adults Present and Working 123 D.9 Distribution (%) of Household Members (age ≥ 16 and ≤ 65) and Household Heads by Occupation in Sri Lanka 123 D.10 Adult Labor Force Participation Rate and Labor Force Composition 124 D.11 Accounting for Economic Activities of Households and Household Members 126 D.12 Variation in the Structure of the Labor Force by Region 127 D.13 Characteristics of Groups of Labor Force Participants 128 D.14 Average per Household Landownership in the Selected Countries 128 D.15 Distribution (%) of Households by Value of Land 129 D.16 Distribution (%) of Households by Total Value of Agricultural Assets 129 D.17 Distribution (%) of Households by Value of Durable Assets 130 D.18 Distribution of Households by House Value 130 D.19 Distribution (%) of Households by Total Assets 131 D.20 Measures of Inequality in Total Assets 131 D.21 Household Investment by Category 132 D.22 Consistency in Reports on Household Enterprise Activity 133 D.23 Total Household Income and Its Components 134 D.24 Per Capita Income and Its Components 136 D.25 Measures of Inequality in per Capita Total Income 137 Contents vii D.26 Income Total, per Capita and per Worker, by Category and Presence of Nonfarm Enterprise 137 D.27 Household Income and Employment by Benchmark Quintiles and Population Size 138 E.1 Distribution of Connectivity Index Over Communities 144 E.2 Distribution of Infrastructure Service Index Over Communities 146 E.3 Distribution of Business Service Index Over Communities 147 E.4 Distribution of Governance Indexes Over Communities 148 E.5 Distribution of Human Capital Indexes Over Communities 151 E.6 Distribution of Finance Service Indexes Over Communities 151 F.1 Definitions and Descriptive Statistics of Variables Used in the Econometric Models 157 F.2 Determinants of Entrepreneurship Choices 159 F.3 Contributions to Explanation of Entrepreneurship Choice 161 F.4 Determinants of Household Activity Choices 161 F.5 Determinants of Enterprise Start-Up 163 G.1 Missing data for Communities and Components 166 G.2 Nicaragua: Correlation Coefficients Between Benchmark Indicators and Prices at Community Level 169 G.3 Sri Lanka: Correlation Coefficients Between Benchmark Indicators and Prices at Community Level 169 G.4 Tanzania: Correlation Coefficients Between Benchmark Indicators and Prices at Community Level 170 G.5 Nicaragua: Correlation Coefficients Between Benchmark Indicators and Community Characteristics 171 G.6 Sri Lanka: Correlation Coefficients Between Benchmark Indicators and Community Characteristics 172 G.7 Tanzania: Correlation Coefficients Between Benchmark Indicators and Community Characteristics 173 G.8 Descriptive Statistics of Variables Used in Regression Analysis 174 G.9 Regressions at the Community Level 175 H.1 Detailed Responses to EICO Questions, by Country and Category 183 H.2 Descriptive Statistics of Explanatory Variables 186 H.3 Determinants of Top Ten EICO Responses: Models with Benchmark Indicators 187 H.4 Determinants of Top Ten EICO Responses: Models with Specific Community Characteristics 193 H.5 Effect of Enterprise Size and Productivity on EICO Responses 199 H.6 Effect of Enterprise and Community Characteristics on Total Investment Climate Burden, by Country 200 H.7 Opinions About Predictability and Manipulability of Laws 201 H.8 Effect on Business of Kickbacks, by Target: Nicaragua 202 H.9 Determinants of Perceptions About the Legal System 202 H.10 Corruption as An Investment Climate Constraint 203 H.11 Alternative Ways of Estimating EICO Models: Interest Rate of Loans in Sri Lanka 204 I.1 Ratio of Standard Errors of Parameter Estimates Obtained Through V2 and V1 211 I.2 Ratio of Standard Errors of Parameter Estimates Obtained Through V3 and V1 212 I.3 Effect of Sampling Weights and Random Effect Specification on Enterprise Performance Regression Results 215 viii Contents I.4 Effect of Sampling Weights and Random Effect Specification on Entrepreneurship Selection Results 219 I.5 Effect of Sampling Weights and Random Effect Specification on Selected EICO Regression Results 221 J.1 Nicaraguan Communities 245 J.2 Sri Lankan Communities 246 J.3 Tanzanian Communities 247 K.1 International Benchmarking of the Nonmetropolitan/Rural Investment Climate 250 Boxes 1.1 What Is Investment Climate? 2 7.1 Some Options for Exploring Improvement of Benchmark Indicators 69 7.2 Operational Applications: The Sri Lanka RIC Pilot 72 A.1 Specifications of Labor Input in Enterprise Questionnaires 80 C.1 Variance Explained by Enterprise and Community Variables 94 E.1 Definition of Formal Financial Instruments (Including Insurance) 152 F.1 RIC Survey Design and the Undercount of Enterprise Start-Up 154 Foreword Improving the investment climate constitutes one of the strategic pillars in the World Bank Group’s strategy to reduce poverty. Methods for assessing the investment climate have been developed over the past decade, and assessments are being mainstreamed in dozens of countries. Although the concept of investment climate is a generic one, the assessment tools developed have focused prima- rily on larger formal enterprises in the manufacturing sector. However, the applied sampling strategy for registered enterprises has been ineffective in rural areas, where most enterprises are small and informal. In 2002, when the World Bank Group was preparing its Agricul- ture and Rural Strategy “Reaching the Rural Poor,� conditions in rural areas, where 70 percent of the world’s poor people are living, were viewed as being significantly different from those in urban areas. This view was subsequently endorsed in the World Develop- ment Report 2008, which focused on rural development problems in developing nations. Hence, the idea was born to modify invest- ment climate assessment methodology to capture much needed information on rural enterprises and their particular growth con- straints. A program of six survey-based pilot assessments was launched, initially in Nicaragua, Sri Lanka, and Tanzania and later, when more resources became available, in Benin, Ethiopia, and Indonesia. Creating tools for assessing the investment climate is very much a process of learning by doing. It includes three aspects: creating a cost-effective survey methodology, establishing an efficient methodology for analyzing survey results, and applying survey findings in rural development policy analysis and intervention design. The lessons learned from the three first pilots, as well as from two other pilots then in advanced stages, were the focus of the first study in this series—The Rural Investment Climate: It Differs and It Matters (2006). The present report details the second comprehen- sive study on the rural investment climate. It builds on the valuable databases now available from the first three pilot studies as well as their assessment reports and a training course and toolkit for practitioners developed by ARD in 2007. Grounded in the use of advanced methodologies, the study explores the use of these ix x Foreword databases in designing better strategies for rural I hope this work will inspire more and better-tar- development. Several recommendations are made geted efforts to promote private-sector develop- to help task managers refine survey designs and ment in rural areas. methods of analysis in future RIC assessments—a vital aspect of using the results of these important Jeurgen Voegele assessments in country and sector dialogues. Director, Agriculture and Rural Development Acknowledgments This report was prepared by a team led by Kees van der Meer (consultant) and consisting of Wim Vijverberg and Richard Burcroff (consultants). Valuable support was provided by Yyannu Cruz- Aguayo, Andres Lopez, Isabel Rodriguez Tejedo, and Cateryn Vucina Banjanin (consultants). Peer reviewers were Regina Birner (IFPRI), Mary Hallward-Driemeier (DECRG), Stephen Mink (AFTSN), and Andrew Stone (MNSED). This work had three task-team leaders. It started with Kees van der Meer, and Renate Kloeppinger-Todd led the work following Mr. Van der Meer’s retirement from the Bank. The work was concluded under the leadership of John Lamb, who ac- tively participated in the final stages of the report. Mona Sur (ARD), Josef Loening (AFTAR), and Fan Qimiao (WBIFP) provided many helpful comments during the preparation of the report. The task team thanks Jeurgen Voegele (Director, ARD), Mark Cackler (Sector Manager, ARD), Kevin Cleaver (former Director, ARD), and Sushma Ganguly (former Sector Manager, ARD) for their support and guidance. This study is based on findings from rural investment climate pilot studies led by Mona Sur and Ismail Radwan (Sri Lanka), William Leeds Lane and Josef Loening (Tanzania), and Francisco Pichon and Norman Bentley Piccioni (Nicaragua). Renate Kloep- pinger and Ram Ramaswami provided background papers on rural finance. Information about the pilot studies is available at http://intranet.worldbank.org/WBSITE/INTRANET/SECTORS/ INTARD/0,,contentMDK:20592746~pagePK:210082~piPK:210098~ theSitePK:335808,00.html. From the beginning of this research project, colleagues from the Bank’s Financial and Private Sector Development Network and DECRG, Andrew Stone in particular, regularly provided advice on methodology based on their experiences with generic investment climate assessment tools. The econometric modeling forming the analytical core of this study was developed by Wim Vijverberg (consultant). Preparation of the report and the further development of RICS methodology were supported by the Bank Netherlands Partnership Program (BNPP); the pilots were supported by BNPP and the Norwegian Trust Fund Private Sector and Infrastructure (NTF-PSI). xi Abbreviations and Acronyms ADB Asian Development Bank ANOVA Analysis of variance BI Benchmark indicator DPR Development Policy Review EA Enumeration Area EICO Enterprise investment climate outcomes EICI Enterprise investment climate interactions GDP Gross domestic product GVA Gross value added IC Investment climate IFPRI International Food Policy Research Institute LSS Living standards survey NFHE Nonfarm household enterprise NGO Nongovernmental organization NVA Net value added PICS Productivity and Investment Climate Surveys PRSP Poverty Reduction Strategy Plans RIC Rural investment climate RICS Rural Investment Climate Survey RIC1 The first study in this series: The Rural Investment Climate: It Differs and It Matters (2006) RIC2 The present study: The Rural Investment Climate: Analysis and Findings RNFE Rural nonfarm enterprise WBG World Bank Group WDR World Development Report xiii Executive Summary CONTEXT AND BACKGROUND Interest in investment climates has emerged relatively recently. In the 1960s and 1970s, governments in many countries believed they should play a direct role in rural credit, input supply, production, trade, transport, distribution, and even marketing. However, in the 1980s and 1990s, government-dominated systems fell into disgrace because of poor performance. This led to a push for privatization and liberalization, after which conditions for the development of markets and private enterprises were addressed—in short, the investment climate. Initially, invest- ment climate surveys were developed to determine empirically the constraints inhibiting entrepreneurs from expanding their economic activities. Yet early efforts focused primarily on formal and larger manufacturing enterprises in urban areas. Only in recent years, and especially after the World Conference on Sustainable Development ratified the Millennium Development Goals, has the attention of the development community broadened to include other sectors and smaller enterprises. Recognizing the car- dinal significance of the investment climate to economic growth and poverty reduction, the World Bank devoted to the topic its 2005 World Development Report—A Better Investment Climate for Everyone (World Bank 2005). For the rural sector, the primary focus had traditionally been on agriculture, particularly commercial agriculture and agribusiness, which were perceived to be the main drivers of rural growth. There was not much interest in other rural enterprises, since they were thought to be almost fully dependent on agriculture and unimpor- tant to the dynamics of rural economies. This changed in the 1990s, however, as donors focused more on poverty reduction. Household surveys designed to gather new information on the sources of rural household income were then formulated and conducted. As results came in, more attention was given to the importance of rural non- farm enterprises (RNFEs) to rural livelihoods. Work by Reardon et al (2001) initiated a stream of research efforts recently summarized in a major study led by IFPRI (Haggblade et al. 2007). Within the World Bank, the 2003 rural strategy “Reaching the Rural Poor� realized the importance of the RNFE, and the 2008 WDR Agriculture for Develop- ment embraced the view that a sound rural investment climate and a rapidly growing agriculture sector are basic ingredients for a dynamic rural economy. xv xvi Executive Summary With the recognition of the importance of RNFE, CONTENTS OF THE RIC2 STUDY the direct policy question becomes: what can be done to further stimulate it? In most countries very What Has Been Covered? little information is collected on rural enterprises, their constraints, and their potential. To fill some Four analytical chapters form the core of this study. of these policy-critical gaps, the Bank in 2004 Chapter 3 addresses the question of how the launched pilot studies in Sri Lanka, Nicaragua, investment climate affects enterprise performance, Tanzania, Indonesia, Benin, and Ethiopia. The as indicated by both sales and net value added. main task was to develop a methodology with The investment climate is measured by conditions three salient features: in the community, as shown in the benchmark indicators and other community characteristics. (i) design and conduct of cost-effective RIC Community factors, including benchmark indica- surveys; tors, are in general highly influential, but, not sur- (ii) reliable analysis of survey results and inter- prisingly, they capture only part of the variance. pretation of policy implications; and Chapter 4 investigates household entrepreneur- (iii) effective intervention in rural development. ial choices, which are twofold. One set of choices For each RIC pilot study, a country team was refers to the activities pursued by a household. Of charged with conducting the surveys, analyzing primary interest is the choice to engage in entre- the data, and compiling a RIC Assessment report. preneurial activity, but this choice interacts with The assessments add substantive information on the choices to seek wage employment and to oper- several issues relevant to the rural IC that were not ate a farm. The second set of household choices appropriately addressed in the urban IC surveys refers to the decision to start up an enterprise. The and that were overlooked in agriculture-based models applied explain each of these choices, with studies. Yet further use can still be made of the reference to household characteristics (structure, available data. In these analytical efforts only parts skills, ethnicity, non-labor income, and assets), of the collected data were explored. Important benchmark indicators and their components, and avenues of analysis were not fully covered due to other community characteristics. limitations in time and budget and to the lack of Chapter 5 examines responses of entrepreneurs readily available tools for analysis. concerning their perceptions of the investment cli- Initial results from this survey-based assess- mate. One set of responses concerns a list of poten- ment were published in June 2006 under the title tial barriers to the operation and growth of the The Rural Investment Climate: It Differs and It Mat- respondent’s enterprise; the other describes inter- ters (World Bank 2006). That report mainly focused actions with and observations about government. on survey methodology1 and descriptive findings. Explanatory variables are the characteristics of the The present second study was announced as a fol- enterprise, the benchmark indicators and their low up that would delve mainly into analytical selected components, and community characteris- questions and the use of findings in shaping devel- tics. One prominent conclusion of the econometric opment policies. The present study systematically analysis is that the community random effect is explores the RIC databases for three of the six pilot always important: community “unobservables� countries—Nicaragua, Sri Lanka, and Tanzania— influence the perceptions of the entrepreneur with the following objectives: regarding the investment climate. Chapter 6 explores differences between commu- • to provide broader and deeper understand- nities. It makes a cross-country comparison of ing of rural nonfarm activity in rural areas, its benchmark indicators and explores how benchmark constraints, and possible ways to mitigate indicators are correlated with community-level those constraints; characteristics and prices collected at the commu- • to initiate a method of benchmarking the nity level. It also tries to explain community-level investment climate in rural areas; and indicators of economic activity (outcomes), such as • to advance and sharpen the methodology of enterprise density, income per capita, nonfarm such analysis and to provide guidance for income shares, and enterprise productivity analysis conducted by future survey teams through simple regression on benchmark indica- and policy analysts. tors and a number of other economic descriptors. Executive Summary xvii What is New About This Study? 1. The household data show that simultaneous involvement in RNFE self-employment, Several features of this study are innovative, and wage employment, and farming is common each serves to broaden the methodological frontier among rural households. Available choices available for rural investment climate studies. for individual households and household Innovative features include: members differ with family structure, human 1. A comprehensive analysis of the RICS house- capital, assets, and community characteris- hold data. tics. The RNFE self-employment choice fits 2. More systematic use of community data than within a broader livelihood strategy of pur- in previous studies. suing opportunities and managing risks. The 3. Development of benchmark indicators and view held by some analysts that RNFE self- their use in the analyses. employment results solely from push factors 4. Analysis of responses about constraints in felt by the poor is much too narrow. This the investment climate, using the concepts study provides evidence that households of Enterprise Investment Climate Outcomes engaged in RNFE tend to be better off than (EICO) and Enterprise Investment Climate farming families. Interactions (EICI). 2. The bulk of rural enterprises are very small 5. Novel Stata programs developed for random in size, with only one or a few workers. The effect analysis with weights. large majority of enterprises (68 to 86 percent) 6. Use of estimates of the contribution of groups only employ unpaid family workers. Enter- of variables—enterprise, industry, and com- prise population is not static, however. By munity characteristics—in explaining varia- comparing the three countries and differ- tion of outcome variables. ences within the countries, a general pattern 7. Household and enterprise weights used for emerges showing that, with increased per Nicaragua and Sri Lanka. (For Nicaragua, capita income, both enterprise density per household and enterprise weights lost 1,000 inhabitants and average enterprise size during data processing were reestimated. For increase. Sri Lanka, the enterprise weights used by the 3. Large numbers of small rural enterprises buy Sri Lankan team were readily available, but and sell mainly locally, which is understand- household weights had to be reestimated.) able from the nature of their businesses: ser- 8. Collection of community-based data that vices, retail trade, repair, and so on. Yet, allowed RIC to introduce a spatial dimension evidence from Sri Lanka shows that enter- not prominent in PICS. This expansion opens prises selling mainly outside their communi- new options for analysis of differential enter- ties have higher productivity. prise development in a heterogeneous rural 4. The share of agro-processing in rural enter- space and along the rural-urban axis. The prises ranges from only 2 percent in Sri Lanka clustering of observations by community to 14 percent in Nicaragua. This shows that a raises complex analytical and methodological development focus on agribusiness as the questions. transformation of raw products represents too narrow a target for private-sector devel- The methods and tools developed herein can be opment in rural areas. employed by policy analysts for future IC studies. 5. In all three countries, enterprise productiv- ity appears to differ with enterprise and community characteristics, leading to major CONCLUSIONS AND differences within and between countries. RECOMMENDATIONS Generally, enterprise age and entrepreneurial experience matter, indicating the impor- What Has Been Learned About tance for increased productivity of adapta- the Three Economies? tion and innovation through learning by The following important findings were made about doing. the economies of Nicaragua, Sri Lanka, and 6. As diverse as are the three countries in this Tanzania: analysis, the list of their most important xviii Executive Summary investment climate concerns shows remark- Policy Perspectives able overlap. On the top-ten list of concerns The RIC results can contribute in several ways to averaged across the countries, six items are preparing effective policy interventions. found on the individual top-ten list of each country and three more occur among two 1. The RIC survey results provide considerable of the three countries’ top-ten concerns. The otherwise unavailable information about the common concerns involve (i) the cost, avail- economic activities of rural households, local ability, and procedures of finance; (ii) electric- enterprise populations, and institutional, pol- ity and water (access, cost, or reliability); icy, and infrastructural constraints. These and (iii) road quality and access. The other findings can importantly inform the policy top-ten concerns are more varied: market dialogue between civil society—especially demand, economic policy uncertainty, and private-sector stakeholders—and govern- telecommunications. At the other end of the ment. The quality of dialogue can be spectrum, probably because of their enter- enhanced by good RIC analysis, and the fur- prises’ small scale, few rural entrepreneurs ther assessment of options facilitated by the in any country see as obstacles: food safety RIC surveys. regulation, regulation on land use, customs 2. The various regressions—enterprise per- rules, or work permits for expatriates, to formance and entrepreneurial choice—show mention a few. that, from micro-perspectives, various factors 7. It is often said that access to credit is the fac- constrain productivity, employment, invest- tor most constraining to RNFEs. This study ment, and income earning. Since many of found that the large majority of enterprises these constraints can be alleviated (some (58 to 79 percent) reported having no formal more readily than others), the RIC analyses or informal debt, and only 1 to 31 percent can identify and document challenges for reported formal debt of more than 50 percent policy intervention. of equity. Although entrepreneurs may gen- 3. In many cases, general RIC analysis as out- uinely desire inexpensive credit, the bench- lined in this report can and should be fol- mark indicator and benchmark components lowed up by further exploration of the for finance services do not reveal credit RIC databases, with regard, for example, to as having a strong effect on enterprise the specificity of constraints, locations, and productivity. groups of enterprises affected. RIC surveys 8. Improved infrastructure provision alleviates create multipurpose databases of a public- perceived investment climate obstacles goods nature that for several years hence can related to infrastructure services. When more be used as information sources to support households in a community have access to policy and project design. Moreover, addi- protected water sources, for example, entre- tional information can be obtained from other preneurs complain less about water. The sources and through targeted interviews and same result was found with respect to small-scale follow-up surveys. In fact, infor- electricity, telecommunication, and postal mation from RIC databases will also serve to services. improve the designs of other kinds of rural 9. Enterprise performance; household choices surveys. about entrepreneurship, farming, and wage employment; and perceptions about invest- ment climate obstacles are all affected by Statistical and Analytical Limitations the community’s economic environment. The economic environment of enterprises consists Observed community characteristics (includ- of large numbers of factors that individually and in ing benchmarks) matter, but apart from these, combination shape the options for entrepreneurial regression analysis shows that unmeasured decisions. Many of these factors interrelate in com- community factors are highly influential in plex ways. This basic situation creates major un- causing similar choices and behavior within avoidable challenges offering analysts no simple a community. solutions. The main challenges are: Executive Summary xix • The number of potentially relevant variables 1. Improvement of the quality of data collection and that have been collected is generally too large processing.3 Poor quality data collection and to include simultaneously in econometric processing has a major effect on the outcomes models. The pilots cover only 100 to 150 com- and efficiency of RIC work. It limits potential munities, roughly equal to the numbers of output and the precision of findings. It also variables; hence either information must be increases the cost of analysis, since the time condensed or variables must be handpicked. required for analysis is much greater than for • The investment climate concept is broad and good-quality data sets. If fewer records must multifaceted. Because of restrictions on ques- be discarded because of data-quality prob- tionnaire length, each facet can only be cov- lems, sample sizes can be reduced somewhat ered in limited detail if all facets are to be or coverage increased. addressed. If one facet is found to be influen- 2. Reducing the length of questionnaires. Many tial, further details always remain to be data have been collected in the pilot RIC sur- discovered. veys that were not used in the country reports • Many relevant factors cannot be well or in the analysis presented in this report. observed or adequately measured using sur- This is understandable because of the pilot veys, and these play a role in the analyses as character of the three surveys. With the bene- unobserved background variables. fit of experience, however, a careful revision • Communities often pursue improvements in leading to a substantially shortened question- the economic environment in many ways naire would be the logical next step. simultaneously. The interrelationships among variables easily results in problems of multi- 3. Further standardization of questionnaires. Coun- collinearity, which, together with the prob- try teams were generally inclined to design lems of unobserved variables, can result in country-specific variations of the survey biased and imprecise parameter estimates. designs. This contributed to an excessive collection of unused data while reducing There are no quick solutions to these challenges: inter-country comparability. Similarly, stan- analysis requires use of different tools, focused on dardization of the structure of the RIC data- different questions, and addressing different bases across countries will enhance the aspects of the database. productivity of empirical investigations. Implications for RIC Methodology Further Research Below are some of the conclusions and recommen- Many potential future research topics present dations that follow from consideration of the RIC themselves. Among these are: methodology and its further application. Survey design 1. The analytical tools described in this report deserve to be applied with priority to other 1. Stratified sampling is indispensable for available data sets, such as those for drawing a cost-effective sample. This implies Bangladesh, Benin, Ethiopia, Indonesia, and that sampling weights are absolutely neces- Pakistan. This will add to understanding sary in the analysis for obtaining unbiased of the similarities and variance in rural results. With the loss of weights, as in the case economies and to further sharpening of the of Tanzania, much of the value of the RICS is analytical methods applied. lost.2 2. A further build-up of analytical tools is war- 2. The length of the standard questionnaire ranted. Future studies should elucidate the should be substantially reduced through interplay between perceptions about the careful revision. investment climate and the actual conditions in the community and how this determines Quality of data collection enterprise performance and entrepreneur- Experience shows that the cost-effectiveness of RIC ship choice, with careful attention to bidirec- surveys can be greatly improved by the following: tional causality and self-selection issues. xx Executive Summary 3. Further exploration of the rural urban contin- of communities have on the reliability of uum and rural urban divides is warranted. results? Given the RICS methodology of One approach would be to conduct compara- using enterprise response aggregates to tive studies using the RICS and PICS data- describe aspects of the investment climate, bases. Another is to make use of the RICS what is, in fact, the optimal number of obser- database codes for communities’ geographic vations per community? longitude-latitude information. These fix 5. Just as for other large area surveys, such as communities in rural space, but their spatial governance, cost of doing business, and implications have not yet been explored at all. PICS, benchmark indicators are needed for Spatial regression methods could exploit comparison and analysis. The experience correlations among nearby communities, an with use of the benchmark indicators in the approach that has been widely ignored in regressions is encouraging, but further applied research to date. analysis is needed to sharpen and optimize 4. The various trade-offs (in terms of statistical their conceptualization and measurement efficiency) between number of communities and to render the indicators more compara- and number of observations per community ble across countries. The present estimation can be analyzed using existing models and models can be used to test the effect of databases. In particular, a few communities changes in definitions of the benchmark indi- in the existing databases contain an unusu- cators. The addition of other countries to the ally small number of households and enter- sample, say Bangladesh, Benin, Ethiopia, prises, while a few others contain unusually Indonesia, and Pakistan, could greatly large numbers. This aspect of sampling raises enhance the effectiveness of this study’s the questions: What effect do these types work on benchmarks. 1 Introduction 1.1 ASSESSING THE RURAL INVESTMENT CLIMATE A sound rural investment climate and rapidly expanding agricul- ture are basic ingredients of a dynamic rural economy, according to the 2008 World Development Report, Agriculture for Development (World Bank 2008b). Recognizing the cardinal signi�cance of invest- ment climate to economic growth and poverty reduction, the World Bank devoted to the topic its 2005 World Development Report, A Better Investment Climate for Everyone (World Bank 2005). Interest in investment climates has emerged relatively recently. In the 1960s and 1970s, governments in many countries believed they should play a direct role in rural credit, input supply, and the pro- duction, transport, distribution, marketing and trade of certain prod- ucts, especially agricultural. In the 1980s and 1990s, however, government-dominated systems fell into disgrace because of their poor performance. Although privatization and liberalization were often necessary to stabilize economies and provide a basis for eco- nomic growth, in many cases these did not result in a quick respo- nse from private investors. The long legacy of state-controlled and parastatal-managed markets left underdeveloped the institutions and policy frameworks for privately led markets. For those reasons, renewed attention is now being given to the conditions under which markets and private enterprises develop—in short, to the invest- ment climate (Box 1.1). Market and policy failures can present obstacles to private enter- prise development, which governments can help to overcome through direct and generic interventions. Direct interventions target individual enterprises or groups of enterprises, aiming, for example, to improve access to finance, technology, markets, or information. By contrast, generic interventions aim to strengthen the enabling environment for enterprises with, say, improvements to the legal and regulatory framework, fewer administrative burdens, improved infrastructure, and a better functioning financial sector. Direct and generic interventions do share areas of overlap. Even so, with the exception of infrastructure and agriculture policies, the general ten- dency in rural development has mostly been to support enterprise development through direct interventions while neglecting broader investment climate issues. Because of very limited information, the largely informal �rms in rural areas have been absent from the more analytically oriented 1 2 The Rural Investment Climate Box 1.1 What Is Investment Climate? The investment climate (IC) consists of the political, administrative, economic, and infrastructural con- ditions for getting a reasonable return on investment as perceived by potential private investors. It is a subjective appraisal by entrepreneurs of the enabling environment. Peace is crucial, as are prevailing conditions of law and order conducive to private-sector development. There must be reasonable macroeconomic stability, �nancial stability, a realistic exchange rate, and low inflation. Supportive business laws, property rights, and bankruptcy laws should be in place and enforced by clearly de�ned judicial authorities. The tax system and taxation should not be disruptive for business. Important elements in the regulatory framework are free entry for new enterprises; freedom to operate without the need for dif�cult-to-obtain permits and without heavy administrative burdens; freedom to trade domestically and internationally without serious administrative obstacles; and functional competition laws, auditing requirements, industrial standards, and market regulations. Policies should enable, but not distort markets. Good governance should also be evident, including transparent policymaking and absence of corruption among and harassment by public servants. Also of importance are the quality and availability of public services; availability of a healthy, educated, and skilled labor force; availability of �nance services; and availability of infrastructure- based services such as transport, telecommunications, post of�ce, power, water, and sanitation. Source: The Authors. studies of rural growth and employment. As a con- Determining the shortcomings in the rural sequence of this lack of attention, as well as the investment climate poses empirical questions. almost complete reliance in the literature on What hinders enterprises in one country may not household-based samples, the profession until bother those in another. What constrains them in only recently knew very little about the nature and one region or community may differ from what size distribution of rural �rms, the constraints they holds them back in another. Such questions are not face when trying to expand or even to survive, the easily answered. Partly because of limits on avail- significance of their effect on the broader rural able sample frames, the universe of largely infor- economy, and even about policies that could spur mal �rms in rural areas has been practically absent investment by rural nonfarm enterprises (RNFEs). from any type of analysis.5 As a consequence of Though still in the pilot stage, the Bank’s small this lack of attention, the profession until only portfolio of RIC assessments has already identi�ed recently knew very little about the nature and size concrete actions that can improve the investment distribution of rural �rms, the constraints they face climate in rural areas (Box 7.2). To date, the when trying to expand or even to survive, the sig- methodology for assessments has been tested, ni�cance of their effect on the broader rural econ- developed, and re�ned, and directions for overall omy, and the policies that could spur their growth. improvements can now be teased out of the pilots’ To �ll some of the policy-critical gaps in infor- designs for data analysis. mation about the nonfarm rural sector, the Bank in Improving the rural investment climate likely 2004 launched a project, the �rst of its kind, to col- will involve a differing set of remedial actions, con- lect enterprise-level data on the rural investment sisting of a tableau of nuanced priorities for the climate. Initial results from this survey-based kinds of public-sector support needed for RNFEs, assessment were published in June 2006 under the distinct from and in contrast to intervention title The Rural Investment Climate: It Differs and It geared toward stimulating investment by the more Matters (World Bank). The study featured proto- formal urban-based enterprise sector.4 Targeting type instruments and designs for data analysis the rural investment climate is consistent with a grounded in surveys at the community, enterprise, rural development policy that stresses the impor- and household levels. Supplementary information tance of a thriving private sector in rural areas as about prices and unit costs were also surveyed to the best means of achieving growth, creating gain a handle on factors contributing to the prof- employment, and reducing poverty—a policy that itability of the surveyed enterprises as well as is, in fact, the focus of the World Bank Group’s the transactions costs of doing business in the approach to rural development. surveyed areas. With adaptations to satisfy local Introduction 3 concerns, the survey instruments were adminis- provide expertise and financial resources to the tered as pilot studies in Sri Lanka, Nicaragua, Tan- teams implementing projects in the regions. zania, Indonesia, and Benin. Currently, the final pilot assessment is underway in Ethiopia. The 1.2 IMPLICATIONS FOR mainstreaming of RIC assessments by the Bank’s Regional Of�ces is underway or in advanced plan- THE RURAL AGENDA ning stages in Bangladesh, Cameroon, Mozam- The International Food Policy Research Institute led bique, Nigeria, Pakistan, and Zambia. These RIC a worldwide, stage-setting review of developments assessments have added or will add substantive in the rural nonfarm economy, resulting in Trans- information on several issues relevant to the rural forming the Rural Nonfarm Economy (Haggblade et al. IC that were not appropriately addressed in urban 2007). The publication provided comprehensive, IC surveys and were overlooked in agriculture- in-depth assessments of off-farm employment based studies. growth and the expansion of rural enterprises in Developing tools for assessing the investment several developing regions. It also discussed incen- climate has been very much a process of learning by tives and constraints, including the fostering role doing. It includes three aspects: a methodology for played by dynamic supply chains as enabling cost-effective surveying, a methodology for analyz- factors for enterprise start-up and expansion, and it ing survey results, and application of the �ndings concluded with several general recommendations to policy analysis and the design of rural develop- for development policy. For policy environments ment interventions. The lessons learned from the “where the basic components of an incentive pilots are most encouraging and have provided the system favorable to rural business are in place activ- basis for both the present study and for follow-up ity,� the study identi�ed speci�c promotional activ- work to the 2006 study, including a RIC Implemen- ities that can “accelerate rural growth as well as the tation Manual (World Bank 2007c) summarizing participation of poor households in the rural non- the lessons learned while implementing the pilot farm economy.�7 surveys and offering revised prototype survey Key promotional activities cited were: (i) identi- instruments for use in future RIC assessments.6 fication of potential engines for rural growth, The need remains, however, to improve the (ii) diagnostic evaluation of supply-chain dynamics RICS by developing more rigorous analytical to identify points and types of strategic interven- methodologies than those used in many of the tions, and (iii) construction of flexible institutional pilots. New analytical tools have been developed coalitions for implementation. The report then for this study for advanced modeling of investment offered detailed guidelines for effectuating determinants, factors underlying entrepreneurial promotional activities, inclusive of the role (and decisions to open new enterprises, and the analy- limitations) of decentralized implementation, the sis of reported IC constraints. This study also devel- government-centric role contained in most Poverty ops benchmark indicators that scale community Reduction Strategy Plans (PRSPs), and involve- factors (features) along several axes to test their ment of NGOs. Such promotional activities do not influence as potential facilitators of or exogenous directly address the main elements of RIC reform, constraints on the local RIC and thus on enterprise however, as it is presumed that the necessary performance and entrepreneurial behavior. improvements will already be in place. Nonethe- The ongoing modifications also aim at stan- less, as RIC improvements are likely to be intro- dardization, which will make RIC data more com- duced in phases, subsequent reforms would no parable and encourage the analysis of RIC data doubt be influenced and supported by the above sets as they become available—both inside and promotional activities, which in turn would likely outside the Bank. Expanding the base of available influence the identi�cation and nature of programs RIC assessment results will also help make the and investments that could further improve the sum of all results more usefully inform policymak- rural investment climate. In sum, IFPRI’s report ing. For this purpose, the RIC study team recom- provides a valuable reference for use during the mended that ten additional RIC assessments be “stage-setting� phase of RIC assessment design. commissioned over the next four years (World Generally, mainstreaming improved RIC assess- Bank 2007c). Within the Bank, the Rural Anchor ments into World Bank operations and rural devel- and its component units are well placed to dissem- opment programs will ultimately lead to a more inate the methodology as it develops and to informed policy dialogue, built on rigorous 4 The Rural Investment Climate supporting analysis. The analysis will also provide useful for any team designing and conducting sur- private-sector actors with improved information veys, analyzing the data obtained, and using the on which to base investment decisions—the �ndings in policy dialogue and preparation of pol- increased con�dence that comes from better infor- icy interventions. mation is in itself an important element of the rural investment climate. For the World Bank, the infor- Audience mation the assessments provide will inform the design and monitoring of operations that support The following chapters contain several key lessons rural enterprise development. It will also improve learned from the pilots, along with suggestions for our understanding of the structure and functions robust approaches to data analysis and pointers of the rural economy generally, beyond the para- for improving the rural investment climate. meters of individual economies—providing Although it was constructed primarily for the ben- insight into agricultural and rural economics and efit of the development community—including their place in the global economy. governments, donor agencies, and larger NGOs— this publication constitutes a useful reference for survey managers and staff, for the agencies 1.3 STUDY OBJECTIVES responsible for the assessments, and for indepen- dent policy research entities. RIC1, the initial study in this series (World Bank 2006) documented the need to address the rural investment climate separately from on-going Bank surveys of the general investment climate known 1.4 ORGANIZATION OF as Productivity and Investment Climate Surveys (PICS).8 Its findings and conclusions were pre- THE REPORT sented mostly in the form of cross-tabulations and This study’s essential �ndings and policy implica- derived mainly from subjective assessments of the tions are organized into the following six chapters. rural investment climate by enterprise owners and Annexes provide supporting material. Chapter 2 managers. The study reviewed technical and presents an overview of related work and of the methodological issues encountered by the country literature; it also describes the subsequent chap- survey teams in data collection and provided guid- ters’ methodological framework, including new ance for future surveys. The study indicated that ways of addressing questions of endogeneity in statistical analysis of the data sets would be pur- these kinds of surveys while seeking to isolate sued in a subsequent study—this study, RIC2. cause and effect. Chapter 3, by applying econo- RIC2 provides a comprehensive, largely econo- metric analysis, measurably extends the examina- metric analysis of the data collected in the pilot tion of enterprise performance and investment surveys with respect to household, enterprise, and climate constraints initiated in RIC1. A rigorous community characteristics. Focusing mainly on the examination of enterprise dynamics and entrepre- data sets for Nicaragua, Sri Lanka, and Tanzania, it neurial choice is developed in Chapter 4. Aiming analyzes relations between household and com- to highlight the differing effects on RNFEs, Chap- munity characteristics and enterprise dynamics ter 5 draws together the main implications RIC2 and performance. The objectives are: findings on the rural investment climate in the three country pilots. Community-level influences • to provide broader and deeper understand- also matter, and Chapter 6 examines how the local ing of nonfarm activity in rural areas, its con- IC and other community characteristics shape the straints, and possible ways to mitigate those environment for economic activity. Conclusions constraints; and recommendations appear in Chapter 7, • to initiate a method of benchmarking the including suggestions for using RIC results for investment climate in rural areas; and policy reform and for targeting the rural public • to advance and sharpen analysis methodol- expenditures needed to foster improvements ogy and to provide guidance on analysis to in the rural investment climate. The Annexes future survey teams and policy analysts. describe the databases employed and the method- RIC1, the Implementation Manual (World Bank ologies used in the study, as well providing 2007c), and RIC2 together will provide guidance detailed regression results. 2 Approach and Methodology RIC2 is grounded in methodological extensions of the existing litera- ture, but it also offers novel approaches to modeling enterprise per- formance and dynamics, investor behavior, and the outcomes of entrepreneurial choices in the face of rural investment climate con- straints. This chapter �rst reviews the �ndings of prior research, both in the general literature and in speci�c assessments of RIC data, and then explains the twofold methodological contributions of RIC2: (i) the construction of benchmark indicators and the performance testing of these indicators as explanations for enterprise performance, entrepreneurship choice, and community-level influences on the IC faced by RNFEs, and (ii) new empirical models and methodologies addressing previously unexplored features of the RIC database. 2.1 LAYING THE GROUNDWORK 2.1.1 General Literature Between two-thirds and three-quarters of the estimated 1.2 billion people living below the one dollar a day poverty line are estimated to live in rural areas (World Bank 2008b). While agricultural produc- tivity can make an important contribution to improving rural wel- fare, recognition is growing that the poor rely on a diversified income portfolio to which the rural nonfarm sector makes an increas- ingly important contribution (Reardon et al. 2001; Barrett et al. 2001; Lanjouw and Lanjouw 2001). Along with Haggblade et al. (2007) and the recent World Develop- ment Report (World Bank 2008b), Lanjouw and Lanjouw (2001) offer a thoughtful, comprehensive review of the rural nonfarm sector. Among its insights was the observation of the widening recognition in recent years of the contributions of the rural nonfarm sector to eco- nomic growth and rural employment. The authors review and docu- ment the size and heterogeneity of the sector in developing countries worldwide, pointing to evidence that the sector in many countries is expanding rather than declining and that a positive effect on the poor, while by no means inevitable, can be considerable. The literature reveals differences, however, concerning the role of geography. Larson and Cruz-Aguayo (2008) point out that research has often emphasized the distinction between rural and urban incomes rather than sectoral differences. This distinction becomes more than a convenient expository device when one recalls that explicit motivations for migration are often tied to urban settings. 5 6 The Rural Investment Climate Some arguments relate to the accumulation of of informal enterprises seek to reduce informal human capital, viewed as an important determi- operating costs by remaining beneath the radar of nant of a successful transition from agriculture to the licensing authorities and the scrutiny of tax employment in other sectors. Also, cities may offer system operatives. The authors view informality a range of supporting markets and economies of as essentially a transitional phenomenon, a condi- scope that are lacking in rural areas. tion expected to fade away as economies develop Methodologically, most studies of nonfarm and state regulation and services transition in par- income are set up to explain labor participation allel from obstructive to more facilitating stances. decisions or indicators of specialization, such as But this will take time. Thus a more sympathetic shares of income derived from off-farm sources. understanding of the reasons why so many �rms Although less typical, a few household studies elect to remain informal—coupled with a more address the complementary issue of how house- supportive system of governance and visible, sus- holds faced with productivity constraints allocate tained efforts to discourage rent-seeking by gov- existing resources among a portfolio of activities. ernment officials—will likely increase the appeal Such studies take a structural approach that allows of registering small enterprises as formal, legal comparison of average returns to quasi-�xed and entities. flexible classes of labor and capital assets.9 Vijverberg (2005) reviewed the adequacy of Much recent literature on enterprise perfor- using living standards survey (LSS) databases to mance focuses (understandably) on Sub-Saharan examine the rural investment climate and nonfarm Africa. Poor business environments often mean entrepreneurship in Ghana, Guatemala, the Kyrgyz high costs for certain services important to manu- Republic, and Vietnam. The study asked whether facturers. Eifert, Gelb, and Ramachandran (2005) “a multi-purpose household survey [is] a suitable show that African �rms face high costs (for trans- vehicle in order to measure the effect of investment port, logistics, telecommunications, water, electric- climate (IC) variables on non-agricultural selection ity, land and buildings, marketing, accounting, and performance,� and it concluded that “while security, and bribes) compared with Asian firms the IC content is informative, it remains limited.� and that African firms suffer substantial losses Vijverberg found that the main gaps in the LSS from power outages, crime, shipment losses, and community questionnaire concern the topics of the like. credit, business organization, and local gover- African financial markets are the least devel- nance. Further, the most elaborately designed oped in the world, and development economists enterprise module, while recording enterprise have long held that this impedes growth. Bigsten outcomes such as size, sales, income, and employ- et al. (2003) examined formal credit market partici- ment, failed to shed light on the effect of the pation and credit constraints based on 1991–95 economic environment on the performance of survey data. They found that the demand for formal the enterprise.11 loans among African manufacturers was quite low Two of the most recent and more comprehen- and that credit market imperfections translate into sive contributions to the literature are the WDR binding constraints only if firms have a desire (2008b) and Haggblade et al. (2007).12 Focused to invest.10 As uncertainty hampers investment, mainly on the policy and program requisites for Bigsten and Soderbom (2006) conclude that gover- stimulating primary agricultural production, the nance is likely central to reducing risk. Meanwhile, WDR offers conclusions and recommendations for an important issue is how to accommodate Africa’s improving the rural investment climate that tend large informal manufacturing sector, given the to be thematic. Nonetheless, many are germane to high costs associated with formality. the present study. The question of informality has also been stud- The WDR (2008b) stressed the importance of ied in the Latin American context. Building on off-farm income generation and employment and seminal work by de Soto (de Soto 1989), Perry, promoted the potential for stimulating RNFEs Maloney, et al. (2007) review the ubiquity and through reforms to improve the rural investment context of enterprise informality in the LAC climate. This approach is new—no primary rural Region and offer several thought-provoking development document by the World Bank insights and conclusions. In practice, informality Group or any other international organization has discourages growth in firm size, as the managers focused so centrally on the importance of the rural Approach and Methodology 7 investment climate. In the overview section, the 20 percent of the firms examined in the studies report posits: recorded in Haggblade (2007), Chapter 5, were new entrants. But large numbers (over 50 percent • the basic ingredients of a dynamic rural non- in some cases) disappeared within three years. The farm economy are a rapidly expanding agri- annual growth rate of the surviving RNFEs was culture and a good RIC, which is as essential high, but in most countries a minority of enter- for agriculture as it is for RNFE develop- prises fueled this expansion. ment; and Jobs created by the expansion of existing enter- • the role of the state in promoting agri-food prises are more likely than start-ups to reflect market development includes improving the increasing ef�ciency and to demand pull forces in investment climate for the private sector. the economy. The expanding enterprises show The report goes on to note that indirect costs several common characteristics. They tend to be stemming from a poor IC appear to be highest on younger, to have started smaller, to be in the average in Africa. manufacturing or service sector, and to be oper- SME development was also a central focus of ated by male entrepreneurs from the home. The the WDR, especially in the agro-processing sectors. IFPRI report thus posits that government inter- Here the WDR argues that an improved RIC can ventions should focus more on existing enter- foster contestability and competitiveness and thus prises than on new start-ups and that unblocking should be considered as a key policy instrument market constraints further up the supply chain for stimulating SME development in rural areas. will be key to expanding opportunities for these Examples cited include the seeds industry and the growing �rms. contribution to R&D and to the operation of rural An intriguing typology is presented in the labor markets. Noting the still limited capacity of study’s exposition on household income diversifica- rural financial markets to serve rural financial tion. The authors note that household motives for needs, the WDR observed that much of the invest- income diversification, as well as the available ment needs for RNFE development and growth opportunities, differ significantly across settings must continue to be financed from rural savings and income groups. This suggests that an impor- and the gains from private-sector development. tant distinction should be made between diversi�- The rural investment climate will be an important cation undertaken for accumulation objectives, determining factor, governing both the robustness driven mainly by pull factors, and diversification of investment demand and the volume of “infor- undertaken to manage risk or supplement stag- mal� savings made available to new or expanding nant or even declining incomes from farming enterprises. Lastly, the report signaled the lack of a activities, driven by push factors. generic mandate imbued in most Agricultural The observed patterns of diversification in the Ministries to deal with issues and reforms aimed several studies reviewed in the IFPRI report tend at improving the RIC, while other ministries gen- to validate the much larger body of anecdotal evi- erally have limited interest in stimulating agricul- dence now available. In richer areas, household tural value chains. specialization in farm or in nonfarm activities is Although the IFPR policy recommendations much more pronounced than in poorer zones. (Haggblade 2007) presume that most of the pre- While pluriactive households in income deficient requisites for improving the rural investment cli- areas may not be able to engage in either pursuit mate will already be in place (see Chapter 1 above) very ef�ciently, they are able to spread risk, com- the study offers many pointers germane for pensate for a limited asset base, and generally advancing the analysis of the RNFE phenomena in survive—albeit at a semisubsistence level. This developing countries. These pointers helped to specialization is also reflected when comparing establish the agenda for the analytics presented in individual households. While many wealthier the following chapters. Key among these are impli- households are genuinely pluriactive, the individ- cations concerning (i) enterprise dynamics, and ual members tend to specialize to a far greater (ii) factors that stimulate rural household income extent than do the less well off cohort. Further- diversi�cation into rural nonfarm activities. more, the available evidence suggests that rural The review of enterprise dynamics found that nonfarm income greatly exceeds the value of farm new �rms were being created at a high rate; over wage-earning, by factors of 4.5–5 to 1 in India and 8 The Rural Investment Climate Latin America and more than 20 to 1 in some parts nonfarming wage labor versus nonfarm self- of Africa.13 employment, farming, or farm wage labor, the The determinants of household capacity to pay-off from education was highest in rural non- diversify into rural nonfarm activities were exam- farm wage labor, although less so in cash cropping ined. Skill-based and �nancial barriers to entry and and self-employment. Because of the close rela- expansion typically do not deter asset-rich house- tionship between education and land ownership, holds, whose members generally appear able to the contribution of the latter to the above noted cream off the more lucrative RNFE (and other non- patterns usually autocorrelates with the influence farm) activities. Conversely, asset-poor house- of education. holds remain confined to the low-return segment In summary, the IFPRI study infers that from rural nonfarm earnings. Referring to “static Because of initial differences in asset endow- and dynamic capital holdings,�14 the study essen- ments, rich and poorer household diversify tially partitioned the householders’ asset base into differently. The rich typically engage in more meso- and micro-level assets. Infrastructure avail- capital- (including human capital) intensive and ability and its quality and reliability are probably more remunerative activities, leaving the poor the single most important meso-level asset. Nearly confined to labor-intensive, highly contested all studies reviewed showed infrastructure to be a niches with low barriers to entry and low signi�cant correlate of rural nonfarm employment, returns.15 especially roads, power, and water supply. Prox- imity to urban marketing centers can be another Haggblade (2007) also discusses the potentially meso-asset. Few studies have used household data transformative contributions of agricultural devel- to analyze disaggregated rural nonfarm activities opment to RNFE development via intersectoral with reference to spatiality, but Haggblade et al. growth linkages. From several cross-sectional record one instance from Nepal showing that rural studies, the authors conclude that sustained RNFE nonfarm wage employment fell away quickly from growth and development is most likely to be cen- peri-urban areas to the rural hinterland, while a tered in rural locales experiencing a vibrant agri- U-shaped pattern for self-employment revealed cultural sector. Backstopped by evidence from East some rural nonfarm activity serves local needs not Asia and from several locales in India, Africa, and met by supply from urban areas. Latin America having a more dynamic agriculture, The micro-assets typically resemble private ser- Haggblade et al. also note that in time the growing vices or are privately held physical assets (for strength of urban demand linkages for rural man- example, land). Good examples of the former are ufacturing will eventually replace stimuli emanat- organizational and social assets. These have been ing from agriculture and may even eventually relatively underexplored in the rural nonfarm lit- absorb or displace RNFEs, as the larger urban erature but they deserve much more emphasis, as firms relocate to the urban periphery or deeper such social linkages can play an important role in into rural areas. reducing risk and transaction costs for households operating RNFEs. The available evidence, though 2.1.2 Prior Studies Using Data paltry, points to a strong link between member- from the RIC Pilots ship in organizations and other connections and successful participation in nonfarm rural activities. RIC2 provides comprehensive analysis of the data Some studies also show that the better endowed collected in the pilot surveys with respect to house- households are more likely to benefit from such hold, enterprise, and community characteristics. social capital. Deininger, Jin, and Sur (2007) analyzed data sets Perhaps most signi�cant, however, is the strong from the Sri Lanka pilot, and Sundaram-Stukel, correlation shown in several of the studies Deininger, and Jin (2006) did similar analysis for between human capital in general and participa- the Tanzania data set. Both studies explore charac- tion in and returns from rural nonfarm employ- teristics of the nonfarm sector and demonstrate ment. Education is key, although there may be quantitatively the sector’s economic importance, signi�cant interactivity differences in skills. When identifying enterprise characteristics and IC schooling effects are disaggregated by types of obstacles to sector expansion and productivity. rural income generation, that is, returns from Deininger et al. estimate that RNF value added Approach and Methodology 9 amounts to 80 percent of agricultural GDP. The indicative of market constraints. On the basis of incomes of participating households were found both parametric and nonparametric conditional to be signi�cantly higher than those of nonpartici- productivity measures, entrepreneurial activity in pating households. Barriers to entry are low, and rural Nicaragua was found to have a dual charac- nonfarm development only modestly affects ter: sometimes representing opportunity and inequality, implying a large potential contribution sometimes serving as a refuge for family labor. to growth and poverty reduction. Infrastructure These results are consistent with the scale-related constraints negatively affect new start-ups, as well constraints on asset-poor households affecting as investment in and productivity of existing rural incomes across all levels of specialization. For enterprises, with small enterprises being especially most rural households, labor appears to be better affected. rewarded by formal wage markets than by family The Sri Lanka and Tanzania studies employed businesses or farms. In contrast, for wealthier probit regression models to estimate and explain households, differences in productivity do not factors behind the decision to start up an enterprise relate significantly to sector or self-employment and probit and tobit models to explain determi- choices. nants of new investment in existing enterprises. Explanatory factors included both objective mea- sures from community-level surveys (for example, 2.2 CONTRIBUTIONS OF RIC2 availability and adequacy of physical infrastruc- RIC2 offers several methodological extensions ture) and subjective perceptions from enterprise of the existing literature and suggests novel managers about the relative importance of several approaches to modeling enterprise performance investment climate constraints. Similarly, esti- and dynamics, investor behavior, and the out- mates of total factor productivity were made to comes of entrepreneurial choices in the face of portray the extent to which investment climate perceived rural investment climate constraints. It constraints reduce the ef�ciency of resource use in also develops a study of perceived investment the rural nonfarm economy. climate obstacles in light of objectively measured In stark contrast to most of the �ndings for the community characteristics. The findings are pre- formal sector in most countries, where taxation sented so as to provide methodological guidance and other regulatory constraints were identi�ed as for future survey teams and policy analysts. key constraints, both IC studies found that infra- As a first step, benchmark indicators were structure constraints (but not regulatory obstacles) developed to summarize the many community- pose a formidable barrier to rural households’ par- level factors commonly associated with the invest- ticipation in rural nonfarm enterprises and to ment climate. Benchmarks serve two purposes: investment and increased productivity by existing (i), they may be used to compare countries, and ones. These barriers being particularly harmful for (ii) they are potentially useful devices for explain- small enterprises, policies to improve delivery of ing enterprise performance, entrepreneurial the public services in question will be important choices, and patterns in entrepreneurial com- foundations for a flourishing rural nonfarm sector, plaints about IC constraints. Benchmark indicators which in turn will have an important role in are described in subsection 2.2.1. poverty reduction. Lacking proper sampling RIC data are gathered in such a way that weights, the reliability of the estimates for Tanza- assumptions behind standard econometric models nia is open to question; however, the general do not apply. The peculiarities, owing to endo- econometric �ndings accord well with more anec- geneity and �xed effects, are shared by other com- dotal evidence. mon surveys, such as Living Standards and Motivated by the question of whether policies agricultural surveys. Accommodating these pecu- supporting nonfarm businesses can reduce rural liarities requires important modifications to stan- poverty, Larson and Cruz-Aguayo (2008) exam- dard econometric techniques, one of the signal ined the consequences for income of household features (outcomes) of this report. Section 2.2.2 decisions about family labor, farms, and busi- below discusses these, as well as introducing in nesses. Using survey data from the Nicaragua RIC greater detail the topics to which new models are pilot, the authors identified sector-related differ- applied: enterprise performance, entrepreneur- ences in returns on allocated household assets ship choice and enterprise start-up, perceived 10 The Rural Investment Climate investment climate obstacles and interactions exploit the information contained in the large between entrepreneurs and government, and dif- number of community-level variables to achieve ferences in the rural investment climate between (i) cross-country comparison, and (ii) analysis of communities. enterprise performance, entrepreneurial choices, and patterns in entrepreneurial complaints about IC constraints. 2.2.1 Benchmarking The RIC assessments yielded six benchmark The RIC household, enterprise, and community indicators: questionnaires ask several hundred questions about various aspects of the geographic, institutional, eco- (i) connectivity, that is, indexes measuring nomic, and social environment faced by RNFEs. rural-urban connectivity and regional eco- The variables that measure these aspects are far too nomic integration, including the quality numerous for direct comparison between countries, and capacity of communications and times among regions within a country, or even among required to travel to main markets and communities in the data set. They are also far too nearby cities or to access postal and other numerous for simultaneous inclusion in the study’s government services; regressions. A common approach in policy analysis (ii) the quality and availability to community is to compute benchmark indicators for which infor- residents of infrastructure services, includ- mation from many variables—often covariant ing power, water, sewerage, and other ones—is scaled and synthesized.16 household services, and of local roads; Of necessity, developing benchmark indicators (iii) access to business services, including man- is an iterative process. It starts with the choice of agement, marketing, accounting, legal, insur- potential subindicators and indicators based on ance, technical, and information technology; theoretical and empirical considerations. Subindi- (iv) the quality, reliability, and transparency of cators must combine related variables that point in governance; the same direction. Also required is suf�cient good (v) the level of human capital, assessed via an quality empirical information that can be scaled to index that scales education and experience similar quantitative ranges and aggregated into levels in the sampled communities; and subindicators, which in turn should allow for (vi) availability, access, and costs of �nance ser- meaningful aggregation into main benchmark indi- vices, including the number and kinds of cators. The subindicators and indicators should be formal financial and insurance service scaled so they can be compared between geo- institutions and services provided, access graphic areas and countries. To be usable, indica- to commodity futures and options, and tors must provide consistent, robust information; the general availability of formal lending cover sufficiently wide ranges; be empirically services. meaningful; allow for unbiased comparisons; and Definitions for each benchmark indicator and serve as useful input variables in a broader analyt- subindicator indexes are provided in Annex E of ical analysis. Missing observations and unreliable this report, with illustrations from the Sri Lanka answers will jeopardize the reliability of indicators. pilot. The indicators are mainly derived from data During data exploration for this study, sets of collected by the community questionnaires, but variables were subjected to a factor analysis to help also in part from aggregations at the community identify patterns that could help define bench- level of data collected through the household and marks. Owing to differences in the pilot surveys’ enterprise questionnaires. Although the RICS designs and de�nitions, factors extracted from the community surveys have no parallel in the PICS, data were unusable for benchmarking, as they the benchmark indicators ii, iii, iv, and vi for RIC were country-specific and not replicable across assessments resemble to some extent four of the countries. But had essentially uniform survey indicators developed by the World Bank Group for instruments been employed, at least for the core PIC assessments: variables, a common factor analysis could have been applied to each pilot’s data set and bench- 1) access to infrastructural services and service marks synthesized using common algorithms. delivery; Benchmarking was nonetheless undertaken to 2) availability of business development services; Approach and Methodology 11 3) governance and corruption, covering quality 2.2.2 Econometric Methods and RIC of public-service delivery and aspects of cor- Applications ruption; and An econometric study of enterprise performance 4) access to financial services and service and entrepreneurship choices must accommodate delivery. the structure of the sample data. Two main features In this study, benchmarks iv and v and one are prominent. First, survey data should be col- subindicator of vi are aggregates of household and lected using a two-stage sampling process. In the enterprise data. Provided communities are large �rst stage, communities are selected; in some coun- enough and contain suf�cient sampling units, the tries they may be stratified by region to ensure benchmark indicators will be mainly exogenous adequate regional representation. Then, a list of and, in principle, useful as explanatory variables buildings in each of the selected communities is in regressions on entrepreneurship and enterprise made, and a random sample of enterprises and performance. The use of benchmark indicators households is determined for each community.17 thus provides an alternative to the ad hoc selection In most countries, enterprises may usefully be strati- of variables from the hundreds of potential vari- �ed by enterprise size and/or activity sector. Most ables that could be chosen for the regressions. The of the selected communities will normally include subindicators also form a systematic set of speci�c very many small one- or two-employee units. indicators. These can easily overwhelm the sample. It is there- An accurate presentation of countries’ rural fore desirable, indeed necessary, to also sample as investment climate and their comparison across many of the larger enterprises as can be found— countries are crucial aspects of policy analysis and perhaps by designing a separate, purposive sam- RIC reform. From the perspective of the donor ple. To correct for oversampling any segment of the community (and the Bank), international bench- enterprise population, sampling weights must be marking constitutes one of the most important computed when the survey is completed and then contributions made by a country-based RICA, as used in any statistical evaluation of the data.18 Sam- international comparison can contribute signifi- pling weights also adjust for the relative size of the cantly to the continuing policy dialogue over pri- selected communities, in comparison to the popu- orities for reform and investment. The benchmarks lation of the stratified regions, to maintain repre- de�ned in this study have already been applied in sentation at the national rural level. RIC studies for Bangladesh and Pakistan. Table 2.1 In every country, the sample of households is provides a comparison. stratified by entrepreneurship status. The rate of RIC2 forms a first step in the development of entrepreneurship falls around 25 percent of house- RICA benchmark indicators, used in the empirical holds in most countries, but it will vary substan- analysis discussed in Chapters 3, 4, and 5. Chapter tially between rural communities. Some remain 6 assesses the strengths and weaknesses of the indi- primarily agricultural; in others, agriculture has cators for descriptive comparison and analysis. become secondary to nonfarm sectors. Without Table 2.1 International Comparison of RIC Benchmark Indicators Bangladesh Nicaragua Pakistan Sri Lanka Tanzania Index Index Index Index Index Connectivity 0.45 0.34 0.38 0.40 0.20 Infrastructure services 0.44 0.51 0.58 0.35 0.19 Business services 0.05 0.24 0.40 0.15 0.07 Governance 0.62 0.67 0.72 0.67 0.50 Human capital 0.24 0.21 0.21 0.33 0.21 Finance services 0.19 0.17 0.39 0.50 0.18 Average 0.33 0.36 0.45 0.40 0.22 Source: Annex K, Table K.1. 12 The Rural Investment Climate strati�cation, visits to mainly agricultural commu- for RICS data,19 inserting sampling weights into nities will yield an abundance of nonentrepreneur- panel-econometric models; the Annex also offers ial households and few observations for the an Appendix with programs in Stata that permit enterprise sample. RICS surveys typically aim for estimation in the presence of random effects and a better than 50-50 proportion of entrepreneurial sampling weights. The development and applica- households, using sampling weights to correct for tion of these models is a major contribution of oversampling households with an enterprise. RIC2, as typical studies of rural investment climate Second, standard econometric models assume using similar kinds of data ignore the issues that observations (enterprises, households) are addressed here. independent of each other, that random distur- Chapter 3 addresses the question of how the bances in their behavior are uncorrelated. Study of investment climate affects enterprise performance, the effect of the rural investment climate, however, indicated by both sales and net value added. The is founded on the notion that the economic envi- investment climate is measured by conditions in ronment influences individual (that is, enterprise the community, as given in the benchmark indica- or household) behavior. Some aspects of the eco- tors and other community characteristics. Com- nomic environment are known: benchmark indica- munity factors in general are highly influential, tors, indicator components, and other community but, not surprisingly, the benchmark indicators or characteristics such as size and seasonality. their subindicators capture only a portion. But it would be presumptuous to claim that Chapter 4 investigates household choices. These observations made about a community’s economic are twofold. One set of choices refers to the activi- environment are the sum total of that environment. ties the household pursues. Of primary interest, of Rather, it must be assumed that relevant aspects course, is the entrepreneurship choice: are mem- of the investment climate are unobserved—and bers of the household operating an enterprise? But indeed probably unobservable, even with the best- because of speci�c opportunities, a desire to diver- designed questionnaire. The unobserved aspects sify, the skill mix in the household, or other fac- enter the econometric model as community-level tors, the entrepreneurship choice interacts with the “disturbances.� By the nature of the sample data choice to seek wage employment or to operate a design, where enterprises and households are farm. Thus, the econometric analysis expands to clustered into communities, all observations in a cover those choices as well, not in the least because given community share the same community-level some development economists view the distinc- disturbance. Thus, the overall random influences tion between wage-employment and nonfarm self- on an individual’s behavior are in part idiosyn- employment as less relevant than the distinction cratic and in part communal. This violates the between farm and nonfarm activities. As the independence assumption underlying standard analysis will show, such a two-way division is less econometric models. informative than a three-way one. The second In the econometrics literature, the community household choice refers to the decision to start up disturbance is called a random effect (sometimes an enterprise. Whereas entrepreneurship status referred to as a “fixed effect�). The literature, may be viewed as a stock, enterprise start-up is a specifically the subfield of panel econometrics, flow. Successful policy changes to the investment offers many techniques for dealing with random climate will quickly be reflected in the number of effects. Panel data are repeatedly observed cross- enterprise start-ups; only over time will they affect sections: a random effect that stays with a person the rate of entrepreneurship.20 (or household or enterprise) over time would be Explanatory variables in the analysis of entre- an unobservable. Econometrically, a community preneurial household behavior include household random effect that attaches itself to all enterprises characteristics (structure, skills, ethnicity, nonla- in a community is similar to a random effect that bor income, assets), benchmark indicators and attaches itself to an enterprise over time. The panel their components, and other community charac- econometrics literature, however, never concerns teristics. Household assets may be endogenous to itself with sampling weights; common statistical this behavior: past entrepreneurship may have software does not permit estimation of panel generated incomes that led to increased savings econometric models with sampling weights. and greater assets at present. Among the commu- Annex I outlines econometric models appropriate nity characteristics, a higher per capita income Approach and Methodology 13 may attract households to entrepreneurship, but per capita, nonfarm income shares, and enter- their success in turn leads to higher incomes in the prise productivity, through simple regression on community. The potential endogeneity of these benchmark indicators and a number of other two variables could bias the estimated entrepre- economic descriptors. This analysis plays itself neurship relationship. Thus, the econometric model out at the community level, and the econometric of entrepreneurship choice is augmented with an model must only account for sampling weights allowance for endogeneity of one household-level that represent the relative importance of each variable and one community-level variable (see community. Annex I). Empirically, however, endogeneity does No claim is made here that the analysis in the not appear to be a relevant issue. following four chapters is definitive. Much work Chapter 5 examines two sets of responses from of great interest remains to be done. In principle, entrepreneurs about the investment climate. One for example, it is arguable that enterprise perfor- concerns a list of potential barriers to the operation mance and entrepreneurship behavior is driven and growth of the enterprise; the other describes by perceptions about the economic conditions in interactions with and observations about govern- the community rather than by actual, factual ment. These responses are in part subjective and in conditions—and that perceptions are driven in part factual; they represent the entrepreneur’s part by the actual conditions and in part by edu- perspectives about the investment climate. cation, networks, and opportunities. Thus, mod- Explanatory variables are the characteristics of the els estimated in Chapters 3 and 4 substitute actual enterprise, the benchmark indicators and selected conditions for entrepreneurial perceptions: they components, and community characteristics. One constitute reduced-form rather than structural prominent conclusion of the econometric analysis models. Chapter 5 develops this issue further and is that the community random effect is always outlines the next steps that beckon the analyst. important: community unobservables influence In brief, perceptions about the investment climate entrepreneurs’ perceptions about the investment may well be endogenous to the performance of the climate. enterprise, as growth may reveal constraints that Chapter 6 delves into differences between com- are not apparent to smaller-scale entrepreneurs, munities. It makes a cross-country comparison of while entrepreneurs’ evaluation of the investment benchmark indicators and explores how bench- climate may differ from that common among mark indicators are correlated with community- those chosing not to operate an enterprise, sug- level characteristics and with prices collected gesting that a signi�cant problem of self-selection at the community level. It also tries to explain must be addressed before investment climate per- community-level indicators of economic activity ceptions can be considered as potential determi- (outcomes), such as enterprise density, income nants of entrepreneurial choice. 3 Enterprise Performance and Investment Climate Constraints This chapter analyzes the economic performance of the enterprises sampled in the Nicaraguan, Sri Lankan, and Tanzanian RIC pilot sur- veys. Four performance measures are used: gross productivity, net factor productivity, employment generation, and capital generation. A brief review of enterprise characteristics is followed by summaries of the methods used and the main �ndings. The chapter concludes with an economic interpretation of the regression results and a dis- cussion of the general �ndings as they relate to RIC survey methods. 3.1 THE ENTERPRISES AND THEIR ENVIRONMENTS Nicaragua, Sri Lanka, and Tanzania offer rural enterprises different economic environments. Over the past �ve years, Nicaragua’s econ- omy grew by about three percent annually, Sri Lanka’s by almost 4 percent, and Tanzania’s by over 6 percent. Of the three countries, Sri Lanka had the highest GDP per capita in 2006. Sri Lanka’s non- agricultural value added per worker was also about 30 percent higher than Nicaragua’s and more than 60 percent higher than Tanzania’s (Table 3.1). Agricultural value added per worker in Nicaragua, how- ever, is more than 2.5 times higher than in Sri Lanka. Tanzania scores lowest in both indicators. Sri Lanka’s population density is relatively high, and a proportionately higher percentage of its roads are paved. Among enterprises for which suf�cient data was available for analy- sis, the countries show different sectoral compositions (Table 3.2). At 50 percent, Tanzania has the most trading enterprises, while Sri Lanka has the most production enterprises. The share of service enterprises is about the same for the three countries. Contrary to common belief, agro-processing enterprises form only 2 to 14 percent of the rural enterprises for these countries. The number of enterprises per 1,000 people is about the same in Nicaragua and Sri Lanka. More than half of Sri Lanka’s rural enterprises are registered, rep- resenting a signi�cantly larger share than in the other two countries (Table 3.3). In Tanzania only 20 percent of enterprises report being registered. The average age of enterprises in Nicaragua is 10.7 years, more than 2 years older than RNFEs in Sri Lanka and Nicaragua. The large majority of enterprises in Nicaragua and Tanzania oper- ate without paid labor and without debt. In Sri Lanka, 32 percent have paid labor, in Nicaragua and Tanzania only 17 and 14 percent respectively. Although the absolute numbers are small, twice as 15 16 The Rural Investment Climate Table 3.1 Basic Economic Characteristics of Selected Countries Nicaragua Sri Lanka Tanzania GDP Growth Rate (%) (annual average 2000–04) 3.1 3.9 6.2 GDP / cap (2006) (PPP constant 2000 US$) 3340 4034 620 Inflation (GDP deflator 2000–04) 6.9 8.6 7.3 Agric. Value Added (AVA) (% of GDP) 19.2 17.8 44.8 AVA / worker (constant 2000 US$) 1946 746 290 Non-Agric. VA / worker (constant 2000 US$) 2280 3048 1885 Rural population density (persons / sq. km) 117 1659 596 Population density (people / sq. km) 43 298 42 Roads paved (% of total roads) 11 81 9 Source: World Bank development indicator database (World Bank 2007d). Table 3.2 Sector of Operations, Sales, Net Value Added, and Productivity of Enterprises Nicaragua Sri Lanka Tanzania Observations 846 1018 947 Of which (%) Trade 40 39 50 Service 19 20 18 Agricultural processing 14 2 4 Other production 11 32 7 Mixed 16 7 21 Enterprise density (per 1,000 people) 50 52 NA Source: Annex D, Annex B. many enterprises in Sri Lanka operate with debt similar in both countries, Sri Lankan managers, on than is the case in Tanzania and Nicaragua. This average, have almost two more years of experience may reflect the greater density of formal �nancial than do their Tanzanian counterparts. In both services in rural Sri Lanka, where more than a countries more than three-quarters of the man- quarter of Sri Lankan RNFEs have obtained for- agers are male. mal-sector loans equivalent to 50 percent or more The three countries’ enterprises clearly vary in of owner’s equity. Only 7 percent of the enterprises size. Using sales as an indicator, Sri Lanka’s enter- surveyed in Nicaragua relied substantially on prises are much larger than Nicaragua’s; enterprises loans from formal sources; in Tanzania the num- in Tanzania are by far the smallest (Table 3.4). Using ber is negligible. net value added (NVA), perhaps a more useful indi- Data on the experience, education, and gender cator of enterprise size, the differences between of enterprise managers was collected by the Sri Nicaragua and Sri Lanka were smaller, but those Lankan and Tanzanian pilots. While the formal edu- for Tanzania were larger. It is likely that the differ- cation of managers in terms of years of schooling is ences in size reflect levels of income per capita. Enterprise Performance and Investment Climate Constraints 17 Table 3.3 Characteristics of Surveyed Enterprises Nicaragua Sri Lanka Tanzania Number in sample 846 1018 947 Registered / formal (%) 30 58 20 Age of enterprise (years) 10.7 8.7 8.5 Stand-alone (%) 7 65 37 Enterprises with paid labor (%) 17 32 14 Enterprises with debt (%) 21 42 23 Formal loan > 50% of equity 7 31 1 Experience of manager (years) NA 7.0 5.0 Education of manager (years) NA 9.5 7.8 Female manager (%) NA 23 22 Source: Annex D, Annex B. The difference between the countries suggests Finally, Table 3.4 shows the size distributions of that enterprise size increases with the level of eco- RNFE sales and net value added for the three nomic development measured by GDP per capita. countries, and Table 3.5 contrasts these two indica- In Nicaragua and Sri Lanka, larger communities tors of enterprise size with indications of commu- had larger median enterprise sizes. Assuming that nity size. Average enterprise sizes measured income per capita tends to be higher in more by sales or NVA are almost six times larger in urbanized areas, enterprise size would seem to Sri Lanka than in Tanzania and almost three increase with level of income. Tanzania, however, times larger than in Nicaragua. The maximum shows no difference in enterprise size between enterprise size in the Sri Lankan sample affects smaller and larger communities. the average. Table 3.4 Enterprise Size Distribution in Terms of Sales and Net Value Added Nicaragua Sri Lanka Tanzania Observations 846 1018 947 Sales (US$) Average 2438 6588 1186 Standard deviation 5980 41584 2153 Distribution points: Minimum 65 83 73 20 percent 452 456 193 40 percent 753 984 367 60 percent 1369 2072 576 80 percent 2881 4973 1378 Maximum 102954 1864962 18368 NVA (US$) Average 1975 2458 381 Standard deviation 4688 28636 2230 Distribution points: Minimum 11752 22094 37382 20 percent 324 75 28 40 percent 646 262 110 60 percent 1140 613 241 80 percent 2275 1579 551 Maximum 88378 1388692 15732 Source: Annex D, Annex B. 18 The Rural Investment Climate Table 3.5 Median Sales and Net Value Added, by Size of Community (US $) Nicaragua (n 98) Sri Lanka (n 147) Tanzania (n 154) Community: Smaller Larger Smaller Larger Smaller Larger Sales 753 1157 881 1554 459 459 NVA 734 1004 406 477 166 165 Source: RIC Surveys. The comparison in Table 3.6 of indicators of aver- negative or very low NVA, often presumably age productivity suggests that rural enterprises in because of incorrect reporting, were omitted.21 Nicaragua are considerably more productive than Although the loss of potentially good observations are enterprises in Sri Lanka and Tanzania. These reduces ef�ciency in the regression analysis, there results must be treated with caution, however, as is no reason to assume that the data losses caused the differences between countries may reflect struc- biases in the estimates. tural differences in supply chains and dissimilari- From the data collected in the RIC surveys, ties in the survey instruments. some variables were taken and others calculated for an analysis of enterprise performance. Table A.2 in Annex A provides definitions of data 3.2 ENTERPRISES AND VARIABLES grouped as financial data, other enterprise char- The database for the enterprise sample and related acteristics, and community characteristics. The variables is described in Annex A. RIC surveys col- numerical values of the variables entered in lected data for about 1350 enterprises in each coun- the regressions are provided in Annex B. try, although as explained in the Annex, the actual number of enterprises is smaller because of survey and data deficiencies (Table A.1). For Nicaragua, 3.3 SPECIFICATION OF data for households with more than one enterprise ENTERPRISE PERFORMANCE were consolidated during data processing. For Sri Lanka, essential community variables in war-torn REGRESSION MODELS areas are incomplete, leading to a signi�cant loss in Four enterprise performance measures are used: the number of observations. In Nicaragua, a data- total productivity, net factor productivity, labor entry failure resulted in loss of observations. In a generation, and capital generation. Total produc- few cases in all three countries, some explanatory tivity is measured at the sales level (gross output) enterprise-level variables are missing. In addition, and at the net value added level (net enterprise a threshold was applied: enterprises reporting income). Table 3.6 Enterprise Productivity in Terms of Gross Sales and Net Value Added Nicaragua Sri Lanka Tanzania Observations 846 1018 947 Average Pro�tability (US$) Ratio: Sales/Total Cost 2.59 1.09 1.01 Ratio: NVA/ Total Factor Cost 2.79 1.41 1.48 Average Productivity (US$) Ratio: Sales/day of labor 3.55 9.73 4.59 Ratio: NVA/day of labor 2.88 3.63 1.47 Source: Annex B. Enterprise Performance and Investment Climate Constraints 19 Productivity Community characteristics present direct incen- The primary empirical speci�cation for measuring tives and disincentives for enterprises and hence productivity uses a Cobb-Douglas production may affect their productivity. For Nicaragua and function, which is loglinear in output and inputs, Sri Lanka, these include data about seasonality in augmented with variables hypothesized to agricultural labor use and the number of enter- enhance enterprise productivity. Throughout this prises per 1,000 inhabitants (enterprise density). In report reference will be made to sales variants and all three countries, community population size is net value added variants. included as an indicator of the degree of urbaniza- The choice of empirical variants for the regres- tion within which the enterprise operates, and sions is discussed in detail in Annex C. Because of agricultural land cultivated per capita and the rate inherent interest in the contribution of enterprises of illiteracy are also noted. In Sri Lanka, informa- of different sizes to the rural economy, quadratic tion about where households from the community and interactive specifications of the factor input buy and sell most of their goods was used to con- variables labor and capital can better describe dif- struct dummies that to some extent indicate the ferences related to size than loglinear forms do. community’s degree of commercial openness. In The aim in formulating the regression models Sri Lanka and Tanzania, data about the commu- is to choose variables that describe enterprise nity’s main source of income, that is, agriculture, characteristics and the community environment wages, or nonfarm enterprises, were used to create and to avoid variables that could be seen as largely dummies that help capture differences in markets endogenous. Since the main objective of the analy- and income between communities. sis is to shed light on the contribution of the All regression models incorporate benchmark investment climate, three stepwise regressions are indicators, which are community-level composite carried out. This reveals the structure of the data indexes of characteristics of location, infrastruc- sets and illustrates the variance captured by enter- ture availability, utilities, public and private ser- prise variables, general community variables, and vices, governance and corruption, human capital, benchmark indicators. In addition, a fourth variant and finance services, as described in Chapter 2 is estimated in which the benchmark indicators are above. Benchmark indicators and their compo- replaced by those of their components providing nents describe both the community-level environ- the highest individual explanation. ment and the investment climate. Underemployment of household labor may lead to a lower marginal productivity for household Employment and Capital Formation labor than for paid labor. Such differences may also What conditions generate enterprises with more exist for household-owned assets and borrowed employment and what conditions generate enter- assets. One way to deal with this is to utilize sepa- prises with more capital? Or, expressed differently, rate speci�cations in the regression model for self- what explains differences in enterprise size, mea- owned and for borrowed inputs. A �fth regression sured by labor and by capital input respectively? variant separates imputed and paid labor and cap- Regression models, using the same enterprise and ital. Table C.1 in Annex C provides an overview of community data used for productivity measure- the regression variants for productivity analysis. ment, explain employment and capital generation Table A.2 in Annex A fully speci�es the variables from industry mix, enterprise characteristics, used. Table B.1 to B.6 in Annex B list descriptive benchmark indicators, and other community char- statistics for the variables used. acteristics. The parameter estimates appear in The selection of variables includes the factor Annex C. inputs—labor and capital—for the NVA variants and the factor and nonfactor inputs and deprecia- tion for the sales variants. Also considered are a 3.4 REGRESSION RESULTS number of enterprise characteristics that could 3.4.1 Variance Explained add to gross and net productivity. They include enterprise age, registration status, and line of busi- Productivity analysis. A dominant characteristic ness and entrepreneur experience, gender, and of the enterprise data sets is that enterprise charac- education. teristics explain by far most of the variance in the 20 The Rural Investment Climate data set; industry and community levels explain Tanzania, and almost 4 percent in Sri Lanka. The only a small part.22 Table C.2 in Annex C illustrates higher percentage of explanation in Sri Lanka this for productivity. The regressions on bench- could well be the result of its larger number of en- mark indicators and enterprise dummies for the terprise and community variables as compared to control of industry differences explain only the other countries. Differences in capital genera- between 3.6 and 9.3 percent of the variance. When tion can be better explained than can differences in enterprise variables are added, the explanation employment generation. Partly this could be caused jumps by a factor of five to eight. General com- by greater weaknesses in the data measuring labor munity variables explain 0.2 to 1.8 percent. If the input. Enterprise characteristics contribute most to benchmark indicators are replaced by those of differences; other community variables contribute their components that add most, the explanation much less. In Nicaragua adding community vari- of the regression increases by another 0.5 to ables results in a lower percentage of explanation; 2.0 percent. for Sri Lanka and Tanzania the explanation increases The NVA regressions have generally a one-third by 1.2 and 2.4 percent. lower explanation in Sri Lanka and Tanzania; in Nicaragua the difference is smaller. When imputed and paid labor and capital are separated (regres- 3.4.2 Productivity Analysis sion variant 5), the explained variation increases Production elasticities and economies of scale. by a few percent only, and for Sri Lanka’s NVA and The Cobb-Douglas model augmented with qua- Tanzania’s sales model it even decreases. dratic and interactive terms (log of labor and log of Although no comparable data are available for capital) has greater overall explanatory power urban enterprises, these findings may not be than do models with linear terms only. Results unique to rural enterprises. Most of the variance from this specification are recorded in Table 3.7 will relate to enterprise size, and IC variables will Annex C, as variant (4). explain a modest percentage in productivity dif- The economies of scale vary with enterprise ferences. Yet the addition of these few percentage size. For the sales variant (4) in Sri Lanka, the points may represent an important contribution to implied returns to scale at mean input levels pro�tability. equals 0.911 (Table 3.7) and rises with increasing Employment and capital. The �ndings for em- levels of labor and capital. For Nicaragua and Tan- ployment generation and capital formation show a zania it is lower, at 0.677 and 0.355 respectively, similar pattern. Industry mix and benchmark indi- but unlike the other countries, in Tanzania it cators together can explain 4 to 5 percent of the vari- decreases with factor input. For the NVA variants ation in employment generation. (For details see (4), the returns to scale are about the same for Annex C, Table C.3.) Addition of enterprise charac- Nicaragua and Tanzania and lower for Sri Lanka teristics doubles the explanation in Nicaragua, (0.733), in all cases increasing with the level of fac- triples it in Sri Lanka, and increases it by half tor input. in Tanzania. Other community variables add Separate speci�cations (referred to as variant 5) 1.5 percent in Nicaragua, almost 2 percent in for paid and imputed labor and capital input lead Table 3.7 Economies of Scale and Production Elasticities for Variant (4) Nicaragua Sri Lanka Tanzania Variant Sales (4) NVA (4) Sales (4) NVA (4) Sales (4) NVA (4) Production elasticity labor 0.422 0.432 0.220 0.369 0.013 0.055 Production elasticity capital 0.193 0.269 0.255 0.364 0.148 0.323 Economies of scale 0.677 0.700 0.911 0.733 0.355 0.379 Source: Annex C, Table C.4 to Table C.9. Notes: Elasticities are evaluated at the log of the mean input values, where the mean is computed under the assumption that the inputs are lognormally distributed in order to reduce the effect of outliers. Enterprise Performance and Investment Climate Constraints 21 Table 3.8 Economies of Scale and Production Elasticities for Variant (5) Nicaragua Sri Lanka Tanzania Variant Sales (5) NVA (5) Sales (5) NVA (5) Sales (5) NVA (5) Production elasticity labor Total labor input 0.386 0.391 0.216 0.400 0.239 0.142 Imputed labor input 0.316 0.312 0.015 0.032 0.215 0.084 Paid labor input 0.070 0.079 0.231 0.432 0.024 0.058 Production elasticity capital Total labor input 0.154 0.196 0.202 0.258 0.096 0.352 Imputed labor input 0.201 0.252 0.190 0.225 0.072 0.225 Paid labor input 0.046 0.055 0.012 0.033 0.025 0.127 Economies of scale 0.687 0.587 0.865 0.646 0.590 0.493 Source: Annex C, Table C.10 to Table C.15. Notes: Elasticities are evaluated at the log of the mean input values, where the mean is computed under the assumption that the inputs are lognormally distributed in order to reduce the effect of outliers. to better explanations, except for sales in Tanzania tant contribution to productivity in Nicaragua and and NVA in Sri Lanka. The models allow for dif- Sri Lanka, and in Sri Lanka manager education is ferences in production elasticities for imputed and important as well. In Tanzania manager experience paid labor and capital costs. Evaluation of the contributes to total productivity and manager edu- parameter estimates shows that economies of scale cation to net factor productivity. These findings are roughly similar for variants (4) and (5) for suggest that learning by doing and innovation are Nicaragua and Sri Lanka (Table 3.8). Similarly, important for survival and increased productivity. total labor and capital elasticities do not differ Manager gender is known for Tanzania and much either, but differences appear in the compo- Sri Lanka. In Tanzania male managers appear to nent effects: in Nicaragua, an increase in imputed induce higher total productivity, but not net pro- (household) labor contributes more to output than ductivity. Sri Lanka shows no signi�cant produc- does an increase in paid labor; in Sri Lanka, an tivity differences. Registered enterprises have increase in imputed labor does nothing for output, higher productivity in Tanzania and higher net whereas paid labor substantially raises output; factor productivity in Nicaragua, but for Sri Lanka and in both countries output responds more no difference appears. Registration may encour- strongly to variations in imputed capital than in age higher productivity in two ways. Enterprises paid capital. The Tanzania results differ from the with higher net income or sales may be unable to other two countries in that the labor elasticities in escape registration, but registration may also con- variant (5) are substantially higher, causing a fer particular benefits that encourage growth. higher estimate of the economies of scale (though Small, low-income enterprises may simply escape still much less than 1); and while imputed labor registration because their officers and entrepre- and capital inputs are more influential than paid neurs don’t bother with it. inputs, the difference is less pronounced than in Line of business explains few productivity dif- Nicaragua and Sri Lanka. While Table 3.8 focuses ferences. In Tanzania no differences were found, only on the implications of the alternative speci�- and in Sri Lanka only service enterprises had cation of inputs, it is important to note that this higher total productivities than trading enterprises modeling exercise can strongly afect parameter (serving as the comparison). Nicaragua shows estimates for other variables. more differences: service enterprises have signi�- cantly lower productivity and mixed enterprises Contribution of enterprise and community vari- have higher productivity; total productivity is ables. Table 3.9 provides an overview of variables higher for manufacturing enterprises; and net with statistically signi�cant contributions to enter- factor productivity is lower for other enterprises prise productivity. Enterprise age makes an impor- (mainly mining and construction). 22 The Rural Investment Climate Table 3.9 Signi�cant Contributions to Productivity by Enterprise and Community Characteristics Nicaragua Sri Lanka Tanzania Variant Sales (4) NVA (4) Sales (4) NVA (4) Sales (4) NVA (4) Enterprise characteristics Age of enterprise Experience of manager NA NA Male manager NA NA Education of manager NA NA Registration Line of business Service enterprise Manufacturing of nonagricultural products Agricultural processing Other production Mixed enterprise Community characteristics Agricultural seasonality NA NA Enterprise density NA NA Community population size Agricultural land per capita Illiteracy Main market in Neighboring communities NA NA NA NA Commercial center NA NA NA NA Nearest city NA NA NA NA Main source of income Wages NA NA Self-employment NA NA Source: Annex C, Table C.4 to Table C.9. Note: Statistically signi�cant at 10 percent or higher with positive ( ) or negative ( ) parameter estimate. NA Not available. Most of the community characteristics make no that a higher value between (0 and 1) indicates an significant contributions to productivity differ- improvement for business. Hence, the expectation ences. The only exceptions are communities is that all parameters would have positive signs. Yet where the enterprises’ main markets are in neigh- in Tanzania access has a negative value, and in boring communities and commercial centers. Nicaragua human capital does. This is most likely This suggests that openness and access to more due to unobserved background variables and mul- distant markets plays a positive role. The general ticollinearity. Multicollinearity can explain the con- finding of the limited contribution of community tradictory signs in Nicaragua for “time to near city� characteristics also suggests that the benchmarks and “cost to near city,� for example. The negative and benchmark components effectively describe sign for “sewage system� in Tanzania is most likely differences between communities. caused by unobserved variables in the few commu- Nicaragua shows relatively many significant nities having sewage systems. contributions to productivity by benchmark indica- tors and benchmark components (Table 3.10). Con- 3.4.3 Explaining Employment nectivity is important in Nicaragua and Tanzania, and Capital Generation business services in Sri Lanka, and governance in Nicaragua. The role of �nance services in Sri Lanka Many factors contribute to differences in employ- is limited to loan access. The benchmark indicators ment and capital generation (Table 3.11). Enter- and their components have been constructed such prise age matters only in Nicaragua. Manager Enterprise Performance and Investment Climate Constraints 23 Table 3.10 Signi�cant Contributions to Productivity of Benchmark Indicators and Components Nicaragua Sri Lanka Tanzania Sales NVA Sales NVA Sales NVA Connectivity Connectivity Connectivity Cost of Connectivity Connectivity Time to near city Time to near city Transportation Time to main Rail stop Cost to near city Cost to near city Time to main market (negative) (negative) market Cost to main (negative) Cost to main market market Distance to post of�ce Infrastructure services Percent with Percent with Concrete/ asphalt Access (negative) electricity electricity road (negative) Sewage system Cost of Availability elec- transportation tricity (negative) (negative) Percent with Garbage protected water collection (negative) Sewage system Garbage collection Business services Technology Business services Business services services Management consulting Governance Governance Governance Public services / Insurance institutions service Human capital Human capital Human capital (negative) (negative) Finance Services Access to loans Access to loans Source: Annex C, Table C.4 to Table C.9. Notes: The table lists �ndings from regression variants (3) and (4) with a signi�cance level of at least 10 percent. The benchmarks are printed in boldface. experience matters for capital generation, but not play varying roles. Enterprise density relates espe- for employment generation. Male managers gen- cially to employment generation and in Sri Lanka erate more employment and capital. Manager edu- to capital generation as well. Interestingly, in Sri cation makes no difference. Registration affects the Lanka openness (that is, marketing outside the amount of employment and capital generated for community) has a positive effect on employment two reasons: registered enterprises have better generation but not on capital generation. access to services, and licensing of�cials may make The benchmark indicators and their compo- special efforts to register enterprises with signifi- nents contribute differently to employment and cant factor input. Generation of employment and capital generation (Table 3.12). In Nicaragua and capital shows more differences among sectors than Tanzania, connectivity contributes significantly does productivity. Community characteristics to employment generation and human capital to 24 The Rural Investment Climate Table 3.11 Signi�cant Contributions to Employment and Capital Generation by Enterprise and Community Characteristics Nicaragua Sri Lanka Tanzania Employment Capital Employment Capital Employment Capital Variant (4) (4) (4) (4) (4) (4) Enterprise characteristics Age of enterprise + + Experience of manager NA NA + + Male manager NA NA + + + + Education of manager NA NA + + + Registration + + + + + Line of business/industry dummies Service enterprise – – – + Manufacturing nonagricultural products – Agricultural processing – – + + – – Other production + + + Mixed enterprise + + + Community characteristics Agricultural seasonality NA NA Enterprise density + + + NA NA Community population size + – + – Agricultural land per capita + – + Illiteracy Main market in Neighboring communities NA NA + NA NA Commercial center NA NA + NA NA Nearest city NA NA + NA NA Main source of income Wages NA NA Self-employment NA NA Source: Annex C, Table C.10 to Table C.15. Note: Statistically signi�cant at 10 percent or higher with positive ( ) or negative ( ) parameter estimate. NA Not available. capital generation. In Sri Lanka, similar to �ndings 3.5 GENERAL DISCUSSION for productivity, business services are important for employment and capital generation. Finance 3.5.1 Data Quality services in Nicaragua contribute only to capital As with any empirical study, the value of the generation. As with their contribution to produc- results depends on the quality of the data. In com- tivity, benchmarks and their components at times mon with other surveys, RICS data quality suffers yield counterintuitive outcomes due to unob- from limitations of human memory, respondent served background variables or multicollinearity. errors, and, at times, respondents’ unwillingness Governance, time to market, and time to post of�ce to disclose full financial data. Moreover, some have negative signs in Nicaragua. Concrete or computed variables have conceptual weaknesses, asphalt roads have a positive sign in Tanzania and and perhaps biases, owing to insuf�cient detail in a negative one in Sri Lanka. Obviously, the vari- the collected data (see Annex A). This may hold ables do not include other powerful location choice especially for computed labor input, for which a factors. Remarkably, �nance services play no role flawed stock concept had to be used. For capital, in explaining enterprise size in terms of employ- separation of capital used for consumption from ment and amount of investment, with capital gen- capital used for production can be arbitrary, erated in Nicaragua the only exception. including as it does such elements as housing, Enterprise Performance and Investment Climate Constraints 25 Table 3.12 Signi�cant Contributions to Employment and Capital Generation of Benchmark Indicators and Components Nicaragua Sri Lanka Tanzania Employment Capital Employment Capital Employment Capital Connectivity Connectivity Time to market Connectivity (negative) Distance to post Time to post of�ce of�ce (negative) Infrastructure services Infrastructure Cost of transporta- Availability Percent with Percent with services (negative) tion (negative) electricity protected water cell phone Percent with Percent with �xed Percent with �xed Percent with �xed Concrete / electricity phone phone lines phone lines asphalt road (negative) Garbage collection Concrete / asphalt Percent with cell road (negative) phones Sewage system (negative) Business services Technology Business services Business services Marketing service services Governance Governance Public services Public services / (negative) (negative) institutions Public service Public (negative) institutions Human capital Human capital Human capital Finance services Finance services Number of Number of bank bank services services (negative) Source: Annex C, Table C.10 to Table C.15. Notes: The table lists �ndings from regression variants (3) and (4) with a signi�cance level of at least 10 percent. The benchmarks, if signi�cant, are printed in boldface. equipment, and vehicles; evaluation too can pre- 3.5.2 Statistical Importance of RIC sent difficulties. Furthermore, as already noted, missing information and data entry errors cause The regressions provide many useful results on loss of observations and possibly loss of represen- factors affecting differences in enterprise perfor- tativeness in the analysis subsample. mance, which in this chapter are measured as pro- It is fair to say, therefore, that the observations ductivity and generation of labor and capital. and coded data are likely to contain errors, in part These vary in response to many factors, and the because of the surveys’ pilot character. This of pilots’ findings can provide input to all kinds of course affects the precision and ef�ciency of para- further analysis to help establish the policy agenda meters estimated in the regressions. Generally, and support the policy dialogue. Work on the pol- biases of estimates presented here are considered icy agenda will be discussed in Chapter 7, follow- to be modest, but the usual caution should be ing further discussion in Chapters 4, 5, and 6 of the applied when interpreting the empirical results. analytical work involved. Here some attention will 26 The Rural Investment Climate be given to statistical limitations and counterintu- investment climate variables, can explain only a itive results. modest part of the total variation. This section Variation depends on many factors, some of focuses on productivity analysis. Some benchmark which the data set may not explicitly include but indicators contribute well to explanations of which may still be influential in the background as productivity. In Nicaragua and Tanzania, for unobserved variables. Indeed some factors may example, connectivity contributes significantly, pertain to only a few enterprises or locations. In but not in Sri Lanka: connectivity captures many principle, variation in dependent variables can of the infrastructure deficiencies related to dis- be expected to derive from (i) observed enter- tance, which are more prominent in Nicaragua and prise variables, (ii) unobserved enterprise vari- Tanzania. ables, (iii) observed community variables, and The infrastructure services indicator did not (iv) unobserved community variables. In the pilots, perform as expected; it has negative signs in community-level factors explain a relatively small many cases, although these are significant only part of enterprise productivity variation. Box C.1, for Tanzania’s sales regressions. Reasons could Annex C, shows that, at best, community factors be multicollinearity—infrastructure services are explain around 12 percent of the total variation in correlated with connectivity—and unobserved productivity in Nicaragua, between 7 and 14 per- background variables. cent in Sri Lanka, and about 27 percent in Tanzania. The business services indicator does well in In practice, the explanation will be much lower. Sri Lanka, but it does not appear to matter in The low explanatory contribution of community- Nicaragua and Tanzania. Business services may level factors, including investment climate play a limited role, if any, for many enterprises, in variables, indicates great heterogeneity among particular in less developed situations. For that enterprises, hindering analysis of the effect of the reason the revealed contribution may reflect Sri rural investment climate. Analysts will simply Lanka’s relatively high level of development. have to cope with this fact. Although efforts were Although the governance indicator shows up in made in the pilots to collect data from communi- the tables with more positive than negative esti- ties ranging from typical rural hamlets to larger mated effects, the regression analysis did not yield market towns, future RIC assessments could per- enough evidence to suggest it really matters. Gov- haps capture an even larger and more diverse mix ernance has perhaps the poorest data reliability of of settlements. Also the scant number of communi- all indexces, in part because it is an aggregate of ties was limiting: the RIC pilots sampled in only 99 more than 20 pieces of information gathered from communities in Nicaragua, 151 in Sri Lanka, and sample enterprises in the community; for that rea- 149 in Tanzania. Because of missing data, the son alone it may be a weak indicator. Some enter- enterprise performance regression models were prises, by self-selection or active participation, estimated with enterprise data from even fewer may even profit from weak governance and communities (93 in Nicaragua, 118 in Sri Lanka, corruption. and 136 and 127 respectively for the sales and Human capital does not add much to explain NVA variants in Tanzania [see Annex C, Box C.1]). productivity; it exhibits a few signi�cant but coun- Thus while productivity variation may have many terintuitive negative signs in Nicaragua, and only potential determinants across communities, only a hints at a positive effect in Tanzania’s sales regres- limited number can be explored. In sum, although sions. Community human capital may offer only a future surveys may be successful in capturing weak additional contribution to the productivity of more variation, analysts must cope with data sets individual enterprises in models that already with many interrelated variables, multicollinear- include entrepreneur experience and education as ity, unobserved variables, and a relatively small characteristics of a successful enterprise. Another contribution of community variables. possible reason is that the measured human capital index is a weak indicator of human capital at the community level because it is based on an aggrega- 3.5.3 Benchmark Indicators tion of data from a limited number of households. The conclusion from the previous section is that Finally, many enterprises may not be conducting benchmark indicators, which together combine businesses sensitive to community-level human many of the community-level economic and capital.23 Enterprise Performance and Investment Climate Constraints 27 The �nance services indicator contributed effec- also make positive contributions. Less developed tively in Nicaragua and to a lesser extent in Sri areas require strengthened commercialization; Lanka. In Tanzania �nance services had a signi�- more developed areas require a more mature eco- cantly negative sign, which unfortunately makes nomic and institutional infrastructure. Human little economic sense. capital is a signi�cant factor in capital generation. Replacing benchmark indicators with the sub- In Sri Lanka, the most developed of the three components with greatest explanatory power pro- countries, business services contribute most to vides interesting results. As expected, in general explaining productivity and employment and cap- the added components increase the percentage ital generation. Perhaps because of the much of variation explained. Many components have higher population density and higher level of road significant positive parameters, helping to signal infrastructure, connectivity contributes less to which among the investment climate variables are explaining productivity differences. Some connec- most important. But another �nding is that, in sev- tivity and infrastructure services subcomponents, eral cases, underlying components are signi�cant however, may play a role in specific situations. with opposite signs. Examples include connectivity Further commercialization emphasizing private- components in the NVA regressions in Sri Lanka, sector service provision seems to be priority for infrastructure services components in both regres- Sri Lanka. sions for Nicaragua, business services in both regressions in Tanzania, governance in the sales 3.5.5 Comparison with Other Findings regression in Sri Lanka, and �nance services in the sales model in Tanzania. Understandably, for all The three pilot studies were conducted by country cases, the aggregate benchmark did not have a teams, and the results were published in country signi�cant effect one way or the other. reports for Nicaragua (World Bank/RUTA 2008 forthcoming), Sri Lanka (World Bank and Asian Development Bank 2005), and Tanzania (World 3.5.4 Comparative Observations Bank 2007e). Each of these reports includes some on the Economies effort to describe enterprise performance in the The findings on productivity and generation of light of investment climate constraints. In addi- employment and capital show a wide variety tion, econometric research was carried out by of factors contributing to differences among the Deininger, Jin, and Sur (2007) for Sri Lanka and three countries. This is as expected, as it reflects a Sundaram-Stukel, Deininger, and Jin (2006) for range of factors related to differences in natural Tanzania. conditions, policies, institutions, infrastructure, All teams used a Cobb-Douglas–based model and level of development. for analysis of factor productivity (gross value Tanzania has the lowest level of development, added) and investment climate constraints. The with weak commercialization and small enter- choice of IC variables was guided by the IC con- prises. Many of the enterprises have a local focus straints reported by entrepreneurs in the enter- and function in communities with large subsistence prise survey and by availability of corresponding components. The finding that connectivity most IC variables reported in the community surveys. A affects productivity indicates priority should be practical concern with this approach is the large placed on infrastructure and services to open rural numbers of IC variables covered in the enterprise areas to further commercialization. Connectivity and community surveys, which due to sheer num- also contributes to employment generation. Human bers cannot all be included in a model. Given the capital is important for capital generation. limited number of community observations, the Nicaragua’s economy is more developed but number of community variables should not exceed shows major differences between modern enter- 10 to 15. This requires that researchers use subjec- prises and commercialization in the country’s tive criteria to select variables. advanced areas and low levels of development RIC2 offers a unique parallel analysis of three and commercialization in less advanced areas. countries using a framework of benchmark vari- Improved connectivity adds most to explaining ables and subcomponents based on theoretical productivity and employment generation differ- considerations and factor analysis to explore the ences. Some infrastructure and business services structure of community data. These variables are 28 The Rural Investment Climate more robust than those used in other studies. model empirically more flexible than the basic Moreover, this framework allows control of many Cobb-Douglas form. To improve the quality of the aspects of the investment climate while simultane- analysis, the model reconstructs lost sampling ously exploring the contribution of components in weights for Nicaragua and adds an econometric one benchmark dimension. This reduces the bias analysis for Nicaragua that includes both household- in the representation of components’ influence. based and stand-alone enterprises. Furthermore, several other community character- RIC2 provides a comparative perspective on the istics were added that appear useful for describing three countries. Although direct comparison of the enterprises’ local environments. In contrast results is not possible owing to the surveys’ differ- with previous econometric studies, regression ing model specifications and variable selections, models specifically incorporate the existence of together they provide a broad overview of data, unobservable community-level factors (as a so- methods of analysis, and findings that will help called random effect), as elaborated in Chapter 2 further research on investment climate in these and Annex I. The analysis is carried out for gross three and other countries. output and net value added, with a regression 4 Entrepreneurship Choice and Enterprise Start-Up Across the developing world, rural areas exhibit richly varied pat- terns of activities. Many households, of course, derive their liveli- hoods from farming, but people are also active as traders, service providers, and manufacturers. In fact, rural nonfarm enterprises pro- vide 30 to 45 percent of rural incomes across the developing world and an even higher share among the rural poor (Haggblade et al. 2007). The level of RNFE activity not only depends on growth in agri- cultural production and rural consumption, it also stimulates it: the relation is one of interaction and synergy. RNFE ef�ciency translates directly into lower transaction costs for bringing farm products to consumers and distant markets and for bringing inputs and tradable consumer goods and services from distant markets to rural resi- dents. Hence, RNFEs affect the terms of trade for farmers, their real income, and their incentives to produce. Yet it would be a mistake to think of RNFEs only in terms of the loop between agricultural production and rural consumption. The low cost of rural labor may induce some urban entrepreneurs to establish rural branches, and it may also lead some rural entrepre- neurs to produce for urban markets (Hayami 1998; Haggblade et al. 2007). In such a decision, the cost of transport and communication plays an obvious role. But other relative improvements in the rural investment climate also offer scope for stimulating rural incomes and employment by attracting economic activities. This chapter explores why households engage in the operation of a RNFE and which households choose to do so. Is it only a matter of the skills and resources available within the household? Or do favor- able investment climates attract households to entrepreneurship? Entrepreneurship, defined as the current operation of an enter- prise, is a stock concept (Van der Sluis, Van Praag, and Vijverberg, 2005). But entrepreneurship is more often thought of in dynamic terms: finding market niches, organizing efficient methods of pro- duction, exploring new opportunities. If entrepreneurship means the pursuit of success, some ventures must inevitably fail. Indeed, turnover among small enterprises in most countries around the world is often high; 15 to 20 percent of business activities are aban- doned each year—but about an equivalent number start up (see Vijverberg 2005).24 The RIC survey does not permit analysis of enter- prise deaths, but analysis of enterprise start-ups is feasible. The obvious policy questions about the “flow� concept of entrepreneur- ship are: Which households start up new enterprises? What 29 30 The Rural Investment Climate influence, if any, does investment climate have on exclude relatives who may have temporarily or these decisions? permanently migrated but still belong to the house- Entrepreneurship is just one of several forms hold in a social or economic (breadwinner) sense. of employment and income generation. House- holds may also pursue wage employment or farm Employment activities. It is useful to analyze the three together, Comprehensive data for the rural labor force is since transition from an agrarian society to an portrayed in Annex D, Table D.8. The table indi- industrial economy moves through a phase in cates that the average number of working-age which nonfarm self-employment and small-scale adults (16 to 65 years of age) is higher in Sri Lanka entrepreneurship are important. As enterprises (2.99) than in Nicaragua (2.80) and Tanzania (2.76). become successful, they expand and hire workers The number of working adults, however, is lowest who in an earlier phase of the transition might in Sri Lanka (1.78) and highest in Tanzania (2.48): have set up their own businesses. Aspects of the the labor force participation rate in Tanzania is investment climate may encourage entrepreneur- 90 percent, as opposed to 57 percent in Sri Lanka ship in one society by drawing people away and 65 percent in Nicaragua. The male-female gap from farming, but discourage it in another by is trivial in Tanzania (2 percent) but quite large in drawing people into wage employment at larger Sri Lanka (37 percent) and Nicaragua (34 percent). enterprises. Table 4.1 provides information on household The next section summarizes patterns of members’ involvement in major economic activi- employment, income, and assets among rural ties such as self-employment, wage employment, households surveyed by the three pilots. Section and farming. It is striking that 46 percent of house- 4.2 examines household involvement in nonfarm holds in Sri Lanka, 37 percent in Tanzania, and enterprises through regression analysis; Section 40 percent in Nicaragua have multiple sources of 4.3 broadens the perspective to include other forms income. Considering all households together, of income earning (wage employment and farm- 50 percent of households in Sri Lanka, 90 percent ing); Section 4.4 addresses the issue of enterprise in Tanzania, and only 29 percent in Nicaragua start-up; and Section 4.5 summarizes �ndings. For have at least one member engaged in farming,25 reasons explained in Section 4.1, the multivariate and 61 percent, 25 percent, and 60 percent of analysis in Sections 4.2 to 4.4 utilizes RICS data households in Sri Lanka, Tanzania, and Nicaragua, from Nicaragua and Sri Lanka only, omitting respectively, have at least one member involved in Tanzania. wage employment. 4.1 HOUSEHOLDS Income Total household income is estimated by aggregat- Demographics ing all sources of income minus expenses of The age distribution of the sample households in productive activities.26 Table 4.2 provides total the three pilot countries is skewed to the left, as household income in the selected countries. Aver- shown in Annex D, Table D.2. The percentage age household income in Sri Lanka (US$2,693) share of the male population of the sample house- is more than 25 percent higher than that of holds is consistently lower than that of their Nicaragua (US$2,016) and four times that of female counterparts, likely reflecting emigration Tanzania (US$671). All of the pilots’ household by the male cohort in search of more remunerative surveys show an unequal distribution of income, employment. although less so in Nicaragua (Annex D, Table The rural population of Sri Lanka has completed D.23). In Sri Lanka almost 21 percent of house- more education (7.4 years) than have their counter- holds’ average income is less than US$500, while parts in Tanzania (5.0 years) and Nicaragua this �gure is only 14 percent in Nicaragua. (4.3 years). The male-female gap is 0.4 years in Households tend to engage in several activities Sri Lanka and 0.6 years in Tanzania, while in for their livelihoods and subsistence. Household Nicaragua women have received more (0.6 years) members earn income from wages and salaries; education than men. These statistics obviously from self-employment in RNFEs; and from average only those residing in rural areas; they agriculture, aquaculture, and animal husbandry. Entrepreneurship Choice and Enterprise Start-Up 31 Table 4.1 Economic Activities of Households and Household Members Any incomea from stated source Nicaragua Sri Lanka Tanzania Enterprise only 7.7 0.0 14.6 Enterprise and wage 8.3 9.9 2.2 Enterprise and farming 3.5 1.5 38.9 Enterprise, wage, and farming 2.7 6.6 6.2 Wage only 45.5 36.0 1.5 Farming only 10.0 18.3 26.6 Wage and farming 12.8 26.4 2.7 Not employed 9.5 1.4 7.3 Total 100.0 100.0 100.0 Source: RIC Surveys. Note: a. For Sri Lanka, adults living in 149 households with zero or negative total incomes are excluded. Table 4.2 Average Total Household Income and Its Components Wage Enterprise Farm Remittance Other Total Nicaragua 988 434 228 230 136 2016 Sri Lanka 835 162 343 1088 268 2693 Tanzaniaa 81 261 203 n.a. 126 671 Source: RIC Survey Data. Note: a. Remittance incomes were not recorded by the Tanzanian pilot. Table 4.3 Measures of Inequality in per Capita Total Income Nicaragua Sri Lanka Tanzania Coef�cient of variation 2.347 2.942 3.468 Gini 0.568 0.769 0.757 Source: RIC Surveys. Note: Households with negative total income were omitted from these calculations. In Tanzania and Nicaragua, the sum of wage and Assets and Investments enterprise incomes—earned largely off-farm— Sri Lanka has the highest average total asset hold- constitutes the largest single source of household ing, followed by Nicaragua (Table 4.4). The average income.27 The relative importance of the several asset holding is lowest in Tanzania, where 43 per- income sources in total household income is por- cent of the country’s households own less than trayed schematically in Annex D, Figure D.2. US$500 in assets. In contrast, only 27 percent of In all three pilots, the distribution of rural per households in Nicaragua own less than US$1,000 capita incomes appears to be highly skewed, more worth of assets (Annex D, Table D.19), while so in Nicaragua. In Table 4.3 this is reflected both 47 percent own assets worth between US$1,000 in the large coef�cients of variations for household and US$4,000. In Sri Lanka only 11 percent of income estimates from the RIC surveys and in the households own US$1,000 in assets or less, while high Gini coef�cients. 32 The Rural Investment Climate Table 4.4 Cross-Pilot Comparison of Household Assets by Type of Asset (US$)a Other Household Agricultural Housing Durables Land Assets Nicaragua 3,162 78 1,869 2,292 Sri Lanka 4,400 1,098 2,900 244 Tanzania 1,683 17 667 96 Source: RIC Surveys. Note: a. excludes �nancial assets. 30 percent own between US$1,000 and US$4,000. Tanzania’s Lack of Sampling Weights Average holdings, by type of asset, are summa- In contrast to the databases for Nicaragua and Sri rized in the following table. Lanka, the Tanzania database lacks household Land remains a prominent asset as it strongly sampling weights. This is an unfortunate omis- influences rural households’ economic position. sion. In gathering RICS data, the sample of house- Table D.14 summarizes land holdings per house- holds operating enterprises is augmented with a hold. Table D.15 provides a distribution of house- random sample of households not operating holds by land value. The valuation of other enterprises. To save on costs, this latter sample agricultural assets extends to production-related constitutes a smaller proportion of the total sample tools and equipment, fruit bearing trees, and ani- than is the case in the overall population. The mal inventories. Table 4.4 above provides esti- tables above, containing weighted statistics on mates of such other agricultural assets. Although Nicaragua and Sri Lanka and unweighted statis- land may be the most prominent asset, housing tics on Tanzania, suggest that nonfarm self- remains the most valuable in all three countries. employment and income are more prevalent in Durable assets include durable consumables, Tanzania; the results should be carefully inter- such as electronics, furniture, transport equip- preted, however. Because of the sampling design, ment, and so on. Sri Lanka has the highest durable the rate of entrepreneurship is substantially higher asset base. in the RICS household sample than in the popula- As with per capita incomes, the value of the as- tion: in the Sri Lanka database, 57 percent of sam- set base (especially housing) is most skewed in ple households operate an enterprise, whereas the Nicaragua (see Table 4.5 and Annex D, Table D.19). rate of entrepreneurship in the rural population is Household investment is de�ned as the net addi- estimated to be 29 percent; the two Nicaragua tion to total assets (Annex D, Table D.21). An aver- percentages are 68 percent and 22 percent, respec- age household increased its assets by US$136 in Sri tively.28 By contrast, in the Tanzania sample, Lanka and by US$44 in Tanzania; large standard 62 percent of households derive income from an deviations indicate that household investment enterprise, a �gure probably so much higher than varies widely among households. Unfortunately, the population rate that any analysis of entrepre- the Nicaragua survey data do not permit an esti- neurship determinants likely suffers from a strong mate of household investment. Table 4.5 Measures of Inequality in Total Assets Nicaragua Sri Lanka Tanzania Coef�cient of variation 3.610 1.103 3.034 Gini 0.821 0.590 0.838 Source: RIC Surveys. Entrepreneurship Choice and Enterprise Start-Up 33 bias. Especially in an analysis of the entrepreneur- Regression Results ship determinants, sampling weights are impor- Entrepreneurship is a discrete 0/1 outcome and is tant in maintaining proper sample representation therefore analyzed with a probit regression model. of the population. Thus, whereas most of this Sampling weights are applied. The clustering of report examines data from the RICS in Nicaragua, households in communities makes it plausible that, Sri Lanka, and Tanzania in parallel, the balance of aside from observable community characteristics, this chapter drops the Tanzania data from the mul- unobservable community factors may be common tivariate analysis. to all households, causing correlation among the model’s disturbance terms. This renders the stan- 4.2 ENTREPRENEURSHIP AMONG dard probit model inadequate: a random effect RURAL HOUSEHOLDS must be incorporated. In cases where the estimated standard deviation of the random effect proved to De�nition of the Variables be exceedingly small, however, the estimation Precise de�nitions and descriptive statistics of all strategy was simpli�ed to a weighted probit model variables are provided in Annex F, Table F.1. In (without a random effect).30 Nicaragua, household entrepreneurship is de�ned Table F.2, Annex F, reports the estimated mod- as households deriving a positive income from an els. For Nicaragua, the random effect always van- enterprise. Because of an error in the Sri Lanka ishes in estimation. For Sri Lanka, a random effect questionnaire design, this de�nition would under- model is needed, since the standard deviation of state the incidence of entrepreneurship; thus the community random effect is estimated at about definition used for Sri Lanka is household mem- 0.35, one-third the standard deviation of house- bers working on their own account. hold-level random disturbance. Related to this, The explanatory variables are divided into three compared with Sri Lanka, the Nicaragua regres- sets (for more detail on expected effects on house- sions �nd more community variables with statisti- hold entrepreneurship, see 7Annex F). The first cally significant effects—similar to the regression pertains to characteristics of the household and results for enterprise performance discussed in includes demographics and ethnicity, evidence of Chapter 3. entrepreneurship by the household head’s par- The �rst column of Table F.2 contains only the ents, and access to finance (remittance income, six benchmark indicators; the second column adds other income, and household assets).29 the household characteristics; the third column The second set of variables describes the invest- fills in the remaining community variables. The ment climate in the household’s community as fourth column selects variables from among the summarized by the six benchmark indicators. set of benchmark indicators, their components, These indicators and their components are used in and other community variables making the largest the econometric analysis to measure the effect of contribution to explanations of entrepreneurship investment climate on entrepreneurship, with a choice. higher value expected to favor enterprise perfor- Since parameter estimates of the probit model mance. As mentioned above, however, a robust are tedious to interpret and the scales of the vari- investment climate may expand the nonfarm econ- ables differ, the table shows the percentage point omy while at the same time creating opportunities effect of a one-standard-deviation change in each more lucrative than household-owned and oper- explanatory variable, evaluated for an average ated enterprises, thus leading to enterprise clo- household. For explanatory dummy variables sures. In other words, the effect of investment such as “Head female,� “Head parents entre- climate on entrepreneurship may be ambiguous. preneur,� and “Not Sinhalese,� the change in the The third set of variables describes the commu- explanatory variable is one unit, which is clearly nity apart from immediate associations with the more appropriate. It is useful to keep in mind that policy-related investment climate: community the rate of entrepreneurship in Nicaragua is 22 per- size, level of per capita income, typical male wage- cent. Thus, the second column indicates that a rate, enterprise openness, and agricultural season- household with 0.99 additional male adults (the ality. As for the BIs, the a priori effect of these standard deviation of this variable; Table F.1) is variables on entrepreneurship is ambiguous. 3.08 percentage points more likely to operate an 34 The Rural Investment Climate enterprise. Also, when parents of the household distribution of assets.31 These computations show head were entrepreneurs, the probability that the that household assets are important determinants household operates an enterprise rises by 2.27 per- of entrepreneurship but also that establishing and centage points, although this effect is not estimated operating an enterprise demands large sums of precisely as the t-statistic equals only 0.86 (see resources.32 Table F.2, part A). Among the benchmark indicators, community In Nicaragua, entrepreneurship is more likely in human capital stands out in Nicaragua, where the households with more adults, whether male or standardized effect is about 5 percentage points. female, and in households with older members. Infrastructure services have a positive effect of Human capital does not matter; skilled workers about 5.5 percentage points as well, but with the may be more productive in the enterprise but, as introduction of other community variables the the literature on wage earnings shows, they are also effect diminishes to about 3. Business services capable of earning higher wages when working may reduce the likelihood of entrepreneurship, elsewhere. Female-headed households are 6 per- and connectivity, governance, and finance have centage points more likely to operate an enterprise. no effect. The last column of Table F.2, which con- In Sri Lanka, where 25 percent of households tains only those community variables with the operate an enterprise, the effect of household size, greatest explanatory power, lists positive effects age structure, and human capital is smaller and not of proximity to the post office and access to pro- statistically significant. Female headship has a tected water: for both, the effect is 3 percentage numerically strong negative effect, equal to 9 per- points per standard deviation increase in the centage points, but it, too, is not precisely enough benchmark component. estimated to attain statistical signi�cance. Members The Sri Lanka results list a positive effect for of the ethnic (non-Sinhalese) minorities tend not to connectivity, with the proximity to the post of�ce operate enterprises, although the effect is also not highlighted in this country as well; the effect of precisely estimated. Surprisingly, the occupation of 5 percentage points is quite large. The indexes the head’s parents proves to be irrelevant in both for infrastructure services,, business services, countries. Elsewhere, such as in Vietnam (see, for governance, human capital, and finance do not example, Vijverberg and Haughton 2004), the inter- appear to matter much. The last column high- generational effect is rather pronounced. lights sewage systems, which are found to have a In regard to household financial variables, the strongly negative effect and thus contribute to the speci�cation of the entrepreneurship model varies overall negative effect of the infrastructure services between the two countries owing to the bias in indicator; engineering services, with a negative measured income components referred to above effect; and information technology services, with a for Sri Lanka. In Nicaragua, access to remittance positive effect. The negative effects are counterintu- income and other income sources reduces the itive, but these variables are all simple 0/1 indica- incentives for self-employment. tors. As only 13 communities have sewage systems On the other hand, assets improve the chances. and merely 5 have engineering services available, Specifically, the standardized effect of log-assets the estimation results may simply be the luck of raises the probability of entrepreneurship by 5 per- the draw. centage points in Nicaragua and 10 percentage The last group of variables consists of various points in Sri Lanka. To interpret this for Nicaragua, community characteristics. Here, Nicaragua and where the average of log-assets amounts to Sri Lanka yield opposite results. In Nicaragua, US$1,419, a one-standard-deviation change in log- enterprise openness (that is, selling and buying assets equals US$6,485, which would take a house- outside the community) raises the likelihood of hold from, say, the twenty-�fth percentile to about entrepreneurship in a household by about 5 per- the eighty-fifth percentile of the distribution of centage points. The community income level may assets. For Sri Lanka, the 10 percentage point have a positive effect, but its effect is collinear with increase in the likelihood of entrepreneurship community human capital, the effect of which is comes about in response to a rise of US$11,621 in more robust and therefore enters the last column. assets, relative to a base of US$5,405; this is equiv- Agricultural seasonality, community size, and the alent to moving the household from the twenty- level of male wages do not matter. In Sri Lanka, on fifth percentile to the eightieth percentile of the the other hand, the likelihood of entrepreneurship Entrepreneurship Choice and Enterprise Start-Up 35 is higher in larger communities with higher wages, The top four lines of Table F.1, Annex F, list the although income per capita, seasonality, and enter- percentages of households in Nicaragua and Sri prise openness had no effect. Lanka participating in the various activities. The As mentioned before, Nicaragua’s model is rate of wage employment is surprisingly high, weighted probit without a random community given that these are rural samples; the percentage effect. In Sri Lanka, the random community effect of households that do any farming at all is only is statistically important. A one-standard devia- 27 percent in Nicaragua, which has large numbers tion in the random effect moves the likelihood of of farm laborers, and 48 percent in Sri Lanka.33 entrepreneurship in a household up by almost This raises the question of how rural these com- 10 percentage points in the model with benchmark munities are. indicators and 7 percentage points in the model In any case, just as households may choose to with selected benchmark components. engage in nonfarm entrepreneurship, they may To conclude this examination of household also choose to pursue wage employment or farm- entrepreneurship choice, it is worthwhile to reflect ing. Therefore, this section extends the analysis of on the explanatory power of the model. The evi- entrepreneurship choice to the other wage dence is summarized in Table F.3. The overall employment and farming choices households explanatory power of the model with household make. Moreover, as mentioned above, since dur- variables, benchmark indicators, and community ing the course of economic development people characteristics (Panel A) is 12.42 percent in may switch from self-employed farming into rural Nicaragua and 7.85 percent in Sri Lanka. House- nonfarm self-employment and wage employment, hold characteristics explain (that is, improve the the percentage of households that pursue wage criterion function by) 8.09 percent in Nicaragua employment or nonfarm entrepreneurship is of and 4.58 percent in Sri Lanka, where, it should be interest: it integrates the development of the non- noted, the model does not include remittance and farm economy into a single concept. other income, which are influential variables in Estimation results are reported in Table F.4, Nicaragua. The benchmark indicators help to showing only the speci�cations with selected com- explain 2.98 percent in Nicaragua and 1.46 percent ponents. Each equation is estimated separately in Sri Lanka, and community characteristics with a random effect probit model.34 The �rst col- further add about half as much. In all, bench- umn refers to engagement in nonfarm enterprises mark indicators and community characteristics and is therefore identical to the fourth column of contribute 35 percent ( (2.98 1.35) 12.42) of Table F.2. the explanation of entrepreneurship choice in In Nicaragua, households with more men diver- Nicaragua and 28 percent ( (1.46 0.72) 7.85) in sify into more activities; those with more women Sri Lanka. Further checks confirmed that bench- pursue only nonfarm entrepreneurship. In Sri mark indicators carry relatively less information in Lanka, diversification is stronger yet when there Sri Lanka than in Nicaragua. are more adult males; specialization is stronger when there are more adult women. The age pattern is similar in the two countries: households with older members are more likely to farm and perhaps 4.3 ACTIVITIES OF HOUSEHOLDS operate an enterprise; younger households seek Typically, two-thirds of households with enter- wage employment. In Nicaragua younger house- prises are also involved in other activities; even holds also engage more in entrepreneurship. among individuals, many engage in several modes Human capital draws households into wage employ- of employment over the course of a twelve-month ment, perhaps out of nonfarm entrepreneurship period (see Table 4.1). This parallels findings in and likely out of farming. The entrepreneurial expe- Cote d’Ivoire, Ghana, Guatemala, Kyrgyz Repub- rience of the head’s parents, however, leaves no lic, Peru, and Vietnam (Moock, Musgrove, and sorting imprint on the head’s household. Stelcner 1990; Lanjouw 2001; Vijverberg 1990, 1998, Nicaragua and Sri Lanka differ in the sorting 1999, 2005). Even if some activities—wage, enter- patterns according to the household head’s gender. prise, or farm—are rather insigni�cant, earning less Female-headed households in Nicaragua are more than $60 per year, the level of participation demon- likely to operate an enterprise, whereas those in strates that employment diversity is the norm. Sri Lanka are significantly more likely to have 36 The Rural Investment Climate wage jobs rather than a household enterprise. In remote the community, the more likely house- both countries, however, farming is less likely for holds will be to earn their living by farming, female-headed households; the effect is large but although the estimates are not precise enough to imprecisely estimated. In Sri Lanka, non-Sinhalese imply statistical signi�cance. Finally, the commu- are less likely to participate in the nonfarm econ- nity random effect may be absent in the determi- omy. In particular, they are found in farming nation of entrepreneurship choice, but it is highly rather than in nonfarm businesses. signi�cant in households’ choice of wage employ- Extraneous income sources reduce the incentive ment and farming: the standardized effect ranges for households to participate not only in a busi- from 15 to 19 percentage points. ness, as seen above, but actually in any kind of In Sri Lanka, the list of community variables work. This is consistent with the notion that, in and benchmark components is long and varied. standard labor economics terms, leisure is a nor- Households in larger communities are more likely mal good. Assets strongly sort households into to work for a wage and less likely to farm. Greater both enterprises and farming; households with seasonality is associated with farming choice. few assets seek wage employment. Turning to benchmark components, chances of The third column of Table F.4 focuses on non- wage employment seem higher in communities farm economic activity, combining (both farm and farther from a major city or closer to a main mar- nonfarm) wage employment and nonfarm self- ket. Overall access to infrastructure helps; more employment. For some explanatory variables, the detailed speci�cations show positive effects for the third column sums up contradictory signals from percentage of households with electricity and the �rst and second columns of the table; for exam- the presence of concrete or asphalt roads and a ple, see the effect of age, female headship, ethnic- negative estimate for the percentage of households ity, or asset ownership. Among the strongest with cell phones. Furthermore, according to the results, human capital pulls people into nonfarm estimates, wage employment is more likely with economic activity. In Sri Lanka, this is supple- the presence of marketing services available to mented by a positive effect for adult women and businesses in the community and less likely with negative effects for age and non-Sinhalese ethnic- higher “�nancial penetration,� that is, the percent- ity. But balancing these insights is an equally valid age of entrepreneurs in the community that have inference that an exclusive study of nonfarm eco- considered asking a �nancial institution for a loan nomic activity hides interesting sorting patterns in in years prior to the RIC survey. Farming is more the economy among wage and self-employment. prevalent in communities with poorer access to Searching for the best explanation through electricity or phone service. As the last variable, community variables, benchmark indicators, and �nancial penetration has a positive effect, as farm- benchmark components yields various results. In ing may also bene�t from access to credit.36 Across Nicaraguan communities with higher incomes, all outcome variables in Table F.4B, the random households are more likely to engage in wage community effect is influential. In particular, the employment and less likely to be found in self- prevalence of farming varies greatly from commu- employed farming, and, when seasonality is more nity to community in ways not captured in the pronounced, farming becomes more likely and observable information about communities. This wage employment less. Wage employment dimin- is consistent with the notion that farming, more ishes when more households have access to elec- than other enterprises, depends more on land and tricity or when government agencies are more water availability in the location than on invest- transparent in their dealings with entrepreneurs ment climate variables. in the community: these effects are counterintu- itive, since one would expect the demand for wage labor to flourish under those conditions.35 The 4.4 ENTERPRISE START-UP weaker negative effect of human capital in the community should be seen in the light of the pos- Measuring Start-Up itive private effect of human capital: a compara- This section examines the effect of investment cli- tive advantage in skills disappears when other mate on enterprise start-up. Households are con- members in the community are skilled as well. For sidered to have started up an enterprise if at the farming, the data suggest that the more rural and time of the survey the enterprise they operate is Entrepreneurship Choice and Enterprise Start-Up 37 less than two years old and is based on the estimates, in Nicaragua only households receiving premises of the household’s residence. This under- income from remittances and other sources are less states the true two-year start-up rate, since these likely to start up a new enterprise. In Sri Lanka, enterprises must have survived for up to two years assets had a positive effect; the number of female to be included; the survey cannot count recently adults and the entrepreneurial experience of the started but already defunct enterprises or enter- head’s parents may raise the likelihood of start-up; prises started as or converted into stand-alone chances are lower among older or non-Sinhalese units. households. In Sri Lanka, therefore, where many enterprises Among benchmark indicators, connectivity are stand-alones, the measured start-up rate is only hurts entrepreneurship chances in Nicaragua but 1.2 percent, compared to an estimated true start- has no effect in Sri Lanka. Proximity to a major city up rate that might be as high as 6.05 percent (see has a negative effect in Nicaragua but a positive one Annex F, Box F.1). In Nicaragua, the situation is in Sri Lanka. Infrastructure services has a positive quite different: the enterprise start-up rate is mea- effect in Nicaragua but a negative one in Sri Lanka. sured to be 3.0 percent, probably a fairly accurate Community human capital raises chances in Sri number, because stand-alone enterprises are only Lanka. The remaining indicators do not matter. 7.5 percent of the sample.37 Other community characteristics show a similar If it is assumed that the enterprise population is disagreement between the two countries. Enter- stationary, such that new start-ups merely replace prise start-up is more likely in Nicaragua in larger exiting enterprises, enterprise start-up informa- communities where enterprises are accustomed to tion may be translated into enterprise survival trading with clients outside the community and estimates. If so, the annual enterprise survival rate38 where agricultural seasonality is more pro- among household-based and stand-alone enter- nounced. Of these three variables in Sri Lanka, prises in Sri Lanka equals 86.6 percent and 87.2 per- enterprise openness has an opposite effect, and the cent, respectively, and in Nicaragua, where it is other two do not matter. assumed that the data captured only half the Overall, therefore, the effect of investment start-ups in 2004, 90.7 percent and 84.7 percent. climate varies considerably between the two By comparison, the enterprise survival rate is countries. It is possible that behavior differs; about 83 percent in Vietnam and similar elsewhere as mentioned above, the effect of household (Vijverberg 2005). and community variables is a priori ambiguous. Alternatively, these may simply be random pat- terns that appear when communities are few and Regression Results variation in the dependent variable is limited. Table F.5, Annex F, examines enterprise start-up Moreover, one should expect differences between determinants. These are estimates of weighted the two countries since Sri Lanka has a much probit models, expressing the effect of explanatory greater proportion of stand-alone entities: the variables, again in standardized form. The com- analysis concerns start-up of household-based munity random effect vanished in both countries enterprises only, omitting start-up of stand-alone as its estimated standard deviation converged to 0. enterprises. Moreover, because the number of start-ups is so small, the model could not be populated with a full set of benchmarks and community characteris- tics. This very fact implies that estimates of com- 4.5 GENERAL DISCUSSION munity variable effects should be interpreted Having completed this empirical analysis, it is use- carefully. ful to ask whether a study such as this accom- Actually, few explanatory variables matter. plishes important objectives. What light does it Moreover, as the average start-up rate is only 3.0 shed on entrepreneurship choice? Is the research percent in Nicaragua and 1.2 percent in Sri Lanka, vehicle—the econometric model—appropriate? Is the estimated effects are small, certainly in com- the investment climate properly captured in the parison with Table F.2, where the average rate of analysis? Questions of whether the data are appro- entrepreneurship was 22 percent and 25 percent priate for the objective will be addressed in the respectively. Among the statistically significant following section. 38 The Rural Investment Climate Entrepreneurship Choice An alternative approach to the simple 0/1 mod- This chapter looks at only one facet of entrepre- eling strategy would be to describe how many of neurship choice: whether to set up a business. Entre- the household members and even how much of preneurs make other choices: which sector to enter, their time is devoted to each economic activity. how to arrange for �nancing, which technology to This converts the entrepreneurship decision into a use, where to locate, and how to seek out clients, to continuous variable with a lower bound at 0. It also mention a few. In this sense, this chapter’s scope is extracts more information from the RICS data, limited. The RICS data, however, do offer opportu- placing more demands on their collection. nities for further analysis to better understand the Application of the random effect probit model is full dimension of entrepreneurship choice. complicated by the strati�cation of the household Entrepreneurship choice must be placed in con- sample and the need to use sampling weights. A text. If farming is more profitable than anything typical RICS database draws more heavily from else, rational people will operate a farm. If a wage households with enterprises than from households job pays better, many people will elect to work for without them, whereas the population consists of an employer. If one’s own enterprise is the most more households without enterprises than with pro�table option, it makes sense to start a business. them. Thus, an unweighted analysis of economic Income may not be the only consideration in activities yields biased estimates. Stata software, this comparison, as each option has its own however, does not allow sampling weights to be nonmonetary compensation and imposes its own speci�ed for the panel data procedures used when income risk. Households may pursue diversifica- community random or fixed effects enter the tion strategies. In all, the results obtained in this model. In experimentation with different estima- chapter illustrate how a focus on the 0/1 outcome of tion methods, estimates sometimes varied substan- enterprise operation fails to uncover a rich sorting tially among weighted probit, unweighted random pattern across the different economic activities. effect probit, and weighted random effect probit. This chapter also illustrates that adopting a Estimation is further complicated by potential dichotomy of farm versus nonfarm economic endogeneity of explanatory variables, such as activities, where the latter aggregates wage and income levels in the community and the house- nonfarm self-employment, can lead to an inade- hold’s assets and nonlabor income. The analysis quate insight into the process of development. above failed to disprove the exogeneity of the �rst Self-employment sorting patterns are not the same two variables and did not examine the potential as those for wage employment. At the same time, endogeneity of nonlabor income. But more the RICS data examined here do not distinguish research is needed to successfully incorporate wage employment by sector: in the future, it endogenous explanatory variables. would be useful to separate agricultural and nona- gricultural laborers. Capturing the Community’s Investment Climate Econometric Modeling The estimates in this chapter show that the com- Participation in farming, wage employment, and a munity random effect is frequently important. Its nonfarm enterprise was modeled with three sepa- standardized effect is usually as large as, and rate regression equations, each estimated indepen- sometimes much larger than, any of the observed dently. This actually hides the interdependence. explanatory variables. This shows that community Greater efficiency is feasible by simultaneously factors matter. The question is whether the survey estimating the three equations, but this comes at a data have captured them in meaningful ways. cost of complexity. The structure of the data has Yet, in Nicaragua’s entrepreneurship choice households clustered in communities. In conse- model and in both countries’ enterprise start-up quence, the model must allow for unobservable model, the estimated standard deviation of the community factors (that is, random effects). Esti- community random effect was zero, and the ran- mation of a single random effect probit equation dom effect vanished. As Table F.3, Annex F, shows is already a more laborious procedure than the that in Sri Lanka community benchmark indica- simple, common probit model; simultaneously esti- tors and community characteristics contributed mating three equations will be time consuming. substantial explanation after allowance was made Entrepreneurship Choice and Enterprise Start-Up 39 for a community random effect. Nevertheless, the Implications for RIC Methodology effect of benchmark indicators is mostly muted The data from Nicaragua, Sri Lanka, and Tanzania and varies between the two countries. were gathered during the pilot phase of the RIC The investment climate is a multidimensional project. The analysis in this chapter uncovers at concept. Defining benchmarks does not prove least four important lessons for the RIC methodol- their usefulness, and it is as yet an empirical ques- ogy. First, sampling weights must be calculated tion whether the investment climate actually mat- when data are collected: since the sampling strat- ters. Theory may suggest contrary directions of egy includes clustering by community and strati�- influence. Connectivity, for example, opens up dis- cation by entrepreneurship status, implementation tant markets for local entrepreneurs and local mar- of the survey implies that sampling weights must kets for distant entrepreneurs. Better governance be carefully constructed and constitute an essen- may be good both for small-scale entrepreneurs tial element of the survey data. and for large-scale employers that hire many locals Second, in a survey database that includes both for a wage. But a number of benchmark indicators household and enterprise data, one would expect clearly relate to the likelihood of entrepreneurship, that entrepreneurship in a household would be whether positively or negatively, which may be de�ned by (i) whether a linked enterprise was pre- viewed as evidence that benchmarks matter. sent, (ii) whether any members of the household The aggregative nature of benchmarks may also were employed on their own accounts, and be their weakness. If only one component matters, (iii) whether the household earned income from its effect is weighed down by its pairing with a work on its own account—and that these three number of other ineffective components. It is aspects would be mutually consistent in the data- therefore advisable to search among benchmark base. The three countries’ databases exhibited components for variables that provide strong too many inconsistencies. In future RIC surveys, explanatory power. The drawback is that there are inconsistencies should be flagged and resolved so many components. Each is listed as a candidate while teams of enumerators remain in the �eld. variable because it reflects an aspect of the invest- Third, the investigation of enterprise start-up ment climate and might, on economic grounds, among rural households was greatly hindered by matter for entrepreneurship, enterprise start-up, the design of the RIC survey as implemented in or enterprise performance. The eventual selection Sri Lanka and Nicaragua, which treated stand- is the result of, essentially, a statistical rather than alone enterprises as entities without a household an economic comparison, that is, a stepwise regres- connection, whereas the data (particularly in sion procedure. The selection is then vulnerable to Sri Lanka) show that stand-alone enterprises are accidental statistical signi�cance, as probably illus- under household ownership. As a result, enter- trated by the effect of management consulting and prise start-up measures refer only to household- marketing services on farming in Sri Lanka (Table based enterprises, thus weakening the analysis. F.4B). In retrospect, some aspects of the investment The survey design should never view an enter- climate in play in only a few (say, less than 10) prise separately from its owner but should, rather, communities should perhaps have been discarded interview the associated household whenever from this selection procedure. possible. Nevertheless, experimentation has shown that a Finally, the households in the research samples few well-selected components and community reside in 93 communities in Nicaragua and in 117 characteristics can provide equal or better explana- communities in Sri Lanka. Community-level vari- tory power than a full list of benchmark indicators ables can explain only the between-community and community characteristics. Proximity to the variation in the dependent variable. This means post of�ce, for example, matters for entrepreneur- that the investigation of investment climate effects ship choice in both countries. Access to protected works with roughly 100 data points. Yet, the invest- water raises chances in Nicaragua, as do informa- ment climate has many dimensions: the six bench- tion technology services in Sri Lanka. Harder to mark indicators alone consist of 30 components. explain are the negative effects of the presence of a This makes reliable detection of investment climate sewage system and the availability of engineering effects quite challenging. services in Sri Lanka. 5 Perceptions About Investment Climate Constraints An important objective of the Rural Investment Climate Survey program is to give entrepreneurs the opportunity to voice their interests and concerns regarding the investment climate. How do they view the environment in which they operate their businesses? How do these perceptions vary by sector and by enterprise type? Does the perception correspond to the actual conditions in the community, as assessed by objective measures? Questions of per- ception are relevant, since it may be argued that entrepreneurs base their decisions on their perceptions. They are more likely to invest when they view the climate as favorable; they are more likely to hire workers if they anticipate profitability for their operations. The RIC questionnaire lists features ranging from availability and cost of electricity to agricultural issues to concerns about crime and civil unrest: for each of these features, respondents are asked whether it constitutes a barrier to the operation and growth of their enterprise. In essence, these responses supply subjective measures of the investment climate: they reflect perceptions colored by the respondent’s frame of reference and, presumably, by actual conditions in the community and the region. These variables are dubbed Enterprise Investment Climate Outcomes (EICOs). Apart from these questions, entrepreneur respondents are asked about their interactions with government agencies and their views on the legal system as it relates to operation of an enterprise. Some of the responses are objective, such as the number of inspections or the time spent dealing with government of�cials; others are subjective, such as level of trust. Entrepreneurs do not rate these conditions as barriers; the questionnaire makes no attempt to link these facets of the investment climate to enterprise operation and growth. Rather, whether essentially objective or subjective, factual responses are sought. These variables are dubbed Enterprise Investment Climate Interactions (EICIs). This chapter examines determinants of EICO and EICI percep- tions. Section 5.1examines which investment climate elements rank high as barriers on the lists of entrepreneurs’ concerns; Section 5.2 describes the model; and Sections 5.3 and 5.4 present estimates of regression models for EICO and EICI variables. Section 5.5 offers suggestions for further research, and Section 5.6 discusses implica- tions for the RIC methodology. 41 42 The Rural Investment Climate 5.1 ENTREPRENEURS’ is, listed among the top ten) are unique to that country. INVESTMENT CLIMATE Overall, finance is viewed as the most impor- CONCERNS tant barrier: high interest rates, difficult loan procedures, and problems �nding lenders rank 1, 5.1.1 EICOs 2, and 4, respectively. Electricity ranks third to The questionnaires list nearly 40 investment �fth in each country and third overall. Issues with climate features entrepreneurs may indicate as road quality, road access, and water supply rank barriers to the operation and growth of their enter- 5, 7, and 8, respectively, and appear on each prises. This list is wide-ranging and may be country’s list. Opinions about market demand, divided into the following topics: economic policy uncertainty, and corruption vary sharply across countries, ranking near the top • Public utilities in one country but being rather unimportant in • Transportation another. Telecommunication is ranked more • Financing evenly and thus makes the top-ten list even • Marketing though no country ranks it higher than seventh. • Registration, license, and permits Because of inherent interest, Tables H.8 and H.10 • Taxation elaborate on the measurement of the corruption • Labor EICO. • Land This brief summary indicates both the breadth • Agricultural policy of topics covered in the questionnaire and their • Nonagricultural trade policy lack of detail. Concerns about “electricity,� for • Environmental policy example, can take many forms: obtaining a hook- • Governance39 up, experiencing frequent blackouts, high tariffs, Across the three pilots, the detailed questions and the like. “Water supply� and “telecommunica- were broadly similar; any differences are inconse- tion� are equally vague. But for this study, detailed quential for a comparative analysis.40 In Sri Lanka inquiry into every EICO dimension was simply not and Tanzania, the responses are coded as 0 No feasible. obstacle to 4 Very severe obstacle, whereas in Nicaragua the variation in the responses is limited 5.1.2 EICIs to 0 No obstacle, 1 Moderate obstacle, and 2 Severe obstacle. This hinders direct numerical com- The RIC survey instruments contained questions parison between countries, but it does not affect on numerous aspects of respondents’ business qualitative comparison of the types of barriers experience: applications for and renewals of entrepreneurs highlight most prominently. licenses and permits; time and monetary outlays Figure 5.1 ranks the more important EICOs, (both formal and illicit) when dealing with gov- treating the categorical responses as true ordinal ernment agencies; utilities issues; efficiency of numbers; Table H.1, Annex H, reports detailed government services, including whether govern- frequencies by category for all variables. In ment of�cials can be influenced regarding enforc- Nicaragua and Sri Lanka, entrepreneurs highlight ing and drafting laws and regulations; and the some 10 to 12 barriers; the other EICOs fail to elicit legal system and conflict resolution methods. much of a response. In Tanzania, concerns are Disentangling the many patterns in all this infor- more wide-ranging and, compared to Sri Lanka, mation is beyond the scope of this report, but it is which uses the same coding scheme, more instructive to examine a few of the more accessible intense.41 and comparable items. To get a better sense of the barriers generally Table 5.2 pulls together three elements of gov- important in rural areas, Table 5.1 lists the top ernment efficiency: the predictability of laws and ten EICOs in each country as well as the top ten regulations, the willful or unintentional misinter- overall for these countries on the basis of their pretation of laws and regulations, and influence average ranks. Six of the ten most frequently cited on government officials. Many entrepreneurs EICO barriers are common across Nicaragua, Sri feel insecure about laws and regulations: often, Lanka, and Tanzania. On the individual country enforcement of these is perceived to reflect lists, no more than two important concerns (that government officials’ whims. With regard to Figure 5.1 Enterprise Investment Climate Outcomes A: Nicaragua Econ policy uncertainty 0.90 Interest rate of loan 0.76 Access to loans 0.72 Corruption 0.66 Electricity 0.60 Loan process 0.53 Market demand 0.45 Crime, theft, etc. 0.37 Road quality 0.33 Water 0.33 Road block 0.25 Access to trans facility 0.18 Tax rate 0.16 Forms of transport 0.14 Market information 0.09 Legal system 0.09 Telecommunication 0.07 Labor hiring 0.07 Bureacracy in tax collection 0.06 0.0 0.2 0.4 0.6 0.8 1.0 B: Sri Lanka Interest rate of loan 1.25 Market demand 1.24 Electricity 1.23 Loan procedures 1.15 Road quality 0.86 Road access 0.79 Market information 0.74 Access to transport facility 0.70 Water 0.66 Telecommunication 0.54 Availability of loan sources 0.41 Road block 0.25 Econ policy uncertainty 0.17 Availability of skillful labor 0.17 Postal service 0.15 Tax rate 0.13 Cost of license/permit 0.12 Cost of registration 0.11 Registration/license process 0.09 Environmental rules 0.08 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 C: Tanzania Loan proc. 1.84 Availability of loan sources 1.80 Interest rate of loan 1.79 Electricity access 1.46 Electricity quality 1.17 Water 1.12 Telecommunication 1.04 Road quality 0.96 Postal service 0.91 Road access 0.89 Market access 0.89 Corruption 0.85 Market information 0.82 Market demand 0.79 Debree of access to trans facility 0.77 Crime, theft, etc. 0.71 Econ policy uncertainty 0.62 Tax rate 0.60 Cost of license/permit 0.58 Registration/license process 0.57 Cost of registration 0.55 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 Source: RIC Surveys. 44 The Rural Investment Climate Table 5.1 Ranking EICOs Across Countries Average Rank Nicaragua Sri Lanka Tanzania Overall Rankd 1 economic policy interest rate of loan loan procedures interest rate of loan uncertainty 2.00 2 interest rate market demand availability of loan procedures of loan loan sources 3.67 3 access to loansa electricity interest rate of loan electricityb 4.17 4 Corruption loan procedures electricity access availability of loan sources 5.33 5 Electricity road quality electricity quality road quality 7.33 6 loan process road access water market demand 7.67 7 market demand market information telecommunication road accessc 8.00 8 crime, theft, etc. access to transportation road quality water facility 8.33 9 road quality water postal service economic policy uncertainty 10.33 10 Water telecommunication road access telecommunication 11.33 Source: RIC Surveys. Notes: Items in italics do not appear in the overall list. Bold-faced items occur in the top-ten list for each country. a. Interpreted to be similar to “availability of loan sources.� b. In Tanzania, electricity concerns distinguish between access and quality, which are given equal weights in computing the overall rank. c. Measured in Sri Lanka and Tanzania only. d. Simple average of the country-speci�c rankings. influencing government of�cials in Nicaragua, the the responses varied a little across countries: question referred to the effect of informal pay- wheres Sri Lanka and Tanzania used codes rang- ments by the entrepreneur or others,42 whereas in ing from Strongly agree to Strongly disagree, the Sri Lanka and Tanzania respondents were asked Nicaraguan questionnaire allowed only Agree, Nei- whether they or other entrepreneurs could influ- ther agree nor disagree, and Disagree. Relative inten- ence government officials. In Nicaragua a similar sities are expressed in the table formatting. follow-up question probed the degree of influence, Moreover, the numerical summary uses values but responses were few.43 In any case, about 15 per- of 0/1/2/3 for Sri Lanka and 0.5/1.5/2.5 for cent of Nicaraguan entrepreneurs consider them- Nicaragua. With this proviso, the countries’ selves affected—whether favorably or not—a responses may be compared. considerably higher response than in Sri Lanka or A need to rely on reputation illustrates an Tanzania, where bribes are considered effective incomplete market mechanism: a low average regarding the enforcement of laws and regulations score indicates a more problematic investment cli- but not their drafting. mate. The other two items, protection by means of Entrepreneurial perspectives on the legal sys- contracts and support from the legal system, reveal tem are summarized in three measures: whether a better-developed market mechanism: in this the entrepreneur must rely on the reputation of case, higher scores are evidence of a poorer invest- those with whom they enter into agreements; ment climate. whether entrepreneurs expect a contract to protect One might expect the need to rely on reputa- them from being cheated by others; and whether tion to be lessened by contracts and legal system they expect that the legal system will uphold their support and thus, to the degree that correlation contract and property rights in business disputes. coefficients of these categorical responses are Table 5.3 summarizes the responses. The coding of appropriate, that correlations would be negative. Perceptions About Investment Climate Constraints 45 Table 5.2 Investment Climate Evidenced in Government Ef�ciency A: Predictability of laws and regulations Completely Highly Fairly Fairly Highly Completely predictable predictable predictable unpredictable unpredictable unpredictable Total N Nicaragua 55.3 44.7 100 1139 Sri Lanka 6.5 27.6 35.6 20.9 9.4 100 1324 Tanzania 19.1 16.1 20.7 23.2 11.5 9.4 100 747 B: Laws and regulations can be misinterpreted or manipulated Strongly agree Agree Disagree Strongly disagree Total N Sri Lanka 9.7 33.0 44.6 12.7 100 730 Tanzania 18.6 24.9 44.1 12.5 100 1051 C1: Your establishment or others can influence the contents of laws and regulations No Yes Total N Sri Lanka 90.4 9.6 100 432 Tanzania 88.4 11.6 100 1211 C2: Your establishment or others can influence of�cials in drafting of laws, decrees, regulations, etc. No Yes Total N Sri Lanka 97.9 2.1 100 472 Tanzania 93.1 6.9 100 1216 C3: Nicaragua: Impact on your business from kickbacks No impact Little impact Lot of impact Total N . . . laws 84.4 9.7 5.9 100 904 . . . enforcement 83.8 10.0 6.2 100 904 . . . judge 84.8 9.8 5.4 100 894 . . . credit 87.4 9.0 3.7 100 880 . . . politician 86.1 10.0 3.9 100 875 Source: RIC Surveys. Note: Arrows show contractions in the response coding that led to combining cells. The correlation between reliance on reputation 5.2 MODELING ISSUES on the one hand and contract protection and legal system support on the other is, however, EICOs and EICIs express the investment climate as around 0.6 in Sri Lanka, 0.5 in Tanzania, and seen by the entrepreneur. Perceptions are founded 0.55 in Nicaragua. In comparison, the correlation on the actual environment, on the entrepreneur’s between contract protection and legal system frame of reference, and on encounters occasioned support is 0.4 in Tanzania and 0.74 in Sri Lanka. by business operations. To elaborate, the economic, It is difficult to reconcile thes strong positive social, and political environment in which the correlations with the implications for the mea- enterprise operates should help determine percep- sured RIC. tions: economists often assume that entrepreneurs 46 The Rural Investment Climate Table 5.3 Investment Climate Evidenced in the Legal System Strongly Strongly agree Agree Disagree disagree Total N Mean A: Sri Lanka Need to rely on reputation 38.6 54.7 5.9 0.8 100 1021 0.69 Contract will protect 28.3 58.1 11.1 2.5 100 953 0.88 Legal system will support 26.6 56.0 15.2 2.3 100 829 0.93 B: Tanzania Need to rely on reputation 32.5 44.2 16.8 6.5 100 1194 1.97 Contract will protect 29.9 51.1 12.9 6.1 100 1193 1.95 Legal system will support 22.2 51.6 18.3 7.9 100 1195 2.12 C: Nicaragua Need to rely on reputation \\_ 54.0 _//\\_ 19.5 _//\\_ 26.5 _// 100 1110 1.22 Contract will protect 52.9 18.6 28.5 100 1111 1.26 Legal system will support 50.0 21.0 29.1 100 1091 1.29 Source: RIC Surveys. have perfect information in this regard, but reality Chapter 3, was not attempted since, with ten may fall well short of that. Experience, education, EICOs under study in each country, it would be a and network relationships build a mindset about Herculean task. The descriptive statistics of the the external world. Business operations may bring explanatory variables are provided in Table H.2, the entrepreneur face to face with regulators and Annex H. government officials. One type of business (for EICO responses are discrete categories on an example, transport) may encounter barriers that integer scale from 0 to 4 for Sri Lanka and Tanzania other types (for example, personal services) never and from 0 to 2 for Nicaragua; EICIs are measured deal with. The econometric model developed on various scales, including 0/1. This calls for esti- below attempts to capture some of these features mation with an ordered probit model (or probit if as an explanation of the variations in entrepre- the outcome is 0/1). Since enterprises cluster in com- neurial perceptions of the investment climate. munities, responses are determined by (observed) This enterprise-level information is comple- community-level information and by community- mented by community-level information. Much of level unobservables representing unmeasured and the latter reflects objective, factual conditions, unmeasurable community characteristics, that is, by information on which was gathered from knowl- a community-level random effect. edgeable community respondents including com- Annex I outlines the econometric details of the munity leaders, teachers, business leaders, and estimation of the weighted random effect ordered local bank officials. This information was used in probit model. Annex H contains a further note two ways. First, one set of models employs �ve of about the computation of standard errors of the the six benchmark indicators.44 This approach is estimate. In order to facilitate the interpretation of compact, as benchmark indicators combine a host the results, the parameter estimates are converted of community characteristics, but its drawback is to reflect the effect of a one-standard-deviation that benchmark indicators sometimes mask inter- change of a continuous variable46 or a one-unit esting detail.45 For this reason, a second set of mod- (0 to 1) change of a dummy variable on a scale that els contains the selected benchmark indicator corresponds with the original discrete coding. components that are a priori the most relevant Technically, the ordered probit model estimates a community characteristics for the analysis of a “tendency to report obstacles� equation that is given EICO. This second model therefore varies subject to an arbitrary scaling that disconnects the between EICOs. A search for the best predictive parameter estimates from the coding of the EICOs model, as was done for enterprise performance in and renders comparison between different EICOs Perceptions About Investment Climate Constraints 47 and different countries difficult. For the purpose A summary of the results is provided in Table 5.4 of this report, the average tendency of a typical in regard to enterprise characteristics and Table 5.5 enterprise reporting no obstacles is set at 0, and with respect to community characteristics. Con- the average tendency of a typical enterprise sider first the effect of enterprise characteristics. reporting very severe obstacles is set at 4 (or 2 in Although Tanzanian entrepreneurs raised more the case of Nicaragua). These averages take into complaints (Figure 5.1), the nature of the enterprise account both the randomness in the responses at or the type of entrepreneur rarely relates to the the enterprise level and also the community-level EICO value. In Nicaragua, entrepreneurs operating random effect. household-based enterprises perceive fewer obsta- This scaling facilitates interpretation of the esti- cles than do their peers with stand-alone enter- mation results. The effect of connectivity on elec- prises. In Sri Lanka, Sinhalese managers are more tricity in Nicaragua, for example, is reported to be vocal about obstacles, especially in the area of 0.194: this implies that a one-standard-deviation �nance. Female entrepreneurs complain less about increase in connectivity—an improvement in the electricity and water, and they also see fewer prob- investment climate—reduces the tendency to view lems in the cost of loans. Substantial differences electricity as an obstacle by 0.194 on a 2-point emerge among industries, with service enterprises scale.47 Consider this dummy variable example: in registering more obstacles related to electricity and Sri Lanka the effect on “electricity as an obstacle� water compared to traders and fewer in most other of being a female manager is reported to be 0.361: aspects. Younger and household-based enterprises on a 4-point scale, by 0.361 points female entrepre- view road quality and access as problematic; older neurs view electricity as less of an obstacle than do enterprises may have learned to cope or may have their male counterparts. survived despite this barrier. Expressing all EICO results on the same scale Across countries, Table 5.4 reveals few similari- also facilitates comparison across EICOs. Esti- ties. Mixed enterprises50 register more obstacles mates for Tanzania, for example, indicate that related to electricity and water in both Nicaragua improving connectivity actually matters little for and Sri Lanka, and household-based enterprises the problems entrepreneurs report having with have less trouble with electricity, but that sums up road access ( 0.017) or road quality ( 0.150), but the degree of similarity among significant enter- it helps in locating sources of financial assistance prise characteristics. Inspection of Table H.3 (availability of loan sources) ( 0.358).48 Similarly, reveals additional patterns, even if the individual manufacturers in Sri Lanka find electricity to be estimates are not statistically significant. House- much more problematic (0.989) than do traders, hold-based enterprises experience fewer obstacles but the gap is much smaller in regard to costs of with the loan cost, procedures, and availability as �nance (0.321) or loan procedures (0.231).49 well as with electricity and water; the effect on other EICOs is mixed. Female entrepreneurs in Sri Lanka mention obstacles with loan cost, proce- dures, and availability less often, but the effect for 5.3 ESTIMATION RESULTS: EICOS Tanzanian women is mildly positive. The effect of enterprise age is generally minor. For brevity’s 5.3.1 Facets of the Investment Climate sake, Table 5.4 skips a small number of EICOs Estimation results from the analysis with Enterprise unique to each country. Investment Climate Outcomes are presented in Not included in the list of enterprise characteris- Table H.3 for the model with benchmark indicators tics are measures of size and productivity. There- and in Table H.4 for the model with selected bench- fore, the estimated models represent reduced form mark components. Table H.3 groups EICOs by equations. The estimated effect of, say, female country; Table H.4 tabulates results across countries entrepreneurship on the EICO response is a combi- by EICO. Both comparisons offer useful informa- nation of a direct effect of gender on the EICO and tion. For example, Table H.3 permits quick insight an indirect effect through size and productivity. into the nature of obstacles perceived by female For further discussion on this issue, see 7Annex H. entrepreneurs or household-based enterprises, It is easy to argue that size matters. A large enter- whereas Table H.4 helps establish whether patterns prise, for example, may be severely disrupted by for a given EICO hold across countries. poor-quality electricity delivery; it may have an 48 The Rural Investment Climate Table 5.4 Signi�cant Enterprise Characteristics in EICO Equationsa EICO variable Nicaragua Sri Lanka Tanzania Interest rate of loan Sinhalese manager Mixed enterprise Manufacturing, non- agricultrura; enterprise Household-based enterprise Services enterprise Female manager Loan procedures Sinhalese manager Female manager Household-based enterprise Services enterprise Electricity Mixed enterprise Services enterprise Manufacturing, nonagricultural enterprise Agricultural processing enterprise Mixed enterprise Education of manager Household-based enterprise Female manager Household-based enterprise Availability of loan sources n.e. Agricultural processing enterprise Age of enterprise n.e. Road quality Education of manager Sinhalese manager Household-based enterprise Manufacturing, nonagricultural enterprise Age of enterprise Low market demand Age of enterprise n.e. Services enterprise n.e. Road access n.a. Household-based enterprise n.a. Age of enterprise Other production enterprise Services enterprise Water Mixed enterprise Services enterprise Agricultural processing enterprise Mixed enterprise Household-based enterprise Female manager Telecommunication n.e. Education of manager n.e. Source: Annex H, Table H.3 and Table H.4. Note: a. Variables with at least 10 percent signi�cance level. Variables in models with benchmarks are in regular font, variables in models with benchmark components are in italic font. Variables that are signi�cant in models with benchmarks are not listed twice if they are also signi�cant in models with benchmark components. n.a. Not applicable; n.e. Not estimated. Perceptions About Investment Climate Constraints 49 inside track to bank loans and other financial on the connectivity benchmark indicator. In Tan- instruments; it may face more harassment from zania, proximity to the post of�ce reduces reported government agencies. But if EICO obstacles hinder problems with the postal service. Improvements in the enterprise’s operation and growth, as is indeed the aggregated infrastructure services benchmark the specific question posed to entrepreneurs, one indicate reduced obstacles of electricity quality would expect to �nd that a greater obstacles reduce and access, telecommunications, water, road the enterprise size and productivity. In other access, and postal service. words, the direction of causality goes both ways, Reports on obstacles presented by weak market and estimated effects of size and productivity on demand proved dif�cult to explain using commu- EICO responses would be negatively biased. This nity characteristics. Higher connectivity scores in resulting endogeneity bias could be solved through Nicaragua create a greater obstacle (though not instrumental variables. Alternatively, size and pro- strongly so), but travel time to the nearest city in ductivity measures may be omitted and a reduced Sri Lanka reduces it.53 In Nicaragua and Sri Lanka, form equation may be estimated—which indeed the availability of business services in the commu- was the strategy followed here. The drawback of nity did nothing to alleviate perceived low mar- this approach is that it remains unclear whether the ket demand or, in Sri Lanka, the lack of market estimated effect of, say, female entrepreneurship information. on the EICO response is direct or is manifested In Nicaragua, the EICOs of economic policy indirectly through size and productivity.51 uncertainty, corruption, and crime/theft rose to Table 5.5 describes the findings with regard to the top ten. The models with benchmark indicators the effect of community characteristics as Table 5.4 assign a signi�cant role to connectivity (negative), did for enterprise characteristics. The column for infrastructure services (positive), and business ser- Tanzania registers many significant effects; in Sri vices (positive). Probing with specific benchmark Lanka and Nicaragua the effects are not quite as components yields less conclusive results: proxim- strong. With respect to loans, a strong pattern ity (inverse time) to the nearest city has a negative emerges: in Sri Lanka and Tanzania, the number of effect but only signi�cantly so for crime/theft, and banks has a positive effect on the number of obsta- the share of households with telephone services cles regarding loan cost, procedures, and availabil- has no effect. ity, while the number of bank services has a The human capital benchmark indicator enters negative effect and alleviates obstacles. Yet, sur- all models with selected components because of its prisingly, the effects in Nicaragua are always just presumed cognitive insight into the social and eco- the opposite, even though the Nicaraguan effects nomic working of society and the effect of obsta- are smaller and not statistically signi�cant. cles. In Tanzania, six of the ten EICO regressions Most often, estimation results indicate entrepre- find a significant negative human capital effect, neurs report fewer obstacles with infrastructure and twice it is positive (significant for electricity whenever infrastructure provision at the commu- access and nearly significant for electricity qual- nity level improves. Reported obstacles regarding ity). In Nicaragua and Sri Lanka, the effect is more electricity decline in every country when the per- variable. In Nicaragua, higher levels of human centage of households with electricity rises. In capital in the community raise economic policy every country, road quality is better in communi- uncertainty and concerns about electricity but ties where the travel time to the nearest city is somewhat decrease obstacles of low market lower. A concrete or asphalt road (recorded only demand. All other effects are not only statistically in Tanzania and Sri Lanka) alleviated obstacles of insignificant but also economically small. In Sri road access and quality, even if insignificantly so Lanka, the human capital benchmark reduces in Sri Lanka. In Nicaragua and Tanzania, com- obstacles with electricity and water. Even if statis- plaints about water supply decline when more tically insigni�cant, however, it raises each �nance households obtain drinking water from a piped obstacle—cost, procedure, and availability. source or from a protected well, but in Sri Lanka Table H.3 and Table H.4 report the effect of this variable has no effect.52 Telecommunications unmeasured community factors, that is, the commu- are better in communities having many house- nity random effect. Across all EICO models with holds with �xed-line phones. In Sri Lanka, access benchmark indicators, the average contribution is to transport facilities improves as scores improve 0.45 in Nicaragua on a scale of 0/2, 0.77 in Sri Lanka 50 The Rural Investment Climate Table 5.5 Signi�cant Enterprise Characteristics in EICO Equationsa EICO variable Nicaragua Sri Lanka Tanzania Interest rate of loan Number of banks Number of banks Number of bank services Share income from agriculture Connectivity Finance services Time to near city Number bank services Loan procedures Business services Number of banks Income per capita Insurance service Number of banks Share income from agriculture Connectivity Human capital Electricity Community population size Community population size Human capital Income per capita Connectivity Human capital Share income from Infrastructure services Percent households with agriculture Percent households with electricity Infrastructure services electricity Percent households with electricity Availability electricity Availability of loan sources Business services n.e. Income per capita n.e. Share income from agriculture Connectivity Finance services Time to nearest city Human capital Number of bank services Road quality Connectivity Agricultual seasonality Business services Time to nearest city Share income from Human capital agriculture Time to nearest city Connectivity Concrete/asphalt road Business services Finance services Time to nearest city Low market demand Connectivity n.e. Human capital Time to nearest city n.e. Road access n.a. Income per capita n.a. Finance services Infrastructure services Business services Human capital Time to nearest city Concrete/asphalt road Perceptions About Investment Climate Constraints 51 Table 5.5 (Continued ) EICO variable Nicaragua Sri Lanka Tanzania Water Sewage system Income per capita Connectivity Income per capita Connectivity Percent households with Human capital Infrastructure services protected water Time to main market Percent households with protected water Human capital Telecommunication n.e. Income per capita n.e. Infrastructure services Infrastructure services Business services Business services Percent households with Finance services �xed phone line Share income from agriculture Percent households with �xed phone line Availability electricity Source: Annex H, Table H.3 and Table H.4. Note: a.Variables with at least 10 percent signi�cance level. Variables in models with benchmarks are in regular font, variables in models with benchmark components are in italic font. Variables that are signi�cant in models with benchmarks are not listed twice if they are also signi�cant in models with benchmark components. n.a. = Not applicable; n.e. = Not estimated. on a scale of 0/4, and 1.13 in Tanzania on a scale of informative? Consider two summary statistics: 0/4; all of the estimated effects are statistically (i) EICO Burden, defined as the simple sum of all highly signi�cant. Thus, �rst of all, signi�cant sim- EICO responses, thus treating the categorical val- ilarities in the responses of entrepreneurs in the ues as ordinal scales; and (ii) EICO Obstacle, same community are not explained by the commu- defined as a count of EICO factors rated major nity characteristics. Second, the magnitude of the or very severe obstacles (or in Nicaragua merely effect of unmeasured community factors is rela- severe). tively smaller in Sri Lanka and larger in Tanzania, Table 5.6 reports results of a simple analysis of with Nicaragua falling between—even if in Tanza- variance of these summary statistics across com- nia more measured community characteristics munities. Enterprise characteristics explain at most showed up as statistically significant. Third, the 6.6 percent of the variation in the EICO Burden or magnitude of the effect of unmeasured community EICO Obstacle. Community dummies, which cap- factors is large: the largest effect of any measured ture all observable and unobservable features of a community variable is only one-half as large. Thus, community’s investment climate, explain between the EICO responses reflect much information about 27 and 64 percent of the variation. This is further the community investment climate that is not obvi- evidence that entrepreneurs in a given community ous from the community questionnaire. tend to share similar perceptions about their com- munity’s investment climate. In other words, the investment climate is indeed a community factor. 5.3.2 Total Burden of the A logical next question is to what degree obser- Investment Climate vable community variables can explain the Burden The sheer number of investment climate facets and Obstacle variables. Table H.6 in the Annex covered in the RICS questionnaire may easily over- provides estimates. Household-based entrepre- whelm the analyst who tries to discern patterns neurs tend to perceive fewer obstacles, as do in the responses. Is a simple summary measure female entrepreneurs, especially in Sri Lanka. 52 The Rural Investment Climate Table 5.6 Analysis of Variance of Total Brden of the Investment Climate Nicaragua Sri Lanka Tanzania Proportion of variation EICO EICO EICO EICO EICO EICO explained by Burden Obstacle Burden Obstacle Burden Obstacle Enterprise characteristics 0.026 0.015 0.064 0.066 0.019 0.019 Community dummies 0.313 0.272 0.402 0.365 0.639 0.599 Enterprise characteristics community dummies 0.326 0.279 0.450 0.411 0.642 0.604 Enterprise characteristics benchmark indicators community variables 0.108 0.066 0.089 0.095 0.081 0.073 Source: RIC Surveys and, for the fourth row, Annex H, Table H.6. Sinhalese entrepreneurs mention more obstacles. Climate Interactions (EICI) variables reflect the Nicaragua shows no significant variations by interaction between entrepreneurs and govern- industry, but in Sri Lanka and Tanzania service ment bodies. This section relates three types of enterprises face fewer obstacles than traders, while EICIs to the standard set of enterprise character- in Sri Lanka manufacturing and other enterprises istics and selected community characteristics: list more obstacles. (i) two aspects of government reliability: pre- Benchmark indicators as a group are jointly sig- dictability or variance in the application of laws ni�cant in Nicaragua and Tanzania. In Nicaragua, and the misinterpretation and manipulation of better connectivity lowers the total EICO burden, laws by government officials; (ii) the perceived but business services raise it; in Tanzania, commu- effect of kickbacks on businesses; and (iii) percep- nity human capital is associated with lower total tions about the support received from the legal EICO burden. Overall, most of the estimated system. benchmark effects are negative, even if not statisti- Five community characteristics are inserted into cally signi�cant. the empirical models.54 Population size denotes the Other community variables do not have a sig- size of networks, the degree to which people know ni�cant effect on the reported burden of the invest- each other, the extent to which government offi- ment climate, even though the effect of population cials can hide their discretionary behavior, and so is positive in every country and the effect of the forth. Income per capita represents the level of importance of agriculture in the economy (mea- development in the community. The connectivity sured by seasonality and the share of income benchmark measures the community’s degree of derived from agricultural activities) is negative. integration with other parts of the country, which Together, the community variables and bench- may alternately raise or reduce the scope of gov- mark indicators explain between 2.5 percent and ernment discretionary behavior. The business 8.2 percent of the variation in the Burden and services benchmark is inserted to mark entrepre- Obstacle variables, only a rather small portion of neurs’ potential to plug into commercial networks, the variation between the communities (Table 5.6). receive information, and learn about common A large portion of the investment climate at the practices. Whether this raises or lowers govern- community level is still shrouded from direct ment officials’ perceived discretionary activity is measurement. an empirical question; arguments could be made both ways. The community’s level of human capital denotes cognitive development that on 5.4 ESTIMATION RESULTS: EICIS the one hand prevents rogue exploitation of illit- Where Enterprise Investment Climate Outcome erate members of the population but on the other (EICO) variables consider obstacles to the operation hand allows creativity in bending rules to one’s and growth of enterprises, Enterprise Investment advantage. Perceptions About Investment Climate Constraints 53 5.4.1 Government Reliability: Predictability Income per capita has a borderline-signi�cant nega- and Manipulation of Laws tive effect. Community human capital has no effect. A significant unmeasured community factor is Table H.7 relates enterprise and community char- larger than any measured community variable but acteristics to perceptions of laws as predictable and not dramatically so. manipulable. Only a few enterprise characteristics The similarity between the five equations may matter. Educated Sri Lankan entrepreneurs do not suggest that entrepreneurs tend to respond in the think laws are unpredictable: per year of schooling, same way to all �ve questions. That is quite correct: the probability diminishes by 5.3 percentage 81.2 percent of the entrepreneurs who responded points, which is a strong effect indeed. Similarly, to all questions never felt an effect from kickbacks the gender effect in Tanzania is negative: women in any dimension, and half of the rest gave the same view laws as more predictable than do men. answer to all questions. But the variations are large Tanzania lacks the education effect, however, and enough to allow the effect of business services to Sri Lanka gives no evidence of the gender effect; evaporate from the credit kickback equation and to in Nicaragua, manager characteristics were not raise the effect of population size and connectivity recorded in the survey data. in the politician kickback equation. The effect of community environment is hardly uniform across the three countries either. The only variable that carries the same sign in all three 5.4.3 Ef�ciency of the Legal System columns is per capita income, but its effect is statis- Business dealings take place within an institu- tically insignificant. Connectivity creates pre- tional framework of rules and regulations. The two dictability in Nicaragua but has no effect elsewhere. sides engaging in any transaction implicitly Business services associate with predictability assume that any disagreement or dispute that in Tanzania, but Nicaragua shows an opposite, might arise can be resolved in a predictable man- although statistically insigni�cant, effect. In Tanza- ner as established by that framework. But if the set nia, human capital raises unpredictability sharply, of rules and regulations leaves ambiguity, has but this variable has a minimal effect elsewhere. loopholes, or is inconsistently enforced, transac- One element is common to all these models, tion costs increase. however: the effect of unmeasured community The RIC database includes several items that factors is large in every country. A one-standard- reflect on the institutional framework; three of deviation swing in these factors changes by 20 per- them are examined in Table H.9: reliance on repu- centage points the probability that entrepreneurs tation, protection through a contract, and support believe laws to be unpredictable. This commonal- by the legal system. Considering patterns among ity in the response by community suggests enterprise characteristics across the three sets of strongly that local factors play a highly important estimates, it is striking that older businesses in role in how laws are applied. Unfortunately, these Nicaragua rely less on reputation yet do not factors are not captured well in the community expect that contracts will protect them from cheat- characteristics employed here. ing or that the legal system will uphold that con- tract. In Sri Lanka and Tanzania, the signs of the 5.4.2 Kickbacks parameter estimates of education, gender, and In Nicaragua, a suf�cient number of entrepreneurs ethnicity exhibit similar patterns, although none provided information about the effect of kickbacks of the estimates (with one exception) is statisti- on their businesses that an analysis with an ordered cally significant. Differences between sectors are probit model is warranted. Table H.8 reports few also more or less similar among the three sets of distinctions among enterprises in their views on estimates. As mentioned before, reputation might bribes. There is no age effect, and service and mixed be expected to count more heavily in economic enterprises may experience less of an obstacle than environments where the rod of the legal system do trading and production enterprises. is short, but neither the descriptive statistics in The real action in this model appears, once again, Table 5.3 nor the regression results in Table H.9 in the community variables. The effect of kickbacks bear out this expectation. is felt more strongly in larger communities with Overall, statistical significance is weak, but more business services and poorer connectivity. three patterns are notable in Table H.7, H.8, and 54 The Rural Investment Climate H.9. First, the gender effect is fairly similar in Sri representative of the population at large. From Lanka and Tanzania, even though it was weakly smallest to largest, weights vary by a factor of 13.1 different for opinions about the predictability and in Nicaragua and a factor of 619.5 in Sri Lanka; manipulability of laws. Of course, regardless of sampling weights are not available for Tanzania. what one thinks about the manner in which laws Community random effects capture unobservable and rules are written and enforced, their actually factors affecting the perceptions of all entrepreneurs ability to provide legal protection is an entirely dif- in a given community. Conceptually, it is plausible ferent matter. Women appear to feel more vulner- that they exist; empirically, they prove to be impor- able in this aspect. Second, if the statistically tant.56 Unfortunately, as demonstrated in Annex H insignificant coefficients may be interpreted this on the econometric modeling of RIC data, the com- way, Sinhalese entrepreneurs both rely more on putation of the standard errors of the parameter reputation and expect contract and legal system estimates is not a fully settled issue and warrants support—even though they view laws as manipu- more research. lable by government officials. Third, Nicaragua and Sri Lanka show a pronounced difference in the way manufacturers as opposed to traders respond 5.5.2 Perceptions, Conditions, to these three questions.55 In Nicaragua, traders and Behavioral Outcomes perceive a greater lack of support from the legal The basic premise of investment climate survey system than do manufacturers; in Sri Lanka, man- research in urban and rural areas alike is that the ufacturers feel disadvantaged relative to traders. It investment climate determines enterprise perfor- will be interesting to relate this difference to a com- mance and therefore entrepreneurship incentives. parison of the two countries’ commercial laws. In this regard, it is useful to distinguish investment In Nicaragua, entrepreneurs in larger commu- climate conditions and investment climate percep- nities expect more protection from contracts; in Sri tions. Conditions are factual; perceptions are Lanka and Tanzania, reputation is less important subjective. Conditions impose hard constraints; in larger communities—as indeed transactions perceptions create constraints that affect behavior. become less personalized—but in Sri Lanka large- In a world of perfect information, perceptions are town entrepreneurs do not expect as much sup- identical to actual conditions. In a more realistic port from the legal system. Other community world of imperfect information, perceptions are variables matter hardly at all. Unmeasured com- imperfect reflections of conditions. This raises munity factors are important, but the magnitude interesting questions: How is enterprise perfor- of the effect is smaller than was found in other mance conditioned by investment climate percep- tables: the predictability and manipulability of tions? How do investment climate perceptions laws is a bit more discretionary than are reputation influence entrepreneurship choice? How do invest- and legal protection and appear to vary more sys- ment climate conditions determine investment tematically between communities. climate perceptions? Figure 5.2 describes these posited relationships in a schematic diagram. It also adds one feedback 5.5 PUSHING THE ANALYSIS loop: perceptions may be influenced by enterprise performance. Growth in the enterprise may expose FORWARD: NEXT STEPS entrepreneurs to limitations in their economic environments (for example, limits on credit, poor 5.5.1 Analyzing Discrete Outcomes infrastructure, low-quality labor) that were not In this chapter, the EICO and EICI variables are obvious before. Not expressed in this diagram is obviously discrete, which necessitates the use of the policy feedback loop in which policymakers probit models for simple 0/1 outcome variables act to improve the investment climate in commu- and ordered probit if the outcome has multiple nities with successful enterprises and entrepre- categories. When applying these models, it is neurs who lobby for change. important to account for sampling weights RIC surveys gather information from entrepre- and community random effects. Because of the neurs about their investment climate perceptions. sampling design of the survey data, sampling The surveys also collect information about invest- weights are necessary if empirical results are to be ment climate conditions through community Perceptions About Investment Climate Constraints 55 Figure 5.2 Investment Climate and Entrepreneurship Investment Investment climate Enterprise climate perceptions performance conditions (EICOs, EICIs) Entrepreneurship choice Source: The Authors. questionnaires and, to some degree, from factual measures for the self-selection effect and then information provided by entrepreneurs. The EICO applying the corrected measure as a determinant and EICI models in this chapter target the leftmost of entrepreneurship choice is a useful next step in relationship in Figure 5.2. They sidestep the inter- the study of entrepreneurship. active relationship with enterprise performance.57 In one respect, Figure 5.2 makes a clinical sim- Instead, they estimate a “reduced form� set of pli�cation: both investment climate conditions and equations in which determinants such as enterprise investment climate perceptions have many dimen- characteristics and investment climate conditions sions. This chapter grapples with the methodology (through community characteristics) capture both of how to relate these two multidimensional con- the direct effect on the investment climate and the cepts together. Chapters 3 and 4 faced a similar indirect effect that works through the feedback challenge in relating investment climate conditions loop on enterprise performance. to enterprise performance and entrepreneurship A next step in this research project is to disen- choice—in terms of Figure 5.2, bypassing the center tangle this relationship. This presents a two-part box. The next step in investment climate research challenge: will be to come to terms with the multidimension- • First, �nd variables that determine enterprise ality of perceptions: Should they be summarized performance but not investment climate per- into “benchmark perceptions�? How important is ceptions and find other variables that deter- each facet of investment climate perception? mine investment climate perceptions but not enterprise performance (that is, in economet- ric jargon, find variables to identify the 5.6 IMPLICATIONS FOR RIC investment climate perception and enterprise METHODOLOGY performance equations); and As diverse as are the three countries in this analysis, • Second, develop an econometric model that the lists of their most important investment climate incorporates sampling weights and commu- concerns show a remarkable overlap: six items nity random effects while accounting for the appear in each country’s top ten, and three more discrete nature of the measured investment occur in two of the three. The common concerns are climate perceptions and continuous nature of enterprise performance. (i) �nance cost, availability, and procedures; (ii) electricity and water (access, cost, or relia- Investment climate perceptions also play a role bility); and in the decision to become an entrepreneur and (iii) road quality and access. set up a business. But these are the perceptions held by would-be entrepreneurs, whereas the The other top-ten concerns are more varied: perceptions measured in the survey are those market demand, economic policy uncertainty, and expressed by current entrepreneurs. This creates a telecommunications. self-selection issue: current entrepreneurs may In a number of investment climate facets, coun- have a more positive view of the investment tries differ greatly: ranks vary sharply for corrup- climate than do those who have decided not to tion, economic policy uncertainty, and postal operate a business. Correcting the EICO and EICI service and road blocks. It is possible that the use 56 The Rural Investment Climate of rather general screen questions in Sri Lanka community questionnaires. Extracting this infor- and Tanzania reduced responses. A question mation is a time-intensive exercise, especially since such as “Is governance an obstacle?� may not the three RIC databases are constructed in com- have triggered an affirmative response, for exam- pletely different ways and the questionnaires con- ple, whereas follow-up questions about corrup- tain signi�cant variations. Moreover, the relevant tion, economic policy uncertainty, crime, war, set of community characteristics may well vary and legal systems might have if they had been between the various obstacle and interaction vari- asked. ables studied here. Community characteristics, Many EICO factors are not really relevant to too, are likely intercorrelated, making it dif�cult to rural enterprises. In Nicaragua and Sri Lanka, disentangle the precise reason for variation in more than 95 percent of respondents rated many EICO and EICI responses between communities. variables No obstacle. Tanzania’s entrepreneurs, On the other hand, community characteristics may who are either more vocal or suffer from a large suffer from some degree of measurement error, array of obstacles, include many items of lesser which could limit their statistically signi�cant con- concern, such as import/export, food safety, and tributions to the explanations pursued here. The labor issues. It is unclear whether all of the almost actual contribution then becomes part of the com- 40 EICO variables should be collected—but they munity level random effect. are easy to collect, as opposed to speci�c questions More likely, the community questionnaire has about enterprise performance, labor employment, not yet captured the idiosyncrasies that character- and asset investment. ize local investment climates. One could point to Throughout the econometric analysis, the mag- factors that may have been overlooked: among nitude of the effect of unmeasured community fac- them, the influence of major local personalities, the tors proved to be large. Thus, the EICO and EICI role of elites, and the importance of networks. responses produced much information about the RIC2 shows that something operating at the com- community investment climate not gleaned from munity level has yet to be captured. Future RIC the community questionnaires. practitioners might profitably strive to improve RIC2 may not yet have explored in sufficient the measurement of that “something.� depth the range of information available in the 6 Differences Among Communities 6.1 COMMUNITY CHARACTERISTICS AND THE COMMUNITY ENVIRONMENT The communities within which rural enterprises operate vary greatly. This is true between economies, but also within the pilot countries. To determine the specific influences these differences might impose on the local rural investment climate, community leaders were questioned about the status of several community char- acteristics that ex hypothesis might support or impede enterprise performance and growth. As shown in the following table, the com- munity descriptors scale community size (population), the presence (or absence) of important infrastructure such as transport and com- munications; the levels of education, health, and other social or pub- lic services; the availability of financial and business services; proximity to urban areas and marketing centers; and the number of taxes paid by local residents. Information was also gathered about the communities’ endowments of dry and irrigated agricultural lands. In Nicaragua, an attempt was made to determine gender bias in seasonal agricultural employment, while in Sri Lanka surveyors queried the variability of male and female participation in the agri- cultural labor force. A cross-country summary comparison of these descriptors appears in Table 6.1. Differences in community population size make comparison dif�cult. Sri Lanka’s communities are about 11 percent the size of those in Nicaragua and less than a third the size of those in Tanzania.58 Taking account of these differences, communities in Sri Lanka areseem relatively well supplied with economic and social infrastructure. The concept of market probably differs among the countries: the distance to market is highest in Nicaragua, despite (or perhaps because of) the large community size. At the community level, the other main differences appear in the proximity of villages to markets and urban demand centers and in the availability of household infrastructure services. Here, residents in Nicaraguan and Sri Lankan communities appear much better pro- vided with electricity than their Tanzanian counterparts. Community-level differences are also evident in aggregated enterprise performance results. As shown in Annex J, the signifi- cance of nonfarm rural employment and of earnings per capita and per worker appears to be greater in larger communities than in smaller ones. In Nicaragua and Sri Lanka, employment share in 57 58 The Rural Investment Climate Table 6.1 Comparison of Community Descriptors (Community Averages) Descriptor Nicaragua Sri Lanka Tanzania Population 16887 1837 6414 Distance (km) to market 25 16 9 Distance (km) to city 35 15 19 Road surface* (max 4) 2.4 2.4 1.9 Percent households with electricity 74 95 6 Percent households with gas 31 16 NA School type (highest level)* (max 3) 1.8 1.6 1.5 Health center type (highest level)* (max 5) 3.4 2.9** 3.1 Local availability of banking services* (max 7) 1.0 0.9 0.9 Local availability of business services* (max 7) 1.7 2.3 0.5 Source: RICS data. Notes: * scaled; ** not including hospitals; max 7. agriculture falls dramatically in the larger com- for Nicaragua and Sri Lanka tend to average slightly munities. Absolute magnitudes of both sets of higher, while Tanzania averages much lower. indicators vary considerably across countries, Connectivity, human capital, and �nance services however, probably reflecting differing levels of score higher in Sri Lanka than in Nicaragua, development and degrees of urbanization. which is as expected because of Sri Lanka’s high Benchmark indicators, introduced in Chapter 2, population density and traditionally high level of are defined in Annex E, which outlines their use education. In Tanzania all indicators are lower in the regressions. The following sections show than or at best equal to the other countries’, with how benchmark indicators vary among the the exception of finance services.59 Sri Lanka’s regressions, starting in Section 6.2 with the varia- relatively low index for business services (com- tion among communities and between countries. pared to Nicaragua’s) may explain why this factor Correlations among community variables are scored as significant in the Sri Lanka regressions explored in some detail, especially with the bench- in Chapter 3. mark indicators (BI). Section 6.3 examines rela- The breakdown between communities with tions between the BI and prices and Section 6.4 smaller and larger populations shows, as expected, those between the BI and other indicators of com- that the indexes in larger communities are gener- munity characteristics. Regression results for Sri ally higher. The differences in Nicaragua are much Lanka are presented in Section 6.5 to help explain greater than those in Sri Lanka and Tanzania, community economic performance through a set probably reflecting the greater income disparity of explanatory variables, as measured by outcome in there. The BIs with the relatively highest dif- variables at the community level. The chapter con- ference between smaller and larger communities cludes with Section 6.6, an assessment of the per- are connectivity, infrastructure services, and busi- formance of benchmark indicators in the three ness services indexes. In Nicaragua, the range in data sets. the human capital and finance services indexes is also substantial. It is noteworthy that the gover- nance index in Tanzania and Sri Lanka does not 6.2 CROSS-COUNTRY differ between small and large communities, COMPARISONS OF while in Nicaragua the difference is notable. BENCHMARK INDICATORS Correlations Between Benchmark Indicators Comparisons The BIs for Nicaragua show nearly all statistically One of the purposes of benchmark indicators significant correlations with each other at the is to facilitate comparisons between countries. 10 percent level (Table 6.3). Governance is the only Table 6.2 reveals that the benchmark indicators exception, as it is not related to infrastructure Differences Among Communities 59 Table 6.2 Comparison of Benchmark Indicators by Country and Community Population Size Nicaragua Sri Lanka Tanzania Population size: All Index Index Index Connectivity 0.34 0.40 0.20 Infrastructure services 0.51 0.35 0.19 Business services 0.24 0.15 0.07 Governance 0.67 0.67 0.50 Human capital 0.21 0.33 0.21 Finance services 0.17 0.50 0.18 Average 0.36 0.40 0.22 Population size: Small* =<4800 =<1350 =<3000 Index Index Index Connectivity 0.24 0.36 0.17 Infrastructure services 0.38 0.31 0.14 Business services 0.11 0.12 0.04 Governance 0.64 0.67 0.50 Human capital 0.17 0.33 0.20 Finance services 0.02 0.48 0.18 Average 0.26 0.38 0.21 Population size: Large* >4800 >1350 >3000 Index Index Index Connectivity 0.44 0.44 0.22 Infrastructure services 0.65 0.38 0.23 Business services 0.37 0.18 0.08 Governance 0.71 0.67 0.50 Human capital 0.24 0.33 0.22 Finance services 0.33 0.52 0.19 Average 0.46 0.42 0.24 Source: Annex E. Note: * the indicated size is about the median for each country. services or business services. All signi�cant corre- Completeness of Information lations have a positive sign, thereby pointing for Benchmark Indicators towards a better rural investment climate. Because the BIs and their components are con- In Sri Lanka most benchmark indicators are also structed from raw data, the problem of missing positively correlated at a significance level of observations had to be addressed. Following the 10 percent, but it shows more insigni�cant correla- procedure outlined in Annex E, synthetic values tions than does Nicaragua. Governance is the were substituted for the missing data, computed exception, with no signi�cant correlation with any as averages from available data. This can result in other BI. Human capital and business services are smaller standard deviations and perhaps bias in unrelated as well. the estimates. Annex G, Table G.1, provides an In Tanzania fewer BIs are significantly corre- overview of communities and subcomponents lated than in the other two countries, but the sig- with missing observations. As missing observa- ni�cant ones all have the correct signs. Governance tions for communities occurred in only 0.3, 4.6, and is not correlated at all, and �nance services corre- 1.7 percent of the observations, respectively, in lates only with connectivity and infrastructure Nicaragua, Sri Lanka, and Tanzania, and in the services. The greatest interrelation exists with con- respective subindicators by 2.6, 3.9, and 2.1 per- nectivity and infrastructure services. cent, the resulting biases were likely very small. 60 The Rural Investment Climate Table 6.3 Correlation Coef�cients Between Benchmark Indicators Infrastructure Business Human Finance Connectivity services services Governance capital services Nicaragua Connectivity 1.00 Infrastructure services 0.73* 1.00 Business services 0.34* 0.49* 1.00 Governance 0.17* 0.12 0.09 1.00 Human capital 0.41* 0.37* 0.26* 0.25* 1.00 Finance services 0.41* 0.45* 0.33* 0.23* 0.35* 1.00 Sri Lanka Connectivity 1.00 Infrastructure services 0.48* 1.00 Business services 0.25* 0.24* 1.00 Governance 0.09 0.12 0.06 1.00 Human capital 0.35* 0.32* 0.04 0.07 1.00 Finance services 0.53* 0.45* 0.22* 0.06 0.31* 1.00 Tanzania Connectivity 1.00 Infrastructure services 0.55* 1.00 Business services 0.20* 0.29* 1.00 Governance 0.04 0.04 0.09 1.00 Human capital 0.27* 0.49* 0.15* 0.03 1.00 Finance services 0.21* 0.21* 0.11 0.04 0.05 1.00 Source: RICS data. The most missing observations occurred in the enterprise viability and entrepreneurial choice. In governance indicator in Tanzania and Sri Lanka. remote or not easily accessible villages, prices may differ substantially from those in urban centers. Variation Among Communities Inadequate infrastructure can substantially raise As is shown in Annex E, the BIs are generally costs of products brought from elsewhere and, well distributed over the range between 0 and 1, because of high transportation costs, lower RNFE the main exception being governance, which has “producer� prices for products destined for dis- relatively high minimum and average values. tant markets. Thus improved infrastructure and Human capital has low maximum values, but finance services may encourage private invest- this is consistent with the still relatively low lev- ment in modern technologies, thereby lowering els of education in the three countries’ rural production costs. By implication, improvements areas. The standard deviations shown in Annex in the investment climate reflected in the BI E also provide information about the range of should also correlate with price movements in the index values. The low standard deviations for community. governance and for human capital suggest that The set of prices considered below differs the scaling of responses did not distribute the among the RIC pilots, in part because the lists of index values over the full unit interval and so prices collected were not identical. But more was not optimal. importantly, values were missing for some com- modities in far too many communities to support a 6.3 BENCHMARK INDICATORS proper correlation analysis. Among the BI, in Nicaragua, connectivity, gov- AND PRICES ernance, and human capital relate most to the Apart from production technology, which was different price variables (Annex G, Table G.2). examined with the production function estimates Business and finance services indicators were in Chapter 3, prices are important determinants of not at all related to any of the prices collected, Differences Among Communities 61 however. Infrastructure services related only to 6.4 BENCHMARKS AND the price of diesel. COMMUNITY The prices of staple foods (rice, maize) are signif- icantly related to connectivity. This probably CHARACTERISTICS reflects higher food prices in more urbanized com- Benchmark indicators may affect other character- munities. The price of diesel is negatively related to istics of communities as well. Benchmarked vari- connectivity and infrastructure services, which sug- ables may affect the provision of schooling, for gests higher prices in more remote areas. Finance example, due both to demand for education on the services are not correlated. Amongst subindicators, part of households wishing to connect with the the costs of phone calls and salt are negatively urban world and to supply generated by forward- related to connectivity. The price of transportation, looking community leaders who foresee commu- Coca-Cola (included as a representative mass- nity integration into the region. These variables produced consumer good), and soap are positively also change the incentives for households to related to human capital. In Nicaragua, when gov- engage in farming, in a nonfarm business, or wage ernance is supportive, wages and the price of phone employment. Effective governance reduces the calls tend to be low and the costs of transportation cost of conflict. Thus, the annex tables referred to high, possibly indicating that governance is less of below examine BI correlations with a broad range an issue in more distant rural areas. of community characteristics. As with price vari- Table G.3 shows relations between benchmark ables, the number of characteristics imputed for indicators and prices in Sri Lanka. As expected, each pilot varies between countries because of data the connectivity and infrastructure services index availability. are negatively related to many prices and posi- The BIs for Nicaragua show several statistically tively related to wages and the price of food. significant correlation characteristics (Table G.5). Business services are positively related to wage The data set shows much communality, although levels and the price of wheat flour, which may many of the relations are quantitatively weak. reflect the situation in more urban commercial Most characteristics representing an endowment areas. The governance index does not relate to show positive relations with the benchmark indi- many prices. The relation with rice is positive and cators. Negative correlations are found with immi- that with wheat flour is negative. Human capital grants as percentage of population and with is positively related to wages. The finance ser- distance to market. In an important �nding, com- vices index shows much similarity with connec- munity population size has positive significant tivity: it is negatively related to many price items correlation with all benchmarks, except gover- and positively related to food prices. Distance nance. This shows, as expected, that the IC is gen- indicators are nearly all negatively related to con- erally more supportive in urban than in rural nectivity, infrastructure services, human capital, areas. The highest correlation for connectivity is and finance services. with the number of public services (0.69); for infra- In Tanzania relatively little significant correla- structure services with number of public services tion is found between price levels and the BIs (0.83); for business services with number of ser- (Table G.4). A negative relation exists between vices available for businesses (number of public phone calls to town and connectivity and human services [0.46] and number of banks [0.43]); and for capital; a mobile phone call is negatively related to human capital with percentage of households with connectivity, infrastructure services, and business gas (0.59). Governance only shows a few low cor- services. The price of cement correlates weakly only relations, and �nance services correlate positively with �nance services; rice prices are weakly related with number of banks (0.72, which is almost deter- only to business services; and wage levels are ministic) and highest level of school type (0.53). generally unrelated to the benchmark indexes, In Sri Lanka the number of significant correla- except that payments for rural works appear to be tions with the BIs tends to be less than in lower in areas with good governance. Coca-Cola Nicaragua (Table G.6). Perhaps the differences and kerosene prices are not related to the Tanzanian among communities in Nicaragua are greater BIs, whereas the price for diesel is negatively than in Sri Lanka. The highest significant correla- related to governance and �nance services. In gen- tions are: with connectivity, the number of public eral, patterns are not clear in Tanzania. services (0.61) and distance to city ( 0.56); with 62 The Rural Investment Climate infrastructure services, the number of public ser- dif�cult to attribute between-community variation vices (0.57) and the percentage of households in the dependent variable to any particular BI or with electricity (0.55); with business services, the community variable. But this econometric prob- percentage of conflicts solved by labor court lem originates more with the multidimensional (0.30) and the number of banks (0.28); with gover- nature of communities and less with the manner in nance, the number of business services (0.26); which the communities are described. Empirically, with human capital, the percentage of households connectivity, infrastructure services, business ser- with gas (0.40) and electricity (0.38); and with vices, human capital, measures of governance and finance services, the percentage of households corruption, and �nance services may be correlated, with electricity (0.51) and number of public ser- but conceptually they remain distinct. vices (0.43). The generally negative relationships with connectivity include seasonality (as reflec- ted in the standard deviation of male and female 6.5 REGRESSIONS AT THE agricultural labor use), distances to markets, dis- COMMUNITY LEVEL FOR tance to the nearest city, and agricultural land per SRI LANKA capita. These indicators seem associated primar- ily with the more remote agricultural areas. Infra- An interesting point for analysis is whether structural services also show roughly the same the communities’ economic performance can negative relations. Interestingly, the finance ser- be explained by community data. (It should be vices index is negatively related to agricultural stressed upfront that this kind of analysis is ham- land per capita. pered by the rather small number of communities In Tanzania the BIs are less signi�cantly corre- used in cross-sectional analysis.)60 The analysis in lated with community characteristics (Table G.7). this section of a set of explanatory variables for Sri The highest significant correlation is connectivity Lanka provides an example of regression results to with the number of public services (0.67), followed explain community economic performance as by infrastructure services also with the number of measured by community-level outcome variables. public services (0.72) and with percentage of Table G.8, Annex G, provides information on households having a cell phone (0.70); business the dependent and explanatory variables used services with the number of banks (0.43); and in the community-level regressions. The list of human capital with the number of public services dependent variables starts with enterprise density (0.38). The largest correlation with governance as a measure of the degree of entrepreneurship in is found with road surface ( 0.29, a negative). the community. It continues with seven variables Finance services shows the highest correlation describing labor market outcomes, such as income with the number of banks (0.38) and the number of level, structure of the labor force, labor force par- public services (0.26) and correlates negatively ticipation, and labor productivity measures; two with distance to banking centers ( 0.29). Generally, variables measuring output generated by enter- negative BI relationships in Tanzania include dis- prises relative to their production costs (focusing tance to market, distance to city, and agricultural on community-level enterprise productivity as far land per capita. as this can be captured in the sample); two vari- In conclusion, the BIs and community charac- ables focusing on gross value added (GVA) to teristics show many significant and sometimes examine size and variation in size among enter- quite high correlations. This demonstrates that a prises; and two variables considering seasonality small number of well-selected indicators can rep- in the community from the perspective of enter- resent many facets of community-level influences prise production cycles and agricultural labor on the rural investment climate, although they do demand cycles. not and cannot tell the whole story. Using bench- mark variables in regression analysis, especially in conjunction with other community variables, may Regression Results lead to problems of multicollinearity: too many Table G.9 displays the regression results for all variables, not dissimilar enough from each other, 14 outcome variables in Sri Lanka. The model compete in the explanation. This can undermine shown in Panel A is exploratory: the same set of the precision of the estimated effects and make it explanatory variables is used in every regression Differences Among Communities 63 model. The risk of such a speci�cation is that irrel- The next two variables, rural nonfarm income evant variables reduce the precision of the para- per capita and per worker, are both affected by the meter estimates of relevant variables. For that measurement problem in income, so results must reason, Panel B limits the speci�cation to variables be cautiously interpreted. Both variables rise with with a t-statistic of at least 0.5 in absolute value. human capital, which is entirely plausible. They Enterprise density per household is positively also fall with the governance benchmark, however. related to the share of Sinhalese in the population, Openness of communities61 is not really helpful; who appear to operate relatively more of the enter- and food price is likewise negatively related. prises. Proximity to schools and hospitals is bene- Labor force participation relates positively to �cial; these are higher in more densely populated community population size, upland per capita and areas. Among the benchmarks, connectivity, busi- the benchmark indicators of business services, ness services, and human capital raise enterprise human capital, and governance. The relationship density. After deleting three insignificant vari- also suggests that labor force participation is lower ables, food prices are found to have a significant in communities with a high share of Singhalese positive effect as well, but the finance services located in the more productive lowlands. The infra- index has a negative effect. structure services and finance services indicators Total per capita income from work (farms, tend to reduce the labor force participation rate. RNFEs) is positively related with community pop- The male wage rises rather signi�cantly in com- ulation size, business services, and human capital munities with more uplands. The human capital and negatively with enterprise openness, infra- index also exerts a positive effect on male wage structure services, governance, and the price of rates, as one would expect. Food prices are also rice. Since the price of rice was generally low when positively related, but the price of kerosene shows the pilot surveys were administered, the relation- a negative effect, which may hint at the role of ship suggests that income from work is highest in commercial areas. These two price variables corre- larger communities having higher average levels late only weakly, so this contrasting result should of education. The negative infrastructure services not be attributed to multicollinearity. After remov- effect, however, is more dif�cult to understand. ing a number of insignificant variables, a high The share of rural nonfarm income falls with the share of Sinhalese, proximity to a city, and the share of Sinhalese and proximity to the city. business services benchmark all serve to raise the Improvement in the �nance services index is bene- male wage. ficial, however. When six insignificant variables The average ratio of sales to cost is a proxy for are removed, the effect of the remaining variables average (total) productivity of community enter- becomes more clearly defined, but no other vari- prises. Enterprise productivity shows weak nega- able rises to a 10 percent significance level. It tive association with connectivity and food prices should be noted, however, that a flaw in the Sri and weak positive association with business ser- Lanka questionnaire has led to an understatement vices and infrastructure services.62 Removing of household nonfarm enterprise income. redundant variables suggests a role for several The next dependent variable—the ratio of other variables: productivity may be lower in lar- RNFE workers—is closely related and, because of ger communities with a larger share of Sinhalese. the problem in measuring enterprise income, more The same variables affect net factor productivity appropriate: employment in rural nonfarm enter- (the next dependent variable), but the relationship prises as a share of working adults. This share rises is less well de�ned. in those larger communities having a lower avail- Gross value added per enterprise, a size mea- ability of agricultural lowlands. Dropping ten sure, relates positively to community openness and insigni�cant variables leaves some statistically sig- a high finance index. In communities with more ni�cant variables: relatively more people work in Sinhalese and a higher governance benchmark, their own nonfarm enterprise when illiteracy is enterprises are smaller. The business services lower, schools are nearby, the main city is farther benchmark comes forward as well with a positive away, and the price of kerosene is lower. The latter effect when insignificant variables are removed. probably reflects lower kerosene prices in com- The standard deviation of gross value added, a mercial areas. The other variables reflect the set of measure of size differences, relates to the same opportunities in the community. variables. 64 The Rural Investment Climate A lower value of the enterprise seasonal adjust- 6.6 PERFORMANCE OF ment factor indicates more seasonality. Thus, RNFE seasonality is more severe in communities BENCHMARK INDICATORS with a greater share of Sinhalese and more widely AS DESCRIPTORS FOR THE available business services. Where �nance services INVESTMENT CLIMATE are more available, seasonality is less. Removing The economic structure of rural communities, redundant variables changes little in this picture. including geographic and climatic characteristics Agricultural seasonality is characterized by a and the enabling environment, has many dimen- high share of Sinhalese, a high share of illiteracy, sions. This makes it dif�cult to compare investment and low food prices, and yet, at the same time, also climate conditions and to explain economic perfor- by proximity to a hospital and a main city. Among mance. The BIs are aggregates developed in the the benchmark indicators, better governance is RIC study to facilitate comparison of the invest- associated with greater agricultural seasonality, ment climate among communities and between but connectivity, human capital, and finance ser- countries. This study demonstrates that the BIs are vices are not. Several of these are characteristic of useful aggregates for this purpose, although re�ne- rural areas with paddy land. ment is needed. The evidence is threefold: Using the R2 values to represent suitable mea- sures of fit, it appears that community-level 1. In binary correlations the benchmark indica- analysis best explains agricultural seasonality, tors generally behaved as expected from a enterprise density, and the various measures of theoretical perspective: positive relations enterprise productivity and size. The community with outcome variables and negative rela- analysis corresponds with the high R2 values tions with obstacles, some with strong statis- found in the enterprise performance analysis for tical significance, even though many of the Sri Lanka in Chapter 3. relations were weak. 2. Together in multiple regressions, some of the Factor Analysis parameter estimates of benchmark indicators give counterintuitive relations, probably Factor analyses conducted to explore the full set of because of their intercorrelations and unob- community variables revealed a considerable served background variables. number of factors and mostly low factor loadings. Clearly, communities are multifaceted in their 3. Replacement of a benchmark by its compo- characteristics, and the structure of the variance in nents, as practiced in Chapters 3, 4, 5, and community variables cannot well be explained above in this chapter, can be a useful strat- with only a limited number of factors or bench- egy to see how much explanatory power mark indicators. It is of course possible to “fish� was lost by aggregation. It can also reveal for variables with signi�cant explanation in regres- which are the relatively weaker subcompo- sion analysis, but still it must be expected that only nents. Because some components correlate a small part of the variance at the community level highly with other benchmarks, however, the can be explained using the data sets in hand. potential gains of replacement are reduced. 7 CONCLUSIONS AND RECOMMENDATIONS THE AIMS OF THE STUDY This study explored RIC databases for three countries—Nicaragua, Sri Lanka, and Tanzania—in a wide-ranging look at the effect of investment climate on enterprise performance and entrepreneur- ship. This is not the �rst report written on the basis of these data. For each RIC pilot study, country teams were in charge of conducting the surveys, analyzing the data, and writing a RICA report. These reports contain important �ndings about the investment climate in the pilot countries. RIC1, the �rst comparative study of the surveys, formally titled The Rural Investment Climate: It Differs and It Matters, appeared in 2006. Yet these analytical efforts explored only parts of the collected data, focused on different issues, and used dissimilar methodological tools. Important topics of analysis were not fully covered because of limitations in time and budget and lack of read- ily available tools for analysis; other topics of great interest were not addressed at all. This study augments and extends earlier studies with further systematic exploration of the three countries’ RIC data. Speci�cally, this second comprehensive study, RIC2, was under- taken to: • Provide a broader and deeper understanding of nonfarm activ- ity in rural areas, its constraints, and possible ways to mitigate those constraints; • Initiate and test a method of benchmarking the investment cli- mate in rural areas; and • Advance and sharpen the methodologies for RIC assessments and provide analytical guidance for future survey teams and policy analysts. Developing and refining methodologies for RIC assessments is in part a process of learning by doing. Lessons learned in the �rst pilots and RIC1 have already been taken into account in the draft Imple- mentation Manual (World Bank 2007c). The �ndings of the present study, currently being supplemented with ongoing work in other country assessments, provide important information and pointers for further sharpening survey designs and analytical methodologies, as well for updating the Implementation Manual. To these ends, sev- eral recommendations from the study are discussed below. 65 66 The Rural Investment Climate WHAT IS NEW IN THIS STUDY? individual households and household mem- bers differ with family structure, human This study makes several new contributions to the capital, assets, and community characteris- analysis of the rural investment climate: tics. RNFE self-employment choice is part of 1. A much more comprehensive analysis is a broader array of livelihood strategies, made of data from the RIC pilots than was which includes pursuing opportunities and hitherto available. managing risks. The view among some ana- 2. More systematic use is made of community lysts that RNFE self-employment results data than in previous studies. from push factors for the poor is too narrow. 3. Benchmark indicators are developed and This study provides evidence that house- employed in the analyses. holds engaged in RNFE tend to be better off 4. Responses about constraints in the invest- than farming families and tend to pursue ment climate are analyzed using the concepts entrepreneurship for commercial reasons. Enterprise Investment Climate Outcomes 2. The large majority of rural enterprises are (EICO) and Enterprise Investment Climate very small in size, with one or a few workers Interactions (EICI). only. The large majority of enterprises (68 to 5. Novel Stata programs (documented in this 86 percent) employ only unpaid family work- report) were developed for random effect ers. Enterprise population is not static, how- analysis with sampling weights. ever. After comparing the three countries and 6. Estimates are made of the contribution of differences within the countries, a general groups of variables—enterprise, industry, pattern emerges showing that, with increased and community characteristics—in explain- per capita income, both enterprise density ing variation of outcome variables. per 1,000 inhabitants and the average size of 7. Household and enterprise weights are used enterprises increases. in the analysis of data for Nicaragua and Sri 3. Large numbers of small rural enterprises buy Lanka. For Nicaragua, household and enter- and sell mainly locally, which is understand- prise weights lost during data processing able given the nature of their businesses— were reestimated. For Sri Lanka, the enter- services, retail trade, repair, and so on. Yet, prise weights used by the Sri Lankan team evidence from Sri Lanka indicates that enter- were readily available, but household weights prises that sell mainly outside their commu- had to be reestimated. nities have higher productivity. 8. By collecting community-based data, RIC 4. It is often believed that access to credit is introduces a spatial dimension not prominent the most constraining factor in RNFE. Yet the in PICS. This expansion opens new options large majority of enterprises in the three for analysis of differential enterprise devel- countries (58 to 79 percent) report having no opment in a heterogeneous rural space and formal or informal debt. Only 31 percent or along the rural-urban axis, addressing the less (varying by country) carry formal debt complex analytical and methodological ques- equivalent to 50 percent of equity or higher. tions raised when observations are clustered Although entrepreneurs may have a genuine in communities. desire for inexpensive loans, the benchmark indicator and benchmark components for finance services do not strongly influence WHAT HAS BEEN LEARNED measures of enterprise productivity. ABOUT RURAL ECONOMIES? 5. The share of agro-processing in rural enter- prises ranges from only 2 percent in Sri Lanka RIC2 unearthed several important �ndings about to 14 percent in Nicaragua. This shows that a the three economies that would not otherwise be focus only on agribusiness is too narrow for available: private-sector development in rural areas. 1. The household data show that multiactivity 6. In all three countries, enterprise productivity of self-employment in RNFEs, wage employ- appears to differ with enterprise and commu- ment, and farming is a common feature nity characteristics. Consequentially, major of rural households. Available choices for differences emerge within and between Conclusions and Recommendations 67 countries. Generally, the age of enterprises Lanka, and Tanzania respectively. Registra- and the experience of entrepreneurs do mat- tion status is associated with higher enter- ter, indicating the importance of adaptation prise performance only in Tanzania, but and innovation through learning by doing registered enterprises employ more labor in for increasing productivity. Nicaragua and Sri Lanka and have more 7. As diverse as are the three countries in this invested capital in Sri Lanka and Tanzania. analysis, the lists of most important invest- Thus, the effect of registration in the three ment climate concerns expressed by their countries is mixed. entrepreneurs show remarkable consis- tency. Six items appear on each country’s list of top-ten concerns and three more RECOMMENDATIONS ON RICS occur on two of the three lists. The common METHODOLOGY concerns are: The recommended method for RICS implementa- • �nance cost, availability, and procedures; tion is described in the Implementation Manual • electricity and water (access, cost, or reli- (World Bank 2007c), which is based on previous ana- ability); and lytical work. Findings from the present study indi- • road quality and access. cate bene�ts from updating and further sharpening the manual’s recommended methodologies and The remaining concerns vary more widely: approach. Main emphases include the following. market demand, economic policy uncer- tainty, and telecommunications. At the other end of the spectrum, probably because Sensitivity for Using Weights of their small scale, few rural entrepreneurs and Random Effects in any of these countries report as obstacles The econometric techniques used to estimate the food safety regulations, regulations on land various models address two characteristics of RIC use, customs regulations, or work permits databases: (i) observations have unequal sampling for expatriates, to mention a few. weights, and (ii) observations are clustered in com- 8. Complaints by entrepreneurs about rural munities. The question arises how much difference infrastructure do in fact reflect conditions it makes to account for weights and random effects that register in the community surveys. in the actual empirical analysis of RIC data. Sec- Where improvements in infrastructure are tion 4 of Annex I considers representative exam- reported at the community level, fewer com- ples of enterprise performance, entrepreneurship plaints about infrastructure appear in the selection, and EICO models estimated in alterna- surveys. In communities reporting a higher tive ways, using data from Nicaragua and Sri Lanka. number of households with access to pro- In particular, the effect of benchmark indicators tected water sources, for example, entrepre- and community characteristics is sensitive to the neurs complain less about water. The same speci�cation of the model with regard to applica- result was found with respect to electricity, tion of sampling weights and use of random effects. telecommunication, and postal services. Overall, the evidence suggests that accounting for 9. Enterprise performance; household choices weights should be mandatory and that accounting about entrepreneurship, farming, and wage for clustering is strongly advisable. employment; and perceptions about invest- ment climate obstacles are all affected by the community’s overall economic environ- Use of Community Variables ment. Observed community characteristics The community forms a core element in RICS. (including benchmarks) matter, but apart Selected as the lowest unit of public administra- from these, regression analysis shows that tion, with public services that perform public unmeasured community factors are highly duties, it is also considered to be a basic level of influential in causing similarities in choices economic organization, meaning it shows basic and behavior within a community. economic infrastructure, interaction between 10. Many rural enterprises are not registered: enterprises, and public functions that affect busi- 70, 42, and 80 percent in Nicaragua, Sri nesses. Hence, at this level, investment climate has 68 The Rural Investment Climate unique characteristics that distinguish one com- other cases, a community may lack basic economic munity from another in attractiveness for private infrastructure and public and business services investment and business operation. Because many readily available in neighboring communities. community characteristics can be measured, the On the basis of these observations it can be con- resulting descriptors can be used as exogenous cluded that, by nature, measurement of commu- variables in explaining enterprise performance nity economic characteristics is difficult and the and entrepreneurial choice, assessing investment descriptors to some extent are imprecise and de�- climate outcomes, and explaining outcomes of cient. Some of the variables measured may have community-level economic activity. The informa- heterogeneous effects, and, as already noted, many tion can also be condensed in robust community- variables correlate and interrelate in complex level benchmark indicators, a technique utilized in ways. Moreover, several important variables the previous chapters. remain by necessity unobserved or insufficiently It sounds simple, but in reality there are some measured. When used in regressions, therefore, complicating factors. First, communities exhibit variables are likely to present problems of significant differences in size and, therefore, in multicollinearity and effects from unobserved available services and economic functions. Obvi- variables. ously, the incidence of certain services in a com- Does this represent a weakness in RICS as com- munity increases with population size. If one pared to PICS? No. PICS makes no systematic community consists of five hamlets and another attempt to measure community-level variables, community of ten of similar size, the chance of leaving most of them unobserved and their influ- �nding public infrastructure, public services, and ence unassessed. Variables that go unmeasured business services will be higher in the second com- don’t create problems of multicollinearity. Some of munity. This does not necessarily imply that the the statistical problems encountered in RICS are, second community will offer a better investment in other words, to an extent problems of “luxury� climate, however. rather than “poverty.� It is better to have data with Second, communities differ in character. Some some weaknesses than to have no data. More communities are simply larger hamlets, others are importantly, the present study clearly shows the a cluster of hamlets, and some can be rural market statistical importance of community variables. In towns. This kind of variation can be helpful in rural space, local conditions can differ widely, explaining differences in enterprise activity. Some hence the need to measure their effect accurately. of these differences may be easily influenced by But, having said that, the challenge faced by the policies and public and private investment, but RICS team and country team managers is to design others may reflect geographic features or large and implement surveys in ways that circumvent past investments that are a given fact. avoidable statistical problems and optimize Third, even though communities are considered chances for reliable and useful results. to be the basic units of administration, functions may be performed at lower levels of aggregation International Comparison as well, for example, when some duties are International comparison of benchmark indicators charged to village headmen or councils. Many and their components is imprecise because of two administrative functions, moreover, are per- main factors. First, sometimes major differences formed not at the community level but at higher can exist between countries in their systems of levels of local administration, such as the subdis- public administration, the average size of commu- trict, district, or province level. nities and the functions performed at the commu- Fourth, significant heterogeneity in access to nity level. Second, the concept rural has no infrastructure services can exist within a commu- common international definition, and each coun- nity. Even if a community has a concrete or asphalt try either employs its own definition or simply road, many residents may still have major prob- designates an inventory of jurisdictions and lems getting to that road because of impassable administrative zones as rural or urban. Therefore, feeder roads, lack of a bridge, or other obstacles. the selection of communities for RIC assessments Even in communities with electricity or fixed may likely extend beyond those the host country phone lines, many residents may still be deprived classi�es as rural. Indeed, it will likely prove nec- of such services due to incomplete coverage. In essary for RICS task managers to sample both in Conclusions and Recommendations 69 Box 7.1 Some Options for Exploring Improvement of Benchmark Indicators The purpose of benchmark indicators is to condense information on the investment climate at the community and country level for comparison and analysis. Further optimization of de�nitions and empirical performance should focus on the following: 1. Overlap in de�nition between the benchmark indicators increases intercorrelation and reduces the effectiveness of both in the regressions. Especially the connectivity index and the infrastructure services index seem to have too much overlap. 2. For comparison and good statistical performance in regressions, benchmark indicators and their subcomponents should show suf�cient variation. Options can be explored to increase variation by restructuring and rescaling subindicators. In particular, the governance and human capital indexes deserve attention. Some benchmark components relate closely to community size, which is not necessarily meaningful from an enterprise perspective. Analysis should guide the rescaling of these components. 3. Some determinants of economic outcomes can be changed by policy and investment efforts, others hardly or not at all. Options should be considered on how to distinguish such factors as determinants of economic structure on the one hand and investment climate obstacles on the other. The PICS provides no guidance since it has no focus on spatial diversity; but for rural areas, RICS should consider this, as infrastructural policies and investment programs are often region speci�c. 4. From a users’ perspective, economic policy obviously involves a trade-off between robust generality of benchmark indicators and ability to identify more speci�c weaknesses in the business environment. 5. It is worthwhile to include a few more indicators while narrowing the scope of each of them. This should lead to a smaller within-community statistical variation among the subindicators while increasing the variation of indicators between communities. Source: The Authors. rural market towns and in peri-urban communi- the incidence of business services and roads. ties, regardless of how they are classi�ed. Improved standardization could mitigate this. In addition, adjustments in de�nitions and scal- Moreover, scope remains for sharpening and ing are likely to affect country rankings.63 optimizing BI conceptualization and measure- ment. Rather than making ad hoc adjustments, the Experience with Benchmark Indicators work should be guided by statistical analysis. The present estimation models can be used to test Just as with other large surveys carried out by the the effect of changes in benchmark indicator World Bank Group, including Governance, Cost of definitions, and adding other countries to the Doing Business, and the many PICS, benchmark sample—Bangladesh, Benin, Ethiopia, Indonesia, indicators from the RIC assessments are needed and Pakistan—will greatly enhance the effective- for cross-country comparison and analysis. This ness of this work. Suggestions are summarized study’s experience using benchmark indicators in in Box 7.1. the regressions is encouraging. In binary correla- tions they behaved as expected from a theoretical Questionnaires point of view, although some weaknesses remain to be addressed. 1. Experience in the RIC pilots shows that more The correlation between benchmark indicators, data were collected than were actually used in part stemming from overlap, could be reduced in the analysis. This is understandable and the scaling of some components improved. because of the three surveys’ pilot character. A particular flaw is that several benchmark indi- Questionnaire length comes at a cost, how- cators increase with community size, because ever, not only in monetary terms, due to the some underlying variables likely relate positively amount of time needed to conduct the sur- to population size and land area, for example, veys and process the data, but also in loss of 70 The Rural Investment Climate quality, due to nonresponse and interview enterprises and is likely to improve the quality fatigue among interviewees. Some efforts of regression results. have already been made to reduce the stan- 3. Similar arguments apply in the selection of dard questionnaire included in the Imple- communities. Community variation is desir- mentation Manual (World Bank 2007c). able. In countries with many more or less Based on accumulated experience, however, similar communities—typical rural commu- a careful revision that leads to a substantially nities—a strati�ed sampling of communities shortened questionnaire is recommended. is recommended that undersamples the 2. Further questionnaire standardization should homogeneous group. This adds to the impor- be promoted. Country teams were generally tance of a good sampling system and inclined to include excessive numbers of weights. country-specific variations in their survey 4. Given the targeted number of observations in designs. These additions, however, resulted a survey of, say, 1,000 enterprises, trade-offs mainly in data that were not even subse- arise between the number of communities quently processed and analyzed, while and the number of observations per commu- unintentionally diluting the pilot’s RIC focus nity. Increasing the number of communities and reducing its utility for inter-country is favorable because it lifts constraints on the comparisons. number of community variables that can be 3. Considerably more care should be given to used in regressions. Several benchmark com- measuring labor inputs. Labor is one of the ponents and other community-level variables most important inputs in the rural economy are aggregates of household and enterprise and a crucial variable in many analyses. The data, however. If there are few observations enumeration of labor input in the Sri Lankan per community, the quality of these aggre- and Nicaragua pilots, particularly, had some gates declines and the justi�cation for employ- deficiencies that undoubtedly affected the ing them as independent variables vanishes. analysis. Unpaid family labor, for example, Therefore, for each country, optimizing the could have been measured properly using number of communities and the number of information from the household database if observations per community deserves special the household and enterprise questionnaires attention. The optimal choice of the number had been better linked. Good labor enumera- of communities and the number of observa- tion deserves priority. tions per community is a conceptual matter that should be resolved by a statistical spe- cialist (who should also address the conse- Survey Design quences of varying the number of observations 1. Stratified sampling is indispensable for per community). In particular, it is unclear drawing a cost-effective sample. This study whether the trade-off implemented in the has shown that weights are absolutely neces- pilot RICS or subsequent RIC implementa- sary in the analysis for obtaining unbiased tions has been optimal. results.64 Therefore major effort should be 5. The number of observations per community made in the design and implementation of is small in many cases, which decreases the surveys to establish a proper sampling reliability of aggregates calculated from indi- framework and to collect reliable weights. vidual observations. To the degree that this 2. Because rural space contains so many small is simply caused by the small number of and so few large enterprises, further strati�- observations in the sampling frame (that is, cation by enterprise size may make sense. If the community is just very small), this is an large enterprises appear in a community’s unavoidable situation, but if instead the sam- sampling frame, they should be sampled pling frame contains an adequate number of with a higher probability than the small observations, the likely cause of the problem enterprises, with sampling weights adjusted is nonresponse. As is well-known, nonre- for the differential sampling probabilities. sponse already causes other statistical com- This effectively increases the amount of plexities, such as reduced randomization, variation in, or the heterogeneity among, but the RICS methodology needs aggregates Conclusions and Recommendations 71 calculated from a sufficient number of indi- of self-selection and simultaneity, and interaction vidual observations. High priority should between EICO and EICI variables, on the one therefore be given to reducing nonresponse. hand, and enterprise performance and household choices, on the other hand, remain to be explored (see Section 5.5). Survey Implementation and Processing Experience shows that the cost-effectiveness of RIC Use of RICS for Operational Work surveys can greatly be improved by the following: The purpose of conducting RICS is to support pol- 1. Improved quality of data collection and pro- icy dialogue, to sharpen development policies, and cessing. Poor quality data collection and pro- to improve the effectiveness of interventions. In Sri cessing has major effect on the outcomes and Lanka, the RIC has clearly contributed to this.65 ef�ciency of RIC work. It limits potential out- (See Box 7.2.) put and the precision of findings. It also The RIC results can contribute in several ways increases the cost of analysis, since the to preparing policy interventions: amount of time required is much greater than for good-quality data sets. If fewer 1. The RIC survey results provide much other- records must be discarded because of data- wise unavailable information about the eco- quality problems, sample sizes can be reduced nomic activities of rural households; the somewhat or coverage increased. population of local enterprises; and institu- 2. Related to this point and as recommended in tional, policy, and infrastructural constraints. the Implementation Manual, another objec- This is important information for policy discus- tive should be consistent database design. sions between civil society, especially private- The three RIC databases studied in this sector stakeholders, and Government. The report follow completely different formats. quality of dialogue can be enhanced by good Great effort had to be expended to replicate RIC analysis and further assessment of options. the analysis of one country with data from 2. The various regressions performed— another. Inconsistency leads to substantial enterprise performance and entrepreneur- delays in research output and discourages ship choices—show that from micro- cross-country analysis. perspectives various factors constrain productivity, employment, investment, and income earning. Many of these constraining Data Analysis factors can be changed—some more easily Based on this study’s experiences with estimation, and others—raising challenges for policy country teams are recommended to include in their intervention. These regressions provide sug- studies the following analyses, in order of priority: gestions that will always need further work, however, rather than ready-to-use policy 1. overviews of household data, enterprise recommendations. data, and community data; 3. In general, the RIC analysis as outlined in this 2. calculation of the benchmark indicators; report can and should be followed by further 3. analysis of enterprise performance; exploration of the RIC databases with regard 4. analysis of household economic choices, to speci�city of constraints, locations, groups including enterprise start-up; of enterprises affected, and other concerns. 5. analysis at the community level of rural non- RIC surveys create multipurpose databases farm outcomes, such as enterprise density, having a public goods character that can be nonfarm income, and nonfarm employment; used for several years to come as sources 6. analysis of EICOs; and of information to feed policy and project 7. analysis of EICIs. design. Moreover, additional information can The econometric models and Stata programs pro- be obtained from other sources and through vided in Annex I facilitate items 3, 4, 6, and 7. At targeted interviews and small-scale follow- the same time, just as the RICS pilots represent a up surveys. In fact, information from RIC process of learning by doing, these models remain databases will be helpful when designing a work in progress: they do not yet address issues better tailor-made surveys of rural space. 72 The Rural Investment Climate Box 7.2 Operational Applications: The Sri Lanka RIC Pilot An assessment of IC constraints faced by RNFEs was conducted in the Sri Lanka pilot using both subjec- tive and objective criteria. The ranking of RIC constraints in Sri Lanka, whether determined by objective measures or from subjective responses, differs in several important aspects from constraints experi- enced by the larger urban-based enterprises. Access to road transport and markets, reliable public utilities (electric power), and access to �nance proved to be the main factors influencing (or constrain- ing) employment growth and total factor productivity in the rural areas. Urban �rms did not experience the same level of dif�culty in accessing infrastructure and �nance; instead, they faced governance issues and were far more likely to be negatively affected by inconsistent macroeconomic policymaking. While it is dif�cult to gauge the full policy effect of the Sri Lanka RIC, the �ndings have already proven useful for the WBG’s country operations there. The �ndings were woven into the Development Policy Review (DPR) and the Development Forum Paper for the 2007 Development Forum. The �ndings were also cited by leading politicians and civil servants at various forums in Sri Lanka. The South Asia Region has also applied the �ndings to the design of operations in the proposed public �nancial management and statistical institutional building project as well as the legal and judicial follow-up project. The RIC analysis has also been used by other international agencies and by Sri Lankan think tanks and universities. RIC �ndings were used by the Asia Foundation for their operations designed to improve provincial level investment climates. The Asian Development Bank, which coauthored the Sri Lanka RIC Assessment with the Bank, has relied on the analysis to inform its micro�nance project. Government is drawing on the data on missing infrastructure to develop a PPP capacity for large infrastructure projects. Bank staff also have used the materials to teach courses at the University of Colombo, the Moratuwa MBA in infrastructure, and the University of Sri Jayawardenpura. It is generally agreed that the Sri Lanka RIC did not result in a breakthrough on the frontier of invest- ment climate studies. Rather it has proved useful as a practitioner’s tool. The assessment quanti�ed many of the issues already under discussion and allowed a prioritization and focus on the most important critical areas. It managed to strengthen the evidence base for policy formulation and allowed stakeholders to speak with authority on particular issues covered by the study, such as the losses from electricity outages or the cost of labor regulations. Source: Author communication with Ismail Radwan 2008. 4. RICAs should only be conducted in countries Bangladesh, Benin, Ethiopia, Indonesia, and where Government, relevant donors, and Pakistan. This will add to the understanding Bank operations strongly demand informa- of similarities and variance in rural eco- tion on rural enterprises. RICAs should be nomies and will help sharpen the analytical part of regular country programs and should methods applied. have strong potential for use in lending oper- 2. Relationships between the perceived invest- ations or analytical work for PRSPs, CASs, ment climate, the actual conditions measured and sector work. Moreover, a team with the through the community questionnaire, enter- right qualifications and sufficient budget prise performance, and entrepreneurship should be available to carry out the RICS. choice can be developed further. RIC2 relates actual conditions to perceptions, enterprise performance, and entrepreneurship choice; FURTHER RESEARCH future studies should relate perceptions to enterprise performance and entrepreneur- Many potential future research topics present ship choice, with careful attention to bidirec- themselves. tional causality and self-selection issues. 1. The analytical tools applied in this report 3. In a descriptive manner, perceived invest- deserve to be applied with priority to other ment climate obstacles (termed EICOs in this available data sets such as those for report) are ranked in each country, allowing Conclusions and Recommendations 73 Figure 7.1 Level of Seasonal Activity Among Enterprises in Tanzania A: Production (n = 204) B: Services (n = 228) 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% Very high High Average Low Very low C: Trade (n = 553) D: Mixed (n = 181) 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% Very high High Average Low Very low Source: RIC Surveys. comparisons among countries in terms of rural space, but spatial implications have not their ranked EICOs. Descriptive differences yet been explored at all. Spatial regression are interesting, but a statistical test should be methods could exploit correlations among developed that indicates whether differences nearby communities that are widely ignored in ranking are indeed statistically signi�cant. in applied research to date. 4. The design of the questionnaires makes the 7. Because questionnaires, and therefore data, analysis of EICOs very inviting, but aspects were not comparable between the countries, of interactions with government and society the deeper details on labor input for Tanzania (termed EICIs) are more dif�cult to summa- were not fully utilized. Additional analysis rize and analyze. The RICS data offers far with adjusted labor variables is worth trying. more than could be covered here. In particu- 8. Entrepreneurs were asked about business lar, EICIs might be benchmarked as well. seasonality. Figure 7.1 illustrates its impor- 5. Selective comparative studies of RICS and tance for Tanzania, but this is only the very PICS will originate a better understanding of beginning of a description. How does the the rural-urban continuum and rural-urban cycle compare with the seasonal cycle in agri- divides, provided suitable databases can be culture? Does the degree of seasonality identi�ed. reported by entrepreneurs vary with the 6. Communities in the RICS databases are market served (that is, location of customers), coded with geographic longitude-latitude the type of customers, or the location of of information. This fixes them somewhere in input sources? Annex A. Description of Data Description of variables and enterprises. The RIC analyzing sales, enterprises must have sales at survey procedures and methodology are described least equal to US$100 or a net value added at least World Bank (2006). In total data on about 1,350 en- US$60; when analyzing net value added, enter- terprises were collected for each of Nicaragua, Sri prises are included only if net value added is at Lanka, and Tanzania (Table A.1). For various rea- least US$60.69 As indicated in Table A.1, this sons, however, smaller numbers of enterprise sam- restriction had the greatest effect on Tanzania. As ples were included in the analysis. For Nicaragua, a result, for all reasons together, the number the household-based enterprises may not belong to of enterprises in the analyzed samples is about different households. Some households reported 40 percent less than that in the overall database. on as many as four enterprises, but since sales and expenditures were reported only at the household Data availability. From the data collected in the level, multiple household-based enterprises were RIC surveys, variables are taken and other variables aggregated to the household level, leading to a re- are calculated for analysis of enterprise performance. duction of 308 in the enterprise count. For some Table A.2 provides an overview of data grouped communities in Sri Lanka, essential variables were as �nancial data, other enterprise characteristics, incomplete, leading to a signi�cant loss in the and community characteristics. Community-based number of observations.66 In Nicaragua, a data- benchmark indicators are calculated mainly from entry failure to record the community identi�ca- the community questionnaires and partly from ag- tion number led to a loss of 247 enterprises. In a few gregated household and enterprise data. cases in all three countries, explanatory enterprise- Reliability of verbally reported costs, sales, vol- level variables were missing. A trade-off clearly oc- umes, and prices can be weak; most enterprises curs between the number of variables speci�ed in have no bookkeeping, and reporting can be affected the regression models and the number of enter- by the limitations of human memory, estimation prises available for the regression analysis. More errors, and unwillingness to fully disclose �nancial elaborate speci�cation implies an increased need to data. On the other hand, most small enterprises drop some records because of missing variables. have simple, fairly simple to remember activities. One type of sample reduction is deliberate. A weakness in the data concerns the measure- Some of the enterprises for which sufficient data ment of labor, for which de�nitions differ between are available reported zero sales, and in other countries. Moreover, assessment of part-time labor cases the reported calculated net value added was and employment of household labor is imprecise, very small or even negative. This may happen especially for Sri Lanka (see Box A.1 for details). when nonperforming enterprises are temporarily Thus, despite its possible flaws, a stock rather than closed, have just opened, or are about to exit. It a flow concept is used. The value of paid labor may also represent self-employment on a very input was recorded in the data as the actual paid small scale.67 This data problem may also occur wage sum, a flow concept, but the stock is not when reference-period inputs yield future rather known because the number of part-time workers is than reference-period output, and similarly, when not known exactly. Quality differences of paid reference-period output results from previously labor are presumably reflected in the wage paid per purchased inputs.68 Because of the potentially dis- person. For unpaid household members, no infor- rupting role of this group, a threshold is adopted mation is available about quality of labor; for the for inclusion in the regression analysis: when manager, the number of years of formal education 75 76 The Rural Investment Climate Table A.1 Enterprises Included in the RIC Surveys and the Enterprise Performance Regressions Nicaragua Sri Lanka Tanzania Net Net Net Value Value Value Regression Sales Added Sales Added Sales Added Number of enterprises 1535* 1327 1239 Enterprises dropped because of 667* 278 134 358 insuf�cient data Maximum used in regressions 868 1049 1105 881 Enterprises with reported 22 54 31 208 158 182 sales 100 and/or NVA 60 US$ Remaining enterprises 846 814 1018 841 947 699 Source: RIC Surveys. * includes multiple enterprises owned by one household that were aggregated. and training is available for Sri Lanka and Tanza- One way of dealing with uncertainties about the nia but not for Nicaragua. reliability of imputed labor and capital costs is to For capital, understandable dif�culties arise in specify imputed and paid labor and capital costs separating capital such as a house, equipment, and as separate variables. This allows the estimation of vehicles into use for consumption and use for pro- different elasticities for paid and imputed inputs. ductive purposes. Also, market prices for valua- Parts of the data sets show deficiencies in the tion of assets are not always known, and some collection and processing of raw data. In the selec- liabilities may have been obtained from relatives tion of the enterprise records for the regression and friends at concessionary prices. analysis, interpretations must be made about how Underemployment of household labor may be a to handle missing data, possible data errors, and serious problem, and hence the marginal produc- mistakes in coding. tivity of household labor may be lower than that of In conclusion, it is fair to say that errors, in part paid labor. To some extent, that may also be the the result of the surveys’ pilot character, negatively case with household-owned assets and borrowed affect the precision and efficiency of parameters cash and assets. estimated in the regressions. Table A.2 De�nition of Variables Variable Name STATA Name Description Variables Used in the Enterprise Performance Regressions Recorded and calculated �nancial data Sales Sale Total Sales (US $) Net value added NVA NVA GVA Depreciation (US $) Total labor input1 VLI Value of labor input paid wages HLIV (US $) Total capital input VCI Value of Capital input ICAP PCC (US $) lnVLL (lnL*lnL) lnVLL lnVLI * lnVLI quadratic lnL term lnVLC (lnL*lnC) lnVLC lnVLI * lnVLC interaction lnL lnC term lnVCC (lnC*lnC) lnVCC lnVCI * lnVCI quadratic lnC term LnIL*lnIL (imputed) lL11 lnHLIV * lnHLIV quadratic imputed lnL term Annex A. Description of Data 77 Table A.2 De�nition of Variables (continued) Variable Name STATA Name Description Variables Used in the Enterprise Performance Regressions Recorded and calculated �nancial data LnPL*lnPL (paid) lL22 lnPLC * LnPLC quadratic paid lnL term lnIC*lnIC (imputed) lC11 lnICAP * lnICAP quadratic imputed lnC term lnPC*lnPC (paid) lC22 lnPCC * lnPCC quadratic paid lnC term lnIL*lnIC (imputed) lL1C1 lnHLIV * lnICAP interaction imputed lnL lnC term lnPL*lnPC (paid) lL2C2 lnPLC * lnPCC Interaction paid lnL lnC term Depreciation depr (imputed) 8% percent of the value of Fixed Assets (excluding land)2 (US $) Nonfactor cost NFC Purchase of materials and items for resale; Purchase of electricity, water, gas, and fuels; Transport costs; Cost for telephone, mobile phone, fax, internet, postal and insurance service; Business services (US $) Family labor input HLIV Household labor imputed value family labor cost at imputed wage rate3 (US $) Paid labor input PLC Paid wages (US $) Imputed capital cost ICAP Imputed capital cost net value of assets * imputed interest rates4, 5 (US $) Paid capital costs PCC Paid Capital Costs rental payments for land, equipment, buildings, etc. (US $) Other enterprise characteristics Age of enterprise �rmage �rmage Year of survey year enterprise established Experience of manager exptmng Years of experience reported for manager Gender of manager sextmng Dummy variable for male manager (1 male) Education of manager edutmng Education of manager6 (years) Registration regis Registration status of enterprise (1 registered) Industry dummies/line of business Trade enterprise B1 Dummy (1 Retail or wholesale trading) Services enterprise B2 Dummy (1 Services) Manufacturing, nonagricultural, B3 Dummy (1 Manufacturing nonagricultural goods) enterprise Agricultural processing enterprise B4 Dummy (1 Processing of agriculture, hunting, �shing products) Mining enterprise B5 Dummy (1 Mining and quarrying) Production B345 Dummy (1 Production [combined B3, B4, B5]) Other production B567 Dummy (1 Other production [combined mining, construction, gas water electricity]) Mixed enterprise B8 Dummy (1 Enterprise including at least two lines of business) Community characteristics Agricultural seasonality agseas Average of the standard deviation in monthly agricultural labor input for male and female labor Enterprise density entdense_pop Number of enterprises in community per 1,000 people Community population popnw Number of inhabitants in the community Paddy land per capita apaddylpc Area of paddy land per capita (acre or hectare) Illiteracy shilliterate Share of illiterates (%) Main market in Market in which households buy/sell most of their goods Own community mainmkt1 Dummy (1 Buying/selling in own community) Neighboring communities mainmkt2 Dummy (1 Buying/selling in neighboring communities) Commercial center mainmkt3 Dummy (1 Buying/selling in commercial center) Nearest city mainmkt4 Dummy (1 Buying/selling in nearest city) Main income from Agriculture insoura Dummy (1 Main income source community is agriculture) Wages insourw Dummy (1 Main income source community is wages/salaries) Self-employment insourn Dummy (1 Main income source community is self-employment) (continued on next page) 78 The Rural Investment Climate Table A.2 De�nition of Variables (continued) Variable Name STATA Name Description Variables Used in the Enterprise Performance Regressions Benchmark indicators and components Connectivity conn Connectivity index (index %) Infrastructure services access Access to infrastructure services (index %) Business services devserv Business services index (index %) Governance govern Governance and corruption index (index %) Human capital humcap Human capital index (index %) Finance services �nan Finance services index (index %) Inverse time to near city c1 Inverse time taken by main means of transportation to nearest major city (index %) Inverse cost of transportation c2 Inverse cost of transportation to the nearest major city (index %) Inverse time to main market c3 Inverse time taken by main means of transportation to main market (index %) Proximity post of�ce c5 Proximity to the main post of�ce (index %) Percent with electricity a1 Percentage of households with electricity (index %) Availability of electricity a2 Dummy (1 Electricity availability) Percent with protected water a3 Percentage of households with access to protected water Percent with �xed phone a4 Percentage of households with �xed-line telephone Percent with cellular phone a4B Percentage of households with access to cellular phones Sewage system a5 Dummy (1 Sewage system in the community) Garbage collection a6 Dummy (1 garbage collection or disposal service in the community) Concrete/asphalt road a7 Dummy (1 Most common road surface is concrete or asphalt) Engineering service d1 Dummy (1 Engineering services available for businesses in the community) Management consulting d2 Dummy (1 Management consulting services available for businesses in the community) Marketing service d3 Dummy (1 Marketing services available for businesses in the community) Accounting service d4 Dummy (1 Accounting services available for businesses in the community) Insurance service d6 Dummy (1 Insurance services available for businesses in the community) Information technology d7 Dummy (1 Information technology services available for businesses in the community) Infrastr and services go2 Infrastructure and services (index %) Public services/institutions go3 Dealing with government services and general policy and institutional constraints (combined indexes) (index %) Public services go3a Dealing with government services (index %) Public institutions go3b General policy and institutional constraints (index %) Rule of law go4 Rule of law (index %) Number of banks �1 Number of formal �nancial sources weighed by mean of the distance (index %) Number bank services �2 Number of formal �nance services weighed by mean of the distance (index %) Access to loans �3 Access to loans (index %) Other Variables Gross value added GVA GVA Sale NFC (US $) Total factor input TFI Total Factor Input Value of capital input (VCI) Value of labor input (VLI) (US $) Total cost TC Total Cost Total Factor Input Non Factor Input Depreciation (US $) Net pro�t netp Net Pro�t Sales Total Cost (US $) Pro�tability or total productivity SATC Pro�tability Sales / Total Cost (ratio) Net factor productivity NVFC Net Factor Productivity NVA / TFI (ratio) Net income ninc Net Income sale NFC PCC depr PLC (US $) Total liability debt Loans or credits from banks, private lenders and from suppliers; Money owed to friends and relatives, and others (leasing companies, etc.) (US $) Total assets totA Total Assets5 Fixed assets Inventories and stocks Financial assets (US $) Annex A. Description of Data 79 Table A.2 De�nition of Variables (continued) Variable Name STATA Name Description Variables Used in the Enterprise Performance Regressions Other Variables Net value of assets Netass Net value of assets value of assets liabilities (US $) Employment E E VLI/imputed daily wage rate (proxy for number of persons) Imputed factor costs IFC IFC HLIV ICAP (US $) Pro�tability of own resources NIIFC NIIFC ninc/IFC (ratio) Bank loan > 50 % of debt formloan Dummy that has a value of one if 50% or more of the total liability is from a Bank Stand-alone enterprise standalone Dummy (1 An enterprise not located at the house lot of the owner) Enterprise with paid labor PLC01 Dummy (1 Enterprise with hired labor) Enterprise with debt debt01 Dummy (1 Enterprise with some liabilities 0) Ethnicity of manager ethtmng Dummy (1 Manager is Sinhalese Sri Lanka only) Source: The Authors. Notes: 1. Paid labor is speci�ed as a cost. For household labor, ambiguities about the coverage of household labor are speci�ed in Box A.1. 2. Depreciation was imputed for estimation of net pro�t and net value added. In all countries (an arbitrary) 8 percent depreciation was adopted on Storage facilities (separate from the building); Machinery and equipment (excluding vehicles); Vehicles for transportation; Specialized vehicles; and Other �xed assets. Land was excluded from depreciation. 3. Imputed wage rates (opportunity cost for unpaid labor) chosen after comparing data from the enterprise and community surveys. Paid wages and numbers employed were calculated for groups of registered (formal) and unregistered (informal) enterprises with mainly full-time labor. Estimates that include enterprises with part-time labor will be biased. Registered enterprises in some cases pay high wages, which may not be accessible for most fam- ily labor. More relevant may be wages paid in unregistered enterprises. Wage rates reported from community surveys tend to reflect wage levels in reg- istered enterprises. The following daily imputed wage rates were adopted: Nicaragua US$0.80, Sri Lanka US$1.30, Tanzania US$1.00. 4. Imputed interest rates were chosen after comparing interest paid on loans from formal enterprises and relevant rates reported in IMF �nancial tables and Central Banks. The imputed interest rates for land were corrected for inflation because in most cases, in the long term, land value will go up with inflation. The following imputed rates were chosen: Land Other Assets Nicaragua 10.58% 18.78% Sri Lanka 7.5% 15% Tanzania 12.9% 17% 5. Total assets A. Fixed Assets Land; Building and improvement in leasehold (excluding storage facilities); Storage facilities (separate from the building); Machinery and equipment (excluding vehicles); Vehicles for transportation; Specialized vehicles; Other �xed assets. B. Inventories and stocks Finished goods and commodities for sale; Work in progress; Raw materials excluding fuel; Fuel; Acquisition Cost: Machinery and equipment (including transport). C. Financial Assets Accounts receivable; Cash in hand/in bank. 6. Education Nicaragua: Not applicable: education of manager was not recorded in the survey. Sri Lanka: 0–15 years. Tanzania: Codes were translated into years of schooling: Primary education: Under standard one 0; Standard one 1; Standard two 2; Standard three 3; Standard four 4; Standard �ve 5; Standard six 6; Standard seven 7; Standard eight 8; Adult education 8; Training after pri- mary education 9. Secondary education: Pre form one 9; Form one 10; Form two 11; Form three 12; Form four 13; Form �ve 14; Form six 15. Training after secondary education 16; University & other tertiary education 17. 80 The Rural Investment Climate Box A.1 Speci�cations of Labor Input in Enterprise Questionnaires The questionnaires differ widely in their descriptions of enterprise labor, and consequently possibili- ties for estimating labor input differ widely as well, as shown in these notes on the questionnaire de�nitions. Tanzania has by far the best description of labor input, although it remains unclear whether vari- ous household membersare unpaid or receive a salary. For Nicaragua, data about household labor and other employees in household enterprises are ad- equate. Data for stand-alone enterprises are ambiguous. A root of the problem is the overlooked pos- sibility that stand-alone enterprises are household enterprises. Some questions skip possible relatives of the manager. Probably the manager is not included as an employee. It is possible that family mem- bers and the manager receive salaries. The categories listed ignore part-time temporary labor and assume that permanent labor is either full-time or part-time; this is an unrealistic assumption for sea- sonal enterprises. The questions in the Sri Lanka survey are ambiguous and leave much room for diverging interpre- tations. It is not clear whether the manager is part-time or full-time. Only family members working full-time in the enterprise are listed. In some questions it is not clear whether household labor is in- cluded; in others household labor is included, but the manager may or may not be included. Also unclear is how to aggregate the different categories of labor. Source: RIC enterprise questionnaires. Annex B. Data Used for Enterprise Performance Analyses Table B.1 Nicaragua: Selected Variables of Enterprises and Communities; Value and Dstribution Variable name mean st dev min max obs pct20 pct40 pct60 pct80 Sales# 7.004 1.162 4.193 11.542 846 6.115 6.625 7.223 7.966 Net value added# 6.823 1.260 0.000 11.389 820 5.902 6.521 7.093 7.846 Total labor input# 5.627 1.071 0.000 11.201 846 4.908 5.355 5.820 6.421 Value of capital input# 4.893 1.642 0.000 10.598 846 3.567 4.479 5.277 6.192 lnVLL (ln*ln) 32.816 12.096 0.000 125.460 846 24.091 28.671 33.873 41.230 lnVLC (ln*ln) 28.081 12.461 0.000 101.309 846 18.422 24.275 29.834 36.668 lnVCC (ln*ln) 26.636 16.806 0.000 112.314 846 12.723 20.057 27.847 38.342 Depreciation# 2.424 1.872 0.000 9.291 846 0.695 1.613 2.795 4.029 Nonfactor cost# 3.350 2.520 0.000 11.100 846 0.000 3.095 4.464 5.578 Other enterprise characteristics Age of enterprise# 2.090 0.912 0.000 4.143 846 1.253 1.946 2.398 2.890 Registration 0.298 0.457 0.000 1.000 846 0.000 0.000 0.000 1.000 Industry dummies/line of business Trade enterprise 0.396 0.489 0.000 1.000 846 0.000 0.000 0.000 1.000 Services enterprise 0.188 0.390 0.000 1.000 846 0.000 0.000 0.000 0.000 Manufacturing, nonagricultural, enterprise 0.084 0.278 0.000 1.000 846 0.000 0.000 0.000 0.000 Agricultual processing enterprise 0.139 0.346 0.000 1.000 846 0.000 0.000 0.000 0.000 Other production enterprise 0.032 0.175 0.000 1.000 846 0.000 0.000 0.000 0.000 Mixed enterprise 0.161 0.368 0.000 1.000 846 0.000 0.000 0.000 0.000 Parent operating NFE 0.386 0.487 0.000 1.000 846 0.000 0.000 0.000 1.000 Parent was manager 0.002 0.040 0.000 1.000 846 0.000 0.000 0.000 0.000 Parent occupation. Missing 0.074 0.262 0.000 1.000 846 0.000 0.000 0.000 0.000 Community characteristics Agricultual seasonality 0.684 0.411 0.000 1.477 846 0.389 0.522 0.866 0.996 Enterprise density 49.915 93.972 0.538 445.000 846 2.749 4.067 11.917 43.695 Community population size# 8.785 1.862 5.303 11.775 846 6.553 8.216 9.893 10.528 Agricultural land per capita 0.422 0.707 0.000 4.762 640 0.050 0.094 0.275 0.667 Illiteracy 80.861 14.768 0.000 99.000 740 72.500 80.000 85.000 92.500 Benchmark indicators and components Connectivity 0.358 0.177 0.028 0.828 846 0.178 0.317 0.413 0.511 Infrastructure services 0.551 0.238 0.000 0.983 846 0.357 0.499 0.600 0.814 Business services 0.303 0.429 0.000 1.000 846 0.000 0.000 0.143 1.000 Governance 0.691 0.111 0.372 0.975 846 0.581 0.669 0.721 0.780 Human capital 0.216 0.074 0.067 0.402 846 0.144 0.187 0.233 0.289 Finance services 0.201 0.267 0.000 1.000 846 0.000 0.000 0.217 0.494 Source: RIC Surveys. Note: See Annex A for de�nitions and description of data; # logarithmic value. 81 82 The Rural Investment Climate Table B.2 Nicaragua: Selected Variables of Enterprises and Communities; Value and Distribution Variable name mean st dev min max obs pct20 pct40 pct60 pct80 Nonfactor cost 374 2388 0 66188 846 0 21 86 264 Sales 2438 5980 65 102954 846 452 753 1369 2881 Depreciation 89 449 0 10843 846 1 4 15 55 Gross value added 2064 4765 10482 89209 846 348 678 1201 2483 Net value added 1975 4688 11752 88378 846 324 646 1140 2275 Family labor input 324 280 0 1934 846 115 195 288 493 Paid labor cost 224 1692 0 72288 846 0 0 0 0 Total labor input 548 1747 0 73195 846 134 211 336 614 Net value of assets 4110 14862 2667 221194 846 314 722 1613 3803 Imputed capital cost 581 2529 0 40046 846 34 81 173 461 Paid capital cost 39 345 0 9036 846 0 0 0 0 Value of capital input 619 2613 0 40046 846 34 87 195 488 Total factor input 1168 3451 0 81668 846 231 395 620 1112 Total cost 1630 5417 36 118003 846 290 476 792 1422 Net pro�t 808 4473 43084 49815 846 172 155 624 1506 Pro�tability 2.628 3.242 0.031 68.126 846 0.774 1.345 2.223 3.735 Net factor productivity 3.180 4.679 2.052 76.724 845 0.719 1.421 2.519 4.241 Net income 1713 3819 13619 51671 846 267 606 1123 2224 Total liability 135 987 0 19766 846 0 0 0 4 Total assets 4245 15276 0 221194 846 314 731 1663 3883 Employment 685 2184 0 91494 846 168 263 420 767 Imputed factor costs 905 2544 0 40046 846 218 379 556 927 Pro�tability of own resources 3.705 5.418 16.352 76.853 844 0.817 1.699 2.788 4.726 Age of enterprise 10.715 10.019 0.000 62.000 846 2.500 6.000 10.000 17.000 Bank loan 50% of debt 0.066 0.249 0.000 1.000 846 0.000 0.000 0.000 0.000 Stand-alone 0.074 0.262 0.000 1.000 846 0.000 0.000 0.000 0.000 Enterprise with paid labor 0.171 0.377 0.000 1.000 846 0.000 0.000 0.000 0.000 Enterprises with debt 0.212 0.409 0.000 1.000 846 0.000 0.000 0.000 1.000 Community population size 22199 30080 200 130000 846 700 3700 19800 37335 Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data. Annex B. Data Used for Enterprise Performance Analyses 83 Table B.3 Sri Lanka: Variables Used in Regressions of Enterprise Performance; Value and Distribution Variable name mean st dev min max obs pct20 pct40 pct60 pct80 Sales# 7.355 1.445 4.429 14.439 1018 6.124 6.893 7.637 8.512 Net value added# 6.207 1.634 0.000 14.144 908 4.964 5.863 6.613 7.475 Total labor input# 6.029 0.790 4.190 13.117 1018 5.745 5.745 5.746 6.517 Value of capital input# 5.493 1.532 0.000 13.674 1018 4.254 5.146 5.854 6.752 lnVLL (ln*ln) 36.974 10.753 17.553 172.064 1018 33.009 33.009 33.019 42.476 lnVLC (ln*ln) 33.696 13.003 0.000 168.937 1018 24.675 29.311 34.351 41.326 lnVCC (ln*ln) 32.522 17.288 0.000 186.987 1018 18.097 26.485 34.269 45.590 Depreciation# 4.067 1.715 0.000 13.037 1018 2.598 3.748 4.496 5.520 Non Factor Cost# 6.479 1.907 0.000 13.064 1018 5.012 6.133 6.995 7.944 Other enterprise characteristics Age of enterprise# 1.720 1.080 0.000 4.431 1018 0.693 1.386 1.946 2.773 Experience of manager# 1.941 0.946 0.000 4.043 1018 1.099 1.792 2.197 2.773 Gender of manager 0.773 0.419 0.000 1.000 1018 0.000 1.000 1.000 1.000 Education of manager 9.523 3.082 0.000 15.000 1018 8.000 10.000 11.000 12.000 Registration 0.576 0.494 0.000 1.000 1018 0.000 0.000 1.000 1.000 Industry dummies/line of business Trade enterprise 0.392 0.488 0.000 1.000 1018 0.000 0.000 0.000 1.000 Services enterprise 0.197 0.398 0.000 1.000 1018 0.000 0.000 0.000 0.000 Manufacturing, nonagricultural enterprise 0.313 0.464 0.000 1.000 1018 0.000 0.000 0.000 1.000 Agricultural processing enterprise 0.023 0.150 0.000 1.000 1018 0.000 0.000 0.000 0.000 Mining enterprise 0.005 0.072 0.000 1.000 1018 0.000 0.000 0.000 0.000 Mixed enterprise 0.069 0.254 0.000 1.000 1018 0.000 0.000 0.000 0.000 Community characteristics Agricultural seasonality 0.739 0.316 0.000 1.379 1018 0.504 0.643 0.793 0.998 Enterprise density 52.298 34.753 4.458 172.308 1018 23.182 36.496 54.293 78.604 Community pop. size# 7.273 0.510 6.326 8.631 1018 6.801 7.091 7.321 7.689 Paddy land per capita 0.111 0.153 0.000 0.917 1018 0.016 0.044 0.098 0.157 Illiteracy 0.111 0.107 0.000 0.520 1018 0.020 0.045 0.100 0.200 Main market in: Own community 0.239 0.426 0.000 1.000 1018 0.000 0.000 0.000 1.000 Neighbor communities 0.189 0.392 0.000 1.000 1018 0.000 0.000 0.000 0.000 Commercial center 0.187 0.390 0.000 1.000 1018 0.000 0.000 0.000 0.000 Nearest city 0.385 0.487 0.000 1.000 1018 0.000 0.000 0.000 1.000 Main community income from: Agriculture 0.484 0.500 0.000 1.000 1018 0.000 0.000 1.000 1.000 Wages 0.390 0.488 0.000 1.000 1018 0.000 0.000 0.000 1.000 Self-employment 0.127 0.333 0.000 1.000 1018 0.000 0.000 0.000 0.000 Benchmark indicators and components Connectivity 0.457 0.151 0.051 0.770 1018 0.317 0.440 0.497 0.581 Access 0.363 0.142 0.055 0.741 1018 0.230 0.320 0.395 0.496 Business services 0.180 0.235 0.000 1.000 1018 0.000 0.000 0.143 0.286 Governance 0.664 0.042 0.570 0.756 1018 0.628 0.657 0.678 0.702 Human capital 0.352 0.079 0.133 0.577 1018 0.281 0.341 0.377 0.416 Finance services 0.532 0.138 0.126 0.842 1018 0.406 0.498 0.563 0.658 Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value. 84 The Rural Investment Climate Table B.4 Sri Lanka: Selected Variables of Enterprises and Communities; Value and Distribution Variable name mean st dev min max obs pct20 pct40 pct60 pct80 Nonfactor cost 3768 16603 0 471421 1018 149 460 1090 2818 Sales 6588 41584 83 1864962 1018 456 984 2072 4973 Depreciation 361 6216 0 459195 1018 12 41 89 249 Gross value added 2819 33674 20933 1847887 1018 154 368 746 1792 Net value added 2458 28636 22094 1388692 1018 75 262 613 1579 Family labor input 298 141 39 1248 1018 176 312 312 312 Paid labor cost 548 6936 0 497323 1018 0 0 0 373 Total labor input 846 6932 65 497479 1018 312 312 312 676 Net value of assets 8192 100226 10 5790863 1018 470 1126 2197 5714 Imputed capital cost 1073 13931 0 868319 1018 64 151 280 743 Paid capital costs 71 708 0 31083 1018 0 0 0 37 Value of capital input 1144 13972 0 868319 1018 69 171 348 855 Total factor input 1990 19008 110 889432 1018 383 520 791 1478 Total cost 6119 33095 150 1346158 1018 674 1302 2241 4840 Net pro�t 468 14556 178701 960042 1018 679 300 68 390 Pro�tability 1.052 1.314 0.022 17.550 1018 0.536 0.771 0.948 1.166 Net factor productivity 1.293 4.548 13.199 97.699 1018 0.151 0.462 0.855 1.549 Net Income 1839 24013 92878 1387325 1018 31 188 427 1113 Total liability 439 2103 0 56985 1018 0 0 21 259 Total assets 8631 100482 5 5790863 1018 567 1259 2487 6175 Employment 651 5332 50 382676 1018 240 240 240 520 Imputed factor costs 1371 13929 106 868520 1018 339 449 640 1061 Pro�tability of own resources 2.052 7.643 21.485 357.622 1018 0.271 0.635 1.204 2.325 Education of manager 9.523 3.082 0.000 15.000 1018 8.000 10.000 11.000 12.000 Gender of manager 0.773 0.419 0.000 1.000 1018 0.000 1.000 1.000 1.000 Ethnicity of manager 0.928 0.259 0.000 1.000 1018 1.000 1.000 1.000 1.000 Age of enterprise 8.745 11.093 0.000 83.000 1018 1.000 3.000 6.000 15.000 Bank loan > 50% of debt 0.312 0.463 0.000 1.000 1018 0.000 0.000 0.000 1.000 Standalone 0.648 0.478 0.000 1.000 1018 0.000 1.000 1.000 1.000 Ent with paid labor 0.317 0.465 0.000 1.000 1018 0.000 0.000 0.000 1.000 Enterprises with debt 0.419 0.493 0.000 1.000 1018 0.000 0.000 1.000 1.000 Community pop size 1653 952 558 5600 1018 898 1200 1510 2183 Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data. Annex B. Data Used for Enterprise Performance Analyses 85 Table B.5 Tanzania: Variables Used in Regressions of Enterprise Performance; Value and Distribution Variable name mean st dev min max obs pct20 pct40 pct60 pct80 Sales# 6.309 1.119 4.310 9.818 947 5.267 5.909 6.358 7.229 Net value added# 5.412 1.535 2.506 9.664 807 4.400 5.078 5.741 6.491 Total labor input# 5.317 0.588 2.864 8.741 947 4.849 5.161 5.485 5.521 Value of capital input# 3.965 2.185 0.000 10.167 947 2.065 3.690 4.735 5.823 lnVLL (ln*ln) 28.620 6.636 8.202 76.408 947 23.517 26.636 30.083 30.486 lnVLC (ln*ln) 21.218 12.570 0.000 88.021 947 10.843 18.575 24.434 31.647 lnVCC (ln*ln) 20.494 17.160 0.000 103.375 947 4.266 13.614 22.424 33.902 Depreciation# 2.365 2.050 0.000 7.715 947 0.000 1.542 2.964 4.280 Non Factor Cost# 5.044 2.163 0.000 10.534 947 4.106 5.057 5.717 6.523 Other enterprise characteristics Age of enterprise# 1.841 0.955 0.000 4.007 947 1.099 1.609 2.079 2.708 Experience of manager# 1.606 0.620 0.693 3.970 947 1.099 1.386 1.838 1.946 Gender of manager 0.782 0.413 0.000 1.000 947 0.000 1.000 1.000 1.000 Education of manager 7.844 2.761 1.000 17.000 947 7.000 7.000 7.000 8.000 Registration 0.202 0.401 0.000 1.000 947 0.000 0.000 0.000 1.000 Industry dummies/line of business Trade enterprise 0.497 0.500 0.000 1.000 947 0.000 0.000 1.000 1.000 Services enterprise 0.184 0.387 0.000 1.000 947 0.000 0.000 0.000 0.000 Manufacturing, nonagricultural enterprise 0.051 0.219 0.000 1.000 947 0.000 0.000 0.000 0.000 Agricultural Processing enterprise 0.038 0.191 0.000 1.000 947 0.000 0.000 0.000 0.000 Mining, gas, etc., construction enterprise 0.024 0.154 0.000 1.000 947 0.000 0.000 0.000 0.000 Mixed enterprise 0.206 0.404 0.000 1.000 947 0.000 0.000 0.000 1.000 Community characteristics Community population size# 8.057 0.842 5.820 9.621 947 7.358 7.781 8.243 8.923 Agricultural land per capita 0.596 0.724 0.000 3.901 947 0.042 0.216 0.437 1.013 Illiteracy 27.600 22.644 1.000 100.000 947 8.000 15.000 27.500 49.500 Main community income from: Agriculture 0.844 0.363 0.000 1.000 947 1.000 1.000 1.000 1.000 Wages 0.136 0.343 0.000 1.000 947 0.000 0.000 0.000 0.000 Self-employment 0.020 0.140 0.000 1.000 947 0.000 0.000 0.000 0.000 Benchmark indicators and components Connectivity 0.214 0.143 0.008 0.560 947 0.082 0.148 0.231 0.342 Access 0.195 0.152 0.000 0.863 947 0.038 0.128 0.231 0.331 Business services 0.061 0.146 0.000 0.857 947 0.000 0.000 0.000 0.143 Governance 0.493 0.076 0.250 0.667 947 0.442 0.483 0.514 0.557 Human capital 0.215 0.056 0.092 0.379 947 0.170 0.195 0.220 0.257 Finance services 0.380 0.194 0.000 0.938 947 0.206 0.324 0.440 0.554 Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value. 86 The Rural Investment Climate Table B.6 Tanzania: Selected Variables of Enterprises and Communities; Value and Distribution Variable name mean st dev min max obs pct20 pct40 pct60 pct80 Non Factor Cost 728 2060 0 37562 947 60 156 303 680 Sales 1186 2153 73 18368 947 193 367 576 1378 Depreciation 76 205 0 2241 947 0 4 18 71 Gross value added 456 2230 37382 15769 947 55 138 276 620 Net value added 381 2230 37382 15732 947 28 110 241 551 Family labor input 206 121 0 1200 947 125 155 240 240 Paid labor input 52 362 0 6254 947 0 0 0 0 Total labor input 258 374 17 6254 947 127 173 240 249 Net value of assets 1100 8720 80175 115350 947 23 174 551 1745 Imputed capital cost 276 1042 0 19609 947 3 21 70 231 Paid capital costs 82 468 0 6612 947 0 0 0 33 Value of capital input 358 1269 0 26038 947 7 39 113 337 Total Factor Input 616 1453 33 31789 947 183 250 340 643 Total Cost 1420 2899 55 42717 947 319 492 787 1581 Net pro�t 234 2867 42597 15414 947 383 162 47 184 Pro�tability 1.202 2.579 0.003 61.226 947 0.436 0.653 0.917 1.282 Net factor productivity 1.227 7.077 155.76 61.226 947 0.074 0.386 0.783 1.693 Net income 247 2409 37382 15732 947 1 90 198 487 Total liability 948 5983 0 81737 947 0 0 0 12 Total assets 2020 6566 0 115350 947 46 211 643 1837 Employment 258 374 17 6254 947 127 173 240 249 Imputed factor costs 483 1050 0 19849 947 163 240 287 510 Pro�tability of own resources 1.303 8.134 155.75 61.226 939 0.125 0.440 0.895 1.963 Education of manager 7.844 2.761 1.000 17.000 947 7.000 7.000 7.000 8.000 Gender of manager 0.782 0.413 0.000 1.000 947 0.000 1.000 1.000 1.000 Ethnicity of manager 0.218 0.413 0.000 1.000 947 0.000 0.000 0.000 1.000 Age of enterprise 8.522 8.851 0.000 54.000 947 2.000 4.000 7.000 14.000 Bank loan > 50% of debt 0.008 0.092 0.000 1.000 947 0.000 0.000 0.000 0.000 Standalone 0.365 0.482 0.000 1.000 947 0.000 0.000 0.000 1.000 Ent with paid labor 0.139 0.346 0.000 1.000 947 0.000 0.000 0.000 0.000 Enterprises with debt 0.233 0.423 0.000 1.000 947 0.000 0.000 0.000 1.000 Community pop size 4407 3639 336 15070 947 1568 2394 3800 7500 Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data. Annex C. Enterprise Performance Regressions; Notes and Tables Speci�cation of enterprise performance regression highest individual explanatory power. Moreover, models. Enterprise performance is measured by a fifth regression variant separates imputed and total productivity and net productivity. Total paid labor and capital, as mentioned above. productivity is measured on gross output, which is Table C.1 provides an overview of the regression equivalent to sales; net productivity is measured on variants. Annex A, Table A.2 provides full speci�- net value added (NVA). The primary empirical cation of the variables used. Descriptive statistics speci�cation uses the Cobb-Douglas production of variables used are listed in Annex B, Tables B.1 functions on sales and net value added (NVA) as a to B.6. Regressions analyzing sales are run with a starting point; this is loglinear in output and inputs, data set of enterprises with sales of at least US$100 augmented with variables hypothesized to enhance or net value added of at least US$60; regressions enterprise productivity. Throughout the report ref- exploring net value added use samples of enter- erence is made to sales variants and NVA variants. prises with net value added of at least US$60. Due to inherent interest of the contribution of large The theoretical justification for the selection of enterprises to the rural economy, a particular ques- variables is as follows. The factor inputs—labor tion arises about describing potential size differ- and capital—correspond to the net value added ences. Employing a dummy for larger enterprises is output concept, and the factor and nonfactor a poor strategy, as the size threshold to be used must inputs and depreciation correspond to sales (gross be preselected and in any case represents something output concept). Depreciation was selected as a in the nature of the enterprises that should be de- separate variable since it differs for the capital scribed by variables that can be more meaningfully components of land and �xed assets.71 The use of interpreted. Instead, quadratic and interactive spec- capital is perhaps the most endogenous factor i�cations of the factor input variables labor and on the right hand side. Own capital will depend on capital can better describe differences related to size the owner’s family history and level of income than can log linear forms, and examination of in the past, which is not easily captured in the preliminary regression results indicates that inter- enterprise data set. Borrowed capital, however, active speci�cations signi�cantly improved the will depend on other endogenous variables such statistical �t.70 as solvability, profit level, and size of enterprise The aim in formulating the regression models is (for which we have data), and on supply of avail- to choose variables that describe enterprise charac- ability of sources approximated by the finance teristics and the community environment and to services index and perhaps other benchmark indi- avoid variables that can be seen as largely endoge- cators. It might be worthwhile to pursue this in a nous. Since the main objective of the analysis is to subsequent study. shed light on the contribution of the investment In addition to these production factors, a num- climate, three stepwise regressions are carried out. ber of enterprise characteristics are considered that This reveals the structure of the data sets and can add to gross and net productivity. Enterprise illustrates the variance captured by enterprise characteristics include enterprise age; entrepre- variables, general community variables, and neur experience, gender, and education; enterprise benchmark indicators. In addition, a fourth variant registration status; and line of business characteriz- is estimated in which the benchmark indicators are ing enterprise operations.72 The contribution of reg- replaced by some of their components with the istration status to productivity is not clear upfront 87 88 The Rural Investment Climate Table C.1 Speci�cation of Regression Variants Variants (1) (2) (3) (4) (5) Variable to be explained Sales NVA Sales NVA Sales NVA Sales NVA Sales NVA Speci�cation of Benchmark BI with control BI with control BI and BI- BI and BI- variables indicators (BI) of industry mix of industry mix, components components with control of and enterprise enterprise and with control of with control of industry mix variables community industry mix, industry mix, variables enterprise enterprise variables, variables with and community separation variables of imputed and paid labor, and capital and community variables Homogeneous labo and capital input X X X Separation of imputed and paid labor and capital X Other enterprise variables X X X X Industry dummies X X X X X General community variables X X X BI X X X BI and BI components X X Source: This report. since it may differ with policies and size. In some seasonal locations, enterprise activity may be cases, registration may be a necessary condition for secondary to the household’s main pursuits; on access to public and infrastructural services; in the other hand, highly seasonal fluctuations in other cases, public officers may actively enforce demand cause both highs and lows in production registration even if there are only costs and no ben- activity, increasing downtime and reducing e�ts. Most very small enterprises may not gain any ef�ciency. For Nicaragua and Sri Lanka, the num- benefit from registration and may remain under ber of enterprises per community is available and the radar of registration of�cers.73 Bribery may play was used to compute enterprise density per 1,000 a role as well. Dummy variables for enterprise line inhabitants. A higher density could well result in of business or sector capture differences in produc- higher productivity due to a conglomeration effect tivity not explained by other variables.74 and lower transaction costs. Community popula- Community characteristics present direct incen- tion size may be an indicator for the degree of urban- tives and disincentives for enterprises and, hence, ization in which the enterprise operates. Population may affect their productivity. For Nicaragua and size may also capture the quality of the economic Sri Lanka, data were collected about seasonality in and institutional infrastructure to the degree that agricultural labor use. This variable may indicate these are not directly measured by other commu- variation in the efficiency of labor and capital nity variables in the regression model. The variable use between locations. On the one hand, in highly may not be a good indicator of urbanization in all Annex C. Enterprise Performance Regressions; Notes and Tables 89 cases, however, as a community’s geographic size distance to urban areas) are likely important for can contribute to its population size; the pilot sur- enterprises but not easily affected by economic vey designs did not speci�cally de�ne what could policy. The benchmark indicators have certainly constitute a community. a degree of exogeneity, but questions of self- Area of agricultural land cultivated per capita selectivity remain. The environment described by can affect a community’s enterprise productivity. the benchmark indicators may determine whether High areas per capita indicate relatively important enterprises are established and which investments lowland and upland agriculture and perhaps also are made. This study, however, will not pursue this low population density. A distinction between issue.76 lowlands and uplands makes much sense in mon- Some variables sometimes included in enter- soon Asian countries, but less in many other devel- prise performance models are intentionally oping countries. Paddy land (or irrigated lowland) omitted here. Regional location, for example, is may indicate high population densities and high omitted since the regression model includes many productivity per unit of land. The quality of land variables that describe the community environ- data appears to be poor, allowing inclusion in the ment—precisely what regional dummy variables regression analysis of only paddy land per capita attempt to approximate. Similarly, price variables in Sri Lanka and total cultivated land per capita in are omitted. The surveys measure cross-regional Nicaragua and Tanzania. variation in the price of male daily labor and of The rate of illiteracy is included as a variable in commodities such as staple foods, Coca-Cola, and the regression model, since it can affect enterprise kerosene/diesel. The price of male labor could performance. It may reflect the quality of labor indicate labor scarcity/productivity. Community enterprises can employ, but the indirect effect may prices for staple foods, Coca-Cola,47 and kerosene be more important: illiteracy may be a proxy for could reflect margin differences related to trans- the general level of development in a community port costs, value added, and price competition. or for a less favorable business environment.75 These variables of course also reflect price The Sri Lanka questionnaire includes a question variations inappropriate in a production function, about where households from the community buy however, and the regression model already and sell most of their goods, that is, in the commu- includes measures of infrastructure services and nity, a neighboring community, a commercial cen- connectivity that are proxies for determinants of ter, or the nearest city. The answers are used to margins. Chapter 6 explores the relation between specify dummies that to some extent at least indi- benchmarks and prices. cate the community’s degree of commercial open- ness. In Sri Lanka and Tanzania, questions were Regression results. A dominant characteristic of asked about the main source of income in the com- the enterprise data sets is that by far most of the munity, that is, agriculture, wages, or nonfarm variance is explained by enterprise characteristics enterprises. This says something about the types and that industry and community levels explain only of production in which local enterprises engage. a small part of it.78 This is illustrated in Table C.2 It may also capture some difference in income using the results of the �rst four regression variants between communities. Dummies were created for for the three countries; the �fth is similar to the these communities since they can explain differ- fourth but has a different speci�cation, with im- ences in productivity. puted and paid labor and capital separated. The All regressions models incorporate benchmark contribution of community variables will be further indicators, which are community-level composite analyzed later on. The regressions on benchmark indexes of characteristics of location, availability of indicators and enterprise dummies for the control infrastructure, utilities, public and private services, of industry differences explain only between 3.6 governance and corruption, human capital, and and 9.3 percent of the variance. When enterprise �nance services, as described in Chapter 2 above. variables are added, the explanation jumps by a Benchmark indicators and their components factor of �ve to eight. General community variables describe both the enterprise environment at the explain 0.2 to 1.8 percent. Replacing the benchmark community level and the investment climate. Many indicators with those of their components that add can be directly affected by policy interventions most increases explanation of the regression by and public investments, but some (for example, another 0.5 to 2.0 percent. 90 The Rural Investment Climate Table C.2 Percentage of Variation Explained in Enterprise Performance Regressions Nicaragua Sri Lanka Tanzania Sales NVA Sales NVA Sales NVA Numbers of observations 846 814 1018 841 947 699 Regression variants 1. Benchmark indicators (BI) with control of industry mix 6.7 5.9 9.3 6.3 3.6 3.9 2. BI with control of industry mix and enterprise variables 42.6 33.3 74.1 50.4 33.9 19.4 3. BI with control of industry mix, other community variables, and enterprise variables 42.8 33.6 75.0 52.2 34.2 19.9 4. Mix of BI and BI components with control of other community variables, industry mix, and enterprise variables (preferred model) 44.1 35.3 75.7 54.2 36.0 20.4 5. Mix of BI and BI-components with control of industry mix, enterprise variables with separation of imputed and paid labor, and capital and community variables 46.5 38.2 76.2 53.6 35.5 21.4 Source: Table C.4 to Table C.9. Note: The percentage of variation explained is measured by means of a pseudo-R2, since the regression model is not estimated with Ordinary Least Squares but rather with a weighted random effects model. The pseudo-R2 is computed as the squared correlation between the observed and predicted values of the dependent variables. The NVA regressions have generally a one-third signi�cant coef�cients of almost 0.50. For variant lower explanation in Sri Lanka and Tanzania; (4), the implied returns to scale at mean input levels in Nicaragua the difference is smaller. When equals 0.879 and rises with increasing levels of imputed and paid labor and capital are separated labor and capital. (regression variant 5), the explained variation The log of enterprise age, with a coefficient of increases by a few percent only, and for Sri Lanka’s about 0.12, contributes significantly to produc- NVA and the Tanzania’s sales model it even tivity. The log of manager experience has a coef�- decreases. cient of about 0.07 points in the same direction for the variants (2) through (4), but its signi�cance is Sales regressions: Sri Lanka. Since the Sri Lanka just at or below 10 percent. These �ndings suggest regressions have more variables than do the others, that learning by doing and innovation are im- they will be discussed �rst. The estimated coef�- portant for survival and increased productivity. cients for enterprise characteristics are fairly stable The dummy for male entrepreneurship shows a in the three relevant regression variants (2) through positive coefficient, but its significance is below (4), indicating that community variables do not the 10 percent level, except for variant (5). This have much effect on the parameter estimates. This indicates a tendency toward higher productivity isn’t surprising, given the relatively small share among male-managed enterprises, although the of variance explained by community variables. statistical evidence is weak. The manager’s years Quadratic and interaction terms for logarithmic of education contributes significantly to produc- values of labor and capital dominate, with high tivity; each year of education adds about 2.5 to signi�cant levels in the regressions, and the coef�- 3.0 percent to productivity. Enterprise registra- cients for the log linear variables are not signi�cant. tion status does not contribute to explanations In the variant with separated variables for imputed of productivity. Some literature argues that and paid labor and capital, variant (5), all factor registered enterprises have higher productivity input variables are signi�cant except the loglinear because they receive more public services; and quadratic variables for paid capital. Interaction others maintain that especially for smaller enter- terms were not signi�cant. The estimated coef�- prises the cost of registration may outweigh the cients for depreciation show negative signs but are benefits; leading them to avoid registration. not signi�cant. Nonfactor cost has robust, highly While simultaneously considering many other Annex C. Enterprise Performance Regressions; Notes and Tables 91 characteristics, our findings don’t support either not quite signi�cant, but it is rather robust across of these views. model variants. The likely explanation for this is Industry dummies in the variants (2) through underuse of off-season capital and labor. Enter- (4) reveal that services have about 20 percent prise density, community population size, and higher productivity than does trade (which is illiteracy have unstable parameter estimates. In all included in the intercept), but for variant (5) no regressions, the area of paddy land per capita is significant contribution was found. For the other positively related to productivity, but does not sectors no significant differences are found that reach the 10 percent signi�cance level. Three dum- cannot be explained by enterprise and community mies indicate where households buy and sell most characteristics. of their goods, that is, in the community, a neigh- Community characteristics are less stable in the boring community, a commercial center, or the regressions than are enterprise characteristics, in nearest city. These dummies indicate the commu- part because they are more interrelated among nity’s degree of commercial openness. Enterprises themselves and with the enterprise and industry in communities selling primarily in a neighboring variables, and also because they make a relatively community or commercial center have a 25 to weak contribution to overall explanations of vari- 35 percent higher productivity. Surprisingly, for ance. But, given the IC focus, they are in a sense the 38 percent of communities selling in the near- the core of this study. As already seen, the contri- est city, no significantly higher productivity was bution of benchmark indicators is weak. The busi- found. Also dummies for communities with a ness services index has in all specifications a main source of income from wages (variant (3)) or positive and signi�cant contribution. The connec- from nonfarm enterprises (variants (4) and (5)) tivity and governance indexes are signi�cant only suggest a 15 and 20 percent higher productivity, in regression variant (1), where they apparently respectively, although in most cases below the capture explanations contributed to other vari- 10 percent signi�cance level. ables in the other regressions. As already noted, in variants (4) and (5) benchmark indicators are NVA regressions Sri Lanka. The results from the replaced by their best contributing subindexes. NVA and sales regressions show similarities and Among the connectivity subindexes, cost of differences. In the NVA variants (2) through (4), all transportation to the nearest city is positive and labor and capital variables are signi�cant. For the signi�cant at about the 10 percent level. One infra- variant (5) with separation of imputed and paid structure services component is significant: con- capital and labor, only the loglinear and quadratic crete or asphalt roads, which, surprisingly, has a variables for paid capital are signi�cant. With R2 negative sign. The governance subindicator of lower than for (4), this is not the most attractive dealing with government services has a positive variant. Returns to scale for variant (4) are esti- sign and a signi�cance level around 10 percent. A mated at 0.574 when evaluated at mean values of second governance component, contract resolu- log-labor and log-capital, rising with increasing tion and contract enforcement, has a negative levels of labor and capital. effect, but a parameter value below 10 percent Enterprise age is again a significant variable in significance. The human capital index has no explaining productivity, but not entrepreneur expe- subindexes; its parameter estimates are sensitive rience and gender. Manager education strongly to differences in model speci�cations but are in all contributes to higher productivity, suggesting that cases without statistical signi�cance: in essence, it each additional year of education increases produc- has no effect. The �nance services subindicator for tivity by about 5 percent. loan access is significant. This subcomponent, For NVA no industry dummy is significant in however, is somewhat suspect. Aggregated from the regressions with enterprise variables. Among all enterprises in the community, it could well the community variables, only the dummies for reflect a self-selectivity: more productive enter- marketing in the neighboring community and in prises gain better access to loans. the commercial centre are signi�cant. Agricultural Some of the other community characteristics seasonality has a negative effect of a magnitude show interesting results. Enterprises in communi- similar to the sales regression model. Income ties with high agricultural seasonality in labor use source does not matter to enterprise net value have lower productivity; the effect is statistically added. 92 The Rural Investment Climate Among the benchmark indicators, business becoming signi�cant in variants (4) and (5). Infra- services contributes signi�cantly to higher produc- structure services, business services and gover- tivity in all variants; the other benchmarks nance don’t contribute significantly. Even if an don’t contribute significantly. The benchmark aggregate index with no effect, however, may have indicator components add more. The connectivity subindexes with an effect. This is the case for subindexes for transportation costs to the nearest Nicaragua: except for governance (human capital city and time taken by the main means of trans- has no subindexes) and finance services, many portation to the main market have signi�cant coef- subindexes contribute signi�cantly to productivity ficients, although with different signs. The access differences: dummy for asphalt and concrete roads has a nega- tive sign but is not quite signi�cant at 10 percent. c1 Inverse time taken by main means of trans- Governance and human capital don’t contribute portation to nearest major city (in minutes) signi�cantly. For the �nance services benchmark, c2 Inverse cost of transportation to the nearest the loan access subindex makes a signi�cant con- major city tribution to explaining higher productivity. c4 Inverse cost of transportation to the main market (public transportation) Sales regressions for Nicaragua. The parameter c5 Proximity to the main post of�ce estimates for Nicaragua’s enterprise variables a1 Percentage of households with electricity are stable and consistent. The loglinear and qua- dratic variables for labor and capital in variants a2 Availability of electricity (such as lack of (2) through (4) are all signi�cant; the interaction disruptions) variable is not. For variant (4), the implied returns a6 Garbage collection or disposal service in the to scale at mean input levels is quite low at 0.624, community but it increases in labor and capital. In variant d7 Information technology services available for (5) with separated variables for imputed and paid businesses in the community labor and capital, virtually all labor and capital vari- ables have signi�cant coef�cients. Depreciation and In the cases of c2 and a2, the parameter signs are nonfactor input have strongly signi�cant contribu- negative, contrary to the expectations for an index tions, although the parameter values for deprecia- variable or dummy ranging between zero and one, tion and nonfactor input of about 6 to 8 percent with the value one assumed optimal for enterprise differ surprisingly from their 3.5 and 15 percent of development. the share in sales. Firm age has a strongly signi�cant contribution to total productivity. Data on the NVA regressions for Nicaragua. Tanzania’s NVA top manager’s experience, gender, and education regressions show many similarities and some were unfortunately not collected for Nicaragua. differences. Almost all labor and capital variables Enterprise registration status does not affect pro- are signi�cant in variants (2) through (5) and imply ductivity in any of the regression variants. roughly the same returns to scale as the sales Nicaragua shows signi�cant differences in pro- regressions. Enterprise age has a highly signi�cant ductivity between sectors, with enterprises in ser- contribution of 9 percent to the explanation of pro- vices exhibiting 11 percent lower productivity than ductivity, virtually identical to the sales regression, trade enterprises, and manufacturing nonagricul- but, contrary to the sales regressions, here registra- tural products and mixed enterprises having 10 tion status also contributes strongly, suggesting and 18 percent higher productivity, respectively. a 10 percent gain in productivity. The direction of Nicaragua’s community variables are limited causality is not clear since no signi�cance was compared to Sri Lanka’s. It lacks the measure of found in the sales regression. Perhaps more pro- commercial openness and income source, and their ductive enterprises with high net income register contribution differs much with model speci�cation. for tax purposes. Only for variant (3) are signi�cant parameter esti- The �ndings of differences between sectors are mates found: enterprise density and community similar to the sales regression, with, relative to population size.79 The benchmark indicator for con- trading, lower productivity levels for services and nectivity has a signi�cant positive parameter value; higher levels for mixed enterprises. Also similarly, the effect of human capital tends to be negative, enterprise density yields a significant parameter Annex C. Enterprise Performance Regressions; Notes and Tables 93 value only in variant (3). As for other community Infrastructure services parameters have a negative variables, the sign and magnitude of their estimated sign in all variants, but only in variants (2) and effect always corresponds with the sales regression (3) is statistical significance weak; component a5 results, but none is statistically signi�cant, with the (presence of a sewage system in the community) is exception of illiteracy in variant (4), where a higher more robustly negative. The business services share of illiteracy contributes to lower productivity. index fails to exhibit any explanatory power when Connectivity contributes to higher productivity; entered as an aggregate, but the estimated effects human capital has a negative sign that becomes sig- of two subcomponents do achieve almost 10 per- ni�cant in variants (4) and (5), contrary to expecta- cent significance: legal services in an upward tions. Again, several subindicators have signi�cant direction and insurance services downward. The signs. These are mostly the same as for the sales governance indicator does not play a role. Human regressions, adding a3 (percent of households with capital has a positive sign in all variants, but its access to protected water, with its negative sign, con- significance never quite reaches the 10 percent trary to expectations) and switching d7 (information level. The �nance services indicator has no traction technology services) out for d4 (legal services). The in the variants (1) through (3). In variants (4) and contributions of c1, c4, c5, a1, a6, and d4 remain pos- (5), its component fi1 (number of banks in the itive and are roughly similar to the sales regression; community) and fi3 (access to loans) exert a little the contribution of a2 remains negative and of a more influence, but the estimates are statistically similar magnitude. insigni�cant and of opposite sign. In all, commu- nity variables and benchmark indexes do not Sales regressions for Tanzania. In Tanzania’s sales matter much to enterprise sales. regressions the variables for factor input are all signi�cant except for the loglinear and quadratic NVA regressions for Tanzania. In the NVA regres- variables of paid capital cost. Nonfactor cost has a sions the parameter values of most labor and capi- highly signi�cant parameter, but depreciation does tal input variables are not signi�cant. Only the not contribute to explanation of output. At mean quadratic form of capital input is signi�cant in vari- input values, returns to scale equal only 0.464 ants (2) through (4); returns to scale are estimated at for variant (4) and decrease in labor and capital. 0.292 and rise with increases in labor and capital. In Enterprise age and entrepreneur education do not variant (5) the logarithms of paid labor and imputed signi�cantly contribute to productivity in any of the capital and their quadratic forms have signi�cant regression variants, although both have the ex- parameters. Multicollinearity among these inputs pected positive sign; entrepreneur experience does obscures the relationship: in a speci�cation that help. Male led enterprises have about 15 percent omits all quadratic terms, the parameter estimates higher productivity. Registration contributes about of the linear terms are positive and signi�cant but 33 percent higher productivity with a high level of actually quite small.80 signi�cance. Manager experience has some positive contri- No significant differences in productivity are bution, but the estimated effect does not reach to found between industries. the 10 percent significance level. Contrary to the None of the community variables contributes sales regression, gender is not associated with signi�cantly to productivity differences. Commu- differences in net factor productivity, although nities with nonfarm enterprises as the main source all variants suggest a 10 percent advantage for of income show a positive tendency for higher male managers. Each additional year of manager productivity, but it is not signi�cant at the 10 per- education signi�cantly adds 3 to 4 percent to pro- cent level. ductivity. Registration makes a difference of about The contribution of benchmark indicators and 40 percent. subindexes is unstable and levels of significance Sectoral differences are not signi�cant. All com- are in most cases rather low. Connectivity and its munity variables have the expected sign, but none component c3 (time taken by the main means of contributes signi�cantly to explaining differences transportation to the main market) are signi�cant in productivity. Connectivity adds signi�cantly to with the correct sign in all cases, but in variant productivity in all variants, but none of the other (4) and (5) its component c2 (transportation cost benchmark indicators does. Of the benchmark to the nearest major city) has the opposite sign. components, c6 (rail stop within walking distance 94 The Rural Investment Climate of the community) has a positive effect, and d6 total variation, probably often of characteristics re- (insurance services available for businesses in the ferring to limited locations or few enterprises only. community) contributes with a negative sign. In principle, variation in the explained variable derives from (i) observed enterprise variables, Variation in productivity. Productivity varies in (ii) unobserved enterprise variables, (iii) observed response to many often interrelated factors, some community variables, and (iv) unobserved com- of which may not even be explicitly included in munity variables. By far, most of the variation in the data set but which may still be influential in productivity in the Nicaragua and Sri Lanka data the background as unobserved variables. Many sets is explained by factors at the enterprise level, explanatory variables may capture tiny parts of (combining (i) and (ii)). Box C.1 provides an Box C.1 Variance Explained by Enterprise and Community Variables “How much of the variation in the dependent variable can be explained by location as given by community?� Translated, this means: “How much do community dummy variables explain?� The answer is roughly 20 percent for Nicaragua, 28 percent for Sri Lanka,, and 35 percent for Tanzania. Of course, some of the explanatory enterprise variables may differ systematically from community to community. In some communities, for example, enterprises may employ more labor and capital or more may be registered. Thus, some of the systematic variation in lnsale and lnNVA may be explained by systematic differences in the enterprise variables. Therefore, as a follow-up analysis, simple regressions of lnsale and lnNVA on enterprise variables were run (this resulted for Nicaragua in R2 0.423 for lnsale and R2 0.330 for lnNVA), residuals were retrieved, and a dummy variable model was then run on these residuals. Again, for Nicaragua about 20 percent of the variation not explained by the observable enterprise variables is at the commu- nity level. This means that community variables can explain at best 0.2*(1 0.423) 0.115, which is 11.5 percent of the variation in lnsale, and 0.197*(1 0.33) 0.132 percent, which is 13.2 percent of the variation in lnNVA. Of course, they actually explain much less, because if community variables would explain 11.5 percent of the variation of lnsale, they would provide a perfect prediction of the community effect. For Sri Lanka, the maximum for lnsale is much lower, mainly because such a large part of the vari- ance (73.6 percent) is explained by the enterprise variables. By contrast, for Tanzania the variance explained by enterprise variables is relatively low, and the maximum total variance that can be can be explained by community variables is about 27 percent. Percentage of Variance Explained Nicaragua Sri Lanka Tanzania Variables to be explained* lnsale lnNVA lnsale lnNVA lnsale lnNVA 1. ANOVA on explained variable (%) 20.0 19.7 27.5 29.5 39.6 33.3 2. Regression on enterprise and industry variables (%) 42.3 33.0 73.6 49.5 33.0 18.1 3. ANOVA on regression residuals (%) 19.3 20.4 18.3 22.7 34.9 29.1 4. Maximum possible explanation by community variables (%) 11.5 13.2 7.3 14.7 26.5 27.3 Number of enterprises 846 814 1018 841 947 699 Number of communities 93 93 118 118 136 127 Source: RIC Surveys. *lnsale and ln NVA are the logarithmic values of sales and net value added. Annex C. Enterprise Performance Regressions; Notes and Tables 95 overview of the break-down of variation in the from even fewer communities (93 in Nicaragua, data sets. Line 1 in the box table indicates the vari- 118 in Sri Lanka, and 136 or 127 in Tanzania). ation in productivity explained by community- Therefore, while determinants of the variation in level factors (combining (iii) and (iv)). It ranges productivity across communities may be many, from 20 percent in Nicaragua to 40 percent in only a limited number can be explored. Addition- Tanzania. Line 2 shows that observed enterprise ally, few of these determinants are aggregated variables explain between 18 percent of lnNVA in from households and enterprises: in some commu- Tanzania to 74 percent of lnsale in Tanzania. The nities, the number of household and enterprise remainder, residuals of the enterprise regressions, observations is rather small, which affects the covers variation sources (ii), (iii), and (iv). Line 3 reliability of components aggregated from these shows that between 18 and 35 percent of this micro-data. remainder can be explained by community-level factors (that is, (iii) and (iv)). Line 4 shows the max- Employment and capital generation. What condi- imum share that community variables (source (iii)) tions generate enterprises with more employment can explain when explanation is perfect and source and enterprises with more capital? Or, expressed (iv) vanishes. In Nicaragua it is only around 12 per- differently, what explains differences in enterprise cent of total variation in productivity; in Sri Lanka size, measured by labor and by capital input, between 7 and 14 percent; and in Tanzania about respectively? 27 percent. In practice, the explanation will be Regression models are run to explain employ- much lower. ment and capital generation from industry mix An important reason for the low share of expla- enterprise characteristics, benchmark indicators nation of variation by community-level factors is and other community characteristics. The parame- that the communities are relatively homogeneous: ter estimates are presented in Table C.10 to if nothing else, they are all rural. Moreover, Table C.15, a summary of the explanation found is the number of communities is low: the RIC sam- presented in Table C.3. Industry mix and bench- ples cover only 99, 151, and 149, respectively, in mark indicators together can explain 4 to 5 percent Nicaragua, Sri Lanka, and Tanzania, and because of the variation in employment generation. By of missing data, the enterprise performance regres- adding enterprise characteristics the explanation sion models are estimated with enterprise data doubles in Nicaragua, triples in Sri Lanka, and Table C.3 Explanation of Employment and Capital Generation per Enterprise Employment generation Capital generation Nicaragua Sri Lanka Tanzania Nicaragua Sri Lanka Tanzania Number of observations 846 1018 947 846 1018 947 Industry mix and benchmark indicators 5.2 4.0 4.1 5.3 7.9 7.1 Industry mix, benchmark indicators and enterprise characteristics 10.8 13.8 6.7 21.6 29.4 15.4 Industry mix, benchmark indicators, enterprise characteristics and other community characteristics 12.3 17.6 8.6 21.3 30.6 17.8 Ibid. with benchmark indicators replaced by components 13.6 18.6 10.2 24.2 32.9 18.0 Source: Annex C, Table C.10 toTable C.15. Note: See note to Table C.2. 96 The Rural Investment Climate increases by half in Tanzania. Other community than are trading enterprises. Agricultural process- variables add 1.5 percent in Nicaragua, almost ing, mining, and multiactivity enterprises on the 2 percent in Tanzania, and almost 4 percent in other hand are 54, 95, and 15 percent larger. Enter- Sri Lanka. The higher percentage of explanation in prises in communities with higher enterprise den- Sri Lanka could well be the result of the larger sity and marketing in neighboring communities, number of enterprise and community variables commercial centers, and cities have significantly than in the other countries. Differences in capital more employees. A high level of income from generation can be better explained than can enterprises or employment has no strong effect. differences in employment generation. Partly, this Most surprisingly, in Sri Lanka none of the could be due to greater weaknesses in the data benchmark indicators contributes significantly to measuring labor input. Enterprise characteristics employment per enterprise in any of the regres- contribute most to differences; other community sion variants. Only when selecting among the variables contribute much less. In Nicaragua benchmark components do some investment cli- adding community variables results in a lower per- mate elements show up signi�cantly. In particular, centage of explanation; for Sri Lanka and Tanzania secure provision of electricity and wider availabil- the explanation increases by 1.2 and 2.4 percent. ity of telephone service are associated with larger enterprises, as are business services in this speci�- Employment parameter estimates. The regression cation of the model. Among corruption and gover- parameter estimates for employment in Nicaragua nance components, infrastructure and services are nearly all signi�cant at 10 percent. Employment have negative effect. increases signi�cantly with enterprise. Registered In Tanzania enterprises with a male top man- enterprises employ 52 percent more labor. ager and a more educated manager have more Compared to trading enterprises, enterprises in employees. Differences between industries in services, manufacturing nonagricultural products, numbers of employees are limited, with agricul- and manufacturing agricultural products provide tural processing 22 percent smaller than trading 17, 9, and 13 percent less employment, whereas enterprises and 27 percent bigger. Smaller commu- mining and multiactivity enterprises have 40 and nities tend to have enterprises with more employ- 52 percent more employees. Enterprises in com- ees. None of the benchmark indicators and other munities with seasonality and more agricultural community variables contributes significantly to land per capita have 17 and 12 percent more explaining differences in employment per enter- employees, suggesting that the largest enterprises prise. After replacing benchmark indicators by in terms of employment are in agricultural their components, only three contribute signifi- areas. Community size contributes signi�cantly. cantly with a positive sign to explaining employ- Enterprise density and illiteracy don’t contribute ment generated per enterprise: proximity to post much. Connectivity, infrastructure services, and office, availability of marketing services, and governance make signi�cant contributions, but for governance components related to dealing with infrastructure services and governance the sign is government services and general policy and insti- opposite what is expected. Replacing the bench- tutional constraints. mark indicators by the components providing the Community unobservables are important best explanation, shows that enterprises hire more in each country, as witnessed by the standard workers in communities where more households deviation of the community random effect. In carry cell phone service (a sign of development) but Nicaragua, variation of community observables is hire fewer where more households have electric nearly as large as the idiosyncratic variation at the service or a sewage system. Moreover, enterprises enterprise level, and in Tanzania it is not far are smaller where dealing with government agen- behind. Overall, community effect is largest in cies is easier. Tanzania and idiosyncratic variation most impor- In Sri Lanka employment in enterprises relates tant in Sri Lanka. significantly to male managers, manager educa- tion, and enterprise registration. Enterprise age Capital generated. In Nicaragua enterprise age does not play a role; entrepreneur experience does, and the registration status contribute strongly to but signi�cance remains below 10 percent. Service the amount of capital invested per enterprise. enterprises are 16 percent smaller in employment Enterprises in services and multiactivity enterprises Annex C. Enterprise Performance Regressions; Notes and Tables 97 have most capital invested, and enterprises in pro- accounting services contribute. Three other bench- cessing of agricultural products relatively less. mark components have significant parameter Communities with seasonality have not only more estimates. Positive are percentage of access to pro- employees but also a higher level of investment per tected water and households with fixed phone enterprise. Community variables affecting invest- line; negative is roads with asphalt or concrete ment are quite different from those affecting surface. employment. Here we �nd robust signi�cant expla- In Tanzania registered enterprises have 280 per- nation from the human capital and �nance services cent81 more investment. Enterprises with experi- benchmark indicators. After replacing the bench- enced managers and male managers have more mark indicators by their components, several investment. Among the industries, service enter- subcomponents of the connectivity, infrastructure prises and multiactivity enterprises have relatively services, and governance indicators become impor- high levels of investment, those in agricultural tant. Positive contributors are communities with processing relatively less. Enterprises in commu- �xed phone lines, garbage collection, governance nities with a high amount of agricultural land per on infrastructure and services, and number of bank capita have more investment. Of the benchmark services. Negative parameter values are found for indicators, human capital has a positive, signifi- proximity to main market and to post of�ce and cant parameter estimate. Three subindicators lead stable electricity supply. to higher investment: more widespread cell phone In Sri Lanka investment per enterprises relates use, concrete or asphalt roads, and number of signi�cantly with entrepreneur experience but not banks. Offsetting the latter is the negative effect of enterprise age. A male top manager and a more the number of bank services. educated manager also contribute to investment. In conclusion, capital investment seems to be Service enterprises have 40 percent lower levels of more dependent on enterprise characteristics than investment than trading enterprises, and enter- is employment generation. The community char- prises in agricultural processing have 68 percent acteristics and benchmark indicators that explain more. Community variables contribute to explain- investment and employment levels differ. Several ing differences only in the table’s last column, list- benchmark subindicators contribute to explana- ing selected benchmark components: enterprises tion; however, in other cases the indicators’ signs in larger communities with greater enterprise den- are contrary to expectations derived from theory, sity use more capital. Benchmark indicators are most likely because of multicollinearity and unob- ineffectual except for business services, in which served background variables. 98 Table C.4 Sales Regressions: Nicaragua BI and BI- BI and BI-components components with with control of industry BI with control of control of industry mix, enterprise variables Benchmark BI with control of industry mix and mix, enterprise with separation of indicators (BI) industry mix and enterprise and variables, and imputed and paid labor, with control of enterprise community community and capital and Variable name industry mix variables variables variables community variables Variant (1) (2) (3) (4) (5) N 846 846 846 846 846 r2 0.067* 0.426* 0.428* 0.441* 0.465* sva 0.284 0.286 0.286 0.286 0.288 smua 0.317 0.246 0.247 0.251 0.242 Factor and nonfactor inputs Total labor input# 0.225 4.964* 0.231 5.074* 0.232 5.110* Family labor input# 0.041 0.902 Paid labor input# 0.281 11.994* Total capital input# 0.167 4.117* 0.165 4.072* 0.164 4.048* Imputed capital cost# 0.210 5.915* Paid capital cost# 0.105 4.243* lnVLL (ln*ln) 0.047 8.998* 0.048 9.128* 0.048 9.199* LnIL*lnIL (imputed) 0.024 5.856* LnPL*lnPL (paid) 0.051 15.133* lnVCC (ln*ln) 0.024 8.735* 0.024 8.785* 0.024 8.639* lnIC*lnIC (imputed) 0.028 9.936* lnPC*lnPC (paid) 0.032 6.730* lnVLC (lnL*lnC) 0.010 1.543 0.010 1.460 0.010 1.447 lnIL*lnIC (imputed) 0.011 2.488* lnPL*lnPC (paid) 0.008 3.384* Depreciation# 0.060 6.327* 0.060 6.292* 0.061 6.483* 0.073 7.562* Non Factor Cost# 0.077 12.902* 0.077 12.955* 0.079 13.117* 0.074 12.223* Other enterprise characteristics Age of enterprise# 0.098 7.689* 0.098 7.632* 0.097 7.583* 0.101 7.836* Registration 0.044 1.426 0.045 1.461 0.046 1.492 0.023 0.734 Industry dummies/line of business Services enterprise 0.198 5.990* 0.113 3.388* 0.116 3.482* 0.111 3.340* 0.102 3.027* Manufacturing, nonagricultural enterprise 0.021 0.471 0.112 2.564* 0.114 2.610* 0.110 2.502* 0.096 2.182* Agricultural processing enterprise 0.041 1.126 0.001 0.034 0.000 0.002 0.004 0.097 0.040 1.090 Other production enterprise 0.006 0.092 0.073 1.117 0.072 1.100 0.076 1.164 0.061 0.930 Mixed enterprise 0.582 17.552* 0.184 5.313* 0.183 5.288* 0.172 4.967* 0.181 5.178* Community characteristics Agric. seasonality 0.042 0.559 0.047 0.616 0.053 0.714 Enterprise density 0.001 2.362* 0.000 0.802 0.000 0.333 Community pop size# 0.068 2.161* 0.011 0.298 0.032 0.930 Agric land per capita 0.048 1.164 0.035 0.808 0.052 1.244 Illiteracy 0.000 0.155 0.002 0.965 0.001 0.458 Benchmark indicators and components Connectivity 0.873 2.706* 0.811 3.133* 0.673 2.485* Time to near city 0.402 3.025* 0.423 3.277* Cost to near city 2.000 3.029* 2.091 3.262* Cost to main market 1.800 2.956* 1.770 2.999* Distance to post of�ce 0.259 2.267* 0.253 2.280* Infrastructure service 0.035 0.134 0.060 0.290 0.112 0.508 Percent with electricity 0.493 3.531* 0.390 2.868* Cost of transportation 0.256 2.525* 0.213 2.164* Garbage collection 0.267 2.662* 0.267 2.738* Business services 0.079 0.682 0.133 1.439 0.079 0.846 Technology services 0.160 1.859* 0.186 2.237* Governance 0.054 0.156 0.419 1.495 0.212 0.747 0.627 1.999* 0.446 1.462 Human capital 0.442 0.792 0.634 1.413 0.590 1.257 1.471 2.829* 1.170 2.313* Finance services 0.071 0.406 0.083 0.596 0.163 1.041 0.154 0.878 0.193 1.133 Intercept 6.450 26.831* 5.573 21.956* 5.212 15.198* 5.564 13.962* 5.571 14.313* Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*�: signi�cant at least at 10 percent. 99 100 Table C.5 Net Value Added Regressions: Nicaragua BI and BI- BI and BI-components components with with control of industry BI with control of control of industry mix, enterprise variables Benchmark BI with control of industry mix and mix, enterprise with separation of indicators (BI) industry mix and enterprise and variables, and imputed and paid labor, with control of enterprise community community and capital and Variable name industry mix variables variables variables community variables Variant (1) (2) (3) (4) (5) N 814 814 814 814 814 r2 0.059* 0.333* 0.336* 0.353* 0.382* sva 0.291 0.292 0.292 0.292 0.294 smua 0.311 0.300 0.295 0.295 0.273 Total labor input# 0.238 5.076* 0.242 5.162* 0.247 5.257* Family labor input# 0.039 0.815 Paid labor input# 0.392 16.031* Total capital input# 0.024 0.560 0.024 0.543 0.023 0.538 Imputed capital cost# 0.094 2.466* Paid capital cost# 0.100 3.838* lnVLL (ln*ln) 0.065 11.926* 0.065 12.025* 0.066 12.125* LnIL*lnIL (imputed) 0.024 5.364* LnPL*lnPL (paid) 0.068 19.363* lnVCC (ln*ln) 0.034 11.619* 0.035 11.695* 0.034 11.559* lnIC*lnIC (imputed) 0.029 9.769* lnPC*lnPC (paid) 0.030 6.047* lnVLC (lnL*lnC) 0.022 3.028* 0.022 3.092* 0.022 3.056* lnIL*lnIC (imputed) 0.002 0.416 lnPL*lnPC (paid) 0.011 4.270* Other enterprise characteristics Age of enterprise# 0.091 6.788* 0.090 6.721* 0.090 6.695* 0.095 7.074* Registration 0.108 3.578* 0.108 3.572* 0.110 3.668* 0.209 6.725* Industry dummies/line of business Services enterprise 0.225 6.513* 0.139 3.970* 0.142 4.062* 0.139 3.981* 0.130 3.687* Manufacturing, nonagr icultural enterprise 0.055 1.224 0.034 0.750 0.036 0.801 0.034 0.762 0.023 0.502 Agricultural processing enterprise 0.088 2.339* 0.005 0.140 0.006 0.149 0.008 0.211 0.052 1.364 Other production enterprise 0.102 1.540 0.232 3.454* 0.231 3.449* 0.236 3.523* 0.227 3.370* Mixed enterprise 0.525 15.158* 0.239 6.695* 0.238 6.679* 0.232 6.505* 0.250 6.919* Community characteristics Agricultural seasonality 0.018 0.201 0.046 0.516 0.058 0.695 Enterprise density 0.001 2.476* 0.001 0.894 0.000 0.330 Community population size# 0.067 1.840 0.024 0.583 0.056 1.458 Agricultural land per capita 0.046 0.956 0.044 0.846 0.056 1.166 Illiteracy 0.002 0.831 0.004 1.669* 0.003 1.358 Benchmark indicators and components Connectivity 1.033 3.245* 1.133 3.670* 0.930 2.934* Time to near city 0.371 2.415* 0.391 2.715* Cost to near city 1.551 1.998* 1.625 2.238* Cost to main market 1.878 2.688* 1.797 2.756* Distance post of�ce 0.340 2.558* 0.349 2.810* Infrastructure service 0.128 0.499 0.264 1.064 0.252 0.981 Percent with electricity 0.653 3.975* 0.501 3.244* Availability electricity 0.356 3.050* 0.328 3.005* Percent with protected water 0.299 2.126* 0.224 1.697* Garbage collection 0.294 2.538* 0.320 2.948* Business services 0.060 0.525 0.073 0.662 0.021 0.188 Accounting service 0.154 1.532 0.193 2.050* Governance 0.183 0.530 0.549 1.641 0.357 1.072 0.824 2.268* 0.588 1.727* Human capital 0.313 0.568 0.798 1.492 0.725 1.322 1.798 2.902* 1.481 2.550* Finance services 0.000 0.001 0.143 0.855 0.191 1.037 0.189 0.945 0.236 1.266 Intercept 6.297 26.457* 5.210 18.148* 4.970 12.672* 5.519 12.203* 5.554 12.968* Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*�: signi�cant at least at 10 percent. 101 102 Table C.6 Sales Regressions: Sri Lanka BI and BI- BI and BI-components components with with control of industry BI with control of control of industry mix, enterprise variables Benchmark BI with control of industry mix and mix, enterprise with separation of indicators (BI) industry mix and enterprise and variables, and imputed and paid labor, with control of enterprise community community and capital and Variable name industry mix variables variables variables community variables Variant (1) (2) (3) (4) (5) N 1018 1018 1018 1018 1018 r2 0.093 0.741 0.750 0.757 0.762 sva 1.311 0.737 0.737 0.737 0.729 smua 0.636 0.262 0.210 0.179 0.179 Factor and nonfactor inputs Total labor input# 0.309 1.033 0.325 1.094 0.281 0.948 Family labor input# 1.893 1.750* Paid labor input# 0.080 1.861* Total capital input# 0.095 0.554 0.091 0.532 0.076 0.441 Imputed capital cost# 0.251 2.474* Paid capital cost# 0.063 1.071 lnVLL (ln*ln) 0.081 2.476* 0.081 2.495* 0.077 2.364* LnIL*lnIL (imputed) 0.167 1.772* LnPL*lnPL (paid) 0.024 4.007* lnVCC (ln*ln) 0.046 3.976* 0.047 4.125* 0.047 4.133* lnIC*lnIC (imputed) 0.034 3.841* lnPC*lnPC (paid) 0.007 0.712 lnVLC (lnL*lnC) 0.073 2.200* 0.074 2.239* 0.071 2.165* Depreciation# 0.036 1.066 0.039 1.156 0.043 1.279 0.024 0.718 Non Factor Cost# 0.474 20.513* 0.473 20.607* 0.479 20.950* 0.471 20.806* Other enterprise characteristics Age of enterprise# 0.111 3.027* 0.116 3.178* 0.118 3.234* 0.127 3.464* Experience of manager# 0.075 1.808* 0.075 1.791* 0.066 1.585 0.045 1.085 Gender of manager 0.124 1.523 0.117 1.448 0.110 1.376 0.136 1.722* Education of manager 0.025 2.071* 0.026 2.204* 0.027 2.323* 0.024 2.027* Registration 0.094 1.158 0.094 1.167 0.089 1.124 0.089 1.134 Industry dummies/line of business Services enterprise 0.605 3.553* 0.211 2.044* 0.215 2.082* 0.208 2.023* 0.101 0.920 Manufacturing, nonagricultural enterprise 0.560 3.903* 0.031 0.343 0.020 0.219 0.021 0.234 0.053 0.554 Agricultural processing enterprise 0.316 0.891 0.068 0.332 0.085 0.420 0.086 0.424 0.028 0.141 Other production enterprise 0.483 1.214 0.068 0.298 0.097 0.427 0.027 0.118 0.085 0.373 Mixed enterprise 0.015 0.062 0.154 1.150 0.122 0.912 0.109 0.820 0.141 1.048 Community characteristics Agricultural seasonality 0.131 0.825 0.193 1.349 0.184 1.294 Enterprise density 0.000 0.276 0.002 0.992 0.001 0.725 Community population size# 0.006 0.069 0.017 0.201 0.019 0.233 Agricultural land per capita 0.412 1.628 0.121 0.489 0.157 0.641 Illiteracy 0.064 0.165 0.033 0.087 0.084 0.228 Main market in: Neighboring communities 0.369 2.613* 0.353 2.556* 0.326 2.374* Commercial center 0.108 0.758 0.256 1.809* 0.173 1.232 Nearest city 0.036 0.311 0.112 0.964 0.074 0.642 Main community income from: Wages 0.166 1.664* 0.132 1.425 0.144 1.575 Self-employment 0.156 1.028 0.199 1.364 0.220 1.523 Benchmark indicators and components Connectivity 1.167 1.669* 0.194 0.573 0.094 0.267 Cost of transportation 0.276 1.497 0.320 1.750* Infrastructure service 0.085 0.112 0.271 0.743 0.198 0.534 Concrete/asphalt road 0.241 2.601* 0.217 2.372* Business services 1.018 2.426* 0.394 1.968* 0.438 2.192* Management consult 0.416 2.521* 0.361 2.190* Governance 1.055 2.548* 0.117 0.115 0.190 0.178 Conflict resolution and contract enforcement 0.359 1.211 0.258 0.879 Public services/ institutions 2.420 1.714* 2.187 1.561 Human capital 1.266 0.606 0.083 0.139 0.126 0.208 0.152 0.280 0.118 0.220 Finance services 0.139 0.113 0.262 0.732 0.349 1.027 Access to loans 0.428 2.105* 0.471 2.327* Intercept 7.669 5.155* 2.890 2.564* 2.634 2.052* 2.175 0.685 6.279 1.447 Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value. a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*�: signi�cant at least at 10 percent. 103 104 Table C.7 Net Value Added Regressions: Sri Lanka BI and BI- BI and BI-components components with with control of industry BI with control of control of industry mix, enterprise variables Benchmark BI with control of industry mix and mix, enterprise with separation of indicators (BI) industry mix and enterprise and variables, and imputed and paid labor, with control of enterprise community community and capital and Variable name industry mix variables variables variables community variables Variant (1) (2) (3) (4) (5) N 841 841 841 841 841 r2 0.063 0.504 0.522 0.542 0.536 sva 1.258 0.950 0.950 0.950 0.965 smua 0.652 0.378 0.323 0.243 0.235 Factor and nonfactor inputs Total labor input# 0.840 1.956* 0.946 2.218* 0.912 2.158* Family labor input# 1.523 0.910 Paid labor input# 0.212 3.431* Total capital input# 0.951 3.960* 0.953 3.958* 0.927 3.879* Imputed capital cost# 0.368 0.977 Paid capital cost# 0.031 0.370 lnVLL (ln*ln) 0.211 4.406* 0.218 4.551* 0.213 4.498* LnIL*lnIL (imputed) 0.103 0.718 LnPL*lnPL (paid) 0.054 6.034* lnVCC (ln*ln) 0.057 3.138* 0.059 3.213* 0.059 3.293* lnIC*lnIC (imputed) 0.015 1.110 lnPC*lnPC (paid) 0.020 1.229 lnVLC (lnL*lnC) 0.215 4.289* 0.217 4.327* 0.214 4.313* lnIL*lnIC (imputed) 0.059 0.981 lnPL*lnPC (paid) 0.012 1.559 Other enterprise characteristics Age of enterprise# 0.096 1.836* 0.095 1.825* 0.102 1.962* 0.119 2.219* Experience of manager# 0.055 0.918 0.060 1.006 0.040 0.679 0.018 0.299 Gender of manager 0.026 0.223 0.025 0.213 0.032 0.278 0.076 0.644 Education of manager 0.051 2.968* 0.050 2.903* 0.053 3.087* 0.049 2.813* Registration 0.062 0.546 0.064 0.567 0.073 0.659 0.105 0.928 Industry dummies/line of business Services enterprise 0.082 0.454 0.159 1.157 0.148 1.072 0.128 0.926 0.028 0.182 Manufacturing, nonagricultural enterprise 0.133 0.882 0.070 0.582 0.085 0.705 0.081 0.671 0.173 1.307 Agricultural processing enterprise 1.003 2.766* 0.082 0.294 0.106 0.377 0.094 0.334 0.037 0.128 Other production enterprise 1.578 3.821* 0.093 0.291 0.086 0.270 0.155 0.488 0.206 0.626 Mixed enterprise 0.045 0.177 0.231 1.201 0.207 1.075 0.191 1.007 0.250 1.261 Community characteristics Agricultural seasonality 0.166 0.709 0.235 1.107 0.211 0.983 Enterprise density 0.002 0.992 0.003 1.131 0.002 0.850 Community population size# 0.036 0.280 0.011 0.099 0.027 0.233 Agricultural land per capita 0.491 1.259 0.224 0.622 0.280 0.775 Illiteracy 0.113 0.197 0.178 0.333 0.167 0.310 Main market in: Neighboring communities 0.542 2.642* 0.462 2.501* 0.457 2.459* Commercial center 0.294 1.407 0.417 2.225* 0.335 1.784* Nearest city 0.097 0.563 0.078 0.497 0.042 0.267 Main community income from: Wages 0.043 0.295 0.042 0.322 0.017 0.133 Self-employment 0.166 0.737 0.287 1.360 0.341 1.602 Benchmark indicators and components Connectivity 0.831 1.132 0.298 0.611 0.017 0.033 Cost of transportation 0.700 2.586* 0.723 2.664* Time to main market 0.346 1.815* 0.337 1.758* Infrastructure service 0.176 0.220 0.533 1.005 0.337 0.620 Concrete/asphalt road 0.187 1.434 0.171 1.305 Business services 0.862 1.989* 0.526 1.830* 0.587 1.987* 0.484 1.857* 0.440 1.665* Governance 0.033 0.015 0.207 0.141 0.108 0.069 Public services/ institutions 2.195 1.109 2.001 1.007 Human capital 0.551 0.429 0.325 0.382 0.319 0.365 0.340 0.433 0.100 0.128 Finance services 0.214 0.268 0.151 0.289 0.464 0.926 Access to loans 0.605 2.127* 0.681 2.384* Intercept 5.657 3.637* 2.709 1.687* 2.506 1.357 1.699 0.388 5.559 0.840 Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value. a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*� — signi�cant at least at 10 percent. 105 106 Table C.8 Sales Regressions: Tanzania BI and BI- BI and BI-components components with with control of industry BI with control of control of industry mix, enterprise variables Benchmark BI with control of industry mix and mix, enterprise with distinction of indicators (BI) industry mix and enterprise and variables, and imputed and paid labor, with control of enterprise community community and capital and Variable name industry mix variables variables variables community variables Variant (1) (2) (3) (4) (5) N 947 947 947 947 947 r2 0.036* 0.339* 0.342* 0.360 0.355 sva 0.940 0.800 0.800 0.800 0.797 smua 0.625 0.477 0.476 0.458 0.469 Factor and nonfactor inputs Total labor input# 0.740 1.840* 0.746 1.853* 0.705 1.751* Family labor input# 0.387 3.284* Paid labor input# 0.125 1.861* Total capital input# 0.318 2.923* 0.326 2.992* 0.317 2.914* Imputed capital cost# 0.209 3.050* Paid capital cost# 0.043 0.932 lnVLL (ln*ln) 0.024 0.694 0.024 0.683 0.021 0.597 LnIL*lnIL (imputed) 0.043 2.738* LnPL*lnPL (paid) 0.011 0.970 lnVCC (ln*ln) 0.019 3.344* 0.019 3.295* 0.020 3.422* lnIC*lnIC (imputed) 0.013 2.164* lnPC*lnPC (paid) 0.015 1.798* lnVLC (lnL*lnC) 0.076 3.899* 0.077 3.960* 0.077 3.932* lnIL*lnIC (imputed) 0.022 2.086* lnPL*lnPC (paid) 0.015 2.364* Depreciation# 0.009 0.412 0.009 0.418 0.004 0.167 0.038 1.574 Nonfactor Cost# 0.219 13.680* 0.219 13.613* 0.217 13.517* 0.217 13.484* Other enterprise characteristics Age of enterprise# 0.041 1.292 0.039 1.220 0.040 1.271 0.032 0.997 Experience of manager# 0.087 1.748* 0.091 1.839* 0.097 1.962* 0.092 1.855* Gender of manager 0.155 2.088* 0.150 2.009* 0.146 1.959* 0.152 2.031* Education of manager 0.014 1.242 0.014 1.243 0.013 1.145 0.014 1.259 Registration 0.330 4.067* 0.336 4.128* 0.329 4.059* 0.323 3.930* Industry dummies/line of business Services enterprise 0.011 0.114 0.079 0.937 0.079 0.933 0.069 0.819 0.042 0.492 Manufactruing, nonagricultural enterprise 0.162 0.927 0.047 0.318 0.047 0.318 0.027 0.180 0.051 0.342 Agricultural processing enterprise 0.125 0.653 0.136 0.825 0.119 0.721 0.133 0.808 0.096 0.582 Other production enterprise 0.155 0.706 0.039 0.204 0.036 0.189 0.036 0.186 0.054 0.277 Mixed enterprise 0.118 1.182 0.020 0.231 0.022 0.256 0.018 0.212 0.018 0.205 Community characteristics Community population size# 0.047 0.680 0.011 0.162 0.003 0.038 Agricultural land per capita 0.015 0.181 0.034 0.426 0.018 0.221 Illiteracy 0.001 0.353 0.002 0.884 0.002 0.862 Main community income from: Wages 0.045 0.124 0.189 0.502 0.127 0.334 Self-employment 0.228 1.406 0.092 0.595 0.070 0.441 Benchmark indicators and components Connectivity 0.773 1.410 1.051 2.420* 1.042 2.348* Cost of transportation 0.486 1.659 0.473 1.582 Time to main market 0.244 1.810* 0.251 1.825* Infrastructure services 0.341 0.618 0.759 1.730* 0.878 1.890* Sewage system 0.496 2.319* 0.552 2.535* Business services 0.541 1.183 0.032 0.090 0.021 0.057 Legal services 0.347 1.479 0.383 1.596 Insurance services 0.674 1.526 0.686 1.528 Governance 0.612 0.737 0.270 0.411 0.099 0.148 0.031 0.047 0.108 0.158 Human capital 1.955 1.501 1.181 1.136 0.906 0.836 0.736 0.725 0.838 0.812 Finance services 0.409 0.905 0.048 0.132 0.077 0.214 Number of banks 0.232 1.081 0.239 1.095 Access to loans 0.268 1.111 0.169 0.688 Intercept 5.939 11.902* 1.134 0.896 0.656 0.473 1.286 0.926 5.148 6.741* Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value. a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*�: signi�cant at least at 10 percent. 107 108 Table C.9 Net Value Added Regressions: Tanzania BI and BI- BI and BI-components components with with control of industry BI with control of control of industry mix, enterprise variables Benchmark BI with control of industry mix and mix, enterprise with distinction of indicators (BI) industry mix and enterprise and variables, and imputed and paid labor, with control of enterprise community community and capital and Variable name industry mix variables variables variables community variables Variant (1) (2) (3) (4) (5) N 699 699 699 699 699 r2 0.039* 0.194 0.199 0.204 0.214 sva 1.046 0.980 0.980 0.980 0.968 smua 0.559 0.484 0.482 0.489 0.490 Factor and nonfactor inputs Total labor input# 0.357 0.530 0.383 0.567 0.291 0.432 Family labor input# 0.148 0.925 Paid labor input# 0.190 1.759* Total capital input# 0.072 0.420 0.071 0.411 0.091 0.527 Imputed capital cost# 0.198 1.917* Paid capital cost# 0.046 0.672 lnVLL (ln*ln) 0.065 1.121 0.068 1.172 0.061 1.060 LnIL*lnIL (imputed) 0.018 0.834 LnPL*lnPL (paid) 0.046 2.427* lnVCC (ln*ln) 0.039 4.446* 0.039 4.451* 0.040 4.617* lnIC*lnIC (imputed) 0.032 3.511* lnPC*lnPC (paid) 0.021 1.673* lnVLC (lnL*lnC) 0.045 1.465 0.046 1.472 0.051 1.649* lnIL*lnIC (imputed) 0.006 0.367 lnPL*lnPC (paid) 0.014 1.335 Other enterprise characteristics Age of enterprise# 0.018 0.400 0.012 0.268 0.006 0.129 0.008 0.175 Experience of manager# 0.105 1.482 0.112 1.577 0.110 1.554 0.104 1.485 Gender of manager 0.109 1.037 0.101 0.955 0.094 0.888 0.108 1.022 Education of manager 0.036 2.176* 0.036 2.170* 0.035 2.121* 0.027 1.623* Registration 0.429 3.750* 0.439 3.827* 0.447 3.895* 0.408 3.549* Industry dummies/line of business Services enterprise 0.083 0.685 0.085 0.736 0.082 0.716 0.087 0.752 0.139 1.208 Manufacturing, nonagricultural enterprise 0.080 0.364 0.059 0.285 0.068 0.328 0.105 0.505 0.035 0.171 Agricultural processing enterprise 0.037 0.166 0.068 0.324 0.052 0.245 0.080 0.375 0.043 0.204 Other production enterprise 0.191 0.638 0.189 0.662 0.217 0.753 0.222 0.771 0.168 0.588 Mixed enterprise 0.150 1.157 0.047 0.380 0.081 0.644 0.090 0.720 0.075 0.602 Community characteristics Community population size# 0.044 0.541 0.051 0.627 0.038 0.459 Agricultural land per capita 0.138 1.382 0.093 0.975 0.088 0.922 Illiteracy 0.001 0.366 0.000 0.063 0.000 0.097 Main community income from: Wages 0.084 0.161 0.696 1.185 0.636 1.089 Self-employment 0.224 1.186 0.226 1.211 0.156 0.827 Benchmark indicators and components Connectivity 0.980 1.740* 1.019 2.006* 1.169 2.238* Rail stop 0.285 1.921* 0.256 1.721* Infrastructure services 0.436 0.663 0.104 0.175 0.078 0.126 Concrete/asphalt road 0.140 0.837 0.129 0.773 Business services 0.038 0.076 0.287 0.641 0.310 0.677 Legal services 0.266 0.903 0.285 0.965 Insurance services 1.338 2.323* 1.316 2.296* Governance 0.385 0.446 0.437 0.565 0.319 0.411 Public service/ institution 0.312 0.969 0.307 0.956 Human capital 1.114 0.771 0.039 0.029 0.173 0.126 0.330 0.248 0.142 0.107 Finance services 0.044 0.091 0.158 0.361 0.116 0.265 0.167 0.381 0.294 0.668 Intercept 5.375 10.193* 5.122 2.489* 4.612 2.149* 4.510 2.117 5.256 6.098* Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value. a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*�: signi�cant at least at 10 percent. 109 110 The Rural Investment Climate Table C.10 Labor Input Regressions: Nicaragua Industry mix, benchmark Industry mix, indicators, enterprise As in column 3, but benchmark indicators, characteristics, and with benchmark Industry mix and and enterprise other community indicators replaced Variable name benchmark indicators characteristics characteristics by components Variant (1) (2) (3) (4) N 846 846 846 846 r2 0.052 0.108 0.123 0.136 sva 0.284 0.285 0.285 0.285 smua 0.280 0.277 0.238 0.231 Enterprise characteristics Age of enterprise# 0.079 6.24* 0.077 6.09* 0.079 6.22* Registration 0.510 19.64* 0.516 19.93* 0.516 19.97* Industry dummies Services enterprise 0.167 5.06* 0.167 5.05* 0.169 5.12* 0.174 5.27* Manufacturing nonagricultural enterprise 0.117 2.68* 0.096 2.20* 0.093 2.14* 0.083 1.91* Agricultural processing enterprise 0.131 3.64* 0.126 3.49* 0.128 3.55* 0.124 3.42* Other production enterprise 0.373 5.77* 0.402 6.19* 0.397 6.14* 0.402 6.22* Mixed enterprise 0.549 16.57* 0.523 15.77* 0.518 15.62* 0.522 15.78* Community characteristics Agricultural seasonality 0.173 2.36* Enterprise density 0.001 1.51 0.001 2.17* Community population size# 0.051 1.67* 0.057 2.55* Agricultural land per capita 0.122 3.03* 0.097 2.46* Illiteracy 0.002 0.99 Benchmark indicators and components Connectivity 0.629 2.18* 0.742 2.59* 0.667 2.54* Infrastructure services 0.509 2.19* 0.603 2.62* 0.630 2.95* Percent with electricity 0.503 4.83* Percent with cellar phone 0.190 2.80* Sewage system 0.231 2.12* Business services 0.008 0.08 0.010 0.09 0.050 0.56 Governance 0.573 1.84* 0.324 1.05 0.455 1.65* Public services 0.410 2.05* Human capital 0.610 1.22 0.053 0.11 0.139 0.30 Finance services 0.145 0.93 0.084 0.54 0.003 0.02 Intercept 5.842 27.09* 5.494 25.55* 4.855 16.40* 5.313 21.69* Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value. a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*�: signi�cant at least at 10 percent. Annex C. Enterprise Performance Regressions; Notes and Tables 111 Table C.11 Capital Input Regressions: Nicaragua Industry mix, benchmark Industry mix, indicators, enterprise As in column 3, but benchmark indicators characteristics and with benchmark Industry mix and and enterprise other community indicators replaced Variable name benchmark indicators characteristics characteristics by components Variant (1) (2) (3) (4) N 846 846 846 846 r2 0.053 0.216 0.213 0.242 sva 0.284 0.285 0.285 0.285 smua 0.479 0.368 0.357 0.339 Enterprise characteristics Age of enterprise# 0.134 10.56* 0.134 10.56* 0.136 10.69* Registration 1.329 50.91* 1.330 50.95* 1.324 50.77* Industry dummies Services enterprise 0.065 1.97* 0.058 1.74* 0.059 1.77* 0.053 1.61 Manufacturing, 0.075 1.71* 0.044 1.01 0.043 0.99 0.033 0.76 nonagricultural enterprise Agricultural processing enterprise 0.222 6.12* 0.204 5.63* 0.204 5.61* 0.201 5.55* Other production enterprise 0.063 0.98 0.098 1.50 0.096 1.47 0.101 1.56 Mixed enterprise 0.312 9.40* 0.246 7.39* 0.246 7.40* 0.246 7.39* Community characteristics Agricultural seasonality 0.181 1.75* Enterprise density 0.000 0.33 Community population size# 0.003 0.07 0.067 1.65* Agricultural land per capita 0.084 1.47 0.118 2.03* Illiteracy 0.001 0.24 Benchmark indicators and components Connectivity 0.662 1.40 0.409 1.11 0.409 1.09 Time to main market 0.348 2.27* Distance to post of�ce 0.486 3.32* Infrastructure services 0.407 1.07 0.211 0.71 0.205 0.68 Cost of transportation 0.475 4.30* Percent with �xed phone 0.295 2.71* Garbage collection 0.572 4.35* Business services 0.090 0.53 0.115 0.87 0.086 0.66 Governance 0.363 0.71 0.284 0.71 0.241 0.61 Public services 0.915 3.24* Public institutions 0.696 2.48* Human capital 2.766 3.40* 1.435 2.25* 1.734 2.68* 1.916 3.11* Finance services 0.687 2.68* 0.500 2.50* 0.404 1.85* Number of bank services 0.594 2.50* Intercept 4.346 12.42* 3.568 12.96* 3.411 8.16* 4.762 10.69* Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value; a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*� — signi�cant at least at 10 percent. 112 The Rural Investment Climate Table C.12 Labor Input Regressions: Sri Lanka Industry mix, Industry mix, benchmark indicators, benchmark enterprise As in column 3, but indicators, and characteristics, and with benchmark Industry mix and enterprise other community indicators replaced Variable name benchmark indicators characteristics characteristics by components Variant (1) (2) (3) (4) N 1018 1018 1018 1018 r2 0.040 0.138 0.176 0.186 sva 0.759 0.736 0.736 0.736 smua 0.340 0.276 0.220 0.198 Enterprise characteristics Age of enterprise# 0.006 0.16 0.007 0.18 0.003 0.08 Experience manager# 0.058 1.38 0.063 1.52 0.064 1.55 Gender of manager 0.206 2.58* 0.189 2.38* 0.186 2.37* Education of manager 0.042 3.67* 0.043 3.80* 0.041 3.69* Registration 0.246 3.23* 0.249 3.30* 0.228 2.96* Industry dummies Services enterprise 0.088 0.89 0.144 1.51 0.160 1.68* 0.160 1.68* Manufacturing, nonagricultural enterprise 0.056 0.68 0.128 1.52 0.110 1.30 0.110 1.31 Agricultural processing enterprise 0.594 2.90* 0.532 2.68* 0.543 2.75* 0.515 2.62* Other production enterprise 1.042 4.54* 0.943 4.26* 0.951 4.31* 0.965 4.46* Mixed enterprise 0.238 1.73* 0.154 1.16 0.154 1.16 0.157 1.19 Community characteristics Agricultural seasonality 0.227 1.41 Enterprise density 0.004 2.44* 0.003 2.14* Community population size# 0.021 0.24 Agricultural land per capita 0.258 1.01 0.226 1.03 Illiteracy 0.039 0.10 Main market in: Neighboring communities 0.368 2.55* 0.346 2.58* Commercial center 0.361 2.50* 0.251 1.79* Nearest city 0.202 1.70 0.218 1.93* Main community income from: Wages 0.145 1.43 Self-employment 0.089 0.58 Benchmark indicators and components Connectivity 0.245 0.63 0.211 0.61 0.077 0.22 Infrastructure services 0.277 0.66 0.216 0.58 0.313 0.84 Availability electricity 0.180 1.72* Percent with �xed phone 1.036 2.63* Business services 0.277 1.21 0.284 1.40 0.304 1.50 0.353 1.89* Governance 0.432 0.37 1.016 0.98 0.910 0.84 Infrastructure and services 0.503 1.66* Rule of law 0.709 1.01 Human capital 0.182 0.27 0.322 0.53 0.404 0.67 Finance services 0.022 0.05 0.072 0.20 0.210 0.61 Intercept 6.063 7.36* 5.689 7.65* 5.464 5.70* 4.715 17.71* Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value. a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*� — signi�cant at least at 10 percent. Annex C. Enterprise Performance Regressions; Notes and Tables 113 Table C.13 Capital Input Regressions: Sri Lanka Industry mix, Industry mix, benchmark indicators, benchmark enterprise As in column 3, but indicators, and characteristics, and with benchmark Industry mix and enterprise other community indicators replaced Variable name benchmark indicators characteristics characteristics by components Variant (1) (2) (3) (4) N 1018 1018 1018 1018 r2 0.079 0.294 0.306 0.329 sva 1.403 1.269 1.269 1.269 smua 0.657 0.467 0.410 0.395 Enterprise characteristics Age of enterprise# 0.028 0.445 0.025 0.404 0.010 0.15 Experience manager# 0.152 2.119* 0.155 2.154* 0.156 2.20* Gender of manager 0.654 4.753* 0.641 4.668* 0.603 4.44* Education of manager 0.112 5.697* 0.108 5.532* 0.111 5.65* Registration 0.822 6.277* 0.828 6.345* 0.818 6.30* Industry dummies Services enterprise 0.232 1.274 0.389 2.363* 0.403 2.442* 0.403 2.45* Manufacturing, nonagricultural enterprise 0.438 2.859* 0.193 1.323 0.188 1.286 0.174 1.20 Agricultural processing enterprise 0.851 2.247* 0.690 2.020* 0.682 1.997* 0.635 1.87* Other production enterprise 0.820 1.929* 0.531 1.393 0.581 1.520 0.621 1.63 Mixed enterprise 0.196 0.771 0.022 0.094 0.012 0.052 0.025 0.11 Community characteristics Agricultural seasonality 0.154 0.542 Enterprise density 0.002 0.698 0.005 2.00* Community population size# 0.230 1.457 0.281 2.02* Agricultural land per capita 0.126 0.276 Illiteracy 0.527 0.759 Main market in: Neighboring communities 0.050 0.195 Commercial center 0.047 0.184 Nearest city 0.033 0.155 Main community income from: Wages 0.028 0.159 Self-employment 0.294 1.073 0.385 1.53 Benchmark indicators and components Connectivity 1.050 1.431 0.938 1.597 0.860 1.365 Infrastructure services 0.856 1.079 0.672 1.056 0.426 0.642 Percent with protected water 0.856 2.78* Percent with �xed phone lines 1.640 2.20* Concrete/asphalt road 0.333 2.09* Business services 0.585 1.347 0.609 1.750* 0.770 2.142* Accounting service 0.716 3.42* Governance 0.092 0.042 2.117 1.192 2.448 1.285 Human capital 0.708 0.547 0.305 0.295 0.370 0.343 Finance services 0.348 0.443 0.191 0.304 0.089 0.147 Intercept 4.704 3.015* 3.919 3.078* 2.415 1.422 0.221 0.21 Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value. a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*�: signi�cant at least at 10 percent. 114 The Rural Investment Climate Table C.14 Labor Input Regressions: Tanzania Industry mix, Industry mix, benchmark indicators, benchmark enterprise As in column 3, but indicators, and characteristics, and with benchmark Industry mix and enterprise other community indicators replaced Variable name benchmark indicators characteristics characteristics by components Variant (1) (2) (3) (4) N 947 947 947 947 r2 0.041 0.067 0.086 0.102 sva 0.492 0.486 0.486 0.486 smua 0.324 0.320 0.308 0.305 Enterprise characteristics Age of enterprise# 0.008 0.40 0.007 0.38 0.006 0.33 Experience manager# 0.030 0.99 0.033 1.09 0.036 1.22 Gender of manager 0.158 3.51* 0.159 3.54* 0.155 3.49* Education of manager 0.021 3.18* 0.021 3.12* 0.022 3.30* Registration 0.051 1.10 0.050 1.08 0.047 1.02 Industry dummies Services enterprise 0.068 1.34 0.041 0.81 0.041 0.81 0.043 0.84 Manufacturing, nonagricultural enterprise 0.107 1.16 0.096 1.06 0.086 0.95 0.085 0.94 Agricultural processing enterprise 0.206 2.05* 0.223 2.24* 0.220 2.22* 0.229 2.32* Other production enterprise 0.313 2.73* 0.272 2.37* 0.272 2.37* 0.260 2.27* Mixed enterprise 0.085 1.63 0.064 1.22 0.064 1.21 0.071 1.37 Community characteristics Community population size# 0.109 2.53* 0.077 2.03* Agricultural land per capita 0.052 0.99 Illiteracy 0.001 0.59 Main community income from: Wages 0.174 0.77 Self-employment 0.017 0.16 Benchmark indicators and components Connectivity 0.348 1.22 0.407 1.45 0.377 1.34 Distance post of�ce 0.250 2.66* Infrastructure services 0.121 0.42 0.146 0.52 0.189 0.64 Business services 0.389 1.64 0.318 1.35 0.361 1.56 Marketing service 0.216 2.07* Governance 0.048 0.11 0.068 0.16 0.004 0.01 Public services/ institutions 0.284 1.69* Human capital 0.410 0.61 0.319 0.47 0.701 1.02 Finance services 0.103 0.44 0.067 0.29 0.065 0.28 Intercept 5.083 19.59* 4.867 18.29* 5.719 13.08* 5.346 15.52* Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value. a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*�: signi�cant at least at 10 percent. Annex C. Enterprise Performance Regressions; Notes and Tables 115 Table C.15 Capital Input Regressions: Tanzania Industry mix, Industry mix, benchmark benchmark indicators, enterprise As in column 3, indicators, and characteristics, and but with benchmark Industry mix and enterprise other community indicators replaced Variable name benchmark indicators characteristics characteristics by components Variant (1) (2) (3) (4) N 947 947 947 947 r2 0.071 0.154 0.168 0.180 sva 1.835 1.743 1.743 1.743 smua 1.145 1.103 1.077 1.047 Enterprise characteristics Age of enterprise# 0.016 0.23 0.023 0.34 0.031 0.46 Experience manager# 0.356 3.32* 0.367 3.42* 0.369 3.46* Gender of manager 0.507 3.16* 0.489 3.04* 0.501 3.14* Education of manager 0.024 1.00 0.024 0.98 0.025 1.05 Registration 1.332 7.99* 1.337 8.03* 1.311 7.89* Industry dummies Services enterprise 0.502 2.66* 0.298 1.64* 0.304 1.68* 0.306 1.69* Manufacturing, nonagricultural enterprise 0.179 0.53 0.078 0.24 0.068 0.21 0.093 0.29 Agricultural processing enterprise 0.851 2.28* 0.900 2.53* 0.916 2.57* 0.861 2.43* Other production enterprise 0.033 0.08 0.185 0.45 -0.141 0.34 0.147 0.36 Mixed enterprise 0.565 2.92* 0.459 2.47* 0.505 2.68* 0.497 2.66* Community characteristics Community population size# 0.203 1.33 Agricultural land per capita 0.334 1.79* 0.307 1.88* Illiteracy 0.002 0.43 Main community income from: Wages 0.394 0.49 Self-employment 0.381 1.05 Benchmark indicators and components Connectivity 1.512 1.48 1.238 1.26 0.921 0.93 Infrastructure services 1.466 1.42 1.337 1.35 1.513 1.46 Percent with cell phones 1.367 2.23* Concrete/asphalt road 0.576 2.01* Business services 0.353 0.41 0.119 0.15 0.016 0.02 Legal services 0.595 1.37 Governance 0.979 0.63 1.268 0.85 0.897 0.60 Human capital 5.409 2.23* 5.022 2.14* 4.561 1.89* 4.282 1.92* Finance services 0.180 0.21 0.340 0.42 0.470 0.58 Number of banks 0.926 1.64 Number of bank services 1.591 1.89* Intercept 3.106 3.34* 2.314 2.49* 0.168 0.11 1.355 2.37* Source: RIC Surveys. Notes: See Annex A for de�nitions and description of data; # logarithmic value. a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. “*�: signi�cant at least at 10 percent. Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys INTRODUCTION HOUSEHOLD DEMOGRAPHICS This annex discusses the household demograph- AND CHARACTERISTICS ics and characteristics among the survey house- holds with brief notes on estimation methods Age Distribution used for key variables such as income, assets, The sample households’ age distributions in all three investments, and a human capital index. It also countries skewed to the left, as shown in Table D.2, reviews distribution of households across these meaning the countries’ households have a larger key variables as well as averages reported. Assets, number of younger people. Tanzania has the high- investment, and income are recorded in U.S. est percentage share (42.5 percent) of population dollars using the 2004 exchange rate of each below 16 years of age followed by Nicaragua country.82 (38.4 percent) and Sri Lanka (26.4 percent). In The estimation of statistics about rural house- contrast, Sri Lanka has the highest share (66.8 per- holds and the rural population employed sam- cent) of working age population (16 to 60 years) fol- pling weights for Sri Lanka and Nicaragua, but lowed by Nicaragua (56.1percent) and Tanzania unfortunately not for Tanzania, where weights (53.5 percent). Percentage shares of the male popu- are not available. Sampling weights are neces- lation of the sample households are consistently sary because household sampling was stratified lower than their female counterparts; however the by the presence of a household-based enterprise. share is lowest in Tanzania (48.3 percent) and high- This implies that the statistics for Tanzania must est in Sri Lanka (49.6 percent), with Nicaragua in the be interpreted with caution: statistics of vari- ables expected to be correlated with the pres- Table D.1 Sri Lanka: Availability of Sampling ence of an enterprise might be significantly Weights by Province biased. Sampling weights for Sri Lanka were con- Number of Percentage with structed ex-post from a comparison of sample Province household available weight information with a community listing. Unfortu- Western 219 100.0 nately, the listings of households were not avail- Central 123 91.9 able (or incomplete) for 18 communities, leading Southern 172 100.0 to 106 households (10 percent) for which weights North Western 116 100.0 are not available. As Table D.1 shows, these North Central 71 100.0 Uva 84 100.0 households are primarily concentrated in the Sabaragamuwa 116 100.0 war-torn areas in the east and northeast. Statis- Eastern 79 3.8 tics computed with weights can thus represent North & East 83 75.9 only those communities for which weights are Overall 1,063 90.0 available, therefore excluding the war-torn areas Source: RIC Survey, Tanzania. of Sri Lanka. 117 118 The Rural Investment Climate Table D.2 Age Distribution (%) of Household Members: Sri Lanka, Tanzania, and Nicaragua Sri Lanka Tanzania Nicaragua Age group Male Female All Male Female All Male Female All Household members (including household heads): 16 28.0 24.8 26.4 42.6 42.4 42.5 40.6 36.4 38.4 16–30 28.5 28.8 28.7 26.1 29.4 27.8 27.9 28.8 28.4 31–45 21.1 23.1 22.1 18.0 17.5 17.8 17.0 18.1 17.6 46–60 16.1 15.9 16.0 9.0 6.6 7.8 7.5 9.2 8.4 60 6.2 7.5 6.9 4.2 4.0 4.1 7.0 7.6 7.3 Total 100 100 100 100 100 100 100 100 100 N 2082 2113 4195 3909 4106 8015 3799 4071 7870 Household heads: 16–30 9.9 2.8 8.7 21.9 19.2 21.5 19.3 11.4 16.6 31–45 39.0 18.7 35.4 42.8 35.8 41.7 41.5 30.5 37.7 46–60 36.8 38.2 37.0 24.1 24.2 24.2 20.2 29.9 23.5 60 14.4 40.3 18.9 11.2 20.8 12.7 19.1 28.2 22.2 Total 100 100 100 100 100 100 100 100 100 N 814 143 957 1351 260 1611 963 572 1535 Source: RIC Surveys. middle (48.7 percent). Despite the lower shares of Education males, the distribution by age group does not vary The raw data on education are not comparable signi�cantly when it is partitioned by the male- across the selected countries owing partly to their female divide, except in Nicaragua. different education systems and partly to the ways A fair number of the household heads in the the questionnaires were framed. Sri Lanka fol- three countries are female, ranging between 15 and lowed the standard format: 1 to 5 being primary, 16 percent in Sri Lanka and Tanzania. Surprisingly, 6 to 8 pre-secondary, 9 to12 secondary, and above almost 37 percent of households are headed by 12 tertiary or college. Although Nicaragua has a females in Nicaragua. Although most of the house- system of grades and levels similar to Sri Lanka’s, hold heads are within the working-age cohort, data were collected at two levels. The �rst level col- notable numbers of them are over 60 years of age. A lected by categories, such as preschool, primary, higher proportion of female household heads does secondary, university, and so on, and the second not belong to the working-age category in these level ascertained the number of years in each cate- countries. All together, the percentage share of non- gory. These two levels were converted to grade working-age household heads is the highest in level by adding the number of years in the previous Nicaragua (19.5 percent) and lowest in Tanzania category to the number of years completed in the (12.7 percent), which is consistent with the higher assigned category. If an individual completed three percentage of younger population in the country. years of secondary, for example, the number of years required to complete the primary level Household Size were added to the three secondary-level years. In The average household size is highest in addition to these categories, the Sri Lanka and Tanzania (4.98) and lowest in Sri Lanka (4.26). It Nicaragua surveys also utilized other categories, is slightly less in Nicaragua (4.85) than in such as pre-primary, adult education, and basic, Tanzania (Table D.3). On average, numbers of fe- medium, and superior level technician. The same male members were higher than male members method was followed to obtain a corresponding in all of the pilot surveys. In Sri Lanka most of the grade level for each respondent. In Tanzania, stan- households (92.3 percent) have 1 to 6 members, dard 1 through standard 8 is considered primary while in Tanzania and Nicaragua the percentage and pre-form 1 through form 6 as secondary, fol- share of household members within this range lowed by college or university. For comparability, are 75.5 percent and 81.6 percent, respectively. standards 1 through 5 were converted to primary, Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 119 Table D.3 Distribution (%) of Households by Household Size and Gender Sri Lanka (n 957) Tanzania (n 1611) Nicaragua (n 1535) Household size Male Female All Male Female All Male Female All 0 3.1 0.8 0.0 5.0 4.8 0.0 5.6 3.54 0.0 1 28.4 29.9 2.7 27.6 22.6 5.9 25.6 24.61 3.3 2 33.9 40.7 9.4 23.9 27.3 10.4 27.0 29.89 10.1 3 23.2 18.6 19.2 21.1 21.9 12.7 22.5 21.1 15.3 4 9.2 7.0 26.9 13.1 11.7 16.5 11.5 11.73 18.8 5 1.4 2.6 23.4 5.4 6.9 16.6 5.6 4.97 21.0 6 0.3 0.5 10.7 3.0 3.2 13.4 1.5 1.91 13.1 7 0.6 4.8 0.6 1.1 9.0 0.5 0.98 7.0 8 1.6 0.2 0.4 5.8 0.0 0.25 5.2 9 1.0 0.1 0.1 4.8 0.2 0.84 2.6 10 0.3 0.1 4.9 0.0 0.19 3.6 All 100 100 100 100 100 100 100 100 100 Average 2.15 2.11 4.26 2.4 2.5 5.0 2.36 2.49 4.85 Source: RIC Surveys. standards 6 through nine9 to junior high, and pre- by Bils and Klenow (2000), which is based on the form 1 through form 6 to secondary. classic Mincerian returns to schooling and experi- Table D.4 provides information on the educa- ence. De�ne h as the human capital stock, and let a tion of the household members and household denote age, and let s refer to years of schooling. heads. It is clear from the table that the percentage Then, abstracting from an irrelevant scaling con- of the population that has received at least some stant,83 the log of the human capital stock is mea- schooling is impressive in all three pilot countries, sured with the following formula. ranging from 70 to 90 percent, with Sri Lanka ln(h(s, a)) s 1 (a s 6) (a 2 s 6)2 highest and Nicaragua lowest. In Tanzania most (72 percent) of the eligible members have com- Bils and Klenow (2000) obtain parameter estimates pleted primary level education or more, while in for , 1 and 2 from averages of 52 countries: other two countries most of members have com- 0.099, 1 0.512, and 2 0.00071.84 Thus, the pleted post-primary education, including the natural logarithm of the human capital stock is junior high and secondary levels. More than 10 measured for each individual as percent of household members in Sri Lanka, 4 per- cent in Nicaragua, and 5 percent in Tanzania also ln (h) 0.099s 0.0512 (a s 6) 0.00071(a s 6)2 have a tertiary level education. Both literacy rates and the human capital stock is found simply by and educational achievement among female taking the antilog: household members is lower than for males in h eln(h) both Sri Lanka and Tanzania but higher in The human capital stock is computed for each in- Nicaragua. dividual between 16 and 65 years of age. The low- The education of household heads shows a simi- est possible value of h over all values of s and a lar pattern. Household heads rates of schooling are equals approximately 1.55, achieved at a 16 and highest in Sri Lanka (93 percent) and lowest in s 0; the highest value depends on the maximum Nicaragua (73 percent). Among household heads in attainable level of education: for s 14, the maxi- Tanzania it is 86 percent. In Sri Lanka and Tanzania, mum equals about 10.06 for someone aged a 56 however, female household heads received almost years old; for s 15, it equals about 11.11 for some- two years of schooling less than male heads. one aged a 57 years old. Since the longest length of schooling effort depends on the school system in Human Capital a country (and possibly the precision of the coding A human capital stock value and index is com- system used in the survey), h is scaled to a number puted on the basis of the methodology developed between 0 and 1, and we shall refer to this as the 120 The Rural Investment Climate Table D.4 Distribution (%) of Household Members and Household Heads by Level of Education in Selected Countries Sri Lanka Tanzania Nicaragua Male Female All Male Female All Male Female Total Household members: 0 5.5 9.7 7.6 26.7 30.6 28.7 31.8 28.5 30.1 1–5 27.3 27.0 27.1 26.0 25.6 25.8 32.7 30.8 31.7 6–8 20.4 15.4 17.9 36.9 37.4 37.1 20.4 21.9 21.1 9–12 35.9 38.6 37.2 3.9 2.9 3.4 11.6 13.9 12.8 >=13 11.0 9.4 10.2 6.5 3.5 5.0 3.5 4.9 4.3 Total 100 100 100 100 100 100 100 100 100 Mean years 7.6 7.2 7.4 5.3 4.7 5.0 4.0 4.6 4.3 N 1984 2011 3995 3909 4106 8015 3790 4064 7856 Household Heads: 0 4.8 16.4 6.8 11.3 27.3 13.8 27.8 27.4 27.6 1–5 30.1 35.0 31.0 16.7 15.8 16.5 29.9 37.5 32.5 6–8 20.9 19.5 20.7 57.5 47.7 55.9 24.9 18.5 22.7 9–12 34.1 25.8 32.6 3.6 2.7 3.5 13.0 9.8 11.9 >=13 10.1 3.3 8.9 11.0 6.5 10.2 4.5 6.8 5.3 Total 100 100 100 100 100 100 100 100 100 Mean years 7.4 5.7 7.1 6.5 5.0 6.2 4.5 4.4 4.5 N 810 142 952 1351 260 1611 959 570 1529 Source: RIC Surveys. “human capital index�: At the household level, both the human capital stock and the index can be presented as average, h min(h) hindex sum, or maximum individual stock or index. In max(h) min(h) Table D.6 and Table D.7 provide household level This ensures comparability between countries, averages among adults of human capital stocks while obviously retaining comparability between and indexes, respectively. The average human individuals within a country or between commu- capital stock is the highest (4.66) in Sri Lanka with nities of a country. no significant difference between males and For the variable s, years of schooling, the coding females, lowest in Nicaragua (3.61) with no gen- in the surveys does not measure years of schooling der difference, and in between in Tanzania (3.77) as much as levels of attained education. Thus, the with higher average stock among males than survey information must be transformed to derive females. When the distributions of households by the estimate of years of schooling s used in the average human capital stock are compared, it computations of human capital stock. The coding shows an almost normal distribution for Sri Lanka in the household survey was scaled as indicated in and for Nicaragua and Tanzania distributions Table D.5. Nicaragua is not represented in this highly skewed to the left. The average human table: this questionnaire used a different coding capital stock is less than four for 66 percent of scheme, allowing the respondent to state how households in Tanzania and 69 percent in many years were completed at a particular school- Nicaragua, as opposed to 34 percent in Sri Lanka. ing level. To turn this into years of schooling, In Sri Lanka 15 percent of households have an assumptions are made about the number of years average human capital stock of more than 6; the required to reach the indicated level of schooling: same is only 2 percent in Tanzania and 6 percent 6 for junior secondary, 9 for senior secondary or in Nicaragua. mid-level technical training, and 12 for university Table D.7 provides distribution of households by and advanced technical training, with a peak of their average human capital index. Like human 15 years on the total years of schooling taken.85 capital stock, the average human capital index is also Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 121 Table D.5 Coding of Schooling Level: Original and Converted Sri Lanka Tanzania Assigned years of Measured schooling level Original coding schooling Original coding New coding Year 1 1 1 Standard 1 1 Year 2 2 2 Standard 2 2 Year 3 3 3 Standard 3 3 Year 4 4 4 Standard 4 4 Year 5 5 5 Standard 5 5 Year 6 6 6 Standard 6 6 Year 7 7 7 Standard 7 7 Year 8 8 8 Standard 8 8 Year 9 9 9 Training 9 Year 10 10 10 Pre form 1 9 Year 11 11 11 Form 1 10 Year 12 12 12 Form 2 11 Year 13 13 13 Form 3 12 University 14 15 Form 4 13 Professional 15 14 Form 5 14 Technical College 16 14 Form 6 15 Pre-school 17 0 Training 16 No schooling 18 0 University 16 Other 19 Missing Adult education 1 20 (?) 20 (?) Missing Under std. 1 0 Don’t know 99 Missing Source: RIC Surveys. Note: At the university level, years of schooling are not differentiated any further. Table D.6 Distribution (%) of Households by Average Human Capital Stock Sri Lanka Tanzania Nicaragua Average human capital stock Male Female All Male Female All Male Female All 2.5 4.9 9.0 4.7 10.3 16.8 10.1 22.8 21.1 19.6 2.5–3.0 8.5 8.8 9.9 10.5 15.5 12.7 17.1 19.6 16.7 3.0–3.5 8.7 10.2 6.4 13.8 19.9 19.4 16.4 16.3 22.0 3.5–4.0 14.3 10.3 12.9 20.5 21.0 23.7 12.5 13.0 11.1 4.0–4.5 12.0 9.6 17.0 13.1 12.4 14.4 11.7 10.7 10.3 4.5–5.0 8.5 12.9 12.6 15.0 8.1 9.7 4.5 6.2 7.3 5.0–5.5 10.5 12.2 9.9 6.3 2.7 3.8 4.0 4.8 4.3 5.5–6.0 7.5 7.5 8.7 2.5 1.3 1.9 1.5 1.8 2.6 6.0–6.5 8.8 5.8 6.7 1.7 0.5 2.0 3.4 2.3 2.0 6.5 16.4 13.9 11.1 6.5 1.8 2.4 6.3 4.2 4.1 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Meana 4.83 4.63 4.66 4.11 3.54 3.77 3.68 3.61 3.61 Standard Dev 1.63 1.59 1.45 1.48 1.07 1.10 1.56 1.40 1.30 N 903 919 947 1425 1487 1585 1309 1448 1506 Source: RIC Surveys. Note: The overall average is not a simple average of the averages of the gender-speci�c columns. 122 The Rural Investment Climate Table D.7 Distribution (%) of Households by Average Human Capital Index Sri Lanka Tanzania Nicaragua Average human capital stock index Male Female All Male Female All Male Female All 0.0–0.1 5.2 9.3 4.7 13.3 20.3 12.5 24.1 23.5 20.1 0.1–0.2 16.4 17.9 16.0 29.5 40.4 39.6 32.0 32.6 37.5 0.2–0.3 24.6 19.5 27.8 33.4 30.1 33.9 22.9 23.3 21.2 0.3–0.4 17.5 22.8 21.8 14.5 6.2 9.2 8.9 10.2 11.4 0.4–0.5 17.1 13.8 16.0 4.0 1.8 3.0 4.2 5.8 4.8 0.5–0.6 10.3 11.3 8.0 2.1 0.6 1.1 4.0 2.0 3.2 0.6–0.7 4.6 3.3 3.9 1.7 0.5 0.4 2.3 1.3 1.3 > 0.7 4.3 2.1 1.7 1.5 0.2 0.3 1.7 1.4 0.5 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Mean 0.35 0.32 0.33 0.24 0.19 0.21 0.22 0.21 0.21 Standard Dev 0.17 0.17 0.15 0.14 0.10 0.10 0.16 0.15 0.14 N 903 919 947 1425 1487 1585 1309 1448 1506 Source: RIC Surveys. higher in Sri Lanka (0.33), compared with a value of Distribution of adult members by gender reveals 0.21 in both Nicaragua and Tanzania. The distribu- that 83 percent and 88 percent of the households in tion of the households by average human capital Sri Lanka have 1 to 2 male and female members, re- index is almost perfectly normal in Sri Lanka and spectively, while these �gures are 80 percent and 83 relatively skewed to the left in Nicaragua and percent in Tanzania and 76 percent and 82 percent Tanzania. The distribution in Tanzania is concen- in Nicaragua. trated below 0.30. In Sri Lanka, 48 percent of house- Whereas the average household in Sri Lanka has holds have an index less than 0.30, and in Tanzania 1.5 males and 1.5 females, 1.2 males and 0.6 females and Nicaragua 86 percent and 79 percent of house- are engaged in an economic activity of some kind, holds, respectively, fall below the 0.30 threshold. whether working on the farm, being self-employed in a household enterprise, or being employed for a wage or salary. This implies a labor force participa- LABOR FORCE tion rate of 0.77 for males and 0.39 for females. Potential Labor Force: Number of Adults Occupation at the Time of the Survey in the Rural Population The present surveys offer two types of information As a �rst step in examining the rural labor force, about occupation. First is the activity occupying re- Table D.8 considers the number of working adults spondents at the time of the survey. Actually, only (within the age range of 16 to 65 years) in the sample the Sri Lanka questionnaire asks for “Activities in- households in the pilot countries. It is clear from the volved at present� and does not offer respondent a table that the average number of working-age chance to list several activities. The response adults is the highest (2.99) in Sri Lanka, followed by should therefore indicate the primary activity. Nicaragua (2.80). Tanzania has the lowest (2.76) Other countries did not collect this current occupa- number. Average numbers of working-age female tional pro�le. Occupational pro�les of the house- members are, however, higher than male members hold members and household heads are presented in Nicaragua, with no signi�cant difference be- in Table D.9. It shows that in Sri Lanka about tween male and female adults in both Sri Lanka and 36 percent of eligible household members are either Tanzania. In Sri Lanka, 84 percent of the households salaried or wage employed and 16 percent are have 2 to 4 adult male and female members eligible self-employed. Another 7.5 percent and 8.1 percent to work, while in Tanzania and Nicaragua the �g- of eligible household members are unemployed ures are 77 percent and 74 percent, respectively. or students, respectively. Almost 25 percent of Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 123 Table D.8 Potential and Actual Workforce: Number of Adults Present and Working Sri Lanka (n 957) Tanzania (n 1586) Nicaragua (n 1490) Number of adult members present Male Female All Male Female All Male Female All 0 4.7 2.3 0.0 10.0 6.0 0.0 13.1 5.1 0.0 1 59.1 57.9 4.3 60.5 62.9 11.3 58.9 59.8 12.3 2 24.0 30.0 41.4 19.5 19.8 45.2 17.3 22.2 44.4 3 9.0 6.4 24.1 6.0 8.2 21.4 6.5 9.0 18.4 4 2.2 3.3 18.0 3.1 2.1 10.8 2.9 2.5 10.9 5 0.4 0.2 7.8 0.6 0.6 6.3 1.1 0.9 7.8 6 0.6 2.8 0.1 0.3 2.5 0.2 0.5 4.3 7 0.6 0.0 0.1 1.0 0.1 0.8 8 1.0 0.1 0.1 1.5 1.1 Total 100 100 100 100 100 100 100 100 100 Average 1.48 1.51 2.99 1.35 1.42 2.76 1.31 1.49 2.80 Sri Lanka (n 957) Tanzania (n 1586) Nicaragua (n 1490) Number of working adult members Male Female All Male Female All Male Female All 0 8.8 48.9 0.0 12.4 10.1 1.3 13.1 39.9 0.0 1 72.5 43.3 47.0 64.2 65.5 15.5 64.0 46.8 42.9 2 13.8 7.2 36.8 16.0 16.8 48.9 16.5 10.0 33.1 3 4.7 0.4 11.3 4.3 5.4 17.7 4.3 2.4 13.6 4 0.3 0.2 4.3 2.4 1.5 8.6 1.5 0.3 6.3 5 0.6 0.4 0.3 4.2 0.5 0.2 2.7 6 0.1 0.1 0.3 2.0 0.2 0.5 0.5 7 0.0 0.0 0.1 0.7 0.1 0.6 8 0.1 0.1 1.0 0.2 Total 100 100 100 100 100 100 100 100 100 Average 1.15 0.60 1.75 1.22 1.26 2.48 1.19 0.79 1.98 Labor force participation rate 0.76 0.39 0.57 0.91 0.89 0.90 0.83 0.49 0.65 Source: RIC Surveys. Table D.9 Distribution (%) of Household Members (age ≥ 16 and ≤ 65) and Household Heads by Occupation in Sri Lanka All household members Household heads Sri Lanka Total male female Total male female Salaried/wage employed 36.11 51.45 21.02 52.72 58.81 19.53 Self-employed 15.74 20.93 10.62 28.24 28.87 24.80 Employer 0.04 0.08 0.00 0.02 0.03 0.00 Unpaid family member 1.20 1.34 1.06 2.15 2.36 1.01 Unemployed 7.48 8.22 6.76 0.15 0.02 0.84 Student 8.13 9.43 6.86 0.11 0.13 0.00 Housekeeping 24.82 2.34 46.91 5.14 1.04 27.43 Retired 1.47 2.02 0.94 3.87 3.53 5.71 Various inactive 5.00 4.18 5.81 7.34 5.21 20.67 Total 100.00 100.00 100.00 100.00 100.00 100.00 Source: RIC Surveys. 124 The Rural Investment Climate Table D.10 Adult Labor Force Participation Rate and Labor Force Composition Labor force participation rate Proportion of labor force in: Nonfarm Type of Wage self- Number of household Overall Male Female employment employment Farming adults Sri Lanka All 0.574 0.763 0.388 0.643 0.233 0.310 1816 With NFHE(w) 0.652 0.835 0.474 0.351 0.688 0.233 1161 Without NFHE(w) 0.541 0.732 0.352 0.794 0.000 0.350 655 With NFHE(i) 0.687 0.860 0.514 0.571 0.538 0.195 683 Without NHFE(i) 0.565 0.757 0.375 0.724 0.083 0.359 932 Tanzania All 0.898 0.910 0.887 0.153 0.349 0.900 3937 With NFHE(i) 0.891 0.904 0.879 0.159 0.453 0.887 2454 Without NHFE(i) 0.911 0.920 0.902 0.144 0.176 0.921 1483 Nicaragua All 0.647 0.832 0.485 0.631 0.219 0.375 3286 With NFHE(i) 0.756 0.823 0.698 0.360 0.644 0.219 2280 Without NHFE(i) 0.597 0.828 0.399 0.742 0.038 0.423 718 Source: RIC Surveys. Notes: NFHE denotes “nonfarm household enterprise�. (w) indicates that the presence of a household enterprise is judged by employment in nonfarm self-employment activities; (i) indicates that the presence of a household enterprise is judged by the receipt of income from a nonfarm business or from nonfarm self-employment activities. adult household members report housekeeping as of household heads list their current (primary) their main occupation, of which the majority are occupation as housekeeper, and 11.2 percent are women: more than 47 percent of eligible women retired or inactive. household members engage in housekeeping ac- tivities, against only 2.3 percent or their male coun- Economic Activity During the Previous terparts. Although in smaller numbers than their 12 Months male counterparts, a fair percentage of the women are either self-employed (10 percent) or salaried/ The second measure of economic activity makes wage employed (21 percent). Also of interest is that speci�c reference to the previous 12 months. Table women members are relatively less unemployed D.10 provides an overview. The labor force partici- than are male members. Somewhat more male than pation rate is 57.4 percent in Sri Lanka, with a large female household members are enrolled in school: gender gap (76.3 percent for men, 38.8 percent for 9.4 percent against 6.9 percent. Only 1.5 percent of women), lower than both Tanzania at 89.8 percent eligible members took early retirement, with only a with no gender gap and Nicaragua at 64.7 percent small difference between males and females. also with a large gender gap. The proportion of the Unlike the household members, most of the labor force holding a wage job is similar in Sri Lanka household heads (82 percent) are employed either and Nicaragua, as is the proportion self-employed as self-employed or salaried/wage workers. in a nonfarm business. Nicaragua has a slightly Participation of male household heads in these higher percentage of working adults in farming. two categories of employment together is higher The labor force structure in Tanzania is sharply (88 percent) than those of their female (44 percent) different: 90 percent of the rural labor force works counterparts. A significant proportion of female on farms and only 15 percent hold a wage job. household heads, however, either engages in As with any statistical comparison, it must be household activities (27 percent) or is retired/ kept in mind that statistics for Sri Lanka and inactive (26 percent). Together, only 5 percent Nicaragua are weighted with sampling weights, Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 125 whereas Tanzania’s statistics are unweighted. This on farming. About 43 percent of households in caution is especially appropriate here. Since the Sri Lanka, 56 percent in Tanzania, and 35 percent survey design oversampled households with a in Nicaragua have two or three economic activities nonfarm enterprise, these labor force composition simultaneously. statistics for Tanzania overstate the percentage of The table shows that 9.5 percent of households the labor force active in nonfarm self-employment. in Nicaragua, 1.4 percent in Sri Lanka, and 7.3 per- Table D.10 also distinguishes labor force statis- cent in Tanzania do not have any economic tics according to the entrepreneurship status of the activity. These percentages rise to 11.8 percent, household. The presence of a nonfarm enterprise 6.7 percent, and 27.2 percent, respectively, if may be established by income receipt within the marginally relevant income sources (producing household or activity status among its members. less than $60 per year) are excluded. Ideally, these criteria should make no difference, Much of the diversification across activities but in Sri Lanka’s case a questionnaire flaw makes occurs because different household members do a large difference here.86 Because of this, the different types of work. Still, 10.5 percent of work- income criterion provides the evidence of entre- ing adults in Sri Lanka, 29.1 percent in Tanzania, preneurship in Tanzania and Nicaragua, and work and 14.6 percent in Nicaragua engage in more than status does so in Sri Lanka. In Sri Lanka, then, the one activity, and these percentages are higher for labor force participation rate is higher, for both men than for women. Moreover, these statistics men and women, if the household operates an hide occurrences of people holding two of the enterprise; wage employment is more prevalent same type of activities (for example, two wage among households that do not. Nicaragua resem- jobs). Household heads specialize more than do bles Sri Lanka in this regard, but Tanzania does other household members. not show such a difference. Table D.12 reports the share of the active labor Table D.11 provides information on household force engaged in a particular activity. This sums to members’ involvement in at least one major eco- more than 100 percent because people may be (and nomic activity, such as self-employment, wage often are) active in more than one type of employ- employment, and farming. The central message of ment, as was already illustrated in Table D.11. the table is that few households specialize in only Especially in Sri Lanka, the variation in the struc- one activity, whether a nonfarm enterprise, wage ture of the labor force is large. employment, or farming. This is true however As the last table in this section on the labor households are categorized: by income source force, Table D.13 highlights the educational (either including or omitting marginal income achievement, gender composition, age structure, sources yielding less than US$60 per year) or by and human capital indexes among labor force par- work activity. In Sri Lanka, for example, 29.4 per- ticipants by their activity status. In Sri Lanka, cent of the rural households have at least one those who are self-employed in a nonfarm enter- member working in an enterprise, but only three prise tend to be slightly more educated that those of every ten of these (8.6 percent of the total) do so working on a farm; they are comparable to wage as a specialty; this ratio is the same in Nicaragua, employees, except for the more highly educated, where 22.7 percent of the households operate an who often work for someone else. Wage employ- enterprise. The lack of specialization is even more ees are younger and less likely to be female than pronounced in Tanzania: 7 of 100 among the are the self- or farm-employed. As a result, wage 55.5 percent operating an enterprise. (Note again employees’ human capital lags behind that of the that the income criterion biases the statistics in self- and farm-employed, but not by much. Sri Lanka.) In Nicaragua and Tanzania, the differentiation Proportions of households relying absolutely between the three labor force groups according wage employment are 37 percent in Sri Lanka and to education and gender is similar to that of Sri 45 percent in Nicaragua. Complete dependence on Lanka. Age patterns vary, however. In Tanzania, wage employment is very negligible (1.5 percent) wage employees are the oldest group, farmers in Tanzania. Almost 27 percent of households in the youngest, and the nonfarm self-employed are Tanzania rely on farming as their only source of in between; in Sri Lanka, the pattern is the exact income. In Sri Lanka and Nicaragua 8 percent and opposite; and in Nicaragua, it is the nonfarm self- 10 percent of households, respectively, rely only employed who are clearly the oldest. 126 The Rural Investment Climate Table D.11 Accounting for Economic Activities of Households and Household Members Employmentc of all adult household Employment of heads Overall household membersd of household Any Income incomea from from stated stated source > Any source $60b activity Total Male Female Total Male Female Sri Lanka Enterprise only 0.0 1.5 8.6 8.8 10.2 7.4 12.9 13.5 9.5 Enterprise and wage 9.9 10.8 9.1 1.5 2.7 0.2 3.6 4.2 0.2 Enterprise and farming 1.5 0.3 5.3 2.8 3.2 2.4 6.8 5.9 11.7 Enterprise, wage, and farming 6.6 4.3 6.4 0.4 0.7 0.0 1.0 1.1 0.1 Wage only 36.0 43.1 36.7 29.3 39.4 19.5 35.1 38.5 17.1 Farming only 18.3 13.3 8.4 8.9 9.9 7.9 13.5 12.9 16.9 Wage and farming 26.4 20.1 22.5 5.8 10.2 1.5 14.5 16.8 2.2 Not employed 1.4 6.7 3.0 42.6 23.7 61.2 12.6 7.2 42.4 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Tanzania Enterprise 14.6 25.0 3.8 4.9 5.3 4.5 5.1 4.3 9.9 Enterprise and wage 2.2 3.4 3.4 3.3 4.5 2.2 4.7 4.4 6.7 Enterprise and farming 38.9 18.2 28.2 16.2 20.3 12.3 28.5 30.0 20.2 Enterprise, wage, and farming 6.2 2.4 20.1 6.9 10.7 3.4 14.9 16.0 9.0 Wage 1.5 2.7 0.4 0.8 0.9 0.7 1.1 1.0 1.4 Farming 26.6 19.2 36.7 55.0 46.1 63.6 38.8 37.5 45.7 Wage and farming 2.7 2.0 4.6 2.7 3.4 2.1 5.1 5.1 4.9 Not employed 7.3 27.2 2.9 10.2 9.0 11.3 1.8 1.8 2.2 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Nicaragua Enterprise 7.7 8.7 7.2 10.1 8.5 11.5 13.8 10.7 20.4 Enterprise and wage 8.3 8.9 9.4 2.5 2.7 2.3 2.6 2.5 3.1 Enterprise and farming 3.5 2.7 2.6 1.6 2.3 0.9 2.6 3.2 1.3 Enterprise, wage, and farming 2.7 1.7 3.5 0.5 0.7 0.3 0.9 0.8 1.2 Wage 45.5 50.5 37.3 29.2 36.4 23.1 33.6 36.1 28.5 Farming 10.0 8.5 9.3 13.0 18.1 8.6 13.4 18.3 3.2 Wage and farming 12.8 7.4 19.7 10.0 18.3 2.9 16.0 22.9 1.4 Not employed 9.5 11.8 11.1 33.1 13.0 50.4 17.0 5.6 40.9 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Source: RIC Surveys. Notes: a. For Sri Lanka, adults living in 149 households with zero or negative total incomes are excluded. b. Column 2 counts involvement in an activity only if it generates at least $60 over the previous 12 months. c. Employment refers to the previous 12 months. d. Including heads of households. LAND AND PHYSICAL ASSETS common questions and responses, and measure- ment units of land differ within and across coun- Land Holdings tries. Sri Lanka, for example, has two categories of land using different measuring units (acres and Land is still the most prominent asset in the devel- perches), and Nicaragua measures land in man- oping countries’ portfolio, as it determines house- zanas (mzs). hold economic position. The pilots’ data collection In Sri Lanka, perches were converted into on land ownership and value did not follow acres by the international standard of 65 perches Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 127 Table D.12 Variation in the Structure of the Labor Force by Region Wage Self-employed in Self-employed in farm employment nonfarm enterprise activity A: Sri Lanka Western 72.7 19.0 17.4 Central 73.5 13.5 26.4 Southern 56.6 23.3 39.1 North Western 53.1 27.3 33.5 North Central 58.7 20.9 51.5 Uva 53.2 22.6 54.8 Sabaragamuwa 73.0 17.1 23.4 Eastern 33.3 0.0 83.3 North & East 58.3 53.6 17.9 Overall 64.3 23.3 31.0 Variation across regionsa 12.9 14.2 21.5 B: Tanzania Kilimanjaro 16.7 39.4 85.5 Morogoro 14.8 35.1 92.4 Mtwara 8.8 39.5 91.8 Mbeya 19.8 42.7 87.2 Tabora 16.5 25.6 91.4 Kigoma 11.1 32.5 92.8 Kagera 16.5 30.8 89.9 Overall 15.3 34.9 90.0 Variation across regionsa 3.7 5.9 2.8 C: Nicaragua Cabezera departamental 56.8 13.0 56.5 Municipios del pací�co 73.4 25.2 17.5 Municipios del resto del pais 58.6 29.7 37.7 Blue�elds, Puerto Cabezas, San Carlos 81.7 19.6 4.7 Overall 63.1 21.9 37.5 Variation across regionsa 12.0 7.2 22.8 Source: RIC Surveys. Note: a. Variation across regions is measured by the simple standard deviation of the percentages listed. per acre. For a small number of plots, measure- owns 0.92 acres of land. Not all households ments were reported in both acres and perches. actually own land: 89.8 percent of households This could indicate that the plot size was the reported owning land, and among those that do, sum of them or that the plot size was reported in average holdings are 1.03 acres. In Nicaragua, only both measures. A comparison with plot values 22.7 percent of the sample households own land, as well as an inspection of the ratio of perches which on average equals 84 acres. In Tanzania, over acres showed no reliable pattern, 87 but it almost 90 percent of the sample households own seemed most likely that households offered two on average 2.7 acres of land. alternative measures. Accordingly, land size was For comparison across countries, physical land computed as the average of the two reported units are converted into dollar values. As informa- values. tion on land prices was sometimes unavailable, The land unit measure manzana is in use in Latin this problem was resolved by taking the average American countries, but between countries its price of land per unit and imputing the average exact meaning may vary. In Nicaragua, 1 manzana price where it was missing. In Sri Lanka, when equals 7000 square meters or 1.729 acres.88 the value of the plot was not reported, the sample Table D.14 summarizes land holdings per average of 479,529 rupees (US $4955.35) per acre household. In Sri Lanka, the average household was used for imputation. Table D.15 provides 128 The Rural Investment Climate Table D.13 Characteristics of Groups of Labor Force Participants Self-employed All adult rural Working for in nonfarm Self-employed population wage or salary enterprise on the farm A: Sri Lanka Schooling* No schooling 5.5 3.6 3.2 5.2 Elementary school 22.2 22.0 22.1 27.3 Junior high school 15.5 17.3 18.7 21.4 High school 43.5 38.9 46.6 35.9 Tertiary education/training 13.3 18.1 9.4 10.3 Percent female 50.7 28.9 37.6 33.3 Average age 35.8 36.3 39.9 42.5 Average human capital stock 4.73 4.88 4.95 4.90 Average human capital stock index 0.334 0.350 0.358 0.353 B: Tanzania Schooling* No schooling 13.8 6.6 6.7 15.6 Elementary school 12.5 8.3 10.1 13.3 Junior high school 59.5 53.8 63.3 61.0 High school 5.2 5.5 4.3 3.4 Tertiary education 9.1 25.8 15.5 6.6 Percent female 51.2 31.2 36.6 51.5 Average age 32.9 37.7 35.9 33.9 Average human capital stock 3.76 5.01 4.45 3.73 Average Human capital stock index 0.21 0.32 0.27 0.20 C: Nicaragua Schooling* Able to read, write 83.7 84.8 87.2 73.1 No schooling 18.1 16.2 14.7 30.4 Elementary school 26.2 26.5 22.8 36.0 Junior high school 26.4 24.6 29.5 21.2 High school 21.9 22.8 23.5 9.5 Tertiary education/training 7.5 9.9 9.5 2.9 Percent female 53.2 36.3 55.1 27.3 Average age 32.9 31.8 36.5 33.8 Average human capital stock 3.56 3.80 3.95 3.00 Average human capital stock index 0.21 0.24 0.25 0.15 Source: RIC Surveys. Note: * At each level, this includes people who completed only a part of the indicated level. Table D.14 Average per Household Landownership in the distribution of households by land value. Selected Countries Although most of the households own land in Tan- Sri Lanka Tanzania Nicaragua zania, the value of land is less than US$500 for 79 percent of households. Distribution of value of Unit Acre Acre Acre land is more even in Sri Lanka than in Tanzania Mean 0.92 2.40 16.60 and Nicaragua. Although 85 percent of house- Standard deviation 1.15 3.22 131.13 holds are landless, the average per household land Percent owning land 89.8 89.9 22.7 If owning land: value is highest in Nicaragua. This means the top Mean 1.03 2.67 84.04 5 percent of households own most of the land. Standard deviation 1.17 3.28 319.21 A brief note on land leasing: In Sri Lanka, land Source: RIC Surveys. leasing is relatively uncommon. During a period of three years before the survey, about 7 percent of Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 129 land out. Sharecropping is exceedingly rare; pay- Table D.15 Distribution (%) of Households ment is often in the form of straight rent payments, by Value of Land though land is also sometimes shared between Land asset values households without expectation of any payment. (US $ equivalent) Sri Lanka Tanzania Nicaragua In Nicaragua, the situation is similar: 3 percent of households provided land to someone else; half of 500 45.2 79.2 85.2 the plots provided in this way were explicitly 500–1000 13.8 10.8 3.0 rented out, and the other half were shared or lent 1000–1500 6.5 2.8 2.0 1500–2000 7.1 1.9 1.2 or something else of such nature. Leasing in is 2000–2500 3.6 0.6 1.0 slightly more common: 9 percent received land for 2500–3000 4.0 0.9 0.5 agricultural use. 3000–3500 2.4 0.4 0.5 3500–4000 4.1 0.6 0.4 4000–4500 1.1 0.1 0.3 Agricultural Assets Other Than Land 4500–5000 2.9 0.6 5.9 As farming is the most important economic activ- 5000 9.3 2.2 ity for most of the household respondents, esti- Total 100 100 100 mates were made of the value of farm-related Average 1505 667.3 2156.5 assets held by the sample households. The house- SD 3101 2522.1 18833.3 N of obs 998 holds in all three country samples have multiple agricultural activities, such as crop production, Source: RIC Surveys. livestock rearing, aquaculture, and plantation. Production-related tools and equipments are the households leased land for purposes of crop treated as agricultural assets. Trees whose products production, with the majority paying by means of are not sold, however, are considered household sharecropping arrangements. Another 2 percent of assets. Table D.16 provides estimates of agricul- households lease land out, again usually under a tural assets. The average household agricultural sharecropping arrangement. In Tanzania, of the asset is highest in Nicaragua, although it includes 1,610 households in the sample, 202 (one-eighth) only aquaculture and livestock, followed by lease land in, and 57 (3 percent) household lease Sri Lanka, which includes crops, trees, livestock, Table D.16 Distribution (%) of Households by Total Value of Agricultural Assets Sri Lanka Nicaragua** Value of Tanzania* Agricultural % of % of assets (US $) households households Agricultural asset % of households <50 83.8 80.1 10 60.8 50–00 4.1 5.8 10–20 15.5 100–150 2.0 2.0 20–30 4.8 150–200 1.6 1.5 30–40 2.1 200–250 0.7 1.0 40–50 2.6 250–300 1.3 0.7 50–60 1.5 300–350 0.7 0.7 60–70 1.5 350–400 0.3 0.7 70–80 0.7 400–450 0.5 0.6 80–90 10.6 >450 5.1 7.0 90 14.2 Total 100.0 100.0 100.0 Average 274 339.1 96.0 St. Dev. 1042 3106.8 1123.8 N 998 1535 1611 Source: RIC Surveys. * Does not include tree assets, as data were collected in counts only. ** Includes only aquaculture and livestock assets. Data does not allow estimating crop assets and tree assets. 130 The Rural Investment Climate and aquaculture. Tanzania has the lowest level of by durable asset group. The durable asset base agricultural assets. The distribution of households is the highest in Sri Lanka. The average durable by agricultural asset group show that almost 84 per- asset is more than 13 times higher in Sri Lanka cent of households in Sri Lanka and 80 percent of than in Nicaragua and more than 5 times higher households in Nicaragua own less than US$50, and in Nicaragua than Tanzania. The distribution of only 61 percent of households in Tanzania own less households by durable assets is skewed to the left in than US$50 worth of agricultural assets. all the countries. Durable Assets Value of Residential Houses Durable assets include durable consumables like Table D.18 provides the distribution of value of electronics, furniture, transport equipments, and so houses owned by the sample households. The on. Table D.17 provides distributions of households average value of houses is higher in Sri Lanka Table D.17 Distribution (%) of Households by Value of Durable Assets Sri Lanka Tanzania Nicaragua Durable asset % of households Durable asset % of households Durable asset % of households 500 47.60 5 49.78 10 12.70 500–1000 18.72 5–10 20.36 10–20 14.92 1000–1500 11.57 10–15 9.93 20–30 10.68 1500–2000 6.21 15–20 4.90 30–40 9.38 2000–2500 4.42 20–25 3.10 40–50 6.32 2500–3000 2.82 25–30 1.92 50–60 6.06 3000–3500 1.41 30–35 1.37 60–70 6.84 3500–4000 0.94 35–40 0.87 70–80 4.89 >4000 6.30 40 7.76 >80 28.21 Total 100.00 100.00 100.00 Average 1072 16.57 104.89 SD 1836 56.63 445.42 N 998 1611 1535 Source: RIC Surveys. Table D.18 Distribution of Households by House Value Sri Lanka Tanzania Nicaragua House value % of households House value % of households House value % of households 1000 32.64 100 37.93 500 14.72 1000–2000 17.12 100–200 9.06 500–1000 18.31 2000–3000 11.19 200–300 7.45 1000–1500 9.32 3000–4000 7.71 300–400 4.22 1500–2000 17.65 4000–5000 9.13 400–500 5.34 2000–2500 3.52 5000–6000 4.42 500–600 2.79 2500–3000 7.62 6000–7000 1.41 600–700 1.86 3000–3500 5.21 8000–8000 3.67 700–800 2.61 3500–4000 4.17 8000 12.70 800 28.74 4000 19.48 Total 100 100 100 Average 4481 1683 3851 SD 5821 6364 13782 998 1611 1535 Source: RIC Surveys. Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 131 (US$4,481) than in Nicaragua (US$3,851); it is low- less than US$500 worth of assets. Contrary to this, est (US$1,683) in Tanzania. In Sri Lanka the major- 27 percent of households in Nicaragua own less ity (61 percent) of households live in houses worth than US$1,000 worth of assets, and 47 percent own less than US$3,000, and 13 percent live in highly between US$1,000 to US$4,000 worth of assets. In valued houses worth more than US$8,000. In Sri Lanka 11 percent of the households own Nicaragua, the majority of households (60 percent) US$1,000, and 30 percent own between US$1,000 to live in low-quality houses worth less than US$4,000 worth of assets. US$2,000, and almost 20 percent of households Table D.19 illustrates substantial inequality in live in high-quality houses worth US$4,000. In the households’ asset holdings. It is not clear how Tanzania, the value of houses in general is much countries rank in the level of inequality. Table D.20 lower than in the other two countries. Almost and Figure D.1 shed light on this. As shown in 60 percent of households have houses valued Table D.20, Sri Lanka exhibits a much lower coef�- below US$400; only 29 percent of Tanzania’s cient of variation and Gini coef�cientc; Nicaragua households have houses worth more than US$800. and Tanzania are similar. Figure D.1 makes the comparison visual by standardizing the distri- butions at their mean value. Tanzania and Total Assets Nicaragua’s distributions look very similar; the As shown in Table D.19, the average total asset lowest value is more pronounced in Tanzania, but holding is highest in Sri Lanka, followed by the categories right above it show a higher fre- Nicaragua. The average asset holding is the lowest quency in Nicaragua. Sri Lanka has a much more in Tanzania: 43 percent of households there own even distribution in the diagram’s left half. Table D.19 Distribution (%) of Households by Total Assets Sri Lanka Tanzania Nicaragua Total asset % of households Total asset % of households Total asset % of households 1000 11.38 500 43.08 1000 26.84 1000–2000 9.88 500–1000 20.67 1000–2000 24.95 2000–3000 11.29 1000–1500 7.88 2000–3000 12.51 3000–4000 9.22 1500–2000 5.28 3000–4000 9.12 4000–5000 8.75 2000–2500 4.16 4000–5000 4.50 5000–6000 6.68 2500–3000 2.23 5000–6000 3.78 6000–7000 6.02 3000–3500 1.55 6000–7000 2.41 7000–8000 4.42 3500–4000 1.61 7000–8000 1.76 8000 32.36 4000 13.53 8000 14.14 Total 100 100 100 Average 7332 2462.62 6453.94 St.Dev. 8331 7471.38 24883.92 N 998 1611 1535 Source: RIC Surveys. Table D.20 Measures of Inequality in Total Assets Household Investment Sri Lanka Tanzania Nicaragua Household investment has been considered as net addition to total assets over a one-year period in Coef�cient of 1.103 3.034 3.610 categories of agricultural production, including variation crop production, aquaculture, and livestock; of Gini Coef�cient 0.590 0.838 0.821 net additions to house buildings; and of durable Source: RIC Surveys. assets. Table D.21 provides distribution of house- holds by investment. An average household 132 The Rural Investment Climate Household Income Figure D.1 Comparing Distribution of Total Assets Across Countries Before describing household incomes, it is impor- tant to note that the Sri Lanka questionnaire had a 45 flaw that most likely causes income components 40 other than wages and salaries and farming to be 35 understated. In particular, this flaw caused infor- 30 mation on enterprise income and other income 25 to be gathered only if at least one member of the 20 household was employed for a wage. As a result, 15 incomes are likely understated for a signi�cant 10 number of households. Indeed, 134 households 5 show a total income reported of 0, of which 119 0 0 1.5 3 4.5 report nonfarm enterprise employment. In the Total assets relative to mean statistics reported below, households with 0 or negative incomes are removed from the analysis. Sri Lanka Tanzania Nicaragua Secondly, for Nicaragua, six enterprises with nega- Source: RIC Surveys. tive total household incomes are removed from the Note: a Total assets are standardized by the country’s mean in order to calculation of income-related statistics. Third, the compare the distributions. The right tail is accumulated at a value of 5.0 Tanzania statistics do not use sampling weights, and the strati�cation is explicitly based on a feature (entrepreneurship) related to income. Therefore, Table D.21 Household Investment by Category income statistics of these two countries and com- parisons with the third should be considered with Investment in Sri Lanka Tanzania Nicaragua appropriate caution. Fourth, it should be noted that income statistics were reported by house- Agricultural assets Mean 56 27 n.a. hold respondents, which, in the case of Sri Lanka St.dev. 699 950 % positive 24.9 41.6 and Tanzania, is separate from any information % negative 4.2 14.22 recorded in the enterprise questionnaire. The Durable assets Mean 80 1 n.a Nicaragua questionnaire merged enterprise in- St.dev. 347 16 come and expenditure information into the % positive 35.7 57.2 household questionnaire for small, household- % negative 24.9 32.6 based enterprises. Housing Mean n.a. 16 n.a. Further, employment and earnings do not St.dev. 136 always go hand in hand in the survey data. It is % positive 10.6 fairly typical of household surveys to find dis- % negative 0.0 crepancies, which may be caused by response Total assets Mean 136 44 n.a. error, enumerator error, data-entry problems, St.dev. 783 959 unclear questionnaire design, phrasing of ques- % positive 51.7 68.4 tions, and so on. The degree of inconsistency is % negative 3.3 26.3 demonstrated in Table D.22. The questionnaire Source: RIC Surveys. design error in Sri Lanka is obvious: members of 315 household are involved in nonfarm self- increased its assets by US$136 in Sri Lanka and by employment but did not report income—half US$44 in Tanzania; the large standard deviations of all households with employment in an enter- indicate how widely household investment varies prise. Inconsistencies in Tanzania also run high: across households. Measurement error may of 165 households report nonfarm self-employment course be to blame: it is always difficult to obtain but no enterprise income, and 269 record enter- accurate financial statistics of assets, let alone of prise income but no employment. The Nicaragua investment. Unfortunately, the Nicaragua survey data are more consistent, perhaps because the data do not permit an estimate of household enterprise and household questionnaires were investment. blended. Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 133 Table D.22 Consistency in Reports on Household Enterprise Activity Household reports Household reports activity in nonfarm self-employment income from nonfarm self- Sri Lanka Tanzania Nicaragua employment No Yes Total No Yes Total No Yes Total No 447 315 762 448 165 613 440 36 476 Yes 14 287 301 269 729 998 59 1000 1,059 Total 461 602 1,064 717 894 1,611 499 1036 1535 Source: RIC Surveys. Note: These statistics are raw counts of households and therefore unweighted. In a normal situation, the presence of a house- tail. Median household income in Tanzania is only hold enterprise is judged by the reported receipt of US$247. income from a nonfarm business or from nonfarm In Nicaragua, nearly half of household income self-employment activities. In Sri Lanka, the derives from wage earnings. The next largest con- design of the questionnaire led to nonresponses on tributor is nonfarm enterprise income at nearly self-employment and other income because of an one-fourth. Farm income and remittances each erroneous skip pattern. As a result, the criterion of contribute about 11.5 percent. assuming the presence of a nonfarm enterprise by In Tanzania, nonfarm enterprises contribute income is biased. nearly 40 percent of average household income, Table D.23 describes household income and its farming 30 percent, and wage earnings 12 percent. component for each country. The distribution is It may seem surprising that nonfarm enterprise divided into different intervals for each country, income is so dominant, but it should be remem- which are selected to highlight the contribution of bered that the Tanzanian RICS survey oversam- each component effectively. Separate categories of pled households with enterprises but did not 0 and Negative describe households that received provide sampling weights; this prevents the sam- no income from a given source or for which expen- ple from being representative of the Tanzanian ditures from the given activity (for example, in rural population and may bias any statistics com- Tanzania, farming as well as wage earnings) puted from it, especially if the statistic is related to exceeded the revenues it generated. the entrepreneurial status of the household, as is In Table D.23 income is divided into five com- clearly the case here. ponents. Households earn income from wages and In Sri Lanka, remittances are the largest average salary; self-employment; farming operations, contributor to household income, even though which include crop production, aquaculture and nearly 88 percent of the households do not receive livestock, and plantation; remittance income from any remittance income. Wage income contributes friends and relatives; and “other income,� which about one-third; nonfarm enterprises contribute includes interest income from savings and divi- less than one-tenth. But, as mentioned, these dends, lotteries, gifts from friends and relatives, income statistics for Sri Lanka should be treated receipts from government, and retirement bene- with great caution since they are seriously biased. �ts. Households may also have many indigenous Partly in response to the �nding that remittance expenses that cannot be categorized as productive income appears to be so important in Sri Lanka, expenses and thus were excluded from gross other appearing at the top of the income distribution, income to estimate other net income. Figure D.2 examines the functional distribution of Mean income among rural households equals household income: it considers the contribution of US$2,696 in Sri Lanka, US$670 in Tanzania, and each component to total household income within US$2,016 in Nicaragua. In every country, standard quintiles. Clearly, Sri Lankan households in the deviations exceed the mean, and the median is upper quintile benefit greatly from remittances. substantially less than the mean. These are indica- Enterprise income expands slightly from quintiles tions that the income distribution has a long right 1 to 4; wage income is most important for Table D.23 Total Household Income and Its Components A. Sri Lanka Wage Enterprise Farm Remittance Other Total Negative 0.00 0.00 1.93 0.00 0.00 n.a. 0 21.18 82.09 45.32 87.73 59.53 n.a. 0–500 22.68 8.71 35.51 1.07 33.95 20.85 500–1000 27.24 3.78 6.93 2.01 1.51 22.64 1000–1500 14.11 3.09 3.58 0.38 1.56 17.42 1500–2000 5.93 1.24 1.63 0.16 1.23 9.84 2000–2500 4.88 0.05 2.68 1.67 0.04 6.44 2500–3000 1.89 0.42 0.24 0.54 0.61 6.05 3000–3500 1.06 0.00 0.57 0.41 0.08 3.84 3500–4000 0.17 0.01 0.13 0.75 0.30 1.24 4000–4500 0.03 0.13 0.19 0.15 0.14 1.64 4500–5000 0.42 0.13 0.39 0.00 0.23 1.05 5000 0.41 0.34 0.91 5.11 0.83 8.98 Mean 835 162 343 1088 268 2696 Standard dev 1237 565 904 6386 1544 7883 Median 620 0 10 0 0 1216 B. Tanzania Wage Enterprise Farm Other Total Negative 0.93 0.00 19.12 0.00 4.84 0 86.41 38.05 6.08 58.66 0.37 0–200 4.97 36.00 56.61 28.06 40.29 200–400 1.49 10.80 9.37 4.78 16.82 400–600 1.12 5.77 3.41 2.98 11.17 600–800 1.18 2.79 1.99 1.37 6.83 800–1000 0.93 1.55 0.62 1.30 3.48 1000–1200 0.81 1.06 0.25 0.87 2.86 1200–1400 0.56 0.68 0.31 0.50 2.17 1400–1600 0.37 0.19 0.37 0.25 1.80 1600–1800 0.37 0.50 0.31 0.31 1.99 1800–2000 0.19 0.43 0.19 0.25 1.06 2000 0.68 2.17 1.37 0.68 6.33 Mean 81 261 203 126 670 Standard deviation 392 739 1469 500 1873 Median 0 55 42 0 247 C. Nicaragua Wage Enterprise Farm Remittance Other Total Negative 0.00 0.4 4.47 0.00 0.00 0.23 0 30.69 77.43 66.49 68.94 77.43 0.16 0–600 19.41 7.03 22.1 22.07 15.36 20.69 600–1200 18.41 6.21 3.97 4.72 5.11 26.42 1200–1800 13.05 3.14 0.54 1.72 0.76 18.47 1800–2400 8.13 1.62 0.67 0.78 0.75 10.14 2400–3000 3.46 0.69 0.35 0.75 0.01 7.34 3000–3600 2.18 0.62 0.28 0.16 0.2 3.82 3600–4200 1.98 0.54 0.2 0.02 0.05 4.13 4200–4800 0.66 0.51 0.05 0.39 0.04 1.98 4800–5400 0.41 0.23 0.05 0.02 0 0.92 5400–6000 0.42 0.39 0.14 0.06 0.02 1.14 6000 1.21 1.18 0.7 0.38 0.28 4.58 Mean 988 434 228 230 136 2016 Standard dev. 1363 2247 1527 1089 648 3444 Median 586 0 0 0 0 1237 Source: RIC Surveys. 134 Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 135 Figure D.2 Income Shares as a Percentage C. Nicaragua of Total Household Income, by Quintile 100 Cumulative percentage A. Sri Lankaa 80 100 60 Cumulative percentage 80 40 60 20 40 0 Q1 Q2 Q3 Q4 Q5 20 Quintiles 0 Other income Remittances Q1 Q2 Q3 Q4 Q5 Farm income Enterprise income Quintile of total income Wage income Other income Remittances Farm income Enterprise income Wage income roughly constant across the five quintiles; wage Note: a Households with negative or zero total incomes have been omitted. income expands. In Nicaragua, the share of remit- tance income is surprisingly constant across house- holds. The enterprise share expands; the wage share rises until sharply dropping in the �fth quin- tile, and the share of farm income drops steadily. B. Tanzaniaa The concept of household income is useful when sources are brought together and aggre- 100 gated. After all, households often gather income from all sources together and then allocate it to the Cumulative percentage 80 various needs among household members. But to 60 express the standard of living of these household 40 members better, it is of course common to compute household income per capita. Table D.24 reports 20 those statistics. Averages range from US$164 in 0 Tanzania to US$661 in Sri Lanka, with Nicaragua in between at US$483. Median per capita incomes 20 are actually about the same in Sri Lanka and Q1 Q2 Q3 Q4 Q5 Quintiles Nicaragua, at slightly over US$280; in Tanzania, the median is only US$56. Other income Farm income Enterprise income Wage income In Tanzania, where the median per capita income is US$56, 5 percent of households have a Note: a Households with zero or negative total income are per capita income of over US$600. A similar con- omitted, but a few households reported expenditures for earning a wage income that exceed the wage income trast exists in Sri Lanka, where more than 4 percent itself. This causes a curious negative wage share at the of households exceed US$2,000 in per capita lowest quintile. income when the median is US$282. The contrast seems less pronounced in Nicaragua. Table D.25 formalizes this comparison by means of the coef�- households in quintiles 2 and 3. Farm income is cient of variation and the Gini coef�cient. Indeed, most important in the lowest quintile and dimin- Tanzania and Sri Lanka have more unequally ishes from there on. distributed per capita incomes. Figure D.3 makes In Tanzania, enterprise income dominates, this more visual, describing the distribution of per which is to be expected as the survey oversampled capita income in standardized form by dividing enterprise households. The share of farm income is each household’s income by the country’s mean. Table D.24 Per Capita Income and Its Components A. Sri Lanka Wage Enterprise Farm Remittance Other Total Negative 0.00 0.00 1.93 0.00 0.00 n.a. 0 21.18 82.09 45.32 87.73 59.53 n.a. 0–200 41.12 12.50 41.47 3.05 34.97 37.69 200–400 23.69 2.88 4.46 0.63 2.69 26.51 400–600 8.83 1.21 2.60 0.64 0.72 11.96 600–800 2.18 0.52 2.63 0.68 1.12 7.07 800–1000 1.37 0.54 0.38 0.55 0.14 2.75 1000–1200 0.39 0.21 0.35 2.05 0.18 2.58 1200–1400 0.59 0.02 0.06 0.32 0.00 3.64 1400–1600 0.19 0.00 0.60 0.48 0.00 1.10 1600–1800 0.00 0.02 0.20 0.94 0.00 1.47 1800–2000 0.00 0.00 0.00 0.33 0.00 1.13 >2000 0.46 0.00 0.00 2.60 0.66 4.11 Mean 212 38 85 265 61 661 Standard dev 337 128 209 1576 375 1943 Median 143 0 2 0 0 282 B. Tanzania Wage Enterprise Farm Other Total Negative 0.93 0.00 19.12 0.00 4.84 0 86.41 38.05 6.08 58.66 0.37 0–60 5.77 40.35 60.40 29.55 46.74 60–120 2.11 9.62 8.01 5.71 17.13 120–180 1.61 4.22 2.48 1.92 10.24 180–240 0.81 2.67 0.93 1.37 5.46 240–300 0.68 1.06 0.50 1.06 3.85 300–360 0.50 0.56 0.56 0.43 1.99 360–420 0.43 0.68 0.25 0.19 1.49 420–480 0.25 0.43 0.43 0.19 1.43 480–540 0.12 0.43 0.12 0.12 0.62 540–600 0.06 0.25 0.06 0.00 0.87 >600 0.31 1.68 1.06 0.81 4.97 Mean 16 64 52 32 164 Standard dev 75 222 464 194 570 Median 0 11 9 0 56 C. Nicaragua Wage Enterprise Farm Remittance Other Total Negative 0.00 0.4 4.47 0.00 0.00 0.23 0 30.69 77.43 66.49 68.94 77.43 0.16 0–250 35.45 13.16 25.56 24.86 18.77 41.41 250–500 21.32 4.18 1.71 3.59 2.37 29.89 500–750 6.99 1.82 0.42 0.89 0.75 13.47 750–1000 2.53 0.68 0.18 1.1 0.04 5.96 1000–1250 1.36 0.66 0.22 0 0.36 2.7 1250–1500 1.05 0.75 0.27 0.02 0.02 2.29 1500–1750 0.08 0.21 0.16 0.28 0.05 0.81 1750–2000 0.03 0.2 0.01 0.06 0.00 0.63 2000–2250 0.08 0.03 0.16 0 0.00 0.51 2250–2500 0 0.09 0 0 0.00 0.22 2500 0.42 0.39 0.36 0.25 0.22 1.71 Mean 226 111 49 57 40 483 Standard dev. 325 984 361 221 208 1133 Median 126 0 0 0 0 287 Source: RIC Surveys. 136 Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 137 Table D.25 Measures of Inequality in per Figure D.3 Comparing Distributions of Income Capita Total Income per Capita, Normalized at the Country-Speci�c Mean Sri Lanka Tanzania Nicaragua 35 Coef�cient of 30 variation 2.942 3.468 2.347 25 Gini coef�cienta 0.769 0.757 0.568 20 Source: RIC Surveys. 15 Note: a. Households with negative incomes are omitted 10 from the computation. 5 0 Sri Lanka and Tanzania show a more pronounced 1.5 0 1.5 3 4.5 Total income relative to mean clustering of households at very low income levels. It should be noted that the communities in Sri Lanka Tanzania Nicaragua Nicaragua are somewhat larger than those in Tan- Source: RIC Surveys. zania and Sri Lanka. Another perspective on household income is higher than in those without them, both in total gained by expressing income per worker and by and per capita or per worker. The exception is Sri distinguishing households with and without a Lanka’s (w) definition, but this should be dis- nonfarm enterprise. Table D.26 makes clear that counted, since enterprise income is understated as total income in households with enterprises is a result of the questionnaire design flaw. Table D.26 Income Total, per Capita and per Worker, by Category and Presence of Nonfarm Enterprise Sri Lanka Tanzania Nicaragua hh hh hh hh hh with without hh with without hh with without with without All hh NFHE(w) NFHE(w) NFHE(i) NFHE(i) All hh NFHE(i) NFHE(i) All hh NFHE(i) NFHE(i) Household income Total 2697 2443 2779 3228 2581 670 749 542 2025 3120 1705 Wage income 836 805 846 1132 771 81 71 97 992 677 1083 Enterprise income 163 637 8 906 0 261 421 0 436 1927 0 Farm income 344 358 339 308 351 202 163 265 231 259 222 Other 1355 643 1586 881 1459 126 93 179 367 257 399 Income per capita Total 661 601 680 788 633 164 182 135 485 743 410 Wage income 212 194 217 272 199 16 15 19 227 128 256 Enterprise income 38 148 2 211 0 64 103 0 111 493 0 Farm income 85 86 85 69 89 52 39 71 49 67 44 Other 326 173 376 237 345 32 25 45 98 56 110 Income per worker Total 1572 1095 1731 1424 1605 310 351 241 1260 1541 1166 Wage income 727 805 706 801 706 237 196 314 970 838 998 Enterprise income 545 545 n.a. 770 0 257 315 0 1151 1297 0 Farm income 580 524 602 587 579 117 82 172 384 569 338 Source: RIC Surveys. Note: (w) denotes that the presence of a household enterprise is judged by reported employment in nonfarm self-employment activities; (i) indicates that the presence of a household enterprise is judged by the reported receipt of income from a nonfarm business or from nonfarm self-employment activities. 138 The Rural Investment Climate Relation to Benchmark Indicators uniformly upward or downward, but that many and Population Size suggestions of trends are manifest. These trends are equivalent to simple correlations, but they should be Table D.27 provides mean values of income and em- explored in a multivariate context, as, for example, ployment variables by quintiles of the benchmark household income or entrepreneurship rates are in- indicators and population size. This table contains a fluenced by a conglomerate of factors. Therefore, to wealth of information that will not be discussed quantify these trends more carefully is a topic for a here. Suf�ce it to say that the trend is rarely more advanced regression analysis. Table D.27 Household Income and Employment by Benchmark Quintiles and Population Size A. Sri Lanka 1 2 3 4 5 Total Connectivity Total household income 1726 2048 1553 1652 4621 2457 Enterprise income 141 148 86 106 240 148 Wage income 738 660 810 775 758 756 Farm income 483 459 166 434 112 308 Other income 46 222 240 96 530 246 Remittance income 317 558 252 241 2981 999 Other + remittance income 363 781 492 337 3511 1246 Activity: nfhe 0.187 0.311 0.224 0.286 0.378 0.284 Activity: wage 0.715 0.743 0.782 0.709 0.794 0.752 Activity: farm 0.590 0.398 0.390 0.587 0.232 0.429 Infrastructure services Total household income 2219 1401 1598 1935 3937 2421 Enterprise income 263 94 105 104 172 146 Wage income 823 679 678 760 803 752 Farm income 573 351 299 445 53 302 Other income 92 46 156 184 516 241 Remittance income 468 231 361 441 2393 979 Other + remittance income 560 278 516 625 2909 1221 Activity: nfhe 0.273 0.211 0.349 0.264 0.331 0.293 Activity: wage 0.653 0.704 0.674 0.845 0.802 0.747 Activity: farm 0.575 0.587 0.396 0.543 0.209 0.426 Business services Total household income 1621 5840 3070 1184 2453 Enterprise income 140 232 118 115 147 Wage income 708 593 1186 572 744 Farm income 270 520 241 287 308 Other income 129 916 107 107 247 Remittance income 375 3580 1418 103 1007 Other + remittance income 504 4495 1524 210 1255 Activity: nfhe 0.326 0.315 0.220 0.265 0.299 Activity: wage 0.777 0.687 0.724 0.683 0.740 Activity: farm 0.427 0.408 0.366 0.551 0.433 Corruption/governance Total household income 1643 2930 1768 1505 4979 2557 Enterprise income 97 145 100 206 65 125 Wage income 807 953 724 579 745 759 Farm income 337 503 210 420 116 320 Other income 261 73 197 143 682 264 Remittance income 141 1256 536 157 3371 1090 Other + remittance income 402 1329 734 300 4053 1354 Activity: nfhe 0.191 0.264 0.307 0.314 0.186 0.257 Activity: wage 0.804 0.764 0.712 0.643 0.818 0.744 Activity: farm 0.473 0.627 0.294 0.530 0.340 0.453 Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 139 Table D.27 Household Income and Employment by Benchmark Quintiles and Population Size (continued) A. Sri Lanka 1 2 3 4 5 Total Human capital Total household income 1550 1335 2756 1373 5143 2421 Enterprise income 79 268 193 91 128 146 Wage income 635 670 1082 672 724 752 Farm income 323 231 492 156 291 302 Other income 36 64 229 102 800 241 Remittance income 478 102 761 352 3199 979 Other + remittance income 514 166 990 454 4000 1221 Activity: nfhe 0.219 0.400 0.331 0.278 0.267 0.293 Activity: wage 0.797 0.753 0.747 0.697 0.727 0.747 Activity: farm 0.517 0.324 0.435 0.388 0.432 0.426 Finance Total household income 1635 1971 2479 1699 4889 2557 Enterprise income 143 77 167 137 97 125 Wage income 685 751 968 685 671 759 Farm income 296 200 327 479 317 320 Other income 42 129 179 285 700 264 Remittance income 469 814 838 114 3103 1090 Other + remittance income 510 943 1017 399 3803 1354 Activity: nfhe 0.262 0.214 0.317 0.219 0.260 0.257 Activity: wage 0.792 0.780 0.762 0.714 0.666 0.744 Activity: farm 0.450 0.393 0.504 0.515 0.406 0.453 Population size Total household income 1498 1860 2589 1561 3560 2422 Enterprise income 64 93 249 128 159 146 Wage income 683 652 1053 764 660 753 Farm income 368 386 387 247 225 302 Other income 210 65 124 93 529 241 Remittance income 172 665 776 330 1986 980 Other + remittance income 382 730 900 422 2516 1221 Activity: nfhe 0.200 0.274 0.347 0.256 0.329 0.293 Activity: wage 0.581 0.741 0.694 0.766 0.813 0.747 Activity: farm 0.600 0.463 0.540 0.419 0.295 0.426 B. Tanzania 1 2 3 4 5 Total Connectivity Total household income 501 619 723 839 632 658 Enterprise income 113 255 219 418 256 249 Wage income 74 69 62 85 130 83 Farm income 175 144 367 211 89 196 Other income 138 149 74 124 156 129 Activity: nfhe 0.410 0.492 0.587 0.655 0.673 0.555 Activity: wage 0.242 0.210 0.323 0.297 0.382 0.286 Activity: farm 0.938 0.964 0.910 0.826 0.831 0.897 Infrastructure services Total household income 630 441 756 646 830 658 Enterprise income 119 182 307 295 375 249 Wage income 87 32 114 39 142 83 Farm income 296 127 181 171 190 196 Other income 128 97 153 140 122 129 Activity: nfhe 0.469 0.498 0.534 0.631 0.681 0.555 Activity: wage 0.216 0.243 0.309 0.292 0.392 0.286 Activity: farm 0.963 0.951 0.877 0.854 0.821 0.897 140 The Rural Investment Climate Table D.27 Household Income and Employment by Benchmark Quintiles and Population Size (continued) B. Tanzania 1 2 3 4 5 Total Business services Total household income 628 450 1105 666 Enterprise income 246 198 374 256 Wage income 95 78 26 84 Farm income 150 82 603 198 Other income 136 92 100 126 Activity: nfhe 0.555 0.565 0.529 0.553 Activity: wage 0.267 0.364 0.314 0.284 Activity: farm 0.899 0.913 0.880 0.899 Corruption/governance Total household income 512 545 768 630 859 664 Enterprise income 240 287 282 280 205 259 Wage income 67 43 147 51 107 84 Farm income 126 132 171 170 410 202 Other income 75 81 167 128 137 118 Activity: nfhe 0.599 0.593 0.567 0.548 0.571 0.575 Activity: wage 0.310 0.248 0.317 0.314 0.287 0.295 Activity: farm 0.881 0.894 0.917 0.860 0.924 0.895 Human capital Total household income 445 541 578 625 1233 670 Enterprise income 158 252 223 307 383 261 Wage income 20 31 92 74 205 81 Farm income 170 170 107 125 460 202 Other income 97 87 154 117 184 126 Activity: nfhe 0.397 0.504 0.545 0.646 0.715 0.555 Activity: wage 0.203 0.172 0.266 0.360 0.451 0.284 Activity: farm 0.957 0.939 0.872 0.868 0.831 0.896 Finance Total household income 589 624 519 786 832 664 Enterprise income 191 275 269 316 255 260 Wage income 78 76 45 91 145 85 Farm income 160 136 106 280 329 197 Other income 159 137 99 98 102 120 Activity: nfhe 0.550 0.558 0.605 0.580 0.585 0.575 Activity: wage 0.287 0.293 0.319 0.269 0.307 0.295 Activity: farm 0.886 0.890 0.911 0.926 0.859 0.895 Population size Total household income 581 399 928 673 726 663 Enterprise income 228 206 245 248 349 252 Wage income 72 45 105 144 53 84 Farm income 121 79 425 115 228 196 Other income 160 65 153 165 95 130 Activity: nfhe 0.492 0.540 0.568 0.607 0.608 0.559 Activity: wage 0.293 0.211 0.330 0.311 0.285 0.287 Activity: farm 0.867 0.872 0.940 0.929 0.878 0.897 Annex D. Household Characteristics in Sri Lanka, Tanzania, and Nicaragua: Findings from the RIC Pilot Surveys 141 Table D.27 (continued) C. Nicaragua 1 2 3 4 5 Total Connectivity Total household income 1053 1487 2589 2592 2422 2068 Enterprise income 133 245 667 620 573 458 Wage income 586 880 840 1172 1245 974 Farm income 213 225 422 341 75 252 Other income 68 60 181 247 98 136 Remittance income 53 78 480 212 430 247 Other remittance income 121 137 661 459 528 383 Activity: nfhe 0.143 0.166 0.301 0.266 0.264 0.227 Activity: wage 0.746 0.659 0.611 0.693 0.706 0.685 Activity: farm 0.774 0.531 0.279 0.252 0.071 0.377 Infrastructure services Total household income 1562 1730 2040 2197 2547 2068 Enterprise income 175 318 443 398 791 458 Wage income 732 931 852 1011 1242 974 Farm income 356 261 333 274 95 252 Other income 58 144 251 86 135 136 Remittance income 241 75 162 426 284 247 Other remittance income 299 219 413 512 420 383 Activity: nfhe 0.143 0.197 0.223 0.277 0.280 0.227 Activity: wage 0.838 0.699 0.500 0.632 0.744 0.685 Activity: farm 0.673 0.551 0.359 0.251 0.141 0.377 Business services Total household income 1766 2879 2383 2068 Enterprise income 299 860 638 458 Wage income 858 1107 1185 974 Farm income 221 678 113 252 Other income 142 126 126 136 Remittance income 246 108 320 247 Other remittance income 388 234 446 383 Activity: nfhe 0.202 0.306 0.245 0.227 Activity: wage 0.696 0.625 0.692 0.685 Activity: farm 0.449 0.348 0.213 0.377 Corruption/governance Total household income 1661 1774 2424 2166 2153 2070 Enterprise income 397 412 508 458 489 458 Wage income 868 672 1003 1120 1168 975 Farm income 169 190 457 239 142 253 Other income 100 212 163 118 83 136 Remittance income 127 289 293 232 270 248 Other remittance income 227 500 456 350 354 384 Activity: nfhe 0.195 0.199 0.234 0.249 0.243 0.226 Activity: wage 0.737 0.494 0.744 0.697 0.717 0.685 Activity: farm 0.374 0.490 0.361 0.363 0.317 0.377 142 The Rural Investment Climate Table D.27 Household Income and Employment by Benchmark Quintiles and Population Size (continued) C. Nicaragua 1 2 3 4 5 Total Human capital Total household income 800 1674 2139 2280 3246 2068 Enterprise income 92 238 593 586 707 458 Wage income 439 977 995 1052 1322 974 Farm income 205 147 218 158 522 252 Other income 25 100 146 138 251 136 Remittance income 39 213 188 346 443 247 Other remittance income 64 313 333 484 694 383 Activity: nfhe 0.125 0.172 0.254 0.276 0.303 0.227 Activity: wage 0.675 0.724 0.658 0.667 0.703 0.685 Activity: farm 0.747 0.500 0.232 0.236 0.210 0.377 Finance Total household income 1513 1896 2019 2390 2682 2068 Enterprise income 193 269 492 517 834 458 Wage income 691 1307 1106 994 1257 974 Farm income 263 103 204 380 145 252 Other income 88 180 62 210 185 136 Remittance income 277 37 155 289 262 247 Other + remittance income 366 216 217 499 446 383 Activity: nfhe 0.168 0.166 0.275 0.241 0.275 0.227 Activity: wage 0.686 0.750 0.732 0.625 0.707 0.685 Activity: farm 0.530 0.408 0.280 0.403 0.172 0.377 Population size Total household income 1436 948 2143 2230 2562 1950 Enterprise income 276 91 506 563 686 457 Wage income 786 567 967 816 1375 967 Farm income 138 141 279 320 87 177 Other income 72 61 204 244 129 138 Remittance income 165 88 187 287 285 212 Other + remittance income 236 148 390 531 414 349 Activity: nfhe 0.133 0.140 0.295 0.283 0.263 0.224 Activity: wage 0.711 0.709 0.610 0.607 0.743 0.687 Activity: farm 0.561 0.668 0.383 0.224 0.181 0.382 Source: RIC Surveys. Annex E. Specimen Indexes Derived from RICS Data The six benchmark indicators identi�ed in observation as Xmin/X. If X Xmin, the index is Chapter 2 are de�ned below, with illustrations set at 1. from the Sri Lanka pilot. The six indicators are: 2) Relative cost of transportation to the nearest major 1. Connectivity city (by public transportation). 2. Infrastructure services The higher the cost, the less integrated the commu- 3. Business services nity. Let X be the cost in US$ of a trip to the nearest 4. Governance city, and let Xmin be the minimum of X, and set 5. Human capital Xmin 0.05. Then an index of connectivity for each 6. Finance services observation is de�ned as Xmin/X. If X Xmin, the Each benchmark indicator is based on a number of index is set at 1. If there is no public transportation, subindicator indexes, with values calculated at the this indicator is coded as 0. community level. Most subindicator values are di- 3) Time taken by main means of transportation to the rectly obtained from the community survey; some main market. also use data from the enterprise and household The longer the time, the less integrated the com- surveys. The calculation is carried out in two steps. munity. Let X be the time from a given community First, subindicators are calculated at the commu- in the sample to the main market, and let Xmin be nity level; and second, the subindicators are aggre- the minimum of X, and set Xmin 10. Then an gated at the community level by taking their sum index of connectivity for each observation is de- and dividing by the number of observations per �ned as Xmin/X. If X Xmin, the index is set at 1. community (missing observations are ignored). 4) Cost of transportation to the main market (by bus). 1. Connectivity. The index of rural-urban connectivity/ The lower the cost of transportation, the more inte- regional economic integration is computed as the grated the community. Find the cost of transporta- average of the following eight subindexes, trans- tion in the price questionnaire. Let X be the cost of forming the average value such that the maximum transportation by bus in US$ from a given commu- equals 1 and scaling the value of the index for each nity in the sample to the main market, and let Xmin community relative to this maximum. Note that be the minimum of X, and set Xmin 0.05. Then an each subindex is constructed such that a more re- index of connectivity for each observation is de- mote community is characterized by a smaller �ned as Xmin/X. If X Xmin, the index is set at 1. value of the subindex. A scaling of the indexes has 5) Distance to the post of�ce. been selected in such a way that international com- The lower the distance to the post of�ce, the more parison is possible. integrated the community. Define D as the dis- tance to the post of�ce in kilometers. The index is 1) Time taken by main means of transportation to the of connectivity calculated as 1/D. If D 1, the nearest major city. index is set at 1. The longer the time, the less integrated the commu- nity. Let X be the distance in minutes from a given 6) Rail stop in walking distance. community in the sample to a bigger town, and let The connectivity subindex in this case is a dummy Xmin be the minimum of X, and set Xmin 10. variable: 1 if there is a rail stop within walking dis- Then we establish an index of connectivity for each tance, 0 otherwise. 143 144 Table E.1 Distribution of Connectivity Index Over Communities standard 0 to .1 .1 to .2 .2 to .3 .3 to .4 .4 to .5 .5 to .6 .6 to .7 .7 to .8 .8 to .9 .9 to 1 N average deviation % 2.7 10.3 17.1 19.2 22.6 16.4 6.2 5.5 0.0 0.0 146 0.40 0.17 Connectivity Index mean 0.057 0.160 0.258 0.341 0.453 0.539 0.644 0.733 . . Time taken by main means of % 9.6 15.1 7.5 16.4 4.1 15.1 13.7 0.0 0.0 18.5 146 0.47 0.31 transportation to the nearest mean 0.069 0.143 0.236 0.333 0.409 0.500 0.667 . . 1.000 major city Cost of transportation to the nearest % 1.5 5.9 8.1 7.4 10.3 5.1 14.7 0.0 15.4 31.6 136 0.66 0.29 major city (by public transportation) mean 0.032 0.165 0.242 0.351 0.435 0.538 0.637 . 0.806 0.993 Time taken by main means % 9.7 12.4 14.5 12.4 1.4 11.7 17.2 0.0 0.0 20.7 145 0.49 0.32 of transportation to the mean 0.070 0.146 0.243 0.333 0.427 0.500 0.667 . . 1.000 main market Cost of transportation to % 15.8 2.7 3.4 5.5 11.6 4.8 10.3 0.0 13.7 32.2 146 0.61 0.36 the main market (by bus) mean 0.007 0.146 0.249 0.353 0.433 0.538 0.647 . 0.806 0.995 Distance to the post of�ce % 2.1 6.2 12.4 6.2 0.0 20.7 0.0 0.0 0.0 52.4 145 0.69 0.35 mean 0.081 0.140 0.225 0.333 . 0.500 . . . 1.000 Rail stop in walking distance % 87.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 12.4 145 0.12 0.33 mean 0.000 . . . . . . . . 1.000 Percentage of households with % 67.6 16.6 9.0 2.8 2.1 0.7 0.7 0.7 0.0 0.0 145 0.08 0.12 �xed telephone lines mean 0.022 0.122 0.215 0.333 0.417 0.500 0.600 0.750 . . Percentage of households with % 69.7 13.1 6.2 3.4 2.8 4.1 0.0 0.7 0.0 0.0 145 0.10 0.14 cell phone mean 0.027 0.126 0.217 0.330 0.400 0.508 . 0.750 . . Source: The Authors. Annex E. Specimen Indexes Derived from RICS Data 145 7) Percentage of households with �xed-line telephones. The access subindex in this case is a dummy vari- This connectivity subindex is simply the share able: 1 if the main road surface is concrete/asphalt, of households in the community with �xed-line 0 otherwise. telephones. 3. Business services. The index that measures 8) Percentage of households with a cell phone. the availability of business services is derived The connectivity subindex is the share of house- from seven dummy variables, measured in each holds with a cell phone. community. 2. Infrastructure services. The index that measures 1) Engineering services available for businesses in the access to infrastructure services and service deliv- community. ery is constructed as the average of the following 2) Management consulting services available for busi- eight subindexes. Note that each subindex is con- nesses in the community. structed such that a community with poorer access 3) Marketing services available for businesses in the is characterized by a smaller value of the subindex. community. 1) Percentage of households that use electricity. 4) Accounting services available for businesses in This subindex measures the share of households the community. with electricity. 5) Legal services available for businesses in the community. 2) Availability of electricity. 6) Insurance services available for businesses in the This subindex measures the reliability of the elec- community. tricity supply. Index (4/number of blackouts per 7) Information services available for businesses in month) if the number of blackouts per month is the community. greater than or equal to 4. The index equals 1 if the number of blackouts is less than 4, and it equals 0 if The community index is the sum of the indexes electricity is unavailable. divided by the number of indexes for which data are available. 3) Percentage of households with access to drinking water from a protected source. 4. Governance. This index measures the existence of This subindex measures the share of households governance and corruption. A high index indicates a with access to drinking water from a protected pipe low level of administrative burden and corruption. or well. Five subindexes have been de�ned: 4) Percentage of households with a �xed-line telephone. 1) General policy and institutional constraints. This subindex measures the share of households 2) Infrastructure and services. with �xed-line telephones. 3) Dealing with government agencies. 5) Percentage of households with a cell phone. 4) Rule of law. This subindex measures the share of households 5) Conflict resolution and contract enforcement. with a cell phone. All subindexes are constructed from the enterprise 6) Sewage channels in the community.* survey data by taking the average of scores. Each The access subindex in this case is a dummy vari- community subindex is the average of observa- able: 1 if there are sewage channels, 0 otherwise. tions from all businesses in that community. An aggregated index of corruption and governance is 7) Garbage collection or disposal service in the constructed at the community level from these community.* subindexes. Missing scores are ignored in the This subindex is a dummy variable: 1 if there is calculations. garbage collector/disposal service, 0 otherwise. 1) General policy and institutional constraints. 8) Most common road surface (internal road) is con- This subindex measures general policy and crete or asphalt. institutional constraints. The subindex is high if *Access to sewage channels and garbage collection could poten- there are no constraints. Variables included are: tially be a cost and not a bene�t for businesses, depending on the cost of alternative ways of disposal. In the Sri Lanka report, ac- a. Corruption as a constraint for rural invest- cess to sewage and garbage collection was considered a bene�t. ment climate; 146 Table E.2 Distribution of Infrastructure Service Index Over Communities standard 0 to .1 .1 to .2 .2 to .3 .3 to .4 .4 to .5 .5 to .6 .6 to .7 .7 to .8 .8 to .9 .9 to 1 N average deviation % 6.2 13.0 19.2 28.8 14.4 13.0 2.7 2.7 0.0 0.0 146 0.35 0.16 Infrastructure services Index mean 0.027 0.154 0.253 0.356 0.446 0.545 0.632 0.750 . . Percentage of households that use % 7.6 0.7 4.8 5.5 2.8 4.1 14.5 9.7 29.7 20.7 145 0.69 0.29 electricity mean 0.001 0.150 0.240 0.328 0.443 0.513 0.647 0.755 0.864 0.965 Availability of electricity index % 13.1 9.0 13.8 8.3 9.0 2.8 1.4 0.0 4.1 38.6 145 0.55 0.39 mean 0.026 0.127 0.235 0.331 0.403 0.500 0.667 . 0.800 1.000 Percentage of households with % 4.8 3.4 0.0 1.4 2.8 4.8 7.6 4.8 15.9 54.5 145 0.80 0.28 access to potable water mean 0.003 0.132 . 0.350 0.413 0.510 0.631 0.750 0.839 0.990 Percentage of households with % 67.6 16.6 9.0 2.8 2.1 0.7 0.7 0.7 0.0 0.0 145 0.08 0.12 �xed-line telephone mean 0.022 0.122 0.215 0.333 0.417 0.500 0.600 0.750 . . Percentage of households with % 69.7 13.1 6.2 3.4 2.8 4.1 0.0 0.7 0.0 0.0 145 0.10 0.14 access to cellular phones mean 0.027 0.126 0.217 0.330 0.400 0.508 . 0.750 . . Sewage channels in the community % 89.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.0 145 0.11 0.31 mean 0.000 . . . . . . . . 1.000 Garbage collection or disposal % 88.3 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.7 145 0.12 0.32 service in the community mean 0.000 . . . . . . . . 1.000 Most common road surface (internal % 67.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 32.9 146 0.33 0.47 road) is concrete or asphalt mean 0.000 . . . . . . . . 1.000 Source: The Authors. Table E.3 Distribution of Business Service Index Over Communities standard 0 to .1 .1 to .2 .2 to .3 .3 to .4 .4 to .5 .5 to .6 .6 to .7 .7 to .8 .8 to .9 .9 to 1 N average deviation % 54.2 14.8 16.9 0.0 8.5 0.7 0.0 1.4 1.4 2.1 142 0.15 0.22 Availability Index mean 0.000 0.143 0.286 . 0.429 0.571 . 0.714 0.857 1.000 Engineering services available for % 94.4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 5.6 142 0.06 0.23 businesses in the community mean 0.000 . . . . . . . . 1.000 Management consulting services available for businesses in the % 93.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 7.0 142 0.07 0.26 community mean 0.000 . . . . . . . . 1.000 Marketing services available for % 91.5 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 8.5 142 0.08 0.28 businesses in the community mean 0.000 . . . . . . . . 1.000 Accounting services available for % 89.4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 10.6 142 0.11 0.31 businesses in the community mean 0.000 . . . . . . . . 1.000 Legal services available for % 73.9 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 26.1 142 0.26 0.44 businesses in the community mean 0.000 . . . . . . . . 1.000 Insurance services available for % 60.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 39.4 142 0.39 0.49 businesses in the community mean 0.000 . . . . . . . . 1.000 Information technology services available for businesses in the % 90.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 9.9 142 0.10 0.30 community mean 0.000 . . . . . . . . 1.000 Source: The Authors. 147 148 Table E.4 Distribution of Governance Indexes Over Communities standard 0 to .1 .1 to .2 .2 to .3 .3 to .4 .4 to .5 .5 to .6 .6 to .7 .7 to .8 .8 to .9 .9 to 1 N average deviation % 0.0 0.0 0.0 0.0 0.0 8.0 65.6 26.4 0.0 0.0 125 0.67 0.04 Governance index mean . . . . . 0.583 0.661 0.717 . . General policy and institutional % 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 2.4 97.6 125 0.98 0.03 constraints mean . . . . . . . . 0.878 0.987 Infrastructure and services % 34.4 51.2 13.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 125 0.13 0.06 mean 0.064 0.143 0.228 0.306 . . . . . . Dealing with government % 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 100.0 125 1.00 0 agencies mean . . . . . . . . . 0.999 Rule of law % 1.6 1.6 2.4 17.6 36.8 20.8 9.6 8.8 0.8 0.0 125 0.49 0.14 mean 0.052 0.167 0.273 0.351 0.455 0.556 0.648 0.733 0.813 . Conflict Resolution and % 0.0 0.0 0.0 3.2 3.2 4.0 28.0 27.2 20.8 13.6 125 0.75 0.14 contract enforcement mean . . . 0.370 0.459 0.561 0.665 0.761 0.845 0.952 Source: The Authors. Annex E. Specimen Indexes Derived from RICS Data 149 b. Economic policy as a constraint for rural e. Issues relating to environmental regulation; investment climate; and c. Crime, theft, and disorder as constraints for f. Others. rural investment climate; Note: These variables are not included in the core. d. Legal system as a constraint for rural invest- For each type of deal or issue, the business owner ment climate; and or manager is asked if an unof�cial payment or gift e. War or other social frictions as a constraint was ever expected or requested. When the answer for rural investment climate. to such questions was yes, the value of the index is The variables are classi�ed in the database on a recorded as 0; when the answer was no, the value is scale from 1 to 5 (from no obstacle to very severe). 1. For businesses that did not have a relevant deal For this subindex the variable is reclassi�ed as or issue in the 12 months prior to the survey, the follows: observation was treated as a missing value. 0.75 if item is a minor obstacle 4) Rule of law. 0.50 if item is a moderate obstacle This subindex indicates the rule of law and the pre- 0.25 if item is a major obstacle dictability of its application. A high value indicates 0.00 if item is very severe obstacle strong rule of law. Variables included are: 2) Infrastructure and services. a. Predictability of laws and regulations that This subindex measures reports of unof�cial extra affect the operation and growth of businesses; fees for service delivery. The index is 1 when there and are no extra payments. b. Rules and regulations that can be manipu- Variables included are: lated or misinterpreted by of�cials. a. Extra fee to register/renew; Predictability of laws and regulations that affect b. Extra fee to apply for a basic activity license/ the operation and growth of businesses is scaled as permit/renew; follows: c. Extra fee to apply for a construction permit; 1.00 fully predictable (originally coded as 1) d. Extra fee to apply for an electricity connec- 0.75 highly predictable (originally coded as 2) tion for industry use; 0.50 somehow predictable (originally coded as 3) e. Extra fee to apply for an electricity connec- 0.25 unpredictable (originally coded as 4) tion for domestic use; and 0.00 highly unpredictable (originally coded as 5) f. Extra fee to apply for other services. In cases where the question was answered don’t Note: These variables are not included in the core. know, the observation was treated as a missing Each index at the enterprise level is de�ned as value. 1 (X/X95), where X is the value of the unof�cial, The possibility that rules and regulations can be extra fee and X95 is the value at the ninety-�fth per- manipulated or misinterpreted by of�cials is scaled centile of the distribution of available observa- as follows: tions on a given item. The index is only calculated if businesses applied for the service; other cases 0 if strongly agree (originally 1) will be missing values. The index is set at 0 when 0.3333 if agree (originally 2) X X95. 0.6666 if disagree (originally 3) 1 if strongly disagree (originally 4) 3) Dealing with government agencies. The observation is treated as a missing value if the This subindex measures cost related to dealing answer is don’t know, can’t say. with government agencies. A high subindex indi- cates low costs. 5) Conflict resolution and contract enforcement. Variables included are: This subindex registers conflict resolution and con- tract enforcement. A high value indicates favorable a. Tax-related issues; conditions. The following variables are included: b. Labor-related issues; c. Issues related to �re and building safety; a. Must rely on the reputation of others with d. Issues related to sanitation and epidemiology; whom you enter into agreements; 150 The Rural Investment Climate b. A contract will protect you from being cheated Variable Original New coding Percentage by others; and value coding for index of the sample c. The legal system will uphold your contract and property rights in disputes. Year 1 1 1 2.08 Year 2 2 2 6.31 We reclassify the variable so that the index will Year 3 3 3 3.95 be as follows: Year 4 4 4 4.48 Year 5 5 5 6.63 0 if strongly agree (originally 1) Year 6 6 6 4.96 Year 7 7 7 4.66 0.3333 if agree (originally 2) Year 8 8 8 8.97 0.6666 if disagree (originally 3) Year 9 9 9 4.23 1 if strongly disagree (originally 4) Year 10 10 10 13.49 Year 11 11 11 15.52 An observation is treated as a missing value if Year 12 12 12 4.33 the answer is don’t know, can’t say. Year 13 13 13 8.14 University 5. Human capital. This indicator provides an index (post-graduate) 14 15 0.98 Professional 15 14 0.26 of human capital. A high value indicates a high Technical College 16 14 0.06 level of education and working experience. The Preschool 17 0 2.54 index is constructed on the basis of the household No Schooling 18 0 3.83 survey. Other 19 Missing 0.57 Bils and Klenow (2000) describe human-capital 20 (?) 20 (?) Missing 3.80 stock according to the classic Mincerian returns to Don’t know 99 Missing 0.20 schooling and experience: Note: The schooling variable is censored at 14 years of formal education. s1 ln[h(a, t)] 1(a s 6) 2(a s 6)2 1 In case of missing values of education and/or age, the observation is treated as a missing value. where a is age and s is years of schooling, and the The human-capital stock is computed for each quadratic term is standard in the empirical litera- individual between 16 and 65 years of age. The ture on wages. lowest possible value of stock over all values of s The returns to schooling and experience ( 1 and a equals approximately 1.5, achieved at using and 2) are based on estimates of the sources of a 16 and s 0; the highest equals about 10.0 for wage differences, that is, the following Mincer someone with s 14 and a 57. These two values equation: become meaningful when the community human ln [h(w)] s (a s 6) (a s 6)2 capital index is computed. 0 1 2 3 The human capital index of the community is Bils and Klenow obtain estimates for 1 and 2 from aggregated from individuals between 16 and averages of 52 countries of 2 and 3. The estimated 65 years old, without any restrictions in terms of parameter values are 1 0.0512, 2 –0.00071, occupational status. 0.099, and 0 (which corresponds to the classic Thus, the index is obtained through the follow- case in the labor literature showing no diminishing ing steps: returns to education). 1. Using data from the household survey, The individual human stock index is obtained impute a human capital stock for each indi- as follows: vidual on the basis of the model estimated by ln(stock) (0.099* s) (0.0512*(a s 6)) Bils and Klenow. (0.00071(a s 6)2) 2. By community, compute the average human capital stock, that is, h_ave (sum of human Thus an individual’s stock is computed as the capital/number of individuals). antilog of the result of this expression: 3. Finally, compute the community-level human stock = eln(stock) capital index as (h_ave 1.55)/(10 1.5). For the variable s, years of schooling, we rescaled 6. Finance services. This indicator describes the the coding in the household survey as follows: level of development of �nancial intermediation. Table E.5 Distribution of Human Capital Indexes Over Communities standard 0 to .1 .1 to .2 .2 to .3 .3 to .4 .4 to .5 .5 to .6 .6 to .7 .7 to .8 .8 to .9 .9 to 1 N average deviation % 0.0 6.7 31.3 41.0 17.9 3.0 0.0 0.0 0.0 0.0 134 0.33 0.09 Human resources index mean . 0.165 0.259 0.350 0.427 0.551 . . . . Source: The Authors. Table E.6 Distribution of Finance Service Indexes Over Communities* standard 0 to .1 .1 to .2 .2 to .3 .3 to .4 .4 to .5 .5 to .6 .6 to .7 .7 to .8 .8 to .9 .9 to 1 N average deviation % 1.4 5.4 3.4 15.6 19.7 25.9 19.7 7.5 1.4 0.0 147 0.50 0.16 Finance Index mean 0.039 0.172 0.251 0.359 0.450 0.549 0.655 0.735 0.827 . Number of formal �nancial % 2.7 4.1 6.8 11.6 10.2 12.9 21.8 18.4 8.8 2.7 147 0.56 0.22 sources weighed by mean of mean 0.023 0.141 0.255 0.361 0.447 0.560 0.655 0.748 0.855 0.938 the distance Number of formal �nance % 7.5 8.2 8.2 19.0 10.2 16.3 15.6 10.2 2.0 2.7 147 0.47 0.23 services weighed by mean of mean 0.048 0.162 0.260 0.363 0.444 0.560 0.659 0.739 0.843 0.963 the distance Access to loans % 4.1 6.1 11.6 12.2 15.6 21.8 15.0 3.4 7.5 2.7 147 0.47 0.23 mean 0.023 0.135 0.230 0.337 0.426 0.526 0.637 0.761 0.872 1.000 Source: The Authors. * This table differs from the standard de�nitions used in the text due to differences in the de�nitions of �nancial services used across the questionnaires. (This should be avoided in the next phase.) 151 152 The Rural Investment Climate The �nancial intermediation will be more mature if the number of formal institutions is larger, the Box E.1. De�nition of Formal Financial number of services available is larger, and the share Institutions (Including Insurance) of enterprises that want to borrow money is large. Institutions with a permanent of�ce, proper It consists of four variables: marquee, and regular business hours, 1) Number of formal �nancial and insurance institu- including: tions that offer services in the community, weighted • Commercial banks; by distance from the community; • Savings banks and post of�ces that offer 2) Number of �nancial and insurance services provided �nancial services (mostly savings); in the community by formal institutions, weighted • Micro�nance institutions with full-time pro- by distance from the community; fessional staff; 3) Community access to a commodity exchange for • Government agencies engaged in retailing futures or options contracts; and credit; 4) Access to loans. • Cooperatives, credit unions, and business associations providing �nancial services; 1) Number of formal �nancial and insurance institu- and tions that offer services in the community, weighted • Companies offering life, health, accident, by distance from the community. disability, �re, or livestock insurance. This subindex describes the number of formal institutions that serve customers in the community Source: RICS Enterprise and Community Questionnaires, Appendix C. as reported, weighted by distance to the institution. Step 1: Construct the distance index 1 – {(D)/(30)}, where D is the distance, in kilometers, to each type of �nancial institution in the commu- 9. Term deposits nity. If D 30, set D 30. 10. Current account Step 2: Obtain the mean of the indexes calcu- 11. Money transfer (sending and receiving) lated in step1 for each community; call this “M.� 12. Life and funeral insurance Let X be the number of institutions. Calculate a 13. Insurance for health, accident, disability, variable: A X * (1/M), then calculate the index as �re, and livestock follows: A/max(A). Let X be the total number of formal services avail- 2) Number of �nancial and insurance services provided able in the community; calculate: I X * (1/M), in the community by formal institutions, weighted then calculate the index as follows: I/max I. by distance from the community. 3) Community access to a commodity exchange for fu- The questionnaire asks whether the following tures or options contracts. 13 services or �nancial tools are available: This variable is a binary index. 1. Loan up to 6 months 2. Loan 6 months to 2 years 4) Access to loans. 3. Loan longer than 2 years From the enterprise questionnaire, information is 4. Credit line for business linked to checking obtained for each community concerning the pro- account portion of businesses that obtained formal loans for 5. Warehouse receipts accepted as collateral day-to-day operations and investments in their 6. Leasing contracts available businesses during the last 5 years. The index is 7. Land accepted as collateral measured as the percentage of such enterprises. 8. Savings account offering withdrawal on The aggregate �nance services index is the aver- short notice age of the four subindexes. Annex F. Entrepreneurship: Notes and Tables Speci�cation Issues Moreover, because of the design of the survey, Dependent Variables: Activity Status. In Nicaragua, enterprise start-up must be based on the premises household entrepreneurship is de�ned as deriving of the household’s residence at the time of the sur- a positive income from an enterprise. Because of an vey; it cannot be a stand-alone business. In Sri error in the questionnaire design in Sri Lanka, this Lanka, where stand-alone businesses are common de�nition would understate the incidence of entre- (64.8 percent of the sample), the estimated start-up preneurship. Instead, in Sri Lanka entrepreneur- rate of 1.2 percent is therefore signi�cantly under- ship is de�ned as “household members working on stated. The true start-up rate might be as high as own account.�89 Involvement in farming or wage 6.05 percent (Box F.1). In Nicaragua, the situation employment is de�ned similarly. is quite different: the enterprise start-up rate is The income- and work-based definitions of measured to be 3.0 percent, probably fairly accu- entrepreneurship should largely overlap, leaving rate because the number of stand-alone enterprises a gap only if work “on own account� earned no is rather small (7.5 percent of the sample). income. In practice, some households indeed As mentioned above, an enterprise is consid- report work but no income; however, some others ered a start-up if it was established during the two report income but no work. In Nicaragua, a dis- years prior to the survey. More accurately, this crepancy arises between the two definitions in means that for Sri Lanka, the enterprise began 7.3 percent of the sample households, split evenly operations during 2002 or 2003, and for Nicaragua, between those reporting work but no income and the enterprise was established during 2003 or 2004. those reporting income but no work. In Sri Lanka, The RICS survey was held in Sri Lanka between 1.3 percent reported income but no work, and, to a September and November, 2003, but the actual large degree because of the questionnaire design date was not recorded in the database. Therefore, flaw, 29.6 percent reported work but no income.90 the measured start-up rate covers a little less than The income-based definition is preferable from a two years. Fortunately, in any given year, most perspective of productivity and poverty analysis, start-up enterprises appear to start in the months but for Sri Lanka the work-based definition is an between January and August, as 88 percent of the acceptable alternative. existing enterprises did so. Thus, the start-up rate Dependent Variables: Enterprise Start-Up. Tech- for Sri Lanka is still close to the true two-year nically, information on whether a household start-up rate (after allowing for the caveats men- recently started up an enterprise is not directly tioned above). The Nicaragua survey was held available in the RIC data, but for the purpose of between February and May, 2004, and the date of this analysis, households are considered to have the survey is missing in many of the records. The started up an enterprise if the enterprise they oper- month in which an enterprise began operations is ate is less than two years old. This understates the unknown in Nicaragua, but it is safe to assume true two-year start-up rate, since these enterprises that the Nicaragua statistic may capture only must have survived for up to two years to be one-third to one-half of start-ups in 2004. Allow- included; enterprises that recently started up but ing for this, a better estimate of the two-year had already died before the survey was held can- start-up rate would range from 4.0 percent to not be counted. 4.5 percent. 153 154 The Rural Investment Climate Box F.1 RIC Survey Design and the Undercount of Enterprise Start-Up To gain insight into the effect of survey design on the measurement of enterprise start-up, consider that the rate of enterprise start-up is measured to be 1.20 percent in Sri Lanka.a That is, 1.20 percent of the Sri Lankan households started a household-based enterprise during the two years prior to the RICS survey. The weighted percentage of households with a (household-based) enterprise in the survey is 5.47 percent. Thus, 21.93 percent of the enterprises in the survey are rated as start-ups. But another 20.30 percent of the households report work on own account but did not provide data about the enter- prise. Now in the RICS enterprise database, 35.24 percent of the enterprises are household-based and 64.76 percent are stand-alone enterprises; 24.96 percent and 23.90 percent of them are start-ups, respectively.b If one were to infer that the 20.30 percent of the households that report work on own account but did not provide data about the enterprise are operating a stand-alone enterprise and that the start-up rate of 23.90 percent applies to them uniformly, then another 4.85 percent ( 0.239 20.30) of the households would be operating a start-up enterprise, and the enterprise start-up rate among Sri Lankan rural households would be 6.05 percent.c As for Nicaragua, there are 1,372 household-based enterprises in 1,060 households. Of these, 179 began operations during the previous two years (13.28 percent).d These are aggregated to the house- hold level, and a household is assumed to have started an enterprise when one of its several operations began during the previous two years; this means that 18.15 percent of the households with one of more enterprises began at least one.e As mentioned in the discussion about the de�nition of the entrepreneur- ship variable, there were only a few households with inconsistent enterprise income and work activity in the database. Thus, only a few of the stand-alone enterprises might be under household ownership of one of the households in the sample, and thus any correction would be minor. Source: The Authors. aThis derives from the dataset with which the determinants of enterprise start-up are examined. Households with missing infor- mation or in communities without benchmark indicators or community variables are omitted. b The difference between 24.96 percent and 21.93 percent arises from the fact that not all household-based enterprises are actu- ally linked with a household and also that the second statistic is computed in a subsample that is subject to data availability. c This statistic further assumes that stand-alone start-ups did not �rst begin as a household-based entity that was turned into a stand-alone within the �rst two years of its existence. If this happens, double counting occurs among the start-up percentage. d The start-up rate among stand-alone enterprises is 26 out of 110 or 21.3 percent as a weighted percentage. e While Nicaragua surveyed every enterprise activity in the household, Sri Lanka selected only one activity. As shown, the start- up rate among activities in Nicaragua is 13.28 percent but among households it is 18.15 percent. By virtue of this survey design feature, the start-up rate in Sri Lanka should be relatively lower than in Nicaragua. The difference is further accentuated by the way this one activity in Sri Lanka is selected: the main activity in the household is likely the more established one. Explanatory Variables. The explanatory variables enterprise. Similarly, human capital may signal are divided into three sets. The first pertains to skills used either in the enterprise (directly for household characteristics and includes demo- production or indirectly for procuring inputs) or graphics such as the number of male and female elsewhere in wage employment. adult household members, their ages and human Another variable among household characteris- capital, and the gender and ethnicity of the head. tics describes whether the household head’s par- These variables represent determinants of the ents were entrepreneurs. The household head available supply of household labor that may be would have been able to experience entrepreneur- drawn upon in a household enterprise, the quality ship close up, receiving an in-house apprentice- of this labor, and barriers both to the operation of ship, as it were. Parents may have introduced the an enterprise and to other opportunities of head to their business network. employment; that is, these variables capture any The last group of household variables describes barriers on the basis of gender, ethnicity, or age, �nancial resources: remittance income, income from for example, whether encountered when seeking other sources, and household assets. These clearly wage employment in the labor market or in facilitate enterprise operation. Household assets, in procuring inputs or financial resources for the particular, however, might be endogenously related Annex F. Entrepreneurship: Notes and Tables 155 to the decision to operate an enterprise. To be sure, returns to this issue, but for now it is suf�cient to the assets do not include enterprise assets available temper expectations regarding effects of invest- in the enterprise database, but it is possible that ment climate indicators on entrepreneurship entrepreneurs use household assets for their busi- choice. nesses, especially if the enterprises are based at the The third set of variables describes the commu- residence. Moreover, previous pro�tability may nity apart from immediate associations with the well have generated savings or investments in con- policy-related investment climate. These variables sumer durables that are now observed as household are motivated as follows. Community size and assets. The econometric analysis recognizes house- the level of per capita income measure market hold assets as a potential endogenous household- opportunities an enterprise could exploit inside a level variable. community. Since a successful enterprise in turn The second set of variables describes the invest- generates income, per capita income may be an ment climate in the community where the house- endogenous community-level variable. The male hold resides. As described in Chapter 2, the wage rate measures the cost of labor, the opportu- investment climate is measured by a multitude of nity cost of the entrepreneur and household mem- variables that are condensed into six benchmark bers, and the level of well-being among potential indicators. Both these indicators and their com- customers: the effect on household entrepreneur- ponents are used in the econometric analysis ship is therefore ambiguous. Enterprise openness to measure the investment climate’s effect on indicates connection with markets outside the entrepreneurship. The benchmark indicators cover community and thus, as it broadens the potential characteristics of location, availability of infra- market, it is expected to encourage entrepreneur- structure, utilities, public and private services, ship. Finally, in a community with pronounced governance, human capital, and finance services. agricultural seasonality, households may seek These variables are constructed such that a higher ways to make their slack season more profitable value is expected to favor the enterprise perfor- (Haggblade, Hazell, and Brown, 1989; Lanjouw mance. Connectivity may raise the value of entre- and Lanjouw, 2001); alternatively, the cyclical preneurship if it opens up distant markets; it may unreliable availability of labor may be a stumbling lower it if distant competitors seek out clients in block for many enterprises. the community. Infrastructure services are neces- Endogeneity among Explanatory Variables. It was sary for many business exploits: greater access mentioned above that household assets and com- should increase the value of entrepreneurship. The munity income levels are plausibly endogenous. availability of business services may facilitate Because of the random effects structure of the entrepreneurship. The governance index measures entrepreneurship probit equation, this endogene- rule of law, security, contract enforcement, and so ity may express itself at two levels. Household forth, without which enterprise operation is more assets is a household-level variable for which a risky. The community human-capital index is a random effect model is appropriate. Thus, endo- proxy for the quality of labor, which can raise enter- geneity may express itself through a correlation prise productivity and thus the value of entrepre- between the (community-level) random effects of neurship. At the same time, human capital relates the assets and entrepreneurship equations as well positively to income and thus to potential market as through a correlation of the household-level size. The �nance services index summarizes �nance disturbances. Meanwhile, community income may and insurance services that facilitate enterprise be modeled with a community-level simple regres- start-up and operation and thus should encourage sion model, where its disturbance may be corre- household entrepreneurship. lated with the community-level random effect in As mentioned above, however, all of these argu- the assets and entrepreneurship equations. The ments do not necessarily imply that the benchmark most appropriate way to estimate the entrepre- indicators must favor the entrepreneurship choice. neurship model while allowing for endogeneity is A robust investment climate may create more to build the assets equation into a likelihood func- lucrative opportunities outside the household- tion that also includes the residual of the commu- owned-and-operated enterprise and thus may lead nity income equation. Estimation of this model, to enterprise closure while nonetheless fostering however, which relies on numerical optimization an expanded nonfarm economy. The next section methods, runs into frequent convergence problems. 156 The Rural Investment Climate As an alternative, the assets equation may be esti- case where entrepreneurship is perfectly explained, mated with a fixed effects regression model; the the criterion value equals 0. As reported in Panel A community income equation may be estimated of the table, household characteristics explain (that with the ordinary least squares method; and the is, reduce the criterion function by) 8.09 percent in random effect and household-level residual of the Nicaragua and 4.58 percent in Sri Lanka, where it assets equation and the residual of the community- should be noted, the Sri Lankan model does not income equation may be inserted into the entre- include remittance and other income, which are preneurship model: if these added variables have influential variables in Nicaragua. The community a statistically significant effect on entrepreneur- random effect contributes 1.09 percent in Sri Lanka ship choice, assets and/or community income are but is absent in Nicaragua. The benchmark indica- endogenous. As it turns out, the three added vari- tors help to explain 2.98 percent in Nicaragua and ables are never jointly significant, and in all the 1.46 percent in Sri Lanka, and community charac- models only one of them is statistically signi�cant teristics further add about half as much. In all, once at a 10 percent level and not at the 5 percent benchmark indicators and community characteris- level. Thus, endogeneity does not appear to be an tics contribute 35 percent ( (2.98 1.35)/12.42) of issue. the explanation of entrepreneurship choice in Explanatory Power of the Regression Models. An Nicaragua and 28 percent ( (1.46 0.72)/7.85) in analysis of variance assists in the assessment of Sri Lanka. The overall explanatory power of the the degree to which benchmark indicators and model is 12.42 percent in Nicaragua and 7.85 per- community variables help explain the variation cent in Sri Lanka. Panel B indicates the �t when the in entrepreneurship among rural households. best components are used: relative to the model Since the explained variable (entrepreneurship) is with only household characteristics, a few well- dichotomous, the criterion value is not the sum of selected variables provide significant improve- squared residuals but rather the value of the ments in both countries, and in Sri Lanka this log-likelihood function, which, moreover, is nega- even exceeds the explanatory power of panel A. tive and maximized. The evidence is summarized Panels C and D describe the explanatory power of in Table F.3. If no explanatory variables enter the benchmark indicators and community characteris- model, the criterion value equals –611.8 for tics in isolation: the benchmark indicators carry Nicaragua and –489.8 for Sri Lanka. In the ideal relatively less information in Sri Lanka. Annex F. Entrepreneurship: Notes and Tables 157 Tables Table F.1 De�nitions and Descriptive Statistics of Variables Used in the Econometric Models A: Variable de�nitions Variable De�nition Household Characteristics Entrepreneurshipa Dummy variable, 1 household operates an enterprise, 0 otherwise Wage employmenta Dummy variable, 1 members of the household hold a wage job, 0 otherwise Nonfarm economic activitya,b Dummy variable, 1 household operates an enterprise and/or household members hold a wage job, 0 = otherwise Farminga Dummy variable, 1 household operates a farm, 0 = otherwise Enterprise start-upc Dummy variable, 1 household operates an enterprise that started up at most two years prior to the time of the survey, 0 otherwise Number of male adults Number of male adults in the household Number of female adults Number of female adults in the household Average age Average age among adults in the household Human-capital index Average human capital (de�ned as in Annex E) among adults in the household Head female Dummy variable, 1 if the head of household is female, 0 otherwise Not Sinhalese Dummy variable, 1 if the head of household is not of Sinhalese ethnicity, 0 otherwise Head’s parents were entrepreneur Dummy variable, 1 if the parents of the head of household, 0 otherwise ln(Other income) Natural log of the sum of all types of income except wage, enterprise, farm, and remittance income (US$) ln(Remittances) Natural log of remittance income (US$) ln(Assets) Natural log of household assets (US$) Benchmarks and components Connectivity index Connectivity index, de�ned in Annex E Infrastructure services index Infrastructure services index, de�ned in Annex E Business services index Business services index, de�ned in Annex E Governance index Governance index, de�ned in Annex E Human-capital index Community-level human-capital index, de�ned in Annex E Finance services index Finance services index, de�ned in Annex E Community variables ln(Community size) Natural log of number of residents in the community ln(Income per capita) Natural log of the average income per capita (US$) in the community Agricultural seasonality Average of the standard deviation in monthly agricultural labor input for male and female labor Enterprise openness Index of dealings of the enterprise outside the community with respect to clients and input providers ln(Male wage rate) Natural log of the average male wage rate in agriculture, service, and manufacturing 158 The Rural Investment Climate Table F.1 De�nitions and Descriptive Statistics of Variables Used in the Econometric Models (continued) B: Descriptive statistics Nicaragua Sri Lanka Variable Mean StDev Mean StDev Household Characteristics Entrepreneurshipa 0.219 0.414 0.253 0.435 Wage employmenta 0.704 0.456 0.760 0.427 Nonfarm economic activitya,b 0.817 0.387 0.894 0.308 Farminga 0.272 0.445 0.479 0.500 Enterprise start-upc 0.030 0.171 0.013 0.112 Number of male adults 1.316 0.992 1.496 0.914 Number of female adults 1.459 0.901 1.494 0.783 Average age 33.975 8.726 37.323 7.701 Human capital index 0.209 0.136 0.333 0.158 Head female 0.329 0.470 0.191 0.393 Not Sinhalese n.a n.a 0.140 0.347 Head’s parents were entrepreneur 0.362 0.480 0.272 0.445 ln(Other income) 1.190 2.345 n.a n.a ln(Remittances) 1.574 2.580 n.a n.a ln(Assets) 7.258 1.717 8.595 1.147 Benchmarks and components Connectivity index 0.350 0.186 0.436 0.166 Infrastructure services index 0.535 0.256 0.440 0.162 Business services index 0.296 0.428 0.179 0.241 Governance index 0.686 0.111 0.677 0.037 Human capital index 0.205 0.074 0.324 0.092 Finance services index 0.288 0.244 0.357 0.098 Community variables ln(Community size) 8.702 1.929 7.539 0.594 ln(Income per capita) 5.824 0.702 8.144 0.554 Agricultural seasonality 0.721 0.413 0.731 0.307 Enterprise openness 1.438 0.349 2.227 0.785 ln(Male wage rate) 0.002 0.199 5.542 0.200 Number of households 1163 849 Source: RIC Surveys. Notes: a. Dependent variable. Column percentages do not add up to 100 as households may participate in several activities simultaneously. Nicaragua follows income-based de�nitions; Sri Lanka follows work-based de�nitions. b. Combining nonfarm entrepreneurship and wage employment. c. Dependent variable. In Sri Lanka, enterprise start-up is measured for 790 households, rather than 849. Annex F. Entrepreneurship: Notes and Tables 159 Table F.2 Determinants of Entrepreneurship Choices A: Nicaragua: Weighted probita (1) (2) (3) (4) dP t b dP t dP t dP t Household characteristics Number of male adults 3.08 2.23 3.00 2.24 3.42 2.55 Number of female adults 2.55 2.01 2.82 2.30 2.71 2.20 Average age 4.07 2.95 3.89 2.88 3.83 2.89 Human capital index 0.83 0.58 0.84 0.59 1.06 0.75 Head female 6.14 2.06 5.31 1.81 5.74 1.98 Head parents entrepreneur 2.27 0.86 2.06 0.80 2.19 0.86 ln(Other income) 2.66 2.08 2.82 2.21 2.42 1.91 ln(Remittances) 7.53 5.83 7.26 5.71 7.27 5.71 ln(Assets) 5.14 3.59 4.99 3.56 4.71 3.40 Benchmarks and components Connectivity 0.80 0.38 0.42 0.19 0.53 0.23 Proximity to post of�ce 2.78 2.07 Infrastructure services 5.45 2.32 5.78 2.44 2.78 1.10 Access to water 3.33 2.50 Business services 1.91 1.25 2.28 1.48 1.80 1.16 Governance 0.74 0.59 0.45 0.35 0.09 0.07 Human capital 7.91 4.90 6.29 3.79 4.89 2.45 4.49 2.70 Finance services 0.37 0.25 1.00 0.69 1.29 0.85 Community variables ln(Community size) 0.78 0.36 ln(Income per capita) 2.76 1.36 Agricultural seasonality 1.94 1.40 Enterprise openness 3.67 2.48 5.30 3.91 ln(Male wage) 0.48 0.35 Regression statistics Log-likelihood 580.19 544.10 535.83 535.90 Number of observations 1163 1163 1163 1163 (continued on the next page) 160 The Rural Investment Climate Table F.2 Determinants of Entrepreneurship Choices (continued ) B: Sri Lanka: Weighted random effect probita (1) (2) (3) (4) dP t dP t dP t dP t Household characteristics Number of male adults 2.73 1.26 2.92 1.38 3.25 1.54 Number of female adults 2.17 1.00 2.03 0.94 1.88 0.88 Average age 0.91 0.33 0.84 0.32 0.58 0.23 Human capital index 0.93 0.34 1.04 0.38 1.57 0.60 Head female 8.61 1.03 8.83 1.08 10.40 1.42 Not Sinhalese 4.99 0.51 4.87 0.51 14.24 1.80 Head parents entrepreneur 1.56 0.31 2.11 0.42 2.09 0.43 ln(Assets) 9.74 3.13 9.82 3.18 10.55 3.59 Benchmarks and components Connectivity 5.70 1.29 5.34 1.28 4.29 1.04 Proximity to post of�ce 4.84 1.84 Infrastructure services 2.19 0.64 2.08 0.58 4.14 1.13 Sewage system 7.33 2.54 Business services 3.49 1.13 3.66 1.27 3.36 1.08 Engineering services 4.49 1.69 Information technology services 5.93 1.58 Governance 2.89 1.07 3.91 1.43 3.75 1.40 Human capital 4.83 1.35 2.41 0.60 3.36 0.81 Finance services 2.95 0.67 3.68 0.83 4.26 1.02 Community variables ln(Community size) 4.44 1.25 3.98 1.35 ln(Income per capita) 0.90 0.31 Agricultural seasonality 0.96 0.27 Enterprise openness 0.72 0.20 ln (Male wage rate) 2.59 0.78 4.36 1.67 Regression statistics Standard error of random effect 10.38 3.80 9.92 3.65 9.80 3.49 6.99 2.81 Log-likelihood 474.70 454.90 451.39 438.06 Number of observations 849 849 849 849 Source: RIC Surveys. Notes: a. Columns headed with “dP� report the percentage point increase in the probability that an average household operates an enterprise in response to a one-standard-deviation increase in the explanatory variable. For the variables “Head female,� “Head parents entrepreneur,� and “Not Sinhalese,� which are dummy variables, the change in the explanatory variable is one unit. b. Signi�cance levels implied by the t-statistics are: 1% if t 2.58, 5% if 1.96 t 2.58, 10% if 1.645 t 1.96. Annex F. Entrepreneurship: Notes and Tables 161 Table F.3 Contributions to Explanation of Entrepreneurship Choice Nicaragua Sri Lanka Criterion Increment as % Criterion Increment as % Explanatory variables valuea of base valuea of base None (base) –611.797 –489.841 A. Household characteristics only –562.312 8.09% –467.398 4.58% Community random effect –562.312 0.00% –462.053 1.09% Benchmark indicators –544.097 2.98% –454.898 1.46% Community characteristics –535.835 1.35% –451.391 0.72% B. Household characteristics only –562.312 –467.398 Community random effect and best components –535.900 4.32% –438.063 5.88% C. Benchmark indicators and community random effect only –580.186 5.17% –474.698 3.09% D. Community characteristics and community random effect only –579.148 5.34% –479.544 2.10% Source: RIC Surveys. Note: a. The criterion is the maximized log-likelihood function, which in the case of probit models is always negative and rises as explanation improves. Table F.4 Determinants of Household Activity Choices A: Nicaragua: Weighted probit and weighted random effect probita Nonfarm Nonfarm household Wage economic enterprise employment activity Farming dP tb dP t dP t dP t Household characteristics Number of male adults 3.42 2.55 7.91 3.03 3.41 1.80 4.01 1.40 Number of female adults 2.71 2.20 1.47 0.69 1.03 0.71 0.16 0.05 Average age 3.83 2.89 5.46 2.47 1.61 1.17 10.15 3.64 Human capital index 1.06 0.75 9.11 4.43 7.07 4.26 10.62 3.36 Head female 5.74 1.98 2.60 0.52 0.53 0.15 6.16 0.90 Head parents entrepreneur 2.19 0.86 1.69 0.42 2.03 0.67 8.89 1.37 ln(Other income) 2.42 1.91 4.86 2.32 4.80 3.99 4.59 1.61 ln(Remittances) 7.27 5.71 6.42 3.13 7.89 5.50 4.32 1.36 ln(Assets) 4.71 3.40 18.77 6.87 9.59 4.92 19.33 5.47 Benchmarks and components Proximity to post of�cec 2.78 2.07 Percent of households with electricityd 8.18 2.15 4.84 1.42 Infrastructure services, access to waterd 3.33 2.50 Sewage systemd 8.97 1.47 Dealing with government agenciesf 7.19 2.06 4.09 1.42 Rule of lawf 7.13 1.31 Human capital 4.49 2.70 6.72 1.22 Community variables ln(Income per capita) 11.01 1.99 5.47 1.31 15.14 2.56 Agricultural seasonality 7.31 1.72 8.37 1.62 Enterprise openness 5.30 3.91 3.87 1.15 Regression statistics Standard error of random effect 0.00 15.46 5.41 12.00 4.41 19.07 5.12 Log-likelihood 535.90 544.73 330.56 498.38 Number of observations 1163 1163 1163 1163 (continued on the next page) 162 The Rural Investment Climate Table F.4 Determinants of Household Activity Choices (continued ) B: Sri Lanka: Weighted random effect probita Nonfarm Nonfarm household Wage economic enterprise employment activity Farming dP t dP t dP t dP t Household characteristics Number of male adults 3.25 1.54 5.32 2.27 0.64 0.75 12.26 4.08 Number of female adults 1.88 0.88 10.10 4.72 2.66 3.49 4.40 1.45 Average age 0.58 0.23 5.41 2.45 2.44 3.05 4.84 1.69 Human capital index 1.57 0.60 7.68 3.20 3.29 3.57 3.09 0.91 Head female 10.40 1.42 12.17 2.34 1.68 0.95 7.50 0.87 Not Sinhalese 14.24 1.80 4.33 0.49 5.85 2.23 17.76 1.39 Head’s parents entrepreneur 2.09 0.43 0.21 0.04 2.19 1.21 1.89 0.27 ln(Assets) 10.55 3.59 6.04 2.13 0.42 0.47 20.64 4.87 Benchmarks and components Inverse cost of transport to major cityc 6.55 1.65 Inverse cost of transport to main marketc 5.73 1.63 Proximity to post of�cec 4.84 1.84 Infrastructure services 5.61 1.78 2.28 1.76 Percent of households with electricityd 8.54 1.32 Percent of households with �xed phone lined 13.38 1.93 Sewage systemd 7.33 2.54 Engineering servicese 4.49 1.69 Management consulting servicese 15.62 2.17 Marketing servicese 4.94 2.18 18.92 2.75 Information technology servicese 5.93 1.58 Conflict resolution, contract enforcementf 1.67 1.37 Financial penetrationg 6.38 2.38 9.32 1.56 Community variables ln(Community size) 3.98 1.35 4.90 1.66 3.35 2.34 11.52 1.88 Agricultural seasonality 9.20 1.36 ln(Male wage rate) 4.36 1.67 Regression statistics Standard error of random effect 6.99 2.81 8.79 3.85 5.38 2.86 24.81 5.18 Log-likelihood 438.06 390.69 188.94 401.02 Number of observations 849 849 849 849 Source: RIC Surveys. Notes: a. Columns headed with “dP� report the percentage point increase in the probability that an average household operates an enterprise in response to a one-standard-deviation increase in the explanatory variable. For the variables “Head female,� “Head parents entrepreneur,� and “Not Sinhalese,� which are dummy variables, the change in the explanatory variable is one unit. b. Signi�cance levels implied by the t-statistics are: 1% if t 2.58, 5% if 1.96 t 2.58, 10% if 1.645 t 1.96. c. A component of the connectivity index. d. A component of the infrastructure services index. e. A component of the business services index. f. A component of the governance index. g. A component of the �nance services index. Annex F. Entrepreneurship: Notes and Tables 163 Table F.5 Determinants of Enterprise Start-Upa A: Nicaragua (1) (2) (3) (4) dP t dP t dP t dP t Household characteristics Number of male adults 0.04 0.13 0.10 0.37 0.08 0.29 Number of female adults 0.08 0.21 0.12 0.31 0.14 0.37 Average age 0.39 1.05 0.50 1.31 0.41 1.13 Human capital index 0.14 0.42 0.23 0.74 0.21 0.75 Head female 1.15 1.40 1.09 1.32 1.05 1.33 Head’s parents entrepreneur 0.10 0.14 0.18 0.25 0.06 0.09 ln(Other income) 0.61 1.81 0.68 2.03 0.67 1.96 ln(Remittance income) 1.08 2.85 0.99 2.74 0.94 2.65 ln(Assets) 0.31 1.05 0.33 1.09 0.36 1.18 Benchmarks and components Connectivity 0.96 1.60 0.89 1.55 Inverse cost of transport to major city 0.57 1.62 Infrastructure services 0.72 1.29 0.79 1.45 Percent of households with �xed phone line 0.52 1.20 Business services 0.26 0.51 0.13 0.31 Governance 0.35 1.04 0.35 1.14 Human capital 0.54 1.41 0.38 0.99 Finance 0.26 0.65 0.27 0.74 Community variables ln(Community size) 0.45 1.16 0.72 1.83 ln(Income per capita) 0.17 0.39 Agricultural seasonality 0.50 1.52 0.40 1.24 Enterprise openness 0.64 1.96 0.63 2.11 ln(Male wage rate) 0.44 1.51 Regression statistics lnL 156.51 152.46 151.34 150.91 n 1170 1170 1170 1170 (continued on the next page) 164 The Rural Investment Climate Table F.5 Determinants of Enterprise Start-Upa (continued ) B: Sri Lanka (1) (2) (3) (4) dP t dP t dP t dP t Household characteristics Number of male adults 0.11 0.66 0.19 1.13 0.16 1.06 Number of female adults 0.20 1.45 0.15 1.19 0.17 1.38 Average age 0.27 1.60 0.26 1.48 0.25 1.52 Human capital index 0.17 0.93 0.11 0.71 0.21 1.18 Head female 0.28 0.76 0.33 0.88 0.26 0.77 Not Sinhalese 1.05 1.68 0.77 1.37 1.04 1.89 Head parents entrepreneur 0.24 0.66 0.51 1.41 0.37 1.06 ln(Assets) 0.43 2.11 0.46 2.28 0.43 2.38 Benchmarks and components Connectivity 0.02 0.10 0.03 0.13 Inverse cost of transport to major city 0.33 1.81 Inverse cost of transport to main market 0.26 1.65 Infrastructure services 0.37 1.46 0.47 1.96 0.40 2.27 Business services 0.19 0.83 0.21 1.06 Governance 0.17 1.20 0.17 1.28 Human capital 0.38 2.02 0.44 1.93 0.28 1.47 Finance 0.10 0.53 0.08 0.42 Community variables ln(Community size) 0.16 0.96 ln(Income per capita) 0.00 0.02 Agricultural seasonality 0.23 1.26 Enterprise openness 0.29 1.71 0.30 1.88 ln(Male wage rate) 0.36 2.05 0.24 1.38 Regression statistics lnL 50.54 49.29 49.00 48.48 n 790 790 790 790 Source: RIC Surveys. Notes: a. Estimates of a weighted probit model. Columns headed with “dP� report the percentage point increase in the probability that an average household operates an enterprise in response to a one-standard-deviation increase in the explanatory variable. For the variables “Head female,� “Head parents entrepreneur,� and “Not Sinhalese,� which are dummy variables, the change in the explanatory variable is one unit. b. Signi�cance levels implied by the t-statistics are: 1% if t 2.58, 5% if 1.96 t 2.58, 10% if 1.645 t 1.96. Annex G. Benchmark Indicators 165 166 Table G.1 Missing Data for Communities and Components Nicaragua Sri Lanka Tanzania missing missing missing missing missing missing st commu- compo- st commu- compo- st commu- compo- obs obs average dev nities nents obs obs average dev nities nents obs obs average dev nities nents Number of communities 98 147 149 E. Benchmark Indicators Indicator 1: Connectivity 98 0.34 0.18 0 146 0.40 0.17 1 149 0.20 0.14 0 Subindex 1: Time taken 98 0.32 0.30 0 146 0.47 0.31 1 149 0.26 0.31 0 by main means of transportation to the nearest major city Subindex 2: Cost of 98 0.09 0.08 0 136 0.66 0.29 11 149 0.08 0.17 0 transportation to the nearest major city (by public transportation) Subindex 3: Time 85 0.33 0.31 13 145 0.49 0.32 2 149 0.52 0.39 0 taken by main means of transportation to the main market Subindex 4: Cost of 98 0.07 0.09 0 146 0.61 0.36 1 149 0.08 0.17 0 transportation to the main market (by bus) Subindex 5: Distance 98 0.47 0.46 0 145 0.69 0.35 2 136 0.26 0.33 13 to the post of�ce Subindex 6: Rail stop n.a. 145 0.12 0.33 2 149 0.23 0.42 0 in walking distance Subindex 7: Percentage 98 0.59 0.49 0 145 0.08 0.12 2 146 0.03 0.07 3 of households with �xed telephone lines Subindex 8: Percentage 98 0.51 0.50 0 145 0.10 0.14 2 147 0.13 0.20 2 of households with cell phones Indicator 2: Infrastructure Services 98 0.51 0.24 0 146 0.35 0.16 1 149 0.19 0.16 0 Subindex 1: Percentage 97 0.67 0.31 1 145 0.69 0.29 2 148 0.12 0.22 1 of households that use electricity Subindex 2: Availability 94 0.62 0.39 4 145 0.55 0.39 2 145 0.31 0.43 4 of electricity index Subindex 3: Percentage 98 0.62 0.32 0 145 0.80 0.28 2 149 0.47 0.38 0 of households with access to protected water Subindex 4: Percentage 98 0.59 0.49 0 145 0.08 0.12 2 146 0.03 0.07 3 of households with �xed-line telephone Subindex 5: Percentage 98 0.51 0.50 0 145 0.10 0.14 2 147 0.13 0.20 2 of households with cellular phones Subindex 6: Sewage 98 0.08 0.28 0 145 0.11 0.31 2 148 0.07 0.25 1 channels in the community Subindex 7: Garbage 98 0.49 0.50 0 145 0.12 0.32 2 149 0.13 0.33 0 collection or disposal service in the community Subindex 8: Most n.a. 146 0.33 0.47 1 149 0.22 0.42 0 common road surface (internal road) is concrete or asphalt Indicator 3: Business 98 0.24 0.40 0 142 0.15 0.22 5 146 0.07 0.15 3 Services Subindex 1: Engineering 98 0.23 0.43 0 142 0.06 0.23 5 146 0.04 0.20 3 services available Subindex 2: Management 98 0.22 0.42 0 142 0.07 0.26 5 146 0.16 0.37 3 consulting services Subindex 3: 98 0.19 0.40 0 142 0.08 0.28 5 146 0.13 0.34 3 Marketing services Subindex 4: 98 0.27 0.44 0 142 0.11 0.31 5 146 0.02 0.14 3 Accounting services Subindex 5: Legal services 98 0.31 0.46 0 142 0.26 0.44 5 146 0.07 0.25 3 Subindex 6: Insurance 98 0.21 0.41 0 142 0.39 0.49 5 146 0.03 0.16 3 services Subindex 7: Information 98 0.24 0.43 0 142 0.10 0.30 5 146 0.03 0.16 3 technology (continued on the following page) 167 168 Table G.1 Missing Data for Communities and Components (continued) Nicaragua Sri Lanka Tanzania missing missing missing missing missing missing st commu- compo- st commu- compo- st commu- compo- obs obs average dev nities nents obs obs average dev nities nents obs obs average dev nities nents Indicator 4: Governance 96 0.67 0.13 2 125 0.67 0.04 22 140 0.50 0.08 9 Subindex 1: General 96 0.79 0.16 2 125 0.98 0.03 22 140 0.85 0.19 9 policy and institutional constraints Subindex 2: Infrastructure 60 0.00 0.00 38 125 0.13 0.06 22 140 0.01 0.03 9 and services Subindex 3: Dealing with 53 0.94 0.18 45 125 1.00 0.00 22 n.a. government agencies Subindex 4: Rule of law 94 0.50 0.30 4 125 0.49 0.14 22 140 0.49 0.16 9 Subindex 5: Conflict 96 0.59 0.22 2 125 0.75 0.14 22 140 0.66 0.18 9 resolution and contract enforcement Indicator 5: Human 97 0.21 0.08 1 134 0.33 0.09 13 146 0.21 0.06 3 Capital Indicator 6: Finance 98 0.17 0.27 0 147 0.50 0.16 0 149 0.18 0.15 0 Services* Subindex 1: Number of 98 0.20 0.29 0 147 0.56 0.22 0 149 0.24 0.25 0 formal �nancial sources that serve a community Subindex 2: Number of 98 0.15 0.25 0 147 0.47 0.23 0 149 0.12 0.16 0 �nance services that are provided by formal institutions in a community Subindex 3: Access to loans n.a. 147 0.47 0.23 0 140 0.19 0.22 9 Total missing number of 3 109 42 183 15 95 observations on communities and components respectively Total missing observations 0.3 2.6 4.6 3.9 1.7 2.1 on communities and components respectively (%) Source: the Authors. * See note to Table E.6 Annex G. Benchmark Indicators 169 Table G.2 Nicaragua: Correlation Coef�cients Between Benchmark Indicators and Prices at Community Level Infra structure Connectivity services Business services Governance Human capital Finance services Index, price of food 0.21* 0.11 0.08 0.04 0.13 0.04 Diesel price 0.34* 0.37* 0.18 0.01 0.21 0.04 Fertilizer price 0.17 0.11 0.07 0.02 0.19 0.08 Index, price of inputs 0.08 0.12 0.08 0.07 0.21 0.13 Average price of phone call 0.21* 0.14 0.17 0.28* 0.10 0.05 Index, price of transportation 0.17 0.01 0.07 0.30* 0.30* 0.05 Average price of salt 0.36* 0.16 0.06 0.04 0.13 0.07 Index, price of Coca-Cola 0.02 0.01 0.05 0.13 0.23* 0.04 Price of soap 0.05 0.03 0.04 0.01 0.34* 0.06 Index, wage (male) 0.15 0.12 0.11 0.29* 0.07 0.04 Index, wage (female) 0.16 0.04 0.03 0.22* 0.10 0.01 Source: RIC Surveys. Note: * indicates at least 10 percent signi�cance level. Table G.3 Sri Lanka: Correlation Coef�cients Between Benchmark Indicators and Prices at Community Levela Infrastructure Business Human Finance Connectivity services services Governance capital services Average fertilizer (Rs/kg) 0.17* 0.04 0.02 0.03 0.17* 0.01 Average cement (Rs/50 kg bag) 0.27* 0.10 0.14 0.25* 0.04 0.25* Asbestos sheet (one sheet 96*53) 0.27* 0.20* 0.21* 0.08 0.05 0.37* Lawyer fee for a civil case 0.29* 0.18* 0.16* 0.04 0.11 0.15* Average price rice per kg. 0.16* 0.19* 0.12 0.27* 0.06 0.30* Price of wheat flour per kg. 0.14* 0.15* 0.26* 0.24* 0.07 0.15* Price of Coca-Cola 0.25* 0.21* 0.13 0.04 0.13 0.35* Price of kerosene per liter 0.17* 0.20* 0.14* 0.14 0.11 0.24* Price of toilet soap 0.04 0.21* 0.04 0.12 0.01 0.17* Price of a 40w bulb 0.07 0.26* 0.06 0.13 0.01 0.21* Price of a bike tire 0.14* 0.03 0.02 0.15 0.13 0.07 Male casual wage 0.15* 0.02 0.13 0.05 0.23* 0.02 Female casual wage 0.17* 0.11 0.23* 0.08 0.17* 0.04 Casual wage 0.15 0.06 0.20* 0.05 0.22* 0.01 Time to clear a check in this area (days) 0.30* 0.18* 0.11 0.11 0.19* 0.21* Time to clear a clearance check in this area 0.35* 0.28* 0.12 0.03 0.27* 0.44* Cost of bus fee to commercial center 0.50* 0.36* 0.10 0.15 0.10 0.52* Minutes to commercial center by bus 0.48* 0.17* 0.09 0.01 0.19* 0.20* Cost of 3-wheeler fee to commercial center 0.60* 0.23* 0.19* 0.12 0.19* 0.38* Minutes to commercial center by 3-wheeler 0.45* 0.16* 0.06 0.03 0.25* 0.22* Source: RIC Surveys. Note: * indicates at least 10 percent signi�cance level. a. Several cost indicators are expressed in time required for service. 170 The Rural Investment Climate Table G.4 Tanzania: Correlation Coef�cients Between Benchmark Indicators and Prices at Community Level Infrastructure Business Human Finance Connectivity services services Governance capital services Diesel per liter 0.17 0.02 0.27 0.59* 0.22 0.37* Cement 0.07 0.05 0.02 0.05 0.05 0.19* Galvanized steel sheet for roo�ng 0.05 0.08 0.00 0.07 0.21* 0.02 Ordinary rice per kg. 0.04 0.08 0.17* 0.03 0.09 0.03 Coca-Cola per can 0.06 0.02 0.09 0.06 0.11 0.07 Kerosene per liter 0.12 0.11 0.10 0.07 0.06 0.11 Bar of toilet soap 0.12 0.12 0.19* 0.18* 0.08 0.06 Light bulb 0.13 0.18 0.09 0.08 0.25* 0.07 Agriculture wage per day 0.09 0.10 0.05 0.06 0.07 0.03 Construction wage per day 0.05 0.07 0.12 0.13 0.00 0.05 Rural public works per month 0.02 0.05 0.07 0.21* 0.01 0.08 Agriculture wage per day (female) 0.06 0.04 0.06 0.02 0.04 0.14* Construction wage per day (female) 0.03 0.12 0.05 0.03 0.06 0.11 Rural public works per month (female) 0.01 0.03 0.08 0.10 0.06 0.05 Telephone call to the capital town of the nearby region 0.22* 0.10 0.04 0.09 0.26* 0.03 Cell phone/mobile phone call to the capital town of the nearby region 0.26* 0.22* 0.22* 0.04 0.15 0.17 Average transport cost 0.08 0.08 0.01 0.20* 0.03 0.19* Source: RIC Surveys. Note: * indicates at least 10 percent signi�cance level. Annex G. Benchmark Indicators 171 Table G.5 Nicaragua: Correlation Coef�cients Between Benchmark Indicators and Community Characteristics Infrastructure Business Human Finance Connectivity services services Governance capital services Population 0.53* 0.51* 0.46* 0.08 0.27* 0.51* Road surface 0.37* 0.42* 0.27* 0.08 0.32* 0.06 Percent households with gas 0.44* 0.57* 0.32* 0.09 0.59* 0.17 Number of public services 0.69* 0.83* 0.46* 0.04 0.45* 0.31* Highest level of school type 0.32* 0.51* 0.34* 0.13 0.43* 0.53* Highest level of health center type 0.28* 0.44* 0.30* 0.08 0.29* 0.39* Index: Highest level of health center type, weighted by distance 0.46* 0.56* 0.38* 0.15 0.38* 0.47* Number of types of banks available 0.42* 0.50* 0.43* 0.20* 0.40* 0.72* Percentage of households that buy outside Immigrants (as % of population) 0.29* 0.38* 0.31* 0.04 0.23* 0.39* Number of business services provided by chamber of commerce 0.27* 0.24* 0.27* 0.03 0.03 0.20* Number of business services provided by overall business association 0.15 0.13 0.15 0.16 0.11 0.17* Number of business services provided by sector business association 0.35* 0.27* 0.20* 0.06 0.14 0.25* Number of services available to businesses 0.36* 0.30* 0.28* 0.09 0.12 0.28* Number of taxes 0.50* 0.44* 0.27* 0.20* 0.41* 0.45* Percent households with electricity 0.29* 0.62* 0.22* 0.10 0.37* 0.08 Quality of electricity 0.22* 0.16 0.10 0.02 0.15 0.04 Distance to market 0.23* 0.08 0.12 0.18* 0.08 0.14 Distance to city 0.06 0.16 0.28* 0.01 0.22* 0.38* Seasonality of male labor in agriculture 0.06 0.01 0.09 0.03 0.12 0.08 Seasonality of female labor in agriculture 0.06 0.03 0.02 0.08 0.11 0.05 Percent of households that bene�t from a series of programs (cumulative, might be over 100%) 0.14 0.11 0.25* 0.20* 0.22* 0.25* Percent of households that bene�t from infrastructure programs (cumulative, might be over 100%) 0.12 0.09 0.21* 0.19* 0.22* 0.22* Percent of households that bene�t from social programs (cumulative, might be over 100%) 0.14 0.12 0.34* 0.14 0.19* 0.29* Index: Formality of conflict-solving institution 0.10 0.05 0.06 0.01 0.10 0.13 Number of crimes 0.36* 0.25* 0.11 0.04 0.17* 0.25* Percent of reported incidents solved 0.20* 0.39* 0.27* 0.11 0.26* 0.45* Percent of conflicts solved by neighbors/friends 0.19* 0.14 0.17* 0.06 0.16 0.03 Percent of conflicts solved by community leaders 0.15 0.08 0.01 0.04 0.15 0.17* Percent of conflicts solved by local judge 0.18* 0.06 0.02 0.04 0.24* 0.12 Percent of conflicts solved by district 0.01 0.02 0.05 0.01 0.07 0.06 judge Percent of conflicts solved by magistrate court 0.14 0.12 0.01 0.04 0.14 0.13 Percent of conflicts solved by labor court 0.10 0.16 0.06 0.02 0.05 0.11 Total cultivated land per capita 0.14 0.12 0.12 0.08 0.00 0.04 Amount of lowland per capita 0.03 0.02 0.07 0.05 0.01 0.08 Total upland per capita 0.23* 0.20* 0.13 0.11 0.06 0.18 Source: RIC Surveys. Note: * indicates at least 10 percent signi�cance level. 172 The Rural Investment Climate Table G.6 Sri Lanka: Correlation Coef�cients Between Benchmark Indicators and Community Characteristics Infrastructure Business Human Finance Connectivity services services Governance capital services Population 0.19* 0.37* 0.27* 0.04 0.04 0.17* Road surface 0.19* 0.16* 0.09 0.20* 0.19* 0.14 Percent households with gas 0.41* 0.52* 0.26* 0.02 0.40* 0.35* Number of public services 0.61* 0.57* 0.21* 0.00 0.34* 0.43* Highest level of school type 0.01 0.06 0.12 0.11 0.04 0.02 Number of hospital services 0.08 0.14 0.08 0.09 0.14 0.04 Highest level of health center type 0.05 0.06 0.02 0.16* 0.12 0.02 Index: Highest level of health center type, weighted by distance 0.15* 0.17* 0.08 0.05 0.18* 0.16* Number of banks 0.32* 0.19* 0.28* 0.05 0.16* 0.32* Percent of households series of programs bene�t (cumulative, might be over 100%) 0.04 0.04 0.02 0.11 0.15* 0.02 Percent of households infrastructure programs bene�t (cumulative, might be over 100%) 0.05 0.01 0.05 0.15 0.03 0.24* Percent of households social programs bene�t (cumulative, might be over 100%) 0.02 0.02 0.05 0.03 0.03 0.16* Index: Formality of conflict-solving institution 0.03 0.03 0.19* 0.13 0.04 0.10 Percent of reported incidents solved 0.13 0.01 0.09 0.10 0.14 0.10 Number of business services provided by chamber of commerce 0.00 0.02 0.17* 0.02 0.00 0.02 Number of business services provided by overall business association 0.02 0.14 0.12 0.15 0.05 0.12 Number of business services provided by sector business association 0.01 0.06 0.10 0.08 0.10 0.06 Immigrants, as % of population 0.04 0.03 0.03 0.03 0.08 0.12 Number of business services (overall) 0.01 0.13 0.21* 0.26* 0.07 0.01 Number of taxes 0.40* 0.19* 0.21* 0.06 0.20* 0.26* Percent households with electricity 0.44* 0.55* 0.16* 0.02 0.38* 0.51* Quality of electricity 0.00 0.26* 0.10 0.14 0.06 0.07 Distance to market 0.38* 0.23* 0.09 0.02 0.07 0.05 Distance to city 0.56* 0.33* 0.08 0.03 0.06 0.03 Standard deviation of agricultural male labor 0.22* 0.27* 0.21* 0.23* 0.21* 0.14 Standard deviation of agricultural female labor 0.29* 0.28* 0.17* 0.20* 0.17* 0.16* Percent conflicts solved by neighbors/friends 0.11 0.11 0.02 0.08 0.12 0.00 Percent conflicts solved by community leaders 0.03 0.06 0.08 0.18* 0.05 0.18* Percent conflicts solved by local judge 0.05 0.03 0.05 0.21* 0.08 0.11 Percent conflicts solved by court of appeal 0.01 0.13 0.06 0.11 0.02 0.02 Percent conflicts solved by magistrate court 0.01 0.02 0.03 0.05 0.06 0.02 Percent conflicts solved by labor court 0.16* 0.05 0.30* 0.03 0.08 0.03 Number of crimes 0.06 0.02 0.03 0.03 0.03 0.08 Total cultivated land per capita 0.35* 0.34* 0.13 0.06 0.14 0.30* Total lowland per capita 0.29* 0.34* 0.15* 0.09 0.11 0.29* Total upland per capita 0.31* 0.28* 0.10 0.11 0.13 0.25* Source: RIC Surveys. Note: * indicates at least 10 percent signi�cance level. Annex G. Benchmark Indicators 173 Table G.7 Tanzania: Correlation Coef�cients Between Benchmark Indicators and Community Characteristics Infrastructure Business Human Finance Connectivity services services Governance capital services Population 0.00 0.01 0.01 0.14* 0.02 0.02 Road surface 0.10 0.27* 0.06 0.29* 0.01 0.20* Percent households with cell phone 0.47* 0.70* 0.17* 0.10 0.36* 0.15* Number of public services 0.67* 0.72* 0.26* 0.01 0.38* 0.26* Highest level of school type 0.11 0.24* 0.17* 0.02 0.17* 0.02 Highest level of health center type 0.19* 0.25* 0.13 0.07 0.26* 0.05 Index: Highest level of health center type, pondered by distance 0.26* 0.35* 0.27* 0.05 0.21* 0.04 Index: Highest level of health center type, weighted by distance 0.25* 0.42* 0.27* 0.01 0.22* 0.12 Number of banks 0.18* 0.27* 0.43* 0.08 0.14* 0.38* Number of banks available, weighted by distance 0.05 0.07 0.12 0.04 0.10 0.24* Index: Mean distance to banking centers 0.06 0.06 0.12 0.03 0.08 0.29* Index: Formality of conflict-solving institution 0.02 0.05 0.08 0.10 0.02 0.01 Percent of conflicts solved by neighbors/friends 0.13 0.20* 0.01 0.22* 0.13 0.00 Percent of conflicts solved by community leaders 0.09 0.01 0.08 0.16* 0.05 0.00 Percent of conflicts solved by ward tribunal 0.18* 0.13 0.00 0.14* 0.13 0.14* Percent of conflicts solved by magistrate court 0.04 0.02 0.04 0.06 0.02 0.05 Percent of conflicts solved by resident court 0.05 0.10 0.02 0.01 0.10 0.02 Emigrants, as % of population 0.06 0.15* 0.14* 0.06 0.04 0.01 Number of business services provided by chamber of commerce 0.20* 0.07 0.24* 0.09 0.06 0.18* Number of business services provided by overall business association 0.08 0.04 0.40* 0.10 0.02 0.06 Number of business services provided by sector business association 0.06 0.01 0.26* 0.09 0.04 0.12 Number of business services (overall) 0.14* 0.05 0.35* 0.11 0.05 0.15* number of taxes 0.18* 0.16* 0.18* 0.06 0.17* 0.22* Number of disasters/events 0.09 0.04 0.09 0.02 0.17* 0.11 Number of crimes 0.21* 0.35* 0.15* 0.08 0.14* 0.10 Percent of reported incidents solved 0.10 0.14 0.12 0.14 0.13 0.01 Percent households with electricity 0.46* 0.68* 0.12 0.06 0.35* 0.14* Quality of electricity 0.16 0.19 0.10 0.18 0.09 0.04 Distance to market 0.40* 0.27* 0.02 0.07 0.17* 0.05 Distance to city 0.31* 0.08 0.02 0.14 0.04 0.14 Percent households that bene�t from investment programs 0.07 0.03 0.01 0.11 0.01 0.04 Total cultivated land per capita 0.27* 0.28* 0.08 0.04 0.11 0.07 Amount of lowland per capita 0.14 0.13 0.10 0.05 0.13 0.20* Amount of upland per capita 0.18* 0.23* 0.07 0.19* 0.11 0.15 Source: RIC Surveys. Note: * indicates at least 10 percent signi�cance level. 174 Table G.8 Descriptive Statistics of Variables Used in Regression Analysis A: Dependent variables Variable name Obs Mean St.Dev. Min Max De�nition entdense_hh Enterprise density 147 0.162 0.132 0.018 0.935 number of enterprises per household lnworkincpca Income from work# 130 5.510 0.662 1.562 7.376 log of per capita income from work (farm, nfhe, wage) a share_rn�nca Ratio RNF income 130 0.787 0.586 0.000 6.706 ratio of rural nonfarm income over total income from worka share_rnfworka Ratio RNF workers 130 0.808 0.213 0.263 1.000 ratio of rural nonfarm workers over total number of workersa lnrn�ncpca RNF income per capita# 128 5.166 0.748 1.044 7.324 log of rural nonfarm income per capitaa lnrn�ncpwa RNF income per worker# 128 6.320 0.646 3.285 8.474 log of rural nonfarm income per workera lfpra Labor force participation rate 130 0.592 0.127 0.337 0.995 labor force participation ratea lnwmalea Male wage# 136 5.456 0.242 4.828 5.991 log of reported male wagea ratio_salecostb Ratio sales / total cost 147 2.643 7.156 0.577 86.06 ratio of sales over total costb ratio_nvatfcb Ratio NVA / factor cost 142 21.12 67.41 0.455 658.0 ratio of net value added over total factor costb lnGVAb Gross Value Added (GVA)# 147 7.166 1.386 4.284 12.51 log of gross value added per enterpriseb lnSD_GVAb St dev GVA# 144 7.307 1.908 2.083 13.39 log of standard deviation of gross value addedb seasadjb Seasonal adjustment 147 10.11 1.387 6.167 12.00 seasonal adjustment factor in enterprises (higher is less seasonal) b agseas Agricultural seasonality 141 0.744 0.293 0.000 1.379 agricultural seasonality (higher is more seasonal) B: Explanatory variables lnpopnw Community population size# 142 7.312 0.596 6.310 8.990 log of population size shsinh Share Sinhalese population 142 77.148 36.950 0.000 100.000 share of Sinhalese among the population shilliterate Illiteracy 142 13.173 13.032 0.000 70.000 share of illiterate dschoolindex Proximity to school 142 0.360 0.208 0.000 1.000 proximity to school index dhospindex Proximity to hospital 142 0.227 0.136 0.000 0.571 proximity to hospital index distcityindex Proximity to city 142 0.594 0.372 0.000 1.000 proximity to city index opennessb Openness 124 2.135 0.689 1.048 4.834 openness (inputs, customers) among enterprisesb lowlandpc Lowland per capita 139 0.123 0.192 0.000 1.019 acres of lowland per capita uplandpc Upland per capita 139 0.247 0.441 0.000 4.587 acres of upland per capita conn_index Connectivity 146 0.398 0.167 0.026 0.781 connectivity benchmark infra_index Infrastructure services 146 0.345 0.165 0.000 0.795 infrastructure services benchmark devs_index Business services 142 0.153 0.224 0.000 1.000 business services benchmark gov_indexb Governance 125 0.670 0.041 0.569 0.756 governance benchmarkb hc_indexa Human capital 134 0.329 0.086 0.133 0.577 human capital benchmarka �nance_indexc Finance services 147 0.501 0.162 0.002 0.842 �nance services benchmarkc pricerice Price rice 131 30.319 2.640 20.000 38.000 consumer price of rice pcoke Price Coca-Cola 147 20.755 3.035 12.870 32.540 consumer price of can of Coca-Cola pkerosine Price kerosene 146 27.248 2.452 21.500 36.000 consumer price of kerosene Source: RIC Surveys. Note: a. Aggregated from household data. b. Aggregate from enterprise data. c. In part, an aggregate from enterprise data. # Logarithmic value. Table G.9 Regressions at the Community Level A: Full set of explanatory variables RNF income per Enterprise density Income from work# Ratio RNF income Ratio RNF workers capita# b t b t b t b t b t Community population size# 0.0211 0.98 0.2108 2.19 0.0356 0.62 0.0972 2.93 0.1261 0.95 Share Sinhalese population 0.0008 1.92 0.0003 0.14 0.0035 2.61 0 0 0.0044 1.39 Illiteracy 0.0015 1.61 0.0033 0.73 0.0011 0.46 0.0022 1.25 0.0036 0.63 Proximity to school 0.1431 2 0.1754 0.6 0.1033 0.64 0.0725 0.73 0.0264 0.07 Proximity to hospital 0.3049 3.15 0.4815 1.18 0.2021 0.92 0.1955 1.45 0.3559 0.8 Proximity to city 0.01 0.23 0.0857 0.46 0.2781 2.62 0.061 0.89 0.2076 0.94 Openness 0.0318 2.61 0.1983 2.53 0.0413 1.29 0.0066 0.23 0.2262 1.99 Lowland per capita 0.0139 0.3 0.1571 0.39 0.2954 1.38 0.4269 2.64 0.6903 1.35 Upland per capita 0.0202 1.12 0.1302 0.84 0.0827 1.57 0.0108 0.23 0.1847 0.86 Connectivity 0.1865 2.14 0.8791 1.75 0.2499 1.08 0.0448 0.3 0.2344 0.39 Infrastructure services 0.025 0.29 1.0754 1.83 0.1675 0.63 0.0524 0.34 0.7345 0.98 Business services 0.1486 3.45 0.8895 4.03 0.0755 0.53 0.08 1.24 0.5357 1.63 Governance 0.2631 0.8 4.09 2.41 1.0194 1.52 0.1503 0.34 3.7865 1.94 Human capital 0.3354 2.26 2.1749 3.64 0.1016 0.28 0.1634 0.67 2.28 2.92 Finance services 0.1443 1.71 0.1243 0.22 0.8128 2.3 0.0104 0.06 0.6406 0.83 Price rice 0.0104 2.3 0.0561 2.27 0.0154 1.37 0.0031 0.35 0.0793 2.56 Price Coca-Cola 0.0064 0.93 0.0889 1.6 0.0088 0.37 0.0052 0.4 0.0394 0.56 Price kerosene 0.0049 1.28 0.023 0.92 0.0134 1.22 0.0125 1.41 0.0381 1.1 Intercept 0.3773 1.04 11.0703 4.78 2.0168 2.5 0.4588 0.7 11.2461 3.8 N of communities 103 103 103 103 101 R2 0.487 0.41 0.334 0.421 0.258 F 4.43 3.23 3.32 6.23 2.1 p-value 0 0 0 0 0.013 Standard error 0.079 0.48 0.24 0.155 0.626 (continued on the following page) 175 176 Table G.9 Regressions at the Community Level (continued) Labor force Labor force Ratio sales / Ratio NVA / participation rate participation rate Male wage# total cost factor cost b t b t b t b t b t Community population size# 0.0236 0.17 0.0482 1.69 0.0163 0.4 2.945 1.46 34.43 1.89 Share Sinhalese population 0.0034 0.97 0.001 1.79 0.0012 1.37 0.119 1.69 0.88 1.47 Illiteracy 0.0076 1.22 0.0013 1.02 0.0024 1.06 0.061 0.64 1.04 1.32 Proximity to school 0.191 0.47 0.022 0.28 0.0393 0.29 8.987 1.23 70.93 1.13 Proximity to hospital 0.2756 0.58 0.0238 0.21 0.2257 1.28 2.899 0.39 44.93 0.72 Proximity to city 0.0766 0.33 0.008 0.14 0.1102 1.38 1.128 0.36 2.69 0.1 Openness 0.1548 1.33 0.0119 0.69 0.0002 0.01 2.913 1.48 20.25 1.28 Lowland per capita 0.3597 0.66 0.1367 1.72 0.1603 1.09 0.536 0.08 23.16 0.34 Upland per capita 0.344 1.4 0.0607 2.71 0.1726 3.53 1.124 0.4 13.81 0.72 Connectivity 0.2898 0.51 0.01 0.08 0.2423 1.24 23.614 1.87 188.38 1.64 Infrastructure services 0.5713 0.69 0.1792 1.59 0.0122 0.07 21.505 1.65 173.33 1.51 Business services 0.096 0.31 0.1182 2.55 0.1319 1.67 21.117 1.94 148.14 1.75 Governance 4.5566 2.32 0.6151 1.86 0.043 0.07 21.576 0.78 390.36 1.4 Human capital 1.8947 2.57 0.448 2.66 0.58 1.78 6.291 0.6 56.25 0.57 Finance services 0.7399 0.9 0.1622 1.7 0.0752 0.46 20.334 1.57 167.82 1.53 Price rice 0.0589 1.68 0.0073 1.27 0.0301 2.62 1.633 1.81 12.54 1.47 Price Coca-Cola 0.0167 0.21 0.0041 0.39 0.0045 0.31 0.16 0.23 5.81 0.74 Price kerosene 0.0267 0.84 0.0049 0.88 0.0267 2.58 0.647 1.24 6.85 1.63 Intercept 12.2117 3.94 0.0583 0.13 4.7297 5.86 36.941 1.11 302.27 0.9 N of communities 101 103 96 103 101 R2 0.225 0.293 0.45 0.554 0.495 F 1.51 3.01 8.12 0.26 0.47 p-value 0.108 0 0 0.999 0.964 Standard error 0.626 0.107 0.178 9.842 85.452 Gross Value Seasonal Agricultural Added (GVA)# St dev GVA# adjustment seasonality b t b t b t b t Community population size# 0.1179 0.44 0.4663 1.18 0.4713 1.51 0.0425 0.71 Share Sinhalese population 0.0133 2.09 0.0182 1.95 0.0146 2.41 0.0025 2.09 Illiteracy 0.0027 0.23 0.0024 0.16 0.0101 0.98 0.0064 2.54 Proximity to school 0.8479 1.05 1.1159 1 1.2544 1.43 0.1231 0.85 Proximity to hospital 1.4538 1.44 2.8082 2.13 0.4248 0.41 0.3718 1.7 Proximity to city 0.071 0.15 0.4235 0.63 0.1298 0.24 0.3039 3.14 Openness 0.9783 5.53 1.3591 6.14 0.2846 1.73 0.0198 0.64 Lowland per capita 2.0124 1.79 2.336 1.83 0.3681 0.53 0.1994 1.59 Upland per capita 0.5757 1.27 0.6686 1.24 0.0252 0.06 0.045 0.92 Connectivity 1.868 1.64 1.4115 0.88 0.3394 0.26 0.3934 1.79 Infrastructure services 0.8706 0.58 1.7739 0.85 0.7207 0.58 0.0476 0.2 Business services 1.1365 1.92 0.9195 1.23 1.4604 2.29 0.1501 1.16 Governance 11.3944 2.63 16.0493 2.62 3.3118 0.76 3.301 4.33 Human capital 2.1677 1.43 1.7799 0.76 1.4056 0.66 0.9383 2.69 Finance services 3.4432 2.4 4.8838 2.4 3.8692 2.83 0.4954 2.08 Price rice 0.0613 0.97 0.0638 0.74 0.069 1.01 0.0253 2.02 Price Coca-Cola 0.0415 0.3 0.1421 0.7 0.0343 0.27 0.016 0.7 Price kerosene 0.0899 1.38 0.1186 1.42 0.0215 0.39 0.0461 3.75 Intercept 15.758 2.5 13.9641 1.65 17.0432 3.9 1.1687 1.58 N of communities 103 103 103 101 R2 0.585 0.564 0.289 0.596 F 7.53 9.61 1.8 9.24 p-value 0 0 0.039 0 Standard error 1.156 1.603 1.197 0.225 Source: RIC Surveys. Note: # Logarithmic value. (continued on the following page) 177 178 Table G.9 Regressions at the Community Level (continued) B: Reduced set of explanatory variables (removing variables with |t| 0.50) Enterprise Income from Ratio RNF Ratio RNF RNF income density work# income workers per capita# b t b t b t b t b t Community population size# 0.0232 1.14 0.1976 2.14 0.0654 1.96 0.1417 1.03 Share Sinhalese population 0.0008 2.46 0.0036 3.23 0.0042 1.37 Illiteracy 0.0016 1.69 0.003 0.69 0.0021 1.71 0.0034 0.59 Proximity to school 0.1434 2.03 0.1963 0.69 0.1556 1.66 Proximity to hospital 0.2986 3.06 -0.4819 -1.2 0.2173 1.01 0.375 0.94 Proximity to city 0.1039 0.6 0.2844 2.76 0.1199 2.03 0.2442 1.24 Openness 0.0312 2.62 0.1979 2.63 0.0489 1.73 0.215 2.02 Lowland per capita 0.3202 1.59 0.281 1.68 0.7197 1.49 Upland per capita 0.0228 1.54 0.1043 0.88 0.0616 1.21 0.194 0.92 Connectivity 0.1955 2.95 0.8361 1.81 0.2949 1.33 Infrastructure services 1.0952 2.67 0.1849 0.74 0.2464 1.55 0.7354 1.05 Business services 0.146 3.69 0.8849 4.15 0.0349 0.61 0.5003 1.57 Governance 0.243 0.86 3.7898 2.68 0.9992 1.46 3.484 1.95 Human capital 0.3237 2.62 2.2523 4.23 2.1859 2.82 Finance services 0.1458 2.22 0.8022 2.22 0.5812 0.77 Price rice 0.0103 2.35 0.0585 2.49 0.0115 1.09 0.0834 2.8 Price Coca-Cola 0.0065 1 0.0884 1.66 Price kerosene 0.005 1.29 0.02 0.81 0.0128 1.15 0.0146 1.9 0.0367 1.02 Intercept 0.3851 1.17 10.9974 5.4 2.2242 3.14 0.7248 2.36 10.1925 4.7 N of communities 103 103 103 123 101 R2 0.485 0.408 0.325 0.328 0.252 F 5.29 3.76 3.4 7.32 2.22 p-value 0 0 0 0 0.011 Standard error 0.078 0.472 0.234 0.173 0.617 Labor force Labor force Ratio sales / Ratio NVA / participation rate participation rate Male wage# total cost factor cost b t b t b t b t b t Community population size# 0.0438 1.78 3.643 1.72 33.05 1.88 Share Sinhalese population 0.0033 0.97 0.001 2 0.0012 2.02 0.124 1.78 0.9 1.5 Illiteracy 0.0067 1.15 0.0013 1.05 0.063 0.71 0.95 1.27 Proximity to school 8.583 1.24 79.34 1.34 Proximity to hospital 0.2744 0.6 0.177 1.16 49.61 0.84 Proximity to city 0.1355 1.91 Openness 0.1375 1.29 0.013 0.79 2.791 1.63 20.45 1.42 Lowland per capita 0.3791 0.73 0.1354 1.79 0.1756 2.34 Upland per capita 0.3661 1.57 0.0642 3.08 0.163 4.36 Connectivity 0.4177 0.7 0.1985 1.28 23.884 2.06 194.61 1.86 Infrastructure services 0.4966 0.64 0.1776 1.6 20.784 1.68 172.09 1.55 Business services 0.1199 2.74 0.1539 2.31 20.466 2.02 145.48 1.83 Governance 4.4186 2.72 0.6117 2.28 25.002 0.97 344.91 1.39 Human capital 1.8344 2.28 0.4541 2.78 0.6901 2.65 8.856 0.91 63.17 0.7 Finance services 0.6635 0.87 0.1686 1.93 20.084 1.65 171.17 1.63 Price rice 0.0626 1.95 0.0063 1.11 0.0192 1.99 1.616 1.91 13.04 1.75 Price Coca-Cola 5.41 0.7 Price kerosene 0.0261 0.83 0.005 0.88 0.0227 2.74 0.734 1.34 6.57 1.6 Intercept 11.8047 5.61 0.0264 0.09 4.9896 12.37 40.511 1.43 350.93 1.12 N of communities 101 103 99 105 103 R2 0.221 0.289 0.447 0.55 0.494 F 1.99 4.01 16.25 0.36 0.51 p-value 0.03 0 0 0.98 0.928 Standard error 0.61 0.105 0.173 9.38 82.007 (continued on the following page) 179 180 Table G.9 Regressions at the Community Level (continued) Gross Value Seasonal Agricultural Added (GVA)# St dev GVA# adjustment seasonality b t b t b t b t Community population size# 0.4759 1.28 0.4662 1.47 Share Sinhalese population 0.0135 2.2 0.0184 1.94 0.0142 2.38 0.0025 2.24 Illiteracy 0.0111 1.12 0.006 2.28 Proximity to school 0.9567 1.14 1.1068 1.01 1.3255 1.58 0.1545 1.14 Proximity to hospital 1.4444 1.55 2.809 2.14 0.3848 1.7 Proximity to city 0.4091 0.62 0.3033 3.06 Openness 0.9733 6.18 1.3618 6.1 0.2751 1.76 0.0164 0.55 Lowland per capita 1.976 1.77 2.3503 1.85 0.356 0.52 0.2197 1.95 Upland per capita 0.5336 1.26 0.6808 1.27 0.0519 1.04 Connectivity 1.91 1.71 1.4622 0.92 0.418 2.04 Infrastructure services 0.7847 0.55 1.8033 0.87 0.6258 0.55 Business services 1.1788 2.04 0.9036 1.2 1.4742 2.42 0.1536 1.19 Governance 11.4427 2.86 15.9317 2.66 2.9963 0.7 3.2262 4.2 Human capital 2.2286 1.39 1.7947 0.76 1.6403 0.78 0.8808 2.64 Finance services 3.3834 2.58 4.8582 2.44 3.5932 2.94 0.468 1.97 Price rice 0.0509 0.79 0.067 0.78 0.0689 1.05 0.0274 2.02 Price Coca-Cola 0.1451 0.72 0.013 0.58 Price kerosene 0.0907 1.47 0.116 1.46 0.0451 3.76 Intercept 17.1409 4.28 13.873 1.68 16.5634 5.01 1.3894 1.87 N of communities 103 103 103 101 R2 0.583 0.564 0.283 0.596 F 8.74 9.17 2.13 9.24 p-value 0 0 0.022 0 Standard error 1.132 1.593 1.161 0.225 Source: RIC Surveys. Note: # Logarithmic value. Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables MEASUREMENT AND Other than this EICO question, most enterprises have to deal to some extent with government SPECIFICATION ISSUES agencies, for example, to obtain various permits Corruption and services; those who did were asked about pay- ment of informal fees. Information about such Corruption is an investment climate constraint unof�cial fees is incorporated in the infrastructure since it increases the cost of doing business for and services component of the governance bench- producers and traders and may distort incentives. mark indicator (see Annex E). Furthermore, It causes business risk through the ensuing un- responses on the predictability of laws, kickbacks, certainty about the rule of law and the integrity and the manipulability of rent seeking by govern- of the legal system. Consumers may also be af- ment of�cials are examined under the heading of fected through higher prices and inferior quality enterprise investment climate interactions (EICIs) services. in Chapter 5. The RIC enterprise questionnaire aimed at reg- The registration of questions about corruption istering corruption through questions about infor- is sensitive. In general, respondents may be reluc- mal payments for public services (registration, tant to give answers because they may be unsure licenses, permits, inspections) and hooking up to about confidentiality. In Nicaragua, for example, infrastructural services (phone, water, electricity, the survey was conducted before the election, and sewage, gas). In addition, entrepreneur percep- the political fears of the time contributed to low tions were registered about the integrity of the rule responses. But respondents are not the only ones of law. who might worry about the implications of In the three RICS pilot countries, respondents responses to questions about corruption: public were asked whether corruption affected their and private survey agencies may have disincen- businesses. In Nicaragua all interviewees were tives to pursue sensitive questions as well. They directly asked whether corruption was an invest- may fear that their association with a critical report ment climate constraint. In contrast, entrepreneurs may affect future government contracts or involve in Tanzania and Sri Lanka were asked whether political risks. “governance [poses] a problem [for the operation The analysis of corruption and governance, and growth of your establishment],�91 which func- while inconclusive owing to the reluctance of tioned as a screening question for follow-up ques- some respondents, is nonetheless instructive. The tions on corruption, economic policy uncertainty, descriptive �ndings discussed here and in Section crime and theft, the legal system, and war. The 5.1 of Chapter 5 are examined in greater detail by most likely effect of this questionnaire structure means of regression analytical models in Sections was that corruption was relatively underreported. 5.3 and 5.4. Table H.10 provides insights on this issue. In Nicaragua 412 enterprises, or 36 percent of the Computation of Standard Errors total, reported that corruption as a problem. In Sri Lanka only 5 percent and in Tanzania 27 per- The formula that econometric theory advocates for cent did so. the computation of standard errors (labeled “V3� 181 182 The Rural Investment Climate for convenience) is based on asymptotic consid- Effect of Size and Productivity erations, but the number of communities only on EICO Responses ranges between 100 and 150. In the absence of The estimated EICO models omit enterprise size and sampling weights, it is easily shown that the as- productivity as determinants of the entrepreneur's ymptotic formula should simplify to the standard response. But it is easy to argue that size matters. A formula (“V1�) that statistical software always large enterprise may be severely disrupted by poor uses, but actually the asymptotic formula V3 quality of electricity delivery, for example; it may yields estimates of the standard errors that vastly have an inside track to bank loans and other �nancial overstates those computed with the standard for- instruments; it may face more harassment from gov- mula V1. From a limited amount of experimenta- ernment agencies. But if EICO obstacles hinder the tion, it was determined that the use of sampling operation and growth of the enterprise, as is indeed weights typically increases the V1 standard errors the speci�c question posed to entrepreneurs, one in Nicaragua by 10.7 percent for enterprise char- would expect to �nd that a greater obstacle reduces acteristics and 9.7 percent for community char- the size and productivity of the enterprise. In other acteristics and in Sri Lanka by 31.8 percent for words, the direction of causality goes both ways, and enterprise characteristics and 47.4 percent for com- the estimated effect of size and productivity on EICO munity characteristics. The standard error of the responses would be negatively biased. This resulting estimated standard deviation of the random error endogeneity bias could be solved through instru- did not need adjustment in the case of Nicaragua mental variables. Alternatively, the size and produc- but needed adjustment upward by 59 percent in tivity measures may be omitted and a reduced-form Sri Lanka. equation may be estimated-which indeed was The effect of ignoring weights and random the strategy followed here. The drawback of this effects and the difference in using V1 or V3 is approach is that it remains unclear whether the illustrated in Table H.11, showing a typical estimated effect of, say, female entrepreneurship on EICO model for cost of finance in Sri Lanka. the EICO response is direct or manifested indirectly The use of weights propels some explanatory via through size and productivity. variables to greater prominence (for example, To satisfy curiosity, however, Table H.5 reports Sinhalese manager, household-based enterprise, the estimates of the effect of size, as measured by infrastructure services, business services, and the log of sales, and productivity, as measured finance services) and diminishes the measured by the log of the ratio of net value added over total effect of others (female manager). Incorporating factor cost (V/C). The reported values are not a random effect again leads to changes in para- scaled estimates as in the other tables; rather, they meter estimates (industry dummies, income are unscaled estimates of the parameters of the per capita, infrastructure services, and finance weighted random effect ordered probit model. In services). Nicaragua, many of the two effects are statistically The s(V1) standard errors incorporate the signi�cant; in Sri Lanka, the effect is statistically rel- adjustment mentioned above. Among the enter- evant for only one EICO, telecommunication; and prise characteristics, the s(V1) and s(V3) stan- in Tanzania, �ve of the twenty parameter estimates dard errors are not dramatically different, with are statistically signi�cant. It is notable that the size three exceptions (age of enterprise, Sinhalese effect is positive for 23 of the 30 estimates, that the manager, and other production enterprise). But productivity effect is negative for 24 of the 30 esti- major differences are found among all commu- mates, and that the effects have the same sign only nity characteristics: s(V3) standard errors are five times. Taken at face value, this implies that about two and a half times as large as s(V1) operators of larger enterprises tend to complain standard errors, even after the adjustment of more often about investment climate conditions s(V1). T-statistics based on s(V3) collapse: if they and that entrepreneurs overseeing more produc- are to be believed, nothing is statistically signifi- tive businesses tend to view the environment in cant anymore. The adjustment of s(V1) has a less problematic terms. But as mentioned, these degree of arbitrariness, but t-statistics are not effects are probably biased downward: the size implausible. This methodological econometric effect is probably more strongly positive, and the issue must be examined in greater detail at some productivity effect is less negative. point. Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 183 Table H.1 Detailed Responses to EICO Questions, by Country and Category A: Nicaragua (0 No obstacle, . . . , 2 Severe obstacle) Description 0 1 2 Total N Mean Public services electricity 59.97 20.17 19.86 100 1135 0.599 telecommunication 94.93 3.27 1.80 100 1133 0.069 water 78.17 11.12 10.72 100 1135 0.325 postal service 98.83 0.72 0.44 100 1132 0.016 Transport road quality 77.81 11.46 10.72 100 1136 0.329 road block 83.54 8.12 8.34 100 1136 0.248 forms of transport 88.98 7.62 3.40 100 1135 0.144 access to transport facility 87.60 6.89 5.51 100 1135 0.179 Financing interest rate of loan 54.13 15.58 30.28 100 1135 0.762 loan process 68.58 9.46 21.96 100 1119 0.534 access to loans 56.83 14.17 29.01 100 1118 0.722 Commercialization market information 92.39 5.78 1.83 100 1132 0.094 market demand 69.86 15.00 15.14 100 1135 0.453 Permits and licenses obtaining licenses 96.25 2.37 1.38 100 1116 0.051 problems with registry 97.30 1.63 1.07 100 1115 0.038 business permit inef�ciency 97.93 1.38 0.69 100 1114 0.028 Taxes tax rate 89.10 5.39 5.52 100 1113 0.164 bureaucracy in tax collection 95.59 2.69 1.71 100 1105 0.061 Work labor hiring 94.60 4.15 1.24 100 98 0.066 complementary labor cost 97.78 1.11 1.11 100 97 0.033 labor �ring 98.89 1.11 0.00 100 97 0.011 work permit 100.00 0.00 0.00 100 90 0.000 availability of skillful labor 98.07 1.93 0.00 100 96 0.019 Land policy regulation on land use 98.64 0.83 0.54 100 897 0.019 legalizing delay 99.48 0.09 0.43 100 895 0.010 construction permit 99.79 0.21 0.00 100 907 0.002 land collateral 99.06 0.22 0.72 100 898 0.017 Agricultural policy agricultural product movement 98.66 0.86 0.48 100 909 0.018 agricultural cooperative 100.00 0.00 0.00 100 898 0.000 agricultural trade 97.46 2.25 0.29 100 910 0.028 food safety regulation 99.34 0.66 0.00 100 907 0.007 Nonagricultural policies custom rules 99.40 0.11 0.49 100 891 0.011 environmental regulation 99.09 0.91 0.00 100 896 0.009 Governance corruption 58.04 17.47 24.49 100 1127 0.664 economic policy uncertainty 46.47 17.02 36.52 100 1136 0.901 crime, theft, etc. 74.46 14.20 11.34 100 1136 0.369 legal system 94.13 2.67 3.20 100 1134 0.091 (continued on next page) 184 The Rural Investment Climate Table H.1 Detailed Responses to EICO Questions, by Country and Category (continued) B: Sri Lanka (0 No obstacle, . . . , 4 Very severe obstacle) Description 0 1 2 3 4 Total N Mean Public utilities electricity 50.19 11.83 13.22 14.53 10.23 100 1321 1.228 telecommunication 73.80 8.82 9.23 5.88 2.28 100 1280 0.540 water 70.71 9.63 7.45 7.45 4.75 100 1302 0.659 postal service 91.13 5.13 2.41 0.77 0.57 100 1292 0.145 Transportation road access 65.61 9.73 9.24 10.7 4.72 100 1323 0.792 road quality 64.10 10.04 9.08 9.65 7.13 100 1322 0.857 road block 85.69 7.77 3.22 2.88 0.44 100 1296 0.246 access to transport facility 69.82 8.14 6.97 11.94 3.12 100 1305 0.704 Financing interest rate of loan 53.21 5.20 11.56 23.84 6.19 100 1313 1.246 loan procedures 55.48 6.26 11.09 21.91 5.25 100 1314 1.152 availability of loan sources 79.77 7.58 6.23 4.96 1.47 100 1295 0.408 Marketing market information 63.05 13.36 11.78 10.31 1.49 100 1316 0.738 market demand 52.91 6.15 13.11 20.05 7.78 100 1323 1.236 Registration, license and permits cost of registration 95.57 1.18 0.71 1.75 0.80 100 1326 0.110 cost of license/permit 95.17 0.99 1.23 1.62 1.00 100 1324 0.123 registration/license process 96.21 0.94 0.92 1.19 0.74 100 1323 0.093 Taxation tax rate 94.70 0.98 1.54 2.17 0.61 100 1325 0.130 bureaucracy in tax collection 96.61 1.26 1.00 0.75 0.38 100 1323 0.070 Labor flexibility in hiring and �ring 98.05 0.58 0.71 0.57 0.10 100 1323 0.041 government labor policy 98.08 0.42 0.68 0.64 0.18 100 1319 0.044 obtaining work permit for expatriate 99.34 0.32 0.19 0.14 0.00 100 1314 0.011 availability of skillful labor 94.36 0.48 0.75 2.72 1.70 100 1323 0.169 Land regulation on land use 97.85 0.62 0.57 0.76 0.20 100 1324 0.048 obtaining land construction permit 97.37 0.99 0.51 0.50 0.63 100 1321 0.060 land ownership 97.14 0.81 0.26 0.89 0.90 100 1323 0.076 Agricultural policy agricultural subsidy 98.13 0.47 0.25 0.96 0.19 100 1326 0.046 agricultural import/export tariff 99.03 0.09 0.21 0.56 0.11 100 1319 0.026 protection of rice production 99.02 0.19 0.50 0.29 0.00 100 1325 0.020 protection of agricultural price 98.57 0.19 0.23 0.80 0.22 100 1326 0.039 Nonagricultural trade policy import/export regulation 99.15 0.45 0.30 0.07 0.03 100 1326 0.014 custom rules 99.60 0.05 0.26 0.03 0.06 100 1326 0.009 Environmental policy food safety regulation 98.11 0.64 0.35 0.60 0.31 100 1319 0.044 environmental rules 96.36 1.23 1.19 0.48 0.74 100 1325 0.080 Governance corruption 96.98 1.26 1.29 0.33 0.14 100 1318 0.054 economic policy uncertainty 93.57 0.88 1.96 2.10 1.49 100 1326 0.171 crime, theft, etc. 96.61 0.86 1.69 0.64 0.19 100 1323 0.069 legal system 96.71 1.11 0.99 0.59 0.61 100 1318 0.073 war 96.90 1.00 1.01 0.18 0.92 100 1317 0.072 Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 185 Table H.1 (continued) C: Tanzania (0 No obstacle, . . . , 4 Very severe obstacle) Description 0 1 2 3 4 Total N Mean Public utilities electricity access 50.42 6.14 6.40 21.46 15.57 100 1188 1.456 electricity quality 55.75 7.98 8.91 18.00 9.37 100 1078 1.173 telecommunication 58.49 8.49 9.34 17.83 5.86 100 1178 1.041 water 56.98 8.70 7.47 19.29 7.55 100 1218 1.117 postal service 63.87 8.91 7.21 12.55 7.46 100 1179 0.908 Transportation road access 64.58 7.65 7.08 15.07 5.62 100 1228 0.895 road quality 63.46 6.77 7.01 15.82 6.93 100 1226 0.960 road block 85.07 5.68 3.82 3.90 1.53 100 1179 0.311 access to transport facility 69.07 7.30 6.97 11.03 5.64 100 1206 0.769 Financing interest rate of loan 38.46 5.69 6.88 36.76 12.22 100 1178 1.786 loan procedures 36.85 6.20 7.20 35.85 13.90 100 1194 1.838 availability of loan sources 37.75 5.86 7.81 35.96 12.61 100 1229 1.798 Marketing market access 64.94 6.57 6.66 17.86 3.98 100 1232 0.894 market information 66.20 7.47 7.47 15.50 3.36 100 1219 0.824 market demand 66.14 8.81 7.58 15.16 2.31 100 1214 0.787 Registration, license and permits cost of registration 78.18 4.40 3.99 11.56 1.87 100 1228 0.546 cost of license/permit 77.70 3.24 4.70 11.68 2.68 100 1233 0.584 registration/license process 78.37 3.01 4.23 12.03 2.36 100 1230 0.570 Taxation tax rate 78.22 3.00 2.59 12.79 3.40 100 1235 0.602 bureaucracy in tax collection 81.11 4.56 3.40 9.11 1.82 100 1207 0.460 Labor flexibility in hiring and �ring 94.23 1.95 0.81 2.44 0.57 100 1231 0.132 government labor policy 94.47 1.30 1.46 2.44 0.33 100 1230 0.128 obtaining work permit for expatriate 94.63 1.06 1.14 2.52 0.65 100 1228 0.135 availability of skillful labor 93.83 1.30 0.97 3.17 0.73 100 1231 0.157 Land regulation on land use 84.72 4.53 2.51 7.60 0.65 100 1237 0.349 obtaining land construction permit 85.29 2.83 3.40 7.11 1.37 100 1237 0.365 land ownership 84.75 3.49 3.08 7.30 1.38 100 1233 0.371 Agricultural policy agricultural subsidy 83.14 0.89 2.03 10.70 3.24 100 1234 0.500 agricultural import/export tariff 86.95 1.75 3.33 6.32 1.66 100 1203 0.340 protection of rice production 88.89 2.22 3.29 4.94 0.66 100 1215 0.263 protection of agricultural price 84.43 2.85 2.69 7.66 2.36 100 1227 0.407 Nonagricultural trade policy import/export regulation 92.18 0.98 1.55 3.59 1.71 100 1227 0.217 custom rules 92.01 0.90 1.87 3.67 1.55 100 1227 0.218 Environmental policy food safety regulation 93.17 1.73 1.15 3.45 0.49 100 1216 0.164 environmental rules 92.59 1.89 1.89 3.37 0.25 100 1215 0.168 Governance corruption 72.31 2.85 2.52 12.05 10.26 100 1228 0.851 economic policy uncertainty 74.14 6.22 4.83 13.09 1.72 100 1222 0.620 crime, theft, etc. 73.27 4.97 4.65 11.25 5.87 100 1227 0.715 legal system 77.17 7.53 5.54 8.35 1.41 100 1209 0.493 war 86.15 4.65 2.45 4.98 1.77 100 1184 0.316 Source: RIC Surveys. 186 The Rural Investment Climate Table H.2 Descriptive Statistics of Explanatory Variables Nicaragua Sri Lanka Tanzania Meana St. dev.a Meana St. dev.a Meanb St. dev.b Enterprise characteristics Enterprise age 10.739 10.140 9.026 10.426 8.570 9.279 in logarithm 2.085 0.921 1.872 0.931 1.823 0.978 Education of manager 9.549 3.072 7.748 2.704 Female manager 0.250 0.433 0.223 0.417 Sinhalese manager 0.786 0.410 Enterprise sales (US$) 2371 5911 5914 38495 1030 2032 in logarithm 6.912 1.267 7.031 1.630 5.966 1.366 Net value added/total factor cost 2.672 2.564 1.015 1.010 1.211 1.451 in logarithm 0.279 1.657 0.590 1.328 0.529 1.353 Household-based enterprise 0.917 0.275 0.424 0.494 0.610 0.488 Industry dummies Services enterprise 0.193 0.395 0.170 0.376 0.171 0.376 Manufacturing, nonagricultural enterprise 0.085 0.278 0.311 0.463 0.061 0.240 Agricultural processing enterprise 0.138 0.345 0.025 0.155 0.035 0.183 Other production enterprise 0.033 0.180 0.004 0.065 0.024 0.152 Mixed enterprise 0.158 0.365 0.138 0.345 0.200 0.400 Community variables Community population size 21987 30108 1752 1221 4387 3673 in logarithm 8.756 1.870 7.297 0.553 8.050 0.841 Income per capita (US$) 438 236 3666 2367 141 170 in logarithm 5.909 0.649 8.047 0.550 4.518 0.924 Agricultural seasonality 0.685 0.413 0.734 0.303 Share income from agriculture 0.092 0.104 0.153 0.712 0.332 0.302 Benchmark indexes Connectivity 0.356 0.178 0.455 0.165 0.211 0.144 Infrastructure services 0.550 0.239 0.377 0.161 0.195 0.162 Business services 0.305 0.430 0.173 0.241 0.066 0.151 Human capital 0.215 0.075 0.338 0.083 0.213 0.056 Finance services 0.200 0.266 0.535 0.133 0.187 0.143 Number of enterprisesc 1113 869 1178 Source: RIC Surveys. Notes: a. Weighted statistics. b. Unweighted statistics. c. For a few variables the number of observations is less because of missing values. Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 187 Table H.3 Determinants of Top Ten EICO Responses: Models with Benchmark Indicatorsa A: Nicaragua, Part 1 Availability Interest Loan of loan Road rate of loan procedures Electricity sources quality (2) (6) (5) (3) (9) Enterprise characteristics Age of enterpriseb 0.022 0.007 0.028 0.056* 0.003 Household-based enterprise 0.381** 0.255 0.403*** 0.155 0.045 Industry dummies Services enterprise 0.084 0.044 0.080 0.000 0.102 Manufacturing, nonagricultural enterprise 0.022 0.111 0.211 0.025 0.026 Agricultural processing enterprise 0.168 0.232 0.028 0.132 0.099 Other production enterprise 0.221 0.142 0.063 0.181 0.053 Mixed enterprise 0.061 0.016 0.380*** 0.025 0.072 Community variables Community population sizeb 0.033 0.057 0.135*** 0.042 0.091 Income per capitab 0.014 0.066 0.101 0.011 0.012 Agricultural seasonality 0.009 0.115 0.059 0.002 0.074 Share income from agriculture 0.068 0.025 0.022 0.002 0.039 Benchmark indexes Connectivity 0.082 0.004 0.194* 0.037 0.389*** Infrastructure services 0.102 0.139 0.314*** 0.035 0.013 Business services 0.103 0.220** 0.014 0.198** 0.063 Human capital 0.059 0.154 0.095 0.070 0.036 Finance services 0.034 0.099 0.001 0.028 0.034 Regression statistics St.dev of comm. RE 0.515*** 0.576*** 0.421*** 0.495*** 0.439*** lnLikelihood 800.20 612.00 760.99 765.02 540.89 Number of obs 865 853 864 854 865 p (community variables) 0.849 0.491 0.016** 0.936 0.306 p (benchmark indexes) 0.405 0.095* 0.000*** 0.163 0.005*** (continued on next page) 188 The Rural Investment Climate Table H.3 Determinants of Top Ten EICO Responses: Models with Benchmark Indicatorsa (continued) A: Nicaragua, Part 2 Low Economic market policy demand Water uncertainty Corruption Crime (7) (10) (1) (4) (8) Enterprise characteristics Age of enterpriseb 0.054 0.024 0.049 0.041 0.042 Household-based enterprise 0.267 0.324* 0.343** 0.241 0.293** Industry dummies Services enterprise 0.028 0.163 0.151 0.137 0.074 Manufacturing, nonagricultural enterprise 0.252 0.004 0.153 0.179 0.093 Agricultural processing enterprise 0.110 0.074 0.137 0.050 0.032 Other production enterprise 0.124 0.001 0.271 0.311 0.488** Mixed enterprise 0.089 0.240* 0.131 0.112 0.021 Community variables Community population sizeb 0.019 0.035 0.102* 0.065 0.086* Income per capitab 0.026 0.081 0.135 0.039 0.043 Agricultural seasonality 0.011 0.006 0.131 0.054 0.048 Share income from agriculture 0.054 0.016 0.016 0.049 0.117 Benchmark indexes Connectivity 0.130* 0.211** 0.316*** 0.208* 0.302*** Infrastructure services 0.015 0.152 0.279** 0.165 0.286*** Business services 0.005 0.011 0.209* 0.166* 0.192** Human capital 0.194*** 0.019 0.122 0.069 0.038 Finance services 0.028 0.089 0.057 0.041 0.126 Regression statistics St.dev of comm. RE 0.147* 0.314*** 0.575*** 0.576*** 0.489*** lnLikelihood 722.08 568.40 805.06 718.01 492.70 Number of obs 864 864 860 865 865 p(community variables) 0.785 0.735 0.068* 0.762 0.090* p(benchmark indexes) 0.046** 0.005*** 0.005*** 0.012** 0.000*** Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 189 Table H.3 (continued) B: Sri Lanka, Part 1 Interest rate Loan Low market of loan procedures Electricity Road quality demand (1) (4) (3) (5) (2) Enterprise characteristics Age of enterpriseb 0.000 0.054 0.004 0.126** 0.015 Education of manager 0.090 0.004 0.104 0.146** 0.032 Female manager 0.217 0.215 0.361** 0.056 0.053 Sinhalese manager 0.757*** 0.958*** 0.013 0.574** 0.088 Household-based enterprise 0.205 0.168 0.516*** 0.390** 0.205 Industry dummies/line of business Services enterprise 0.462*** 0.519*** 0.615*** 0.134 0.541*** Manufacturing, nonagricultural enterprise 0.321** 0.231 0.989*** 0.323* 0.109 Agricultural processing enterprise 0.494 0.653 0.722* 0.385 0.626 Other production enterprise 0.437 0.406 0.810 0.422 0.779 Mixed enterprise 0.167 0.059 0.412* 0.290 0.290 Community variables Community population sizeb 0.163 0.157 0.082 0.121 0.035 Income per capitab 0.048 0.063 0.159 0.032 0.170 Agricultural seasonality 0.009 0.052 0.069 0.358** 0.271 Share income from agriculture 0.009 0.010 0.093 0.265** 0.045 Benchmark indexes Connectivity 0.064 0.104 0.019 0.299* 0.078 Infrastructure services 0.023 0.053 0.030 0.025 0.063 Business services 0.149 0.035 0.000 0.297** 0.154 Human capital 0.235 0.189 0.276** 0.077 0.017 Finance services 0.137 0.161 0.082 0.342** 0.081 Regression statistics St. dev of comm. RE 0.840*** 0.646*** 0.793*** 0.908*** 0.762*** lnLikelihood 1240.19 1262.36 1426.91 1126.29 1394.49 Number of obs 1104 1108 1110 1110 1111 p(community variables) 0.652 0.369 0.116 0.001*** 0.007*** p(benchmark indexes) 0.051** 0.119 0.030*** 0.000*** 0.600 (continued on next page) 190 The Rural Investment Climate Table H.3 Determinants of Top Ten EICO Responses: Models with Benchmark Indicatorsa (continued) B: Sri Lanka, Part 2 Lack of Availability Easy access Telecommu- market of transport to roads Water nication information facilities (6) (9) (10) (7) (8) Enterprise characteristics Age of enterpriseb 0.090 0.023 0.007 0.046 0.013 Education of manager 0.002 0.022 0.267*** 0.017 0.078 Female manager 0.092 0.443** 0.014 0.050 0.108 Sinhalese manager 0.199 0.074 0.197 0.238 0.290 Household-based enterprise 0.391** 0.212 0.114 0.001 0.264 Industry dummies/line of business Services enterprise 0.293 0.382* 0.050 0.619*** 0.905*** Manufacturing, nonagricultural enterprise 0.170 0.211 0.207 0.008 0.034 Agricultural processing enterprise 0.202 0.904** 0.231 0.393 0.648 Other production enterprise 0.223 0.770 0.565 0.935 0.260 Mixed enterprise 0.291 0.632*** 0.154 0.176 0.074 Community variables Community population sizeb 0.130 0.098 0.152 0.047 0.045 Income per capitab 0.029 0.202* 0.033 0.095 0.169 Agricultural seasonality 0.005 0.208 0.053 0.211** 0.168 Share income from agriculture 0.163 0.072 0.226 0.112 0.092 Benchmark indexes Connectivity 0.192 0.259 0.069 0.029 0.387** Infrastructure services 0.224 0.133 0.320* 0.140 0.021 Business services 0.177 0.162 0.295** 0.246** 0.002 Human capital 0.254 0.115 0.116 0.132 0.002 Finance services 0.392** 0.088 0.008 0.020 0.157 Regression statistics St. dev of comm. RE 0.896*** 0.838*** 0.806*** 0.458*** 0.846*** lnLikelihood 1090.54 1024.65 905.31 1172.45 975.65 Number of obs 1111 1099 1089 1105 1096 p(community variables) 0.296 0.033** 0.389 0.004*** 0.145 p(benchmark indexes) 0.000*** 0.087* 0.004*** 0.007*** 0.003*** Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 191 Table H.3 (continued) C: Tanzania, Part 1 Availability Interest Loan Quality of of loan Road rate of loan procedures electricity sources quality (3) (1) (5) (2) (8) Enterprise characteristics Age of enterpriseb 0.015 0.034 0.005 0.002 0.028 Education of manager 0.052 0.032 0.032 0.043 0.011 Female manager 0.074 0.185* 0.162 0.072 0.143 Household-based enterprise 0.116 0.031 0.062 0.076 0.123 Industry dummies Services enterprise 0.040 0.153 0.191 0.041 0.189 Manufacturing, nonagricultural enterprise 0.155 0.008 0.076 0.176 0.190 Agricultural processing enterprise 0.326 0.220 0.066 0.519** 0.228 Other production enterprise 0.003 0.081 0.143 0.008 0.554 Mixed enterprise 0.263** 0.086 0.192 0.137 0.197 Community variables Community population sizeb 0.071 0.015 0.255*** 0.046 0.049 Income per capitab 0.147 0.257** 0.499*** 0.237** 0.099 Share income from agriculture 0.381*** 0.390*** 0.394*** 0.272*** 0.064 Benchmark indexes Connectivity 0.220** 0.197** 0.103 0.358*** 0.150 Infrastructure services 0.041 0.211 0.410*** 0.046 0.175 Business services 0.049 0.035 0.148 0.063 0.303*** Human capital 0.115 0.082 0.126 0.155 0.319*** Finance services 0.262*** 0.158 0.033 0.254*** 0.004 Regression statistics St. dev of comm. RE 0.988*** 0.963*** 1.329*** 0.974*** 1.008*** lnLikelihood 1283.60 1377.61 1063.97 1361.76 1120.82 Number of obs 1117 1133 1023 1168 1165 p(community variables) 0.000*** 0.000*** 0.000*** 0.001*** 0.697 p(benchmark indexes) 0.000*** 0.000*** 0.004*** 0.000*** 0.000*** (continued on next page) 192 The Rural Investment Climate Table H.3 Determinants of Top Ten EICO Responses: Models with Benchmark Indicatorsa (continued) C: Tanzania, Part 2 Easy access Telecomm- Access to Postal to roads Water unications Electricity service (10) (6) (7) (4) (9) Enterprise characteristics Age of enterpriseb 0.041 0.013 0.040 0.010 0.047 Education of manager 0.011 0.015 0.053 0.002 0.012 Female manager 0.012 0.050 0.009 0.284*** 0.046 Household-based enterprise 0.129 0.075 0.023 0.035 0.001 Industry dummies Services enterprise 0.163 0.005 0.142 0.019 0.254* Manufacturing, nonagricultural enterprise 0.187 0.137 0.266 0.184 0.088 Agricultural processing enterprise 0.142 0.310 0.107 0.065 0.177 Other production enterprise 0.701** 0.031 0.081 0.205 0.037 Mixed enterprise 0.077 0.128 0.120 0.060 0.008 Community variables Community population sizeb 0.075 0.056 0.033 0.014 0.030 Income per capitab 0.237* 0.334** 0.612*** 0.407*** 0.327* Share income from agriculture 0.061 0.005 0.083 0.085 0.057 Benchmark indexes Connectivity 0.017 0.240* 0.027 0.036 0.089 Infrastructure services 0.241** 0.477*** 0.514*** 0.414*** 0.635*** Business services 0.369*** 0.001 0.198** 0.264*** 0.195* Human capital 0.248** 0.034 0.019 0.049 0.002 Finance services 0.028 0.033 0.237** 0.149* 0.094 Regression statistics St. dev of comm. RE 1.056*** 1.117*** 1.177*** 1.505*** 1.216*** lnLikelihood 1101.38 1237.26 1133.80 1181.88 1041.92 Number of obs 1168 1160 1120 1127 1122 p(community variables) 0.265 0.073* 0.001*** 0.016** 0.063* p(benchmark indexes) 0.001*** 0.000*** 0.000*** 0.000*** 0.000*** Source: RIC Surveys. Notes: a. EICO items are sorted by the average rankings across the three countries; the number in parentheses in the column headings is the country-speci�c rank. b. Variable enters the model in logarithmic form. *** Signi�cant at 1 percent; ** Signi�cant at 5 percent; * Signi�cant at 10 percent. Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 193 Table H.4 Determinants of Top Ten EICO Responses: Models with Speci�c Community Characteristicsa A: Interest rate of loan Nicaragua Sri Lanka Tanzania Enterprise characteristics Age of enterpriseb 0.007 0.009 0.014 Education of manager 0.080 0.055 Female manager 0.260* 0.068 Sinhalese manager 0.740*** Household-based enterprise 0.452*** 0.183 0.102 Services enterprise 0.074 0.449*** 0.042 Manufacturing, nonagricultural enterprise 0.017 0.284* 0.043 Agricultural processing enterprise 0.126 0.419 0.419 Other production enterprise 0.220 0.413 0.031 Mixed enterprise 0.084 0.153 0.260** Community characteristicsc Income per capitab 0.158 Share income from agriculture 0.435*** Time to near city 0.056 0.144 0.247** Cost of transportation to city 0.111 0.027 .068 Insurance service 0.082 0.141 0.042 Human capital benchmark 0.068 0.221 0.146 Number of banks 0.195 0.551** 0.215** Number bank services 0.201 0.346* 0.452*** Regression statistics St.dev of comm. RE 0.503*** 0.804*** 1.046*** lnLikelihood 829.21 1291.59 1276.31 Number of obs 893 1146 1117 B: Loan procedures Nicaragua Sri Lanka Tanzania Enterprise characteristics Age of enterpriseb 0.005 0.042 0.040 Education of manager 0.013 0.036 Female manager 0.246 0.145 Sinhalese manager 0.883*** Household-based enterprise 0.302* 0.134 0.023 Services enterprise 0.071 0.496*** 0.122 Manufacturing, nonagricultural enterprise 0.075 0.195 0.002 Agricultural processing enterprise 0.177 0.530 0.312 Other production enterprise 0.318 0.447 0.073 Mixed enterprise 0.080 0.035 0.102 Community characteristicsa Income per capitab 0.290*** Share income from agriculture 0.389*** Time to near city 0.049 0.216 0.152 Cost of transportation to city 0.095 0.104 0.047 Insurance service 0.269* 0.110 0.052 Human capital benchmark 0.130 0.159 0.225** Number of banks 0.126 0.527*** 0.214** Number bank services 0.045 0.353** 0.461*** Regression statistics St. dev of comm. RE 0.623*** 0.651*** 0.972*** lnLikelihood 641.89 1311.99 1372.78 Number of obs 881 1150 1133 (continued on next page) 194 The Rural Investment Climate Table H.4 Determinants of Top Ten EICO Responses: Models with Speci�c Community Characteristicsa (continued) C: Electricity Nicaragua Sri Lanka Tanzaniad Enterprise characteristics Age of enterpriseb 0.019 0.023 0.005 Education of manager 0.132* 0.035 Female manager 0.286* 0.136 Sinhalese manager 0.138 Household-based enterprise 0.390*** 0.501*** 0.052 Services enterprise 0.075 0.570*** 0.159 Manufacturing, nonagricultural enterprise 0.214 1.000*** 0.052 Agricultural processing enterprise 0.069 0.839** 0.043 Other production enterprise 0.125 0.745 0.193 Mixed enterprise 0.375*** 0.469** 0.169 Community characteristicsc Income per capitab 0.415** Share income from agriculture 0.333*** Percent households with electricity 0.268*** 0.300** 0.371*** Availability electricity 0.478*** Information technology 0.066 0.087 0.023 Human capital benchmark 0.164*** 0.199 0.170 Regression statistics St.dev of comm. RE 0.479*** 1.119*** 0.661*** lnLikelihood 792.34 1483.24 1062.10 Number of obs 892 1152 1023 D: Availability of loan sources Nicaragua Sri Lankae Tanzania Enterprise characteristics Age of enterpriseb 0.052 0.003 Education of manager 0.052 Female manager 0.043 Sinhalese manager Household-based enterprise 0.150 0.091 Services enterprise 0.001 0.023 Manufacturing, nonagricultural enterprise 0.027 0.149 Agricultural processing enterprise 0.078 0.577** Other production enterprise 0.168 0.005 Mixed enterprise 0.002 0.175 Community characteristicsc Income per capitab 0.286** Share income from agriculture 0.322*** Time to near city 0.028 0.305*** Cost of transportation to city 0.069 0.076 Insurance service 0.132 0.140 Human capital benchmark 0.106 0.159* Number of banks 0.264 0.135 Number bank services 0.235 0.380*** Regression statistics St.dev of comm. RE 0.481*** 0.935*** lnLikelihood 795.01 1358.78 Number of obs 882 1168 Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 195 Table H.4 (continued ) E: Road quality Nicaragua Sri Lanka Tanzania Enterprise characteristics Age of enterpriseb 0.012 0.109* 0.032 Education of manager 0.174** 0.012 Female manager 0.018 0.166 Sinhalese manager 0.151 Household-based enterprise 0.021 0.415** 0.123 Services enterprise 0.088 0.161 0.089 Manufacturing, nonagricultural enterprise 0.005 0.353** 0.194 Agricultural processing enterprise 0.106 0.354 0.241 Other production enterprise 0.088 0.469 0.560 Mixed enterprise 0.037 0.137 0.191 Community characteristicsc Income per capitab 0.153 Share income from agriculture 0.051 Time to near city 0.331*** 0.362** 0.193** Time to main market 0.061 0.086 0.076 Concrete/asphalt road 0.039 0.252*** Human capital benchmark 0.054 0.007 0.420*** Regression statistics St.dev of comm. RE 0.472*** 0.888*** 1.105*** lnLikelihood 557.59 1186.04 1157.15 Number of obs 893 1152 1191 F: Low market demand Nicaragua Sri Lanka Tanzaniae Enterprise characteristics Age of enterpriseb 0.056* 0.017 Education of manager 0.045 Female manager 0.059 Sinhalese manager 0.015 Household-based enterprise 0.248 0.166 Services enterprise 0.016 0.558*** Manufacturing, nonagricultural enterprise 0.241 0.108 Agricultural processing enterprise 0.118 0.589 Other production enterprise 0.159 0.770 Mixed enterprise 0.088 0.304 Community characteristicsc Time to near city 0.034 0.385** Cost of transportation to city 0.067 0.130 Time to main market 0.022 0.206 Cost of transportation to market 0.037 0.097 Marketing service 0.058 0.007 Human capital benchmark 0.086* 0.053 Regression statistics St.dev of comm. RE 0.133 0.801*** lnLikelihood 749.58 1437.62 Number of obs 892 1153 (continued on next page) 196 The Rural Investment Climate Table H.4 Determinants of Top Ten EICO Responses: Models with Speci�c Community Characteristicsa (continued) G: Road access Nicaraguaf Sri Lanka Tanzania Enterprise characteristics Age of enterpriseb 0.091* 0.030 Education of manager 0.008 0.015 Female manager 0.075 0.022 Sinhalese manager 0.155 Household-based enterprise 0.376** 0.115 Services enterprise 0.340* 0.161 Manufacturing, nonagricultural enterprise 0.156 0.168 Agricultural processing enterprise 0.144 0.060 Other production enterprise 0.283 0.604* Mixed enterprise 0.310 0.093 Community characteristicsc Income per capitab 0.210* Share income from agriculture 0.081 Time to near city 0.147 0.231*** Time to main market 0.193 0.003 Concrete/asphalt road 0.056 0.194** Human capital benchmark 0.001 0.453*** Regression statistics St. dev of comm. RE 0.918*** 1.137*** lnLikelihood 1138.09 1124.01 Number of obs 1152 1194 H: Water Nicaragua Sri Lanka Tanzania Enterprise characteristics Age of enterpriseb 0.040 0.025 0.008 Education of manager 0.039 0.009 Female manager 0.504*** 0.023 Sinhalese manager 0.142 Household-based enterprise 0.256 0.237 0.091 Services enterprise 0.167 0.366* 0.045 Manufacturing, nonagricultural enterprise 0.068 0.235 0.031 Agricultural processing enterprise 0.125 0.908** 0.156 Other production enterprise 0.076 0.565 0.048 Mixed enterprise 0.254* 0.610*** 0.079 Community characteristicsc Income per capitab 0.375*** Share income from agriculture 0.070 Time to main market 0.105 0.142 0.231** Percent households with protected water 0.094* 0.021 0.231*** Sewage system 0.210* 0.353 Human capital benchmark 0.011 0.251* 0.253*** Regression statistics St. dev of comm. RE 0.347*** 0.967*** 1.262*** lnLikelihood 593.01 1047.70 1259.48 Number of obs 892 1141 1185 Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 197 Table H.4 (continued ) I: Telecommunication Nicaraguae Sri Lanka Tanzania Enterprise characteristics Age of enterpriseb 0.005 0.033 Education of manager 0.250 *** 0.074 Female manager 0.080 0.029 Sinhalese manager 0.442 Household-based enterprise 0.122 0.002 Services enterprise 0.063 0.128 Manufacturing, nonagricultural enterprise 0.251 0.227 Agricultural processing enterprise 0.188 0.023 Other production enterprise 0.539 0.074 Mixed enterprise 0.095 0.083 Community characteristicsc Income per capitab 0.569*** Share income from agricultural 0.215** Percent households with �xed phone line 0.569*** 0.545*** Percent households with mobile phone 0.216 0.189 Availability electricity 0.074 0.572*** Information technology 0.168 0.098 Human-capital benchmark 0.008 0.007 Regression statistics St. dev of comm. RE 0.907*** 1.143*** lnLikelihood 935.32 1126.64 Number of obs 1130 1120 J: Other EICO variables speci�c to Nicaragua Economic policy Crime, uncertainty Corruption theft, etc. Enterprise characteristics Age of enterpriseb 0.052* 0.049 0.047 Household-based enterprise 0.364** 0.239 0.267* Services enterprise 0.122 0.104 0.073 Manufacturing, nonagricultural enterprise 0.126 0.195 0.101 Agricultural processing enterprise 0.138 0.046 0.013 Other production enterprise 0.220 0.322 0.485** Mixed enterprise 0.185 0.157 0.033 Community characteristicsc Population sizeb 0.029 Income per capitab 0.096 Time to near city 0.089 0.104 0.222*** Time to main market 0.006 Percent households with �xed phone line 0.143 0.045 Percent households with mobile phone 0.112 0.064 Business services benchmark 0.142 0.163* Human capital benchmark 0.211** 0.047 0.002 Number of banks 0.115 Number bank services 0.171** 0.183** Regression statistics St. dev of comm. RE 0.556*** 0.591*** 0.556*** lnLikelihood 832.30 741.45 501.23 Number of obs 893 888 865 (continued on next page) 198 The Rural Investment Climate Table H.4 Determinants of Top Ten EICO Responses: Models with Speci�c Community Characteristicsa (continued) K: Other EICO variables speci�c to Sri Lanka Market information Access to transport facility Enterprise characteristics Age of enterpriseb 0.060 0.007 Education of manager 0.009 0.104 Female manager 0.041 0.121 Sinhalese manager 0.216 0.210 Household-based enterprise 0.004 0.269* Services enterprise 0.602*** 0.906*** Manufacturing, nonagricultural enterprise 0.005 0.082 Agricultural processing enterprise 0.370 0.620 Other production enterprise 0.974 0.246 Mixed enterprise 0.179 0.060 Community characteristicsc Connectivity benchmark 0.341** Time to main market 0.117 Cost of transportation to market 0.061 Percent households with �xed phone line 0.043 Percent households with mobile phone 0.021 Concrete/asphalt road 0.057 Marketing service 0.091 Information technology 0.034 Human capital benchmark 0.073 0.044 Regression statistics St. dev of comm. RE 0.580*** 0.853*** lnLikelihood 1225.11 1023.88 Number of obs 1147 1137 L: Other EICO variables speci�c to Tanzania Electricity access Postal service Enterprise characteristics Age of enterpriseb 0.001 0.033 Education of manager 0.017 0.021 Female manager -0.318*** 0.175 Household-based enterprise 0.032 0.074 Services enterprise 0.068 0.145 Manuacturing,f nonagricultural enterprise 0.246 0.054 Agricultural processing enterprise 0.092 0.253 Other production enterprise 0.273 0.122 Mixed enterprise 0.106 0.001 Community characteristicsc Income per capitab 0.561*** 0.334*** Share income from agricultural 0.166* 0.006 Distance to post of�ce 0.412*** Percent households with electricity 0.167 Availability electricity 0.875*** Concrete/asphalt road 0.026 Information technology 0.001 Human capital benchmark 0.217* 0.290*** Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 199 Table H.4 (continued ) L: Other EICO variables speci�c to Tanzania Electricity access Postal service Regression statistics St.dev of comm. RE 1.388*** 1.200*** lnLikelihood 1165.04 1068.57 Number of obs 1127 1146 Source: RIC Surveys. Notes: a. Estimates represent the effect of a one-standard-deviation change in a continuous variable and a 0/1 change in a dummy variable on the tendency to report obstacles. For an average enterprise, this tendency is scaled on a range of 2 for Nicaragua and a range of 4 for Sri Lanka and Tanzania. b. Variable is entered in logarithmic form. c. With the exception of “Income per capita� and “Share of income from agriculture,� all community characteristics are coded such that a higher value implies a presumed better investment climate. d. For Tanzania, this measure refers to quality of electricity, whereas access to electricity is covered elsewhere. For Nicaragua and Sri Lanka, the questionnaire was not speci�c about the obstacles that electricity provision might pose to the entrepreneur. e. This variable was not in the top-ten list of most important obstacles. f. This variable was not included in the questionnaire used in Nicaragua. *** Signi�cant at 1 percent; ** Signi�cant at 5 percent; * Signi�cant at 10 percent. Table H.5 Effect of Enterprise Size and Productivity on EICO Responsesab Nicaragua Sri Lanka Tanzania EICO Variable ln (Sales) ln (V/C) ln (Sales) ln (V/C) ln (Sales) ln (V/C) Interest rate of loan 0.187*** 0.075** 0.048 0.060 0.081 0.030 Loan procedures 0.149*** 0.006 0.001 0.010 0.070 0.024 Electricity 0.104** 0.098*** 0.073 0.021 0.164* 0.165** Availability of loan sources 0.124*** 0.015 n.e. n.e. 0.066 0.007 Road quality 0.203*** 0.079* 0.008 0.006 0.059 0.027 Low market demand 0.222*** 0.083** 0.065 0.041 n.e. n.e. Road access n.a. n.a. 0.085 0.034 0.116* 0.084 Water 0.000 0.077* 0.095 0.060 0.013 0.016 Economic policy uncertainty 0.109** 0.020 n.e. n.e. n.e. n.e. Telecommunication n.e. n.e. 0.169*** 0.102** 0.008 0.054 Corruption 0.212*** 0.040 n.e. n.e. n.e. n.e. Crime 0.175*** 0.066 n.e. n.e. n.e. n.e. Market information n.e. n.e. 0.052 0.089 n.e. n.e. Access to trans facility n.e. n.e. 0.063 0.052 n.e. n.e. Electricity access n.e. n.e. n.e. n.e. 0.123* 0.068 Postal service n.e. n.e. n.e. n.e. 0.035 0.121** Source: RIC Surveys. Notes: a. Parameters reported here are estimates of the weighted random effect ordered probit model, not transformed according to the scale of measurement of the EICO (0/2 for Nicaragua and 0/4 for Sri Lanka and Tanzania). Thus, the magnitudes represent a 1-unit change in the log-sales and log-value added on the unscaled tendency to report obstacles. b. ln (V/C) denotes the logarithm of net value added over total factor cost and measures enterprise productivity. *** Signi�cant at 1 percent; ** Signi�cant at 5 percent; * Signi�cant at 10 percent. 200 The Rural Investment Climate Table H.6 Effect of Enterprise and Community Characteristics on Total Investment Climate Burden, by Countrya Nicaragua Sri Lanka Tanzania EICO EICO EICO EICO EICO EICO Burden Obstacle Burden Obstacle Burden Obstacle b t b t b t b t b t b t Enterprise characteristics Age of enterpriseb 0.102 0.119 0.158 0.054 0.049 0.060 Education of manager 0.161 0.024 0.308* 0.114** Female manager 1.960*** 0.608*** 0.929 0.212 Sinhalese manager 2.665** 0.680** Household-based enterprise 2.083*** 0.481 1.225* 0.205 0.249 0.172 Industry dummies Services enterprise 0.148 0.016 2.369*** 0.552** 3.780*** 1.077*** Manufacturing, nonagricultural enterprise 0.502 0.277 2.667*** 0.589*** 1.376 0.505 Agricultural processing enterprise 0.887 0.418 0.325 0.337 3.293 0.867 Other production enterprise 0.762 0.322 8.076*** 1.931*** 0.797 0.190 Mixed enterprise 0.522 0.220 2.380* 0.535 0.733 0.079 Community variables Community population sizeb 0.141 0.061 0.613 0.146 0.533 0.176 Income per capitab 0.470 0.289 0.825 0.356 1.588 0.380 Agricultural seasonality 0.647 0.288 1.232 0.601 Share income from agriculture 0.037 0.689 0.343 0.007 2.676 0.459 Benchmark indicators Connectivity 8.202*** 2.885** 1.336 0.059 5.425 1.641 Infrastructure services 1.828 0.331 0.239 0.130 1.351 0.070 Business services 2.134** 0.886** 0.845 0.215 13.632 4.114 Human capital 4.988 3.332 1.630 1.199 73.193** 18.765** Finance services 1.681 0.603 5.558 1.799 9.113 1.966 Intercept 7.460** 2.012 15.877 4.345 33.090** 7.645** Regression statistics Average of dependent variable 6.682 2.368 11.572 2.353 25.605 6.297 St. dev of comm. RE 2.271 1.055 4.178 1.090 14.840 3.911 St. dev of enterprise error 4.369 2.048 6.710 1.810 13.282 3.902 R-squared 0.108 0.066 0.089 0.095 0.081 0.073 Number of observations 868 868 1113 1113 1177 1177 Signi�cance of benchmark indicatorsc 0.004*** 0.033** 0.716 0.731 0.030** 0.024** Source: RIC Surveys. Notes: a. Weighted random effect regression results. b. Variable is entered in logarithmic form. c. p-value of a joint test that none of the benchmark indicators have an effect on the dependent variable. *** Signi�cant at 1 percent; ** Signi�cant at 5 percent; * Signi�cant at 10 percent. Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 201 Table H.7 Opinions About Predictability and Manipulability of Laws Laws are unpredictablea Laws cannot be manipulatedb Nicaragua Sri Lanka Tanzania Sri Lanka Tanzania Enterprise characteristics Age of enterprisec 0.75 2.09 0.84 0.002 0.015 Education of manager 5.32*** 0.71 0.009 0.010 Female manager 2.97 6.50* 0.014 0.135 Sinhalese manager 3.25 0.551*** Household-based enterprise 0.08 4.24 0.199* 0.093 Services enterprise 0.09 1.80 0.66 0.092 0.009 Manufacturing, nonagricultural enterprise 4.07 3.42 9.82 0.047 0.039 Agricultural processing enterprise 2.94 11.12 3.47 0.210 0.102 Other production enterprise 13.44 24.00 3.03 0.280 0.245 Mixed enterprise 1.95 4.57 3.65 0.087 0.161 Community & benchmark Community population sizec 5.08 1.15 1.49 0.052 0.011 Income per capitac 1.16 4.44 5.28 0.028 0.127 Connectivity 7.81** 1.46 0.98 0.112 0.006 Business services 5.28 0.13 8.61*** 0.048 0.040 Human capital 0.59 0.05 10.94*** 0.038 0.119 Regression statistics St. dev of comm. RE 18.72*** 19.32*** 22.42*** 0.508*** 0.601*** Log-Likelihood 556.58 1546.31 1177.88 658.13 1164.44 Number of observations 868 1154 714 609 999 Source: RIC Surveys. Notes: a. The estimates indicate the change (expressed in percentage points) in the probability that entrepreneur �nd laws to be unpredictable, as a result of a one-standard-deviation change (continuous variable) or unit change (dummy variable) in the explanatory variable. The model is estimated by random effect weighted probit in Nicaragua and random effects weighted ordered probit in Sri Lanka and Tanzania. For Sri Lanka and Tanzania, the computations for this table aggregate the “fairly,� “highly,� and “completely,� categories that are distinguished in Table 5.2. b. The statement posed to the entrepreneur is “Laws and regulations can be misinterpreted or manipulated.� Responses are coded 0 Strongly agree, 1 Agree, 2 Disagree, 3 Strongly disagree. Estimates indicate the effect of a one-standard-deviation change (continuous variable) or unit change (dummy variable) effect on the tendency to disagree, as estimated with a random effect ordered probit model; the effect is scaled on the range 0–3 for a typical enterprise. c. Variable enters the model in logarithmic form. *** Signi�cant at 1 percent; ** Signi�cant at 5 percent; * Signi�cant at 10 percent. 202 The Rural Investment Climate Table H.8 Effect on Business of Kickbacks, by Target: Nicaragua Laws Enforcement Judge Credit Politician Enterprise characteristics Age of enterpriseb 0.015 0.004 0.018 0.045 0.049 Services enterprise 0.111 0.186 0.297* 0.184 0.331* Manufacturing, nonagricultural enterprise 0.410** 0.217 0.075 0.067 0.121 Agricultural processing enterprise 0.163 0.047 0.008 0.002 0.059 Other production enterprise 0.378 0.257 0.317 0.369 0.225 Mixed enterprise 0.211 0.239 0.163 0.287 0.392** Community & benchmark Community population sizeb 0.129*** 0.113** 0.128*** 0.144*** 0.180*** Income per capitab 0.154 0.095 0.167* 0.112 0.063 Connectivity 0.234** 0.230** 0.243** 0.205* 0.296** Business services 0.256*** 0.231*** 0.225*** 0.059 0.210** Human capital 0.061 0.022 0.041 0.027 0.029 Regression statistics St. dev of comm. RE 0.338*** 0.350*** 0.296*** 0.295*** 0.439*** Log-Likelihood 315.98 338.21 318.08 252.97 271.06 Number of observations 701 699 693 681 677 Source: RIC Surveys. Notes: a. EICIs are coded 0 No impact, 1 Little impact, 2 Lot of impact. The estimates indicate the change (expressed in percentage points) in the tendency to believe that kickbacks have an impact, as a result of a one-standard-deviation change (continuous variable) or unit change (dummy variable) in the explanatory variable. The model is estimated by weighted random effect ordered probit. b. Variable enters the model in logarithmic form. *** Signi�cant at 1 percent; ** Signi�cant at 5 percent; * Signi�cant at 10 percent. Table H.9 Determinants of Perceptions About the Legal Systema Reputation is not importantb Contracts do not protectc Legal system does not supportd Nic. S.L. Tanz. Nic. S.L. Tanz. Nic. S.L. Tanz. Enterprise characteristics Age of enterprised 0.053* 0.009 0.014 0.055* 0.046 0.009 0.078** 0.002 0.014 Education of manager 0.004 0.012 0.028 0.010 0.063 0.004 Female manager 0.077 0.133* 0.073 0.035 0.152 0.006 Sinhalese manager 0.107 0.125 0.232 Household-based enterprise 0.000 0.031 0.000 0.077 0.075 0.005 Services enterprise 0.124 0.023 0.020 0.046 0.037 0.060 0.022 0.042 0.087 Manufacturing, nonagricultural enterprise 0.162 0.179* 0.056 0.397** 0.164 0.109 0.451** 0.262** 0.077 Agricultural processing enterprise 0.167 0.264 0.022 0.058 0.298 0.018 0.178 0.122 0.330** Other production enterprise 0.382 0.248 0.010 0.048 0.109 0.072 0.376 0.154 0.173 Mixed enterprise 0.199 0.131 0.104 0.083 0.073 0.040 0.040 0.048 0.020 Annex H. Enterprise Investment Climate Outcomes and Interactions (EICOs and EICIs): Notes and Tables 203 Table H.9 (continued ) Reputation is not importantb Contracts do not protectc Legal system does not supportd Nic. S.L. Tanz. Nic. S.L. Tanz. Nic. S.L. Tanz. Community & benchmark Community population sized 0.055 0.126* 0.111** 0.112*** 0.073 0.022 0.031 0.202*** 0.051 Income per capitad 0.053 0.018 0.045 0.060 0.060 0.133 0.035 0.031 0.110 Connectivity 0.056 0.031 0.074 0.095 0.023 0.093* 0.137* 0.035 0.011 Business services 0.027 0.120 0.006 0.004 0.025 0.033 0.075 0.112 0.003 Human capital 0.018 0.047 0.066 0.011 0.041 0.090 0.058 0.000 0.063 Regression statistics St. dev of comm. RE 0.368*** 0.428*** 0.706*** 0.462*** 0.594*** 0.680*** 0.356*** 0.534*** 0.705*** lnLikelihood 828.0 749.1 1192.1 820.0 747.8 1150.2 842.5 676.0 1191.5 Number of obs 845 892 1134 846 830 1133 829 716 1136 Source: RIC Surveys. Notes: a. Parameters are estimated with a weighted random effect ordered probit model. Estimates represent the effect of a one-standard- deviation (continuous variable) or one-unit (dummy variable) change on the tendency to agree with the statement at the top of the column-or, semantically more accurately, to disagree with the opposite, i.e., that reputation is important, contracts do protect, and legal system uphold the entrepreneur's contract and property rights. Nicaragua values are scaled on a range of 0–2; Sri Lanka and Tanzania results are scaled on a range of 0–3. b. The dependent variable reflects disagreement with the statement “I must rely on the reputation of those I enter into agreement with.� c. The dependent variable reflects disagreement with the statement “A contract will protect me from being cheated.� d. The dependent variable reflects disagreement with the statement “The legal system will uphold my contract and property rights in business disputes.� e. Variable enters the model in logarithmic form. *** Signi�cant at 1 percent; ** Signi�cant at 5 percent; * Signi�cant at 10 percent. Table H.10 Corruption as an Investment Climate Constraint Nicaragua Sri Lanka Tanzania number number number number number number Total enterprises 1174 100 — 1365 100 — 1238 100 — No response 80 7 — — 0 — 2 0 — Governance does not pose a problema — — — 1217 89 — 864 70 — Total direct response 1094 93 100 148 11 100 372 30 100 of which: No obstacle 671 57 61 73 5 45 24 2 6 Minor 171 15 16 53 4 34 66 5 18 Major 241 21 22 14 1 16 274 22 74 Don't know / NA 11 1 1 8 1 5 8 1 2 Source: RIC Surveys. Note: a. A screening question in Sri Lanka and Tanzania that reduces the total number of responses to the actual question about corruption, resulting in low numbers reporting “no obstacle.� 204 The Rural Investment Climate Table H.11 Alternative Ways of Estimating EICO Models: Interest Rate of Loans in Sri Lankaa Standard ordered Weighted ordered Weighted random effect pzrobit probit ordered probit b s t b s t b s(V1) t(V1) s(V3) t(V3) Age of enterpriseb 0.030 0.036 0.83 0.034 0.045 0.76 0.000 0.063 0.00 0.096 0.00 Education of manager 0.018 0.012 1.47 0.010 0.015 0.70 0.025 0.019 1.34 0.025 1.00 Female manager 0.285 0.086 3.31 0.168 0.100 1.68 0.184 0.129 1.43 0.116 1.58 Sinhalese manager 0.369 0.113 3.27 0.860 0.165 5.20 0.643 0.235 2.74 0.524 1.23 Household-based enterprise 0.043 0.079 0.55 0.165 0.110 1.50 0.174 0.130 1.34 0.134 1.29 Industry dummies Services enterprise 0.234 0.105 2.23 0.351 0.137 2.57 0.392 0.153 2.57 0.159 2.47 Manufacturing nonagricultural enterprise 0.058 0.089 0.65 0.172 0.120 1.44 0.273 0.136 2.01 0.128 2.13 Agricultural processing enterprise 0.239 0.203 1.18 0.395 0.209 1.89 0.420 0.348 1.20 0.309 1.36 Other production enterprise 0.094 0.288 0.33 0.034 0.220 0.16 0.371 0.724 0.51 0.300 1.24 Mixed enterprise 0.037 0.126 0.30 0.155 0.160 0.97 0.142 0.204 0.69 0.176 0.81 Community variables Community population sizeb 0.226 0.070 3.21 0.286 0.094 3.05 0.257 0.229 1.13 0.353 0.73 Income per capitab 0.102 0.071 1.43 0.140 0.085 1.64 0.082 0.238 0.34 0.438 0.19 Agricultural seasonality 0.045 0.127 0.35 0.216 0.171 1.26 0.026 0.475 0.05 1.408 0.02 Share income from agriculture 0.016 0.061 0.27 0.033 0.068 0.48 0.013 0.128 0.10 0.248 0.05 Benchmark indexes Connectivity 0.352 0.275 1.28 0.471 0.348 1.36 0.327 0.586 0.56 1.478 0.22 Infrastructure services 0.064 0.325 0.20 0.216 0.417 0.52 0.123 0.903 0.14 3.922 0.03 Business services 0.168 0.154 1.09 0.373 0.211 1.77 0.529 0.296 1.79 0.637 0.83 Human capital 2.315 0.494 4.69 2.124 0.661 3.21 2.414 1.082 2.23 3.639 0.66 Finance services 0.198 0.296 0.67 0.442 0.393 1.13 0.785 0.596 1.32 1.473 0.53 Source: RIC Surveys. Notes: a. Ordered probit parameter estimates, not further scaled. b. Variable enters the model in logarithmic form. Annex I. Econometric Analysis of RIC Survey Data I.1 INTRODUCTION by i, with i 1, . . . , Nj, where the number of obser- vations per community may vary. For shorthand The Rural Investment Climate Surveys gather in- notation, define n = g j = 1Nj. The term individual J formation about enterprises and households lo- is used loosely here to indicate any kind of eco- cated in communities. Outcome variables at the nomic agent within the community: a household, enterprise or household level are examined in the an enterprise, a person, or an organization such as light of determinants speci�c to the individual level a bank, cooperative, or politician. The sampling (enterprise and/or household) as well as factors at weight associated with this observation is denoted play within the community. This has implications as wji. for the econometric approach that should be fol- Consider a dependent variable yji that describes lowed. In particular, acknowledging that observed an outcome for individual i in community j. community variables determine the outcome of in- Explanatory variables are a vector Xji consisting terest invites the conclusion that some unobserv- of variables that vary with i and perhaps also with able factors may also matter. It has been shown that j, and a vector Zj that are community-specific the presence of such factors can seriously bias the and are the same for all observations within statistical inference drawn from basic statistical that community. Random influences consist of a analyses. The standard solution is to compute the community-level component mj and an individual standard errors of the estimated parameters with component nji. Altogether, the empirical model is an adjustment for clustering at the community written as: level, but this approach is not free of critical as- sumptions either. Alternative methods of analysis yji = Xji b + Z¿ g + mj + nji ¿ j (1) borrow from panel-econometric techniques, but We make the following distributional assumptions these usually do not accommodate sampling about the components of the disturbance: mj is in- weights. The following paragraphs outline the dependently and identically distributed with mean proper methods to follow for parameter estimation 0 and variance s2 across communities; nji is inde- m in the context of data samples such as those gener- pendently and identically distributed with mean 0 ated with RIC surveys. and variance s2 across all individuals; and mj and m In principle, a �xed effects regression is a stan- nji are independent of each other. For some estima- dard procedure, but weights are typically ignored tion approaches, we must also assume that mj and in the econometric literature because panel data nji are independent of X and Z. For future reference, pertain to repeated observations; at most, one de�ne u as would assign weights to the whole time series of a particular observation, not to each separate time uji = mj + nji (2) period. This situation is different in a clustered cross-section such as the RIC sample. Many of the The estimation issues for this model are outlined techniques of panel econometrics apply, but each in the following sections. Section 2 considers the individual has (or may have) its own weight. case where the dependent variable y is continuous, Let communities be denoted by index j, with and Section 3 addresses the case where y is j 1, . . . , J, and individuals within the community discrete. 205 206 The Rural Investment Climate I.2 ESTIMATION APPROACHES and the regression model becomes FOR CONTINUOUS y = Xb + Zmm* + n (5) DEPENDENT VARIABLES Let W denote the matrix with weights on the diagonal. With sampling weights, the �xed effects I.2.1 Ordinary Least Squares estimates are found by minimizing the weighted The simplest estimation approach is OLS. The pa- residual sum of squares: rameter estimates are unbiased and consistent, but min they are not ef�cient relative to other estimation N N (N n) { b, m*} n¿W N (6) methods. The usual standard errors of the parame- Usually, a �xed effects model is not estimated in ter estimates are computed under the assumption this way, because the dummy variables identify- of independent disturbances, but in this model the ing each group (community) are too numerous disturbance (uji) of individuals within a given com- and because a simple model transformation elimi- munity are clearly correlated: they share mj. Thus, nates the �xed effects. To investigate the effect of the standard errors must be adjusted for this clus- the weights, however, it is still useful to develop tering of individuals by community. This is the �xed effects estimator in this way. Solving the straightforward to do in Stata; see Section A1 of the �rst order conditions of the maximization process Appendix. of equation (6), the estimator is written as An important assumption made by the OLS N X¿WX X¿WZm - 1 X¿Wy a b = a b a b technique is the independence between the b (7) explanatory variables and the components of the N m* Z¿ WX Z¿ WZm m m Z¿ Wy m disturbance term. This assumption is usually From this, using the Frisch-Waugh-Lovell made in applied econometric analysis, of course, Theorem, it follows that but in this case the community variables may be inadequately measured where at the same time Zj N b FE = (X¿(W - WZm(Z¿ WZm)-1Z¿ W)X)-1 m m represents only some of the relevant variables at * X¿(W - WZm(Z¿ WZm)-1 Z¿ W)y m m the community level. In other words, one may ' ' ' ' '' make a case that Zj and mj are often correlated. In = (X ¿Q X ) - 1X ¿Q y (8) that case, the OLS estimator loses its property of ' ' ' ' ' being unbiased and consistent. where Q = (In - Zm(Zm Z m) - 1Z m) is an idempo- ¿ ¿ tent matrix that creates deviations of y = W 0.5y ' from the within-community weighted average I.2.2 Fixed Effects Estimation multiplied by the square root of the weight, which OLS does not take into consideration the distur- in matrix notation equals W 0.5Zm(Z¿ WZm) - 1Z¿ Wy, m m ' bance structure of the model. As a result, the para- and similar for X = W0.5X. In other words, the meter estimates are inef�cient. One alternative �xed effects estimator with weighted data is found estimation approach is the �xed effects method; an- by (i) de�ning the data transformation y ji = * notated Stata code is provided in Appendix A.2. To w0.5(yji - yji), and a similar method is followed for ji I understand this method better, order the observa- the explanatory variables, where the symbol ˘ de- tions by community and write out the model in ma- notes a within-community weighted average, and trix form: (ii) estimate a regression of y* on X* by OLS, with- out an intercept. The �xed effects may be derived y = Xb + Zg + Zmm + n (3) from equation (7) as well. Alternatively, going back where Zm is a n J matrix of ones and zeroes link- to equation (6), m* follows from the minimizing of Nj ing the community factors mj to each individual. n¿j Wj nj , which is the portion of n¿WN contributed by N N N n One might write Zm = diag(i Nj), where i Nj is a vec- community j, conditional on b: N tor of ones of length Nj. Because Z varies only by Nj -1 Nj community, Z is perfectly correlated with Zm. Thus, m* = a a wji b Nj a a wji(yji - X¿ b)b ji N to estimate this model with the �xed effects i=1 i=1 method, it is necessary to absorb Z into the com- I I N = yj - Xj¿ b (9) munity component: where the symbol ˘ denotes a community-level m* = Z¿ g + mj j j (4) weighted average. Annex I. Econometric Analysis of RIC Survey Data 207 To estimate s2, note that the transformation ' n jth row contains the average of X in community j; matrix Q W0.5 sweeps the community fixed effect moreover, the second and third matrices of (13) out of the equation. Thus consider the inner prod- have non-zero off-diagonal elements as well.) uct of the residuals of the transformed model: To implement a GLS estimator of equation (10), the variance of mj must �rst be estimated. Let Zc be E[n*¿n*] = E[n¿W0.5QQW0.5n] the matrix that combines Zj at the community level = tr(W0.5 QQW0.5E[nn¿]) (one row per community, as opposed to repeated rows as in the matrix Z). Then s2 may be derived Nm = s2 tr(QW) n (10) from the sum of squared OLS residuals of equa- = s2 n 0.5 tr(W - W Zm(Z¿ WZm) Z¿ W W) m m -1 0.5 tion (12), corrected for the part of the variance that comes from (mj* - m*): N j = s2 (tr(W) - tr((Z¿ WZm) - 1(Z¿ W2Zm))) n m m J Nj J Nj Nj s2 = (m*¿(I - Zc(Z¿c Zc)-1Zc)m* Nm N N = s2 a n a a wji - a a a w2 n a wji bb ji - tr(VN (m* - m*)))>J ar N (14) j=1 i=1 j=1 j=1 i=1 Then, de�ne Therefore an estimator for s2 is found as: n N Var(h) K © h = s2 IJ + VN (m* - m*) N Nm ar N (15) J Nj J Nj Nj s2 = n*¿n*n a a a wji - a a a w2 n a wji b b (11) Nn N N ji and the feasible GLS estimator is obtained: j=1 i=1 j=1 i=1 i=1 N -1 N -1 N gFGLS = (Z¿ © h Zc)-1Z¿ © h m* N (16) c c This does not yet yield estimates of g, of course, which represents the effect of community variables N -1 with a variance equal to Var(gFGLS) = (Z¿ © h Zc)-1. N c Zj on individual outcomes yji. For this, connect the As a �nal comment, it should be noted that the estimated �xed effects with the community error fixed effects estimator allows correlation of the component mj using equation (4): community-error component with X and Z (although the correlation with Z becomes irrele- m* = Zjg + mj + ( m* - m*) = Zjg + hj Nj Nj j (12) vant as it is swept out of the regression model). But With m* estimated and Zj observed, estimation of Nj this applies to the estimator of b only: the estima- this equation yields g (see Amemiya 1978). The dis- tor of g does not permit a correlation between m turbance term is not iid, of course, even if mj is, be- and Z, if it is to be consistent. cause the variance of (m* - m*) equals N N Var( m* - m*) = HgH¿ - HgA¿X¿H¿ - HXAgH¿ I.2.3 Random Effects Estimation N + HXVar ( b FE)X¿H¿ (13) To describe the effect of weights on the random ef- fects approach, it is better to write the model in ma- where trix notation at the community level. Thus, let yj be H = (Z¿ WZm)-1Z¿ W m m a Nj * 1 vector of observations in community j, and other variables are de�ned similarly. Then, A = (X*¿WX*)-1X*¿W© -0.5E, yi = Xj b + i Nj Zjg + i Nj mj + nj E = I - Zm(Z¿ WZm)-1Z¿ W, m m = Xj b + i NjZjg + uj © = s2I, (17) v K Tjd + uj X* = © -0.5EX. As is standard, Thus, equation (13) makes clear that (m* - m*) is het- N eroskedastic (for example, the �rst term of equation E3uju¿4 = s2 i Nj i¿Nj + s2INj K Æ j j m n N N (18) (13) equals H©H¿ = diag(s2(g i =j1w2 )>(g i =j1wji)2), v ji This means that Æ j-0.5uj is iid with mean 0 and the elements of which clearly varies with j because variance 1. Thus, in the absence of weights, the rec- of the variation in the weights), and the elements of ommended transformation matrix that leads to the the vector (m* - m*) are correlated. (The fourth term, N GLS estimator is sn Æ j-0.5; that is, the transforma- for example, is a full matrix that depends on the tion de�nes community-averaged values of the explanatory I variables, as HX = X is a K * J matrix where the y* = yji - ujyj ji (19) 208 The Rural Investment Climate where uj = 1 - (sn>(Njs2 + s2)) and yj is the sim- m n I.3 ESTIMATION APPROACHES ple (unweighted) average of observations in com- FOR DISCRETE DEPENDENT munity j. After this transformation, the resulting model is estimated with OLS. Stated in another VARIABLES way, in the absence of weights, the random effects When the dependent variable is discrete, the re- (GLS) estimator minimizes the generalized sum of gression approaches discussed in Section 2 are less N N squared residuals u¿Æ -1u . appropriate because the distributional assump- Application of GLS with weights is equivalent92 tions (in particular, that the disturbances have a to minimizing the weighted generalized sum of constant variance) cannot be maintained. The typi- squared residuals u¿W1>2 Æ -1W1>2 u = u¿° u , where N N N N cal solution is found in the maximum likelihood es- timation (MLE) method, even though this method Æ = diag(Æ j) = diag(t2 Jj + s2Ej) j n (20) requires one to make an arbitrary assumption where Ej = Ij - Jj, Jj = W1>2 i j(i¿Wj i j)-1 i¿ W1>2, about the shape of the distribution (such as normal j j j j or logistic). Wj = diag(wji) and t2 = (i¿ W i j)s2 + s2. The slope j j j m n This section discusses two estimation ap- parameters are estimated as proaches. One applies to situations in which the N dWGLS = (T¿°T)-1T¿°y (21) dependent variable is a simple 0/1 variable. Exam- ples of this abound in the RIC data: whether a with a variance equal to household operates an enterprise, whether the N Var (dWGLS) = (T¿°T)-1 T¿°Æ°T (T¿°T)-1 (22) enterprise is registered, whether the enterprise pays taxes, and so on. The second approach is suit- where Æ = diag(s2 i j i¿ + s2Ij). As should be obvi- m j n able for dependent variables with a discrete order ous, the parameters b and g are both represented 0/1/2/. . . where the order is meaningful but the in equation (21): the effect of the individual-speci�c numerical differences between the values are arbi- and community-speci�c factors is estimated at the trary. The primary example in the RIC data are same time, unlike the situation in the �xed effects entrepreneurs’ responses to questions about the approach. Annotated Stata code is provided presence of a potential investment climate obstacle Appendix A.3. and degree of its effect on the operation and To implement this procedure, estimates of s2 v growth of their enterprises: the responses of 1 and s2 are needed. The fixed effects approach m No obstacle; 2 Minor obstacle; 3 Moderate obsta- yields a suitable estimator of s2 through equation v cle; 4 Major obstacle; 5 Very severe obstacle con- (27) above, and the nature of the correlation among stitute a natural order, but it is impossible to argue the uji inspires the following estimator of s2 , m that a major obstacle is three times as bad as a which is parallel to Baltagi (2005, p. 81): minor obstacle just because 4 is three units above 1 Nj Nj Nj Nj and 2 is only one unit above 1. Stated otherwise, J J s2 Nm = a a a wji wjh uji ujh n a a a wji wjh (23) N N instead of 1/2/. . ./5, the responses could be coded j=1 i=1h=1 j=1 i=1h=1 0/1/2/4/8 or 1/10/100/500/1000 without loss of hZi hZi any of the information carried in the entrepre- N where u is an OLS residual. neur’s response. This model is estimated with As an aside, the weighted GLS procedure is ordered probit. explained above in matrix notation. One might transform the individual variables in a fashion par- allel to the standard (unweighted) by means of the I.3.1 Weighted Random Effect Maximum following equation: Likelihood Estimation The weighted random effect probit and weighted y* = w0.5(yji - tjyj) ji ji I (24) random effect ordered probit models are special where once again the symbol ˘ denotes a within- cases of the general weighted random effect MLE community weighted average, and where tj is the approach. In general, the MLE approach starts with same as that de�ned in equation (20). Because of a formulation of the density function driving the the weights, however, the variance of the OLS pro- generation of disturbances. From this, the contri- cedure that one might use to regress the trans- bution of observation (ji) to the likelihood function formed variables is incorrect. is derived, often written as l(u|yji, Xji), where u Annex I. Econometric Analysis of RIC Survey Data 209 comprises all parameters of the model, and the N wji 1 (w i = 1 l(u|yji, Xji, mj) ) g(mj) dmj cannot be inter- J overall likelihood function is the product of these contributions: preted anymore as a joint probability: for one thing, it will now depend on the magnitude of the sum of the weights. Therefore, de�ne w* = wji>wj as a J Nj L = q q l(u|yji, Xji) (25) ji j=1 i=1 community-standardized weight, such that w* ji does indeed average to 1. With this in hand, the In the present case, disturbances are twofold: vji likelihood function with weights is de�ned as at the individual level and mj at the community level. The contribution of observation (ji) to the J Nj wj likelihood function must therefore be stated condi- L = qa a q l(u|yji, Xji, mj)w* b g(mj) dmj b ji (29) j=1 L i=1 tional on mj: l(u|yji, Xji, mj). The function l is related to the density speci�ed for v. Let the density of mj such that the community-level joint probability be written as g(mj), where the function g depends created by the integration over mj is itself geometri- on additional parameters suppressed for simplic- cally averaged to obtain the overall joint probabil- ity. The contribution of the observations living in ity that constitutes the likelihood function. community j is then the product of the individual The integral in equation (29) may be dif�cult to contributions, averaged over all feasible values of evaluate analytically. This is the case in particular mj, and the overall likelihood function combines all with the probit and ordered probit applications communities: discussed below. As a substitute, the integral may J Nj be simulated as the average of R random draws of L = qa a q l(u|yji, Xji, mj)b g (mj) dmj b (26) mr from the distribution that is characterized by g: j j=1 L i=1 Nj The usual way of introducing sampling weights a q l(u|yji, Xji, mj)wj*bg(mj) dmj i in the common likelihood function is more intu- L i=1 N itive when (25) is written in logarithmic form: 1 R j the log-likelihood function with weights is the L a q l(u|yji, Xji, mr)w j* j i (30) Rr=1 i=1 weighted sum of the log of the individual contri- butions: This is an application of the GHK simulation esti- G Nj mation method (Vijverberg 1997). ln L = a a wji ln l(u|yji, Xji) (27) j=1 i=1 I.3.2 Probit where it is understood that the weights wji average to 1. This is equivalent to the following direct mod- As a context for the probit model, consider house- i�cation of (25): holds that may or may not operate an enterprise. Let entrepreneurship be observed as yji = 1 if the J Nj household operates an enterprise and yji = 0 if it L = q q l(u|yj, Xji)wji (28) j=1 i=1 does not. It is explained by household variables X and community variables Z. Random factors at the If we interpret each l(u|yji, Xji) as a probability, household and community level are incorporated the likelihood function with weights is therefore a with disturbances nji and mj, respectively, which joint probability that is a geometrically weighted are identically and independently normally dis- combination of individual probabilities with, as tributed. These factors are combined into a latent stated, weights averaging to 1. variable y* that measures the tendency towards en- ji When the model contains a community random trepreneurship: effect, it is problematic to raise every l(u|yji, Xji, mj) to the power of wji . In (26), the expression y* = X¿ b + Zjg + mj + nji ji ji (31) Nj 1 (w i = 1 l(u|yji, Xji, mj)) g(mj) dmj can be thought of as the joint probability of members of community j, Then, yji = 1(0) if y* Ú (6 )0 and therefore ji computed as the averaged conditional joint proba- nji Ú ( 6 ) -X¿ b -Z¿ g - mg. As the distribution of ji j bility. If wji does not average to 1 within a commu- nji is symmetric, the range of nji can be stated in nity, an expression such as concert with the outcome variable yji in a simple 210 The Rural Investment Climate way: nji … (2yji - 1)(X¿ b + Z¿g + mj). Let us de- ji j The matrices V1 and V2 are the same in expectation, note the normal cdf as £ and pdf as f. Then, that is, when samples become large. In small sam- ples, the two matrices may well deviate, and V1 is l(u|yji, Xji, mj) = £3(2yji - 1)(X¿ b + Z¿g + mj)4 ji j the preferred one (Greene 2008, p. 496). (32) With weights, using equation (27), the parallel Substitution into equation (30) provides the expressions for V1 and V2 are: weighted random effect probit model. Parameters J Nj 0 2 ln lji(u) N EstVar(u) = a - a a wji b -1 are estimated subject to parameter standardization, N K V1 (38) which typically amounts to setting Var(nji) equal to j=1 i=1 0u 0u¿ 1. Annotated Stata code in Appendix A.4 may be Nj N N J 0 ln lji(u) 0 ln lji(u) EstVar(u) = a a a wji a ba bb ¿ -1 used to estimate a weighted random effect probit N 2 model. j=1 i=1 0u 0u K V2 (39) I.3.3 Ordered Probit It is straightforward to see that V1 and V2 can no longer be the same, since (38) is linear in the In the annotated Stata code in Appendix A.5, the weights and (39) is quadratic. Instead, the variance ordered probit model assumes that categories are N of u is found by a Taylor expansion of the gradient coded as 0/1/2/3/4. Thus, yji takes on one of these of the likelihood function: outcomes. A latent variable y* measures the ten- ji dency to fall into one of these categories and is re- N 0 ln L(u) = 0 lated to explanatory variables in the same way as in 0u the probit model; see equation (31). Then, 0 ln L(u) 0 2 ln L(u) = + N (u - u) (40) yji = c iff ac … y* 6 ac + 1 for c = 0, 1, . . . , 4 0u 0u 0 u¿ ji Thus (33) 0 2 ln L(u) -1 0 ln L(u) or u - u = -a N b (41) yji = c iff ac - X¿ b - Z¿g - mj … nji 6 ac + 1 ji j 0u 0u¿ 0u and therefore - X¿ b - Z¿g - mj for c = 0, 1, . . . , 4 ji j (34) Var(u) = E3(u - u)(u - u)¿4 N N N (42) where the a’s are parameters to be estimated, 2 2 =a b a ba b where a0 = - q and a5 = q , and where stan- 0 ln L(u) -1 0 ln L(u) 0 ln L(u) 0 ln L(u) -1 dardization requires one other a to be �xed: a1 = 0. 0u 0u¿ 0u 0u¿ 0u 0u¿ From this, the function l is written as: The estimated variance is obtained by evaluating l(u|yji = c, Xji, mj) = £3ac + 1 - X¿ b - Z¿g - mj4 ji j N the �rst and second order derivatives at u : - £3ac - X¿ b - Z¿g - mj4 ji j N EstVar(u) (35) N 0 2 ln L(u) -1 0 ln L(u) 0 ln L(u) 0 2 ln L(u) -1 N N N =a b a ba b 0u 0u¿ 0u 0u¿ 0u 0u¿ I.3.4 Estimating Standard Errors -1 of the Estimates = V1V2 V1 K V3 (43) Writing l(u|yji, Xji) as lji, derivatives of the com- This expression is not sensitive to the scaling of the mon likelihood function (25) are used to compute weights. N the variance of the estimate u of the parameter vec- tor u (see, for example, Greene 2008, p. 495). I.3.5 A �nding About Standard Errors J Nj 0 2 ln lji (u) N with RIC Data EstVar(u) = a- a a b -1 N K V1 (36) j=1 i=1 0 u 0u¿ The Stata programs given in Appendices A.4 and or A.5 automatically estimate the standard errors of Nj N N the estimated parameters with equation (43). In es- J 0 ln lji(u) 0 ln lji(u) EstV u) = a a a a ba bb ¿ -1 ar( N timating the ordered probit model for Tanzania, j=1 i=1 0u 0u however, enterprise weights are absent, and thus K V2 (37) wji is set equal to 1. One would have expected that Annex I. Econometric Analysis of RIC Survey Data 211 V3 would be roughly the same as V1 since typically much smaller. The only plausible explana- -1 V2 V1 = I. In practice, the standard errors showed tion for this difference would have to be the number signi�cant differences. of enterprises (over 1,000) as compared to the num- This led to a more general check against the data ber of communities (between 100 and 150): the for- from Nicaragua and Sri Lanka as well from the mer is large enough to roughly equalize V1 and V2, three enterprise investment climate outcome vari- but the latter falls signi�cantly short. ables (cost of �nance, road quality, and electricity). This �nding implies that V1, V2 and V3 are very Table I.1 summarizes the �ndings, showing ratios N different estimators of the variance of u . Columns of standard errors averaged over the three esti- 1 through 3 of Table I.2 illustrate the difference mated models in each country. For enterprise and between V1 and V3. Where V2-based standard industry variables, the standard errors obtained errors for community-level variables are much from V1 and V2 are roughly the same: overall, across smaller than those derived from V1, the V3-based the three countries and all available ratios, the ratio standard errors are much larger. averages out at 0.999. Among parameters of com- As a further test, for one Sri Lanka variable munity variables, however, the ratio averages only (obstacles related to electricity), the t-statistics of 0.616: standard errors obtained through V2 are one of the benchmark indicators (infrastructure Table I.1 Ratio of Standard Errors of Parameter Estimates Obtained Through V2 and V1 Variable Nicaragua Sri Lanka Tanzania Average Enterprise and industry variables Age of enterprise# 1.046 1.022 0.974 Education of manager 1.000 0.975 Female manager 1.019 1.009 Sinhalese manager 0.858 Household-based enterprise 1.049 0.970 1.022 Services enterprise 1.013 1.007 0.929 Nonagricultural manufacturing enterprise 1.014 1.010 0.943 Agricultural processing enterprise 1.019 1.111 0.898 Other production enterprise 1.063 1.081 0.972 Mixed enterprise 1.000 1.038 0.884 Average 1.029 1.012 0.956 0.999 Community and benchmark variables Community population size# 0.623 0.656 0.490 Income per capita# 0.735 0.584 0.527 Agricultural seasonality 0.700 0.592 Share income from agriculture 0.730 0.807 0.513 Connectivity 0.701 0.645 0.498 Infrastructure services 0.665 0.684 0.546 Business services 0.617 0.634 0.556 Human capital 0.689 0.634 0.519 Finance services 0.629 0.641 0.516 Intercept 0.737 0.612 0.495 Average 0.682 0.649 0.518 0.616 Other model parameters St.dev of community random error (sm) 0.862 0.877 0.712 0.817 a2 1.035 1.007 1.004 a3 1.015 0.988 a4 1.002 0.962 Average across a’s 1.035 1.008 0.985 1.009 Source: Authors’ calculation from RIC survey data. Note: # Variable enters in logarithmic form. 212 The Rural Investment Climate Table I.2 Ratio of Standard Errors of Parameter Estimates Obtained Through V3 and V1 Ratio of weighted Unweighted Weighted a over unweighteda Variable Nicaragua Sri Lanka Tanzania Nicaragua Sri Lanka Nicaragua Sri Lanka Enterprise and industry variables Age of enterpriseb 0.979 0.995 1.078 1.094 1.477 1.117 1.485 Education of manager 1.019 1.164 1.481 1.453 Female manager 1.001 1.056 1.321 1.319 Sinhalese manager 1.247 2.459 1.971 Household-based enterprise 0.970 1.066 1.024 0.994 1.323 1.025 1.241 Services enterprise 1.003 1.018 1.219 1.143 1.325 1.140 1.301 Nonagricultural manufacturing enterprise 1.002 1.010 1.123 1.119 1.280 1.117 1.267 Agricultural processing enterprise 0.999 0.917 1.192 1.117 1.014 1.118 1.106 Other production enterprise 0.954 0.949 1.098 1.073 0.644 1.125 0.678 Mixed enterprise 1.016 0.991 1.197 1.127 1.340 1.110 1.352 Average 0.989 1.021 1.128 1.107 1.318 Community and benchmark variables Community population sizeb 1.733 1.588 2.406 2.038 2.194 1.176 1.381 Income per capitab 1.468 1.790 2.117 1.634 2.437 1.113 1.361 Agricultural seasonality 1.497 1.756 1.583 2.875 1.058 1.637 Share income from agriculture 1.427 1.435 2.213 1.664 2.281 1.166 1.590 Connectivity 1.503 1.630 2.232 1.639 2.182 1.091 1.339 Infrastructure services 1.569 1.510 2.070 1.680 3.009 1.071 1.993 Business services 1.711 1.709 1.985 1.792 2.318 1.048 1.357 Human capital 1.527 1.678 2.209 1.634 2.491 1.070 1.484 Finance services 1.655 1.634 2.150 1.750 2.195 1.057 1.343 Intercept 1.542 1.740 2.377 1.729 2.184 1.122 1.255 Ave com char 1.563 1.647 2.195 1.097 1.474 Other model parameters St. dev of com. random error (sm) 1.322 1.224 1.630 1.306 1.946 0.989 1.590 2 0.988 1.002 1.014 1.096 1.386 1.109 1.383 3 0.999 1.044 1.367 1.368 4 1.031 1.094 1.284 1.245 Average across ’s 0.988 1.011 1.051 1.109 1.332 Source: Authors’ calculation from RIC survey data. Notes: a. Tanzania’s survey does not contain sampling weights. b. Variable enters in logarithmic form. services) was compared to the likelihood ratio test weights. With weights, however, V3 is the theoret- value computed if this benchmark indicator were ically correct (asymptotic) estimator of the vari- to be omitted. The p-value of the likelihood ratio N ance of u , and the divergence of V1 and V2 starts to was 0.1288; the t-statistics and p-values were bite into the implications of the analysis. Table I.2 1.518 and 0.1290 when the V1-based standard illustrates this point in columns 4 and 5 with the errors are used, 2.062 and 0.0392 when the ratio of the standard errors derived from V3 over V2-based standard errors are used, and 1.088 and those derived from V1 when the model is estimated 0.2768 when the V3-based standard errors are used. with sampling weights. For Nicaragua, V3-based This illustrates that V1 is the best estimator of the standard errors are about 70 percent larger than N variance of u . the V1-based ones, and for Sri Lanka the V3-based In principle, this difference would not be prob- standard errors are about two and a half times as lematic, since V1 is the recommended estimator of large—and t-statistics are correspondingly much the variance anyway if there are no sampling lower. Annex I. Econometric Analysis of RIC Survey Data 213 But, because V2 apparently is a poor estimator weights, and (ii) observations are clustered in com- of what it is supposed to measure, should V3 actu- munities. The question arises how much difference ally be considered as an estimator of the variance it makes to account for weights and random effects N of u ? One could argue that the ratio of the stan- in the actual empirical analysis of RIC data. This dard errors with weights set to 1, as shown in section considers representative examples of enter- columns 1 through 3 of Table I.2, reflects the con- prise performance, entrepreneurship selection, and sequence of using RIC samples with relatively EICO models estimated in alternative ways, using small numbers of communities. Furthermore, it is data from Nicaragua and Sri Lanka. arguable that the ratio of standard errors using Table I.3 starts off this comparison with enter- actual sampling weights (as shown in columns 4 prise performance models, referring to both sales and 5 of Table I.2) deviates from 1 both because of and net value added. The models contain enterprise the small number of communities and because of characteristics, industry dummies, benchmark the variation in the sampling weights. It may be indicators, and other community characteristics. concluded then that the difference in those ratios The estimates of these models therefore corre- with unit and actual weights—measured addi- spond with those found in column 3 of Table C.4 to tively or, as in columns 6 and 7 of Table I.2, pro- Table C.7. The �rst column of Table I.3 consists of portionally—is the actual effect of using weights unweighted OLS estimates with common t-statistics on the standard errors. labeled as “t(u)� and t-statistics adjusted for clus- Stated otherwise, in Nicaragua, the effect of tering by community (“t(a)�). The second set of sampling weights on standard errors is a correc- estimates use sampling weights. The third and tion of V1-based standard errors of 10.7 percent for fourth columns present unweighted and weighted enterprise and industry variables (and also the ’s) random effect models, respectively. The compar- and of 9.7 percent for the community variables and isons yield the following conclusions: (i) account- benchmark indicators. In Sri Lanka, the use of sam- ing for sampling weights affects estimates more pling weights necessitates a correction of V1-based dramatically than does incorporating random standard errors of 31.8 percent for enterprise and effects; (ii) the estimated effects of community industry variables (and also the a’s), of 47.4 per- variables (including benchmarks) tend to be more cent for the community variables and benchmark sensitive to the application of weights or random indicators, and of 59 percent for sm. N effects than are enterprise variables; (iii) adjusting These rules of thumb are likely to generate stan- t-statistics of OLS estimates for clustering reduces dard errors and t-statistics that are more accurate many but not all t-statistics; and (iv) if the when sampling weights are used. They appear to weighted random effect procedure is not available, differ between countries, and these particular val- weighed OLS with t-statistics adjusted for cluster- ues are suggested only for the ordered probit mod- ing is the best substitute estimation technique. els with which they were computed. Next, Table I.4 employs the same strategy in the In principle, the validity of this correction to context of entrepreneurship selection. Once again, V1-based standard errors may be con�rmed with a the speci�cation of the estimated model is typical Monte Carlo study. Alternatively, standard errors of those reported in Chapter 4, containing house- may be computed with bootstrap methods, but, as hold characteristics, benchmark indicators, and with a Monte Carlo study, this is a very time- other community characteristics. The �rst column intensive undertaking that will be left for future contains standard probit estimates without research. weights with common t-statistics labeled as “t(u)� and t-statistics adjusted for clustering by commu- nity (“t(a)�). The second set of estimates use sam- I.4 THE IMPORTANCE OF pling weights. The third and fourth columns SAMPLING WEIGHT AND present unweighted and weighted random effect RANDOM EFFECTS IN probit estimates. The table reports estimated slopes, rather than standardized effects on the RIC ANALYSIS probability of entrepreneurship as in Table F.2. The techniques discussed in this Annex aim to offer The most important conclusions are similar to econometric solutions to two characteristics of RIC those above: (i) accounting for sampling weights databases: (i) observations have unequal sampling generates more substantial changes in parameter 214 The Rural Investment Climate estimates than does incorporating random effects, reported in Table H.4 of Annex H. But unlike that and (ii) the estimated effects of community vari- table, Table I.5 reports estimates of slopes rather ables (including benchmarks) tends to be more than standardized effects on EICO outcomes. Con- sensitive to the application of weights or random clusions are as follows: (i) accounting for sampling effects than are household characteristics. Further- weights has a greater effect than does incorporat- more, (iii) adjusting standard errors for clustering ing random effects; (ii) estimated slopes of com- is more important for community variables and munity variables (including benchmarks) are benchmark indexes than for household character- especially affected; (iii) t-statistics must be istics, and (iv) if the weighted random effect probit adjusted for clustering, especially for community procedure is not available, weighed probit with t- variables where unadjusted t-statistics are signi�- statistics adjusted for clustering is the best substi- cantly upwardly biased; and (iv) if weighted ran- tute estimation technique. In the case of Nicaragua, dom effect ordered probit procedure is not the weighted random effect probit model yields a available, no alternative procedure is clearly vanishing random effect and thus automatically second-best, as estimates of the unweighted ran- simpli�es to a weighted standard probit model. dom effect ordered probit and weighted standard As the last example, Table I.5 examines two ordered probit models both deviate in substantial enterprise investment climate outcomes (EICOs) ways from the theoretically preferred weighted prominent in all countries: interest charged on random effect ordered probit estimates. loans and issues surrounding electricity. The spec- In sum, the evidence suggests it is mandatory to i�ed models include enterprise characteristics and account for weights, and it is strongly advisable to the selection of community variables (benchmark account for clustering. In particular, the effect of and other) that in the weighted random effect benchmark indicators and community characteris- ordered probit version appeared to contribute tics is sensitive to the speci�cation of the model in most to the explanation of the particular EICO. regard to application of sampling weights and the Thus, the estimated models are similar to those use of random effects. Annex I. Econometric Analysis of RIC Survey Data 215 Table I.3 Effect of Sampling Weights and Random Effect Speci�cation on Enterprise Performance Regression Results A1: Nicaragua: Dependent variable is ln(sales) Ordinary Least Squares Ordinary Least Squares Random Effects Random Effects (unweighted) (weighted) (unweighted) (weighted) b t(u) t(a) b t(u) t(a) b t b t Factor and nonfactor inputs Total labor input# 0.330 2.68 2.16 0.232 1.67 1.54 0.311 2.53 0.231 5.07 Total capital input# 0.071 0.66 0.66 0.158 1.36 1.45 0.086 0.80 0.165 4.07 lnVLL (lnL*lnL) 0.062 4.42 3.51 0.046 2.78 2.61 0.061 4.30 0.048 9.13 lnVLC (lnL*lnC) 0.009 0.51 0.50 0.010 0.51 0.51 0.008 0.43 0.010 1.46 lnVCC (lnC*lnC) 0.026 3.50 3.06 0.024 2.80 2.59 0.027 3.56 0.024 8.79 Depreciation# 0.057 2.29 2.14 0.050 1.77 1.70 0.058 2.30 0.060 6.29 Nonfactor Cost# 0.083 5.58 4.67 0.087 5.27 4.73 0.079 5.16 0.077 12.96 Other enterprise characteristics Age of enterprise# 0.115 3.36 3.13 0.105 2.61 2.83 0.112 3.31 0.098 7.63 Registration 0.051 0.62 0.54 0.040 0.38 0.35 0.048 0.59 0.045 1.46 Industry dummies Services enterprise 0.160 1.82 1.72 0.160 1.62 1.66 0.135 1.54 0.116 3.48 Manufacturing nonagricultural enterprise 0.086 0.73 0.71 0.144 1.04 1.08 0.077 0.66 0.114 2.61 Agricultural processing enterprise 0.097 0.99 0.99 0.022 0.20 0.19 0.074 0.76 0.000 0.00 Other production enterprise 0.197 1.12 1.23 0.120 0.91 0.82 0.178 1.01 0.072 1.10 Mixed enterprise 0.105 1.15 1.29 0.189 1.88 2.08 0.120 1.32 0.183 5.29 Community characteristics Agricultural seasonality 0.149 1.86 1.84 0.010 0.12 0.10 0.141 1.39 0.042 0.56 Enterprise density 0.001 1.71 1.37 0.001 1.67 1.28 0.001 1.51 0.001 2.36 Community population size# 0.031 0.95 0.84 0.047 1.30 0.99 0.038 0.91 0.068 2.16 Agricultural land per capita 0.014 0.31 0.33 0.005 0.11 0.10 0.001 0.01 0.048 1.16 Illiteracy 0.001 0.47 0.43 0.000 0.06 0.06 0.001 0.33 0.000 0.15 Benchmark indicators Connectivity 0.555 2.01 1.99 0.490 1.71 1.48 0.601 1.68 0.673 2.48 Infrastructure services 0.106 0.44 0.43 0.040 0.17 0.14 0.162 0.54 0.112 0.51 Development services 0.110 1.23 1.36 0.048 0.48 0.52 0.142 1.18 0.079 0.85 Governance 0.354 1.20 1.20 0.303 0.94 1.03 0.313 0.82 0.212 0.75 Human capital 0.047 0.09 0.10 0.441 0.87 0.74 0.023 0.03 0.590 1.26 Finance services 0.042 0.29 0.31 0.092 0.62 0.55 0.044 0.22 0.163 1.04 Intercept 5.339 9.88 9.73 5.293 8.79 8.18 5.286 8.96 5.212 15.20 Regression statistics Number of observations 846 846 846 846 R2 0.422 0.431 0.439 0.428 sva 0.887 0.891 0.864 0.286 smua 0.000 0.000 0.241 0.247 (continued on the next page) 216 The Rural Investment Climate Table I.3 Effect of Sampling Weights and Random Effect Speci�cation on Enterprise Performance Regression Results (continued ) A2: Nicaragua: Dependent variable is ln(net value added) Ordinary Least Squares Ordinary Least Squares Random Effects Random Effects (unweighted) (weighted) (unweighted) (weighted) b t(u) t(a) b t(u) t(a) b t b t Factor and nonfactor inputs Total labor input# 0.359 2.62 2.37 0.240 1.62 1.58 0.345 2.52 0.242 5.16 Total capital input# 0.057 0.45 0.36 0.046 0.27 0.27 0.050 0.40 0.024 0.54 lnVLL (ln*ln) 0.077 4.84 4.16 0.060 3.38 3.19 0.076 4.81 0.065 12.03 lnVLC (lnL*lnC) 0.031 1.47 1.31 0.016 0.60 0.64 0.031 1.47 0.022 3.09 lnVCC (ln*ln) 0.033 3.61 2.39 0.035 3.17 2.40 0.033 3.63 0.035 11.69 Other enterprise characteristics Age of enterprise# 0.111 2.87 2.74 0.094 2.05 2.32 0.110 2.86 0.090 6.72 Registration 0.079 0.92 0.73 0.100 0.87 0.78 0.084 0.98 0.108 3.57 Industry dummies Services enterprise 0.175 1.75 1.74 0.178 1.64 1.66 0.158 1.59 -0.142 -4.06 Manufacturing nonagricultural enterprise 0.026 0.20 0.21 0.069 0.47 0.50 0.017 0.13 0.036 0.80 Agricultural processing enterprise 0.089 0.81 0.80 0.006 0.05 0.05 0.077 0.71 0.006 0.15 Other production enterprise 0.385 1.97 2.02 0.284 1.75 1.67 0.365 1.87 0.231 3.45 Mixed enterprise 0.144 1.42 1.54 0.249 2.14 2.50 0.152 1.51 0.238 6.68 Community characteristics Agricultural seasonality 0.155 1.71 1.69 0.006 0.06 0.05 0.147 1.34 0.018 0.20 Enterprise density 0.001 1.83 1.43 0.001 1.89 1.33 0.001 1.67 0.001 2.48 Community population size# 0.034 0.92 0.82 0.047 1.20 0.90 0.040 0.90 0.067 1.84 Agricultural land per capita 0.025 0.48 0.47 0.005 0.08 0.07 0.013 0.22 0.046 0.96 Illiteracy 0.003 1.23 1.12 0.002 0.72 0.66 0.003 1.02 0.002 0.83 Benchmark indicators Connectivity 0.912 2.95 2.91 0.805 2.43 2.01 0.929 2.43 0.930 2.93 Infrastructure services 0.274 1.02 0.98 0.194 0.70 0.52 0.321 1.00 0.252 0.98 Development services 0.041 0.41 0.44 0.017 0.15 0.17 0.071 0.55 0.021 0.19 Governance 0.545 1.64 1.71 0.488 1.45 1.48 0.469 1.15 0.357 1.07 Human capital 0.157 0.27 0.30 0.609 1.11 0.90 0.087 0.12 0.725 1.32 Finance services 0.072 0.43 0.45 0.125 0.74 0.67 0.071 0.33 0.191 1.04 Intercept 5.124 8.44 8.28 5.084 7.30 7.13 5.089 7.82 4.970 12.67 Regression statistics Number of observations 814 814 814 814 R2 0.321 0.339 0.340 0.336 sva 0.982 0.980 0.959 0.292 smua 0.000 0.000 0.238 0.295 Table I.3 Effect of Sampling Weights and Random Effect Speci�cation on Enterprise Performance Regression Results (continued ) B1: Sri Lanka: Dependent variable is ln(sales) Ordinary Least Squares Ordinary Least Squares Random Effects Random Effects (unweighted) (weighted) (unweighted) (weighted) b t(u) t(a) b t(u) t(a) b t b t Factor and nonfactor inputs Total labor input# 0.256 1.20 0.89 0.294 0.80 0.79 0.256 1.20 0.325 1.09 Total capital input# 0.451 3.61 2.30 0.100 0.44 0.39 0.451 3.61 0.091 0.53 lnVLL (ln*ln) 0.091 3.88 2.54 0.080 2.01 1.86 0.091 3.88 0.081 2.50 lnVLC (lnL*lnC) 0.097 4.14 2.80 0.075 1.93 1.85 0.097 4.14 0.074 2.24 lnVCC (ln*ln) 0.033 4.64 2.77 0.047 2.37 2.45 0.033 4.64 0.047 4.13 Depreciation# 0.053 2.08 1.87 0.041 1.01 0.93 0.053 2.08 0.039 1.16 Nonfactor cost# 0.468 27.15 16.28 0.477 14.18 12.88 0.468 27.15 0.473 20.61 Other enterprise characteristics Age of enterprise# 0.069 2.53 2.26 0.115 3.15 3.01 0.069 2.53 0.116 3.18 Experience of manager# 0.098 3.15 2.63 0.082 2.17 2.02 0.098 3.15 0.075 1.79 Gender of manager 0.043 0.69 0.63 0.104 1.10 1.13 0.043 0.69 0.117 1.45 Education of manager 0.019 2.15 2.22 0.026 2.03 2.14 0.019 2.15 0.026 2.20 Registration 0.011 0.18 0.17 0.095 1.11 1.01 0.011 0.18 0.094 1.17 Industry dummies Services enterprise 0.277 3.47 2.92 0.239 2.30 2.48 0.277 3.47 0.215 2.08 Manufacturing, nonagricultural enterprise 0.019 0.28 0.29 0.008 0.09 0.09 0.019 0.28 0.020 0.22 Agricultural processing enterprise 0.064 0.41 0.46 0.089 0.49 0.48 0.064 0.41 0.085 0.42 Other production enterprise 0.177 0.85 1.31 0.152 0.51 1.14 0.177 0.85 0.097 0.43 Mixed enterprise 0.030 0.30 0.28 0.075 0.70 0.59 0.030 0.30 0.122 0.91 Community characteristics Agricultural seasonality 0.064 0.62 0.75 0.142 1.14 1.27 0.064 0.62 0.131 0.83 Enterprise density 0.001 0.63 0.70 0.000 0.17 0.19 0.001 0.63 0.000 0.28 Community population size# 0.016 0.28 0.30 0.001 0.02 0.02 0.016 0.28 0.006 0.07 Agricultural land per capita 0.067 0.38 0.34 0.414 1.81 1.94 0.067 0.38 0.412 1.63 Illiteracy 0.012 0.05 0.05 0.089 0.31 0.32 0.012 0.05 0.064 0.17 Main market in Neighboring communities 0.178 2.02 1.71 0.378 3.36 3.17 0.178 2.02 0.369 2.61 Commercial center 0.142 1.56 1.57 0.120 0.96 1.10 0.142 1.56 0.108 0.76 Nearest city 0.006 0.09 0.09 0.038 0.42 0.48 0.006 0.09 0.036 0.31 Main community income from: Wages 0.001 0.01 0.02 0.162 2.03 2.17 0.001 0.01 0.166 1.66 Self-employment 0.091 0.97 1.29 0.167 1.44 1.92 0.091 0.97 0.156 1.03 Benchmark indicators Connectivity 0.214 0.93 0.87 0.126 0.42 0.43 0.214 0.93 0.094 0.27 Infrastructure services 0.664 2.70 2.45 0.201 0.71 0.68 0.664 2.70 0.198 0.53 Development services 0.210 1.69 2.03 0.430 3.14 3.65 0.210 1.69 0.438 2.19 Governance 0.767 1.08 1.14 0.248 0.26 0.30 0.767 1.08 0.190 0.18 Human capital 0.257 0.68 0.75 0.182 0.46 0.54 0.257 0.68 0.126 0.21 Finance services 0.280 1.25 1.47 0.356 1.32 1.47 0.280 1.25 0.349 1.03 Intercept 2.117 2.44 2.24 2.384 2.01 1.96 2.117 2.44 2.634 2.05 Regression statistics Number of observations 1018 1018 1018 1018 R2 0.762 0.750 0.769 0.750 sva 0.775 0.735 0.771 0.737 smua 0.000 0.000 0.000 0.210 (continued on the next page) 217 Table I.3 Effect of Sampling Weights and Random Effect Speci�cation on Enterprise Performance Regression Results (continued ) B2: Sri Lanka: Dependent variable is ln(net value added) Ordinary Least Squares Ordinary Least Squares Random Effects Random Effects (unweighted) (weighted) (unweighted) (weighted) b t(u) t(a) b t(u) t(a) b t b t Factor and nonfactor inputs Total labor input# 0.401 1.29 1.29 0.862 1.90 1.84 0.459 1.49 0.946 2.22 Total capital input# 0.868 4.82 5.55 0.932 4.06 4.10 0.873 4.88 0.953 3.96 lnVLL (ln*ln) 0.161 4.64 5.25 0.209 4.23 4.45 0.166 4.82 0.218 4.55 lnVLC (lnL*lnC) 0.177 5.00 5.18 0.207 4.15 4.35 0.179 5.09 0.217 4.33 lnVCC (ln*ln) 0.047 4.29 2.98 0.055 2.83 2.87 0.048 4.36 0.059 3.21 Other enterprise characteristics Age of enterprise# 0.054 1.33 1.27 0.088 1.99 1.87 0.061 1.51 0.095 1.83 Experience of manager# 0.065 1.41 1.34 0.068 1.30 1.17 0.060 1.30 0.060 1.01 Gender of manager 0.010 0.11 0.09 0.007 0.06 0.07 0.002 0.03 0.025 0.21 Education of manager 0.043 3.17 3.49 0.049 3.03 3.58 0.042 3.14 0.050 2.90 Registration 0.123 1.40 1.45 0.076 0.79 0.75 0.106 1.20 0.064 0.57 Industry dummies Services enterprise 0.334 3.04 2.93 0.177 1.45 1.34 0.323 2.97 0.148 1.07 Manufacturing, nonagricultural enterprise 0.005 0.05 0.05 0.095 0.83 0.74 0.014 0.15 0.085 0.71 Agricultural processing enterprise 0.277 1.21 1.25 0.145 0.51 0.51 0.262 1.15 0.106 0.38 Other production enterprise 0.005 0.02 0.01 0.036 0.10 0.08 0.053 0.17 0.086 0.27 Mixed enterprise 0.186 1.30 1.06 0.188 0.91 0.96 0.154 1.08 0.207 1.08 Community characteristics Agricultural seasonality 0.232 1.52 1.66 0.166 0.99 1.04 0.260 1.40 0.166 0.71 Enterprise density 0.001 0.39 0.34 0.002 1.15 1.04 0.001 0.38 0.002 0.99 Community population size# 0.035 0.43 0.46 0.041 0.47 0.43 0.032 0.32 0.036 0.28 Agricultural land per capita 0.179 0.64 0.53 0.515 1.53 1.62 0.154 0.47 0.491 1.26 Illiteracy 0.549 1.43 1.44 0.001 0.00 0.00 0.650 1.40 0.113 0.20 Main market in: Neighboring communities 0.213 1.68 1.35 0.526 3.36 2.81 0.206 1.32 0.542 2.64 Commercial center 0.068 0.51 0.54 0.257 1.69 2.00 0.072 0.45 0.294 1.41 Nearest city 0.107 1.00 1.01 0.078 0.62 0.72 0.099 0.75 0.097 0.56 Main community income from: Wages 0.043 0.44 0.44 0.043 0.38 0.37 0.046 0.39 0.043 0.29 Self-employment 0.079 0.57 0.65 0.175 1.17 1.54 0.091 0.54 0.166 0.74 Benchmark indicators Connectivity 0.054 0.16 0.14 0.085 0.22 0.21 0.039 0.10 0.017 0.03 Infrastructure services 0.504 1.41 1.45 0.305 0.68 0.68 0.499 1.17 0.337 0.62 Development services 0.225 1.24 1.65 0.578 2.67 3.36 0.185 0.83 0.587 1.99 Governance 0.426 0.41 0.40 0.286 0.24 0.24 0.422 0.34 0.108 0.07 Human capital 0.622 1.14 1.01 0.318 0.57 0.55 0.564 0.85 0.319 0.36 Finance services 0.665 2.04 2.03 0.438 1.21 1.33 0.689 1.76 0.464 0.93 Intercept 0.960 0.77 0.73 2.219 1.45 1.24 1.190 0.86 2.506 1.36 Regression statistics Number of observations 841 841 841 841 R2 0.590 0.523 0.605 0.522 sva 1.029 0.955 1.003 0.950 smua 0.000 0.000 0.275 0.323 Source: Authors’ calculation from RIC data. Note: a. “sv� is the standard deviation of the enterprise-speci�c disturbance. “smu� is the standard deviation of the community random effect. 218 Annex I. Econometric Analysis of RIC Survey Data 219 Table I.4 Effect of Sampling Weights and Random Effect Speci�cation on Entrepreneurship Selection Results A: Nicaragua Random Effect Random Effect Standard Probit Standard Probit Probit Probit (unweighted) (weighted) (unweighted) (weighted) b t(u) t(a) b t(u) t(a) b t b t Household characteristics Number of male adults 0.095 2.25 2.37 0.112 2.24 2.32 0.096 2.26 0.112 2.24 Number of female adults 0.103 2.32 2.42 0.116 2.30 2.29 0.102 2.31 0.116 2.30 Average age 0.016 3.15 3.22 0.017 2.88 2.85 0.016 3.13 0.017 2.89 Human capital index 0.052 0.15 0.16 0.229 0.59 0.63 0.052 0.15 0.229 0.59 Head female 0.127 1.41 1.44 0.197 1.81 1.71 0.131 1.45 0.197 1.81 Head parents entrepreneur 0.067 0.82 0.89 0.077 0.80 0.83 0.069 0.84 0.077 0.80 ln(Other income) 0.052 3.03 3.55 0.045 2.21 2.67 0.052 3.02 0.045 2.21 ln(Remittances) 0.087 5.22 5.02 0.104 5.71 4.92 0.088 5.28 0.104 5.71 ln(Assets) 0.084 2.94 2.82 0.108 3.56 3.53 0.085 2.95 0.108 3.57 Benchmarks Connectivity 0.137 0.36 0.55 0.105 0.23 0.29 0.123 0.31 0.105 0.23 Infrastructure services 0.315 0.95 1.39 0.403 1.10 1.35 0.318 0.92 0.403 1.10 Business services 0.126 1.08 1.62 0.157 1.16 1.32 0.132 1.08 0.157 1.16 Governance 0.299 0.77 1.18 0.029 0.07 0.08 0.318 0.78 0.029 0.06 Human capital 0.171 0.20 0.29 2.467 2.45 2.28 0.150 0.17 2.467 2.45 Finance services 0.163 0.88 1.54 0.186 0.85 1.21 0.167 0.85 0.186 0.85 Community variables ln(Community size) 0.044 1.25 1.69 0.015 0.36 0.39 0.044 1.18 0.015 0.36 ln(Income per capita) 0.170 1.74 2.44 0.146 1.36 1.49 0.174 1.72 0.146 1.36 Agricultural seasonality 0.030 0.28 0.52 0.175 1.40 1.43 0.030 0.28 0.175 1.40 Enterprise openness 0.032 0.25 0.42 0.390 2.48 2.87 0.039 0.30 0.390 2.48 ln(Male wage) 0.095 0.43 0.70 0.090 0.35 0.47 0.091 0.40 0.090 0.35 Intercept 1.363 2.44 3.11 4.763 7.75 8.60 1.374 2.37 4.763 7.74 Regression statistics Number of observations 1163 1163 1163 1163 Log-likelihood 689.48 535.83 692.24 535.83 smua 0.000 0.000 0.106 0.000 (continued on the next page) 220 The Rural Investment Climate Table I.4 Effect of Sampling Weights and Random Effect Speci�cation on Entrepreneurship Selection Results (continued ) B: Sri Lanka Random Effect Random Effect Standard Probit Standard Probit Probit Probit (unweighted) (weighted) (unweighted) (weighted) b t(u) t(a) b t(u) t(a) b t b t Household characteristics Number of male adults 0.048 0.87 0.84 0.102 1.29 1.23 0.048 0.87 0.114 1.38 Number of female adults 0.072 1.29 1.19 0.087 0.99 0.92 0.073 1.30 0.087 0.94 Average age 0.012 1.79 1.57 0.004 0.31 0.27 0.012 1.78 0.004 0.32 Human capital index 0.338 0.85 0.76 0.185 0.32 0.31 0.336 0.84 0.227 0.38 Head female 0.421 3.28 3.10 0.280 1.07 1.00 0.425 3.29 0.296 1.08 Not Sinhalese 0.022 0.13 0.15 0.103 0.38 0.49 0.019 0.11 0.163 0.51 Head parents entrepreneur 0.291 2.89 2.60 0.047 0.30 0.27 0.295 2.90 0.071 0.42 ln (Assets) 0.239 4.57 4.33 0.265 3.36 3.30 0.240 4.56 0.279 3.18 Benchmarks Connectivity 0.506 1.37 1.41 0.736 1.27 1.39 0.525 1.39 0.879 1.04 Infrastructure services 0.559 1.27 1.29 0.726 1.17 1.09 0.584 1.30 0.924 1.13 Business services 0.367 1.58 1.73 0.457 1.39 1.41 0.373 1.57 0.483 1.08 Governance 0.135 0.11 0.13 3.210 1.77 2.16 0.160 0.13 3.358 1.40 Human capital 0.028 0.04 0.04 0.981 0.79 0.90 0.040 0.06 1.282 0.81 Finance services 0.109 0.29 0.38 0.858 1.26 1.38 0.111 0.28 0.961 1.02 Community variables ln(Community size) 0.106 1.09 1.22 0.228 1.37 1.44 0.109 1.10 0.277 1.25 ln(Income per capita) 0.022 0.25 0.26 0.052 0.38 0.37 0.023 0.25 0.058 0.31 Agricultural seasonality 0.031 0.17 0.17 0.087 0.30 0.31 0.021 0.11 0.106 0.27 Enterprise openness 0.017 0.23 0.25 0.047 0.43 0.43 0.014 0.19 0.032 0.20 ln(Male wage rate) 0.190 0.80 0.95 0.337 0.88 1.12 0.191 0.79 0.411 0.78 Intercept 3.696 2.06 2.35 8.998 3.21 3.19 3.737 2.04 10.139 2.62 Regression statistics Number of observations 849 849 849 849 Log-likelihood 542.20 455.80 542.66 451.39 smua 0.000 0.000 0.102 0.328 Source: Authors’ calculation from RIC data. Note: a. “smu� is the standard deviation of the community random effect. Annex I. Econometric Analysis of RIC Survey Data 221 Table I.5 Effect of Sampling Weights and Random Effect Speci�cation on Selected EICO Regression Results A1: Nicaragua: Dependent variable is “Interest rate of loan� Random Effect Random Effect Standard Ordered Probit Standard Ordered Probit Ordered Probit Ordered Probit (unweighted) (weighted) (unweighted) (weighted) b t(u) t(a) b t(u) t(a) b t b t Enterprise characteristics Age of enterprise b 0.022 0.50 0.47 0.010 0.20 0.17 0.004 0.08 0.012 0.21 Household-based enterprise 0.519 3.46 3.53 0.500 3.42 3.19 0.499 3.07 0.489 3.06 Services enterprise 0.021 0.19 0.25 0.017 0.13 0.18 0.029 0.23 0.080 0.57 Manufacturing, nonagricultural enterprise 0.043 0.28 0.29 0.060 0.37 0.31 0.002 0.02 0.019 0.10 Agricultural processing enterprise 0.183 1.39 1.73 0.137 0.94 1.12 0.175 1.24 0.137 0.89 Other production enterprise 0.069 0.32 0.34 0.204 0.93 0.91 0.036 0.16 0.238 1.03 Mixed enterprise 0.070 0.59 0.60 0.045 0.33 0.33 0.101 0.79 0.090 0.63 Community characteristics Time to nearest city 0.482 2.65 1.73 0.242 1.09 0.76 0.324 1.20 0.201 0.46 Cost of transportation to city 2.190 3.30 2.13 1.204 1.52 1.07 2.292 2.15 1.424 0.84 Insurance service 0.111 1.12 0.58 0.170 1.56 0.88 0.088 0.40 0.202 0.33 Human capital benchmark 0.379 0.63 0.36 0.802 1.20 0.73 0.591 0.61 0.995 0.67 Number of banks 0.081 0.24 0.13 0.146 0.43 0.22 0.554 0.81 0.654 0.53 Number bank services 0.149 0.38 0.20 0.260 0.64 0.34 0.703 0.91 0.804 0.54 Interceptb 0.162 0.55 0.351 0.86 Regression statistics smua 0.000 0.000 0.571 6.24 0.544 5.05 Cut point 1b 0.264 1.23 0.93 0.407 1.81 1.29 Cut point 2b 0.153 0.71 0.52 0.010 0.04 0.03 0.476 12.33 0.473 11.72 Number of observations 893 893 893 893 Log-likelihood 842.897 858.751 812.135 829.213 (continued on the next page) 222 The Rural Investment Climate Table I.5 Effect of Sampling Weights and Random Effect Speci�cation on Selected EICO Regression Results (continued ) A2: Nicaragua: Dependent variable is “Electricity� Random Effect Random Effect Standard Ordered Probit Standard Ordered Probit Ordered Probit Ordered Probit (unweighted) (weighted) (unweighted) (weighted) b t(u) t(a) b t(u) t(a) b t b t Enterprise characteristics Age of enterpriseb 0.054 1.21 1.36 0.078 1.63 1.84 0.026 0.54 0.033 0.64 Household-based enterprise 0.555 3.75 3.38 0.474 3.31 2.72 0.527 3.32 0.455 3.05 Services enterprise 0.098 0.85 0.77 0.072 0.53 0.49 0.129 1.05 0.088 0.65 Manufacturing, nonagricultural enterprise 0.213 1.41 1.35 0.268 1.60 1.44 0.191 1.17 0.250 1.30 Agricultural processing enterprise 0.132 1.01 0.96 0.107 0.76 0.69 0.109 0.79 0.081 0.52 Other production enterprise 0.113 0.50 0.42 0.132 0.59 0.48 0.108 0.44 0.146 0.60 Mixed enterprise 0.377 3.23 2.93 0.450 3.29 2.94 0.393 3.18 0.438 2.97 Community characteristics Percent households with electricity 0.925 5.74 3.68 0.883 4.93 3.32 1.227 4.86 1.129 2.79 Information technology 0.183 2.03 1.34 0.071 0.67 0.44 0.012 0.08 0.167 0.57 Human capital benchmark 2.693 4.63 3.26 2.725 4.25 3.13 2.556 2.82 2.598 1.72 Intercept 0.471 1.63 0.329 0.86 Regression statistics smua 0.000 0.000 0.535 7.48 0.559 5.92 Cut point 1b 0.387 1.74 1.34 0.299 1.32 1.00 Cut point 2b 0.239 1.08 0.79 0.300 1.31 0.97 0.702 14.58 0.681 13.41 Number of observations 892 892 892 892 Log-likelihood 817.17 818.71 793.29 792.34 Annex I. Econometric Analysis of RIC Survey Data 223 Table I.5 Effect of Sampling Weights and Random Effect Speci�cation on Selected EICO Regression Results (continued ) B1: Sri Lanka: Dependent variable is “Interest rate of loan� Random Effect Random Effect Standard Ordered Probit Standard Ordered Probit Ordered Probit Ordered Probit (unweighted) (weighted) (unweighted) (weighted) b t(u) t(a) b t(u) t(a) b t b t Enterprise characteristics Age of enterpriseb 0.027 0.76 0.72 0.041 0.94 0.85 0.022 0.58 0.010 0.19 Education of manager 0.015 1.28 1.20 0.001 0.10 0.08 0.019 1.49 0.022 1.32 Female manager 0.287 3.41 3.09 0.174 1.77 1.52 0.370 4.04 0.219 1.83 Sinhalese manager 0.349 3.39 2.26 0.733 4.80 3.70 0.295 2.29 0.623 2.42 Household-based enterprise 0.033 0.42 0.39 0.198 1.84 1.91 0.029 0.34 0.154 1.17 Services enterprise 0.228 2.22 2.38 0.326 2.44 2.77 0.266 2.42 0.378 2.76 Manufacturing, nonagricultural enterprise 0.040 0.46 0.49 0.157 1.35 1.61 0.113 1.19 0.239 1.83 Agricultural processing enterprise 0.220 1.09 1.08 0.307 1.41 1.25 0.335 1.54 0.353 1.26 Other production enterprise 0.038 0.14 0.22 0.148 0.65 0.81 0.236 0.74 0.348 1.16 Mixed enterprise 0.000 0.00 0.00 0.100 0.64 0.63 0.151 1.12 0.129 0.71 Community characteristics Time to nearest city 0.059 0.40 0.24 0.166 0.94 0.58 0.024 0.09 0.383 0.66 Cost of transportation to city 0.208 1.13 0.74 0.039 0.18 0.12 0.275 0.86 0.082 0.11 Insurance service 0.124 1.74 1.07 0.206 2.24 1.48 0.069 0.54 0.239 0.73 Human capital benchmark 2.088 4.41 2.79 1.718 2.68 1.88 2.052 2.35 2.269 0.95 Number of banks 0.917 2.85 1.93 1.795 4.52 3.26 0.843 1.55 2.271 1.73 Number bank services 0.369 1.28 0.80 1.177 3.27 2.09 0.228 0.48 1.379 1.20 Interceptb 0.799 2.52 1.553 2.29 Regression statistics smua 0.000 0.000 0.568 8.94 0.677 6.07 Cut point 1b 0.984 4.59 2.90 1.518 5.46 4.06 Cut point 2b 1.155 5.37 3.40 1.669 5.97 4.46 0.197 8.86 0.181 6.45 Cut point 3b 1.508 6.97 4.39 2.018 7.16 5.43 0.602 16.04 0.596 11.98 Cut point 4b 2.407 10.89 6.63 3.129 10.44 7.57 1.609 25.03 1.869 19.86 Number of observations 1146 1145 1145 1145 Log-likelihood 1500.05 1343.71 1457.94 1291.59 (continued on the next page) 224 The Rural Investment Climate Table I.5 Effect of Sampling Weights and Random Effect Speci�cation on Selected EICO Regression Results (continued ) B2: Sri Lanka: Dependent variable is “Electricity� Random Effect Random Effect Standard Ordered Probit Standard Ordered Probit Ordered Probit Ordered Probit (unweighted) (weighted) (unweighted) (weighted) b t(u) t(a) b t(u) t(a) b t b t Enterprise characteristics Age of enterpriseb 0.039 1.08 1.19 0.002 0.04 0.04 0.027 0.72 0.022 0.41 Education of manager 0.035 2.93 2.76 0.040 2.45 2.40 0.025 1.90 0.032 1.74 Female manager 0.051 0.61 0.62 0.181 1.73 1.51 0.048 0.54 0.211 1.81 Sinhalese manager 0.110 1.17 0.90 0.143 1.10 0.91 0.095 0.78 0.102 0.39 Household-based enterprise 0.228 2.95 2.63 0.350 3.24 3.01 0.279 3.27 0.369 2.92 Services enterprise 0.307 3.05 3.15 0.396 2.88 2.97 0.340 3.16 0.420 2.91 Manufactruing, nonagricultural enterprise 0.378 4.29 3.86 0.559 4.39 4.04 0.489 5.09 0.737 5.43 Agricultural processing enterprise 0.017 0.08 0.09 0.522 2.67 2.43 0.127 0.61 0.618 3.04 Other production enterprise 0.547 1.48 1.25 0.340 0.88 0.77 0.780 1.93 0.549 1.17 Mixed enterprise 0.334 2.86 2.47 0.347 2.12 1.82 0.327 2.50 0.345 1.87 Community characteristics Percent households 0.812 5.75 3.24 0.804 3.98 3.35 0.939 4.28 0.816 1.29 with electricity Information technology 0.048 0.47 0.50 0.163 1.36 1.06 0.112 0.58 0.203 0.59 Human capital benchmark 0.384 0.83 0.58 1.149 1.71 1.23 0.661 0.91 1.788 1.21 Interceptb 0.650 2.30 0.744 0.97 Regression statistics smua 0.000 0.000 0.484 8.67 0.526 7.12 Cut point 1b 0.424 2.05 1.81 0.577 1.87 2.00 Cut point 2b 0.102 0.49 0.44 0.262 0.85 0.90 0.358 12.39 0.356 7.97 Cut point 3b 0.252 1.22 1.07 0.153 0.50 0.52 0.749 18.31 0.824 12.19 Cut point 4b 0.853 4.10 3.68 0.786 2.50 2.85 1.400 24.08 1.521 16.92 Number of 1152 1152 1152 1152 observations Log-likelihood 1540.68 1518.46 1510.81 1483.24 Source: Authors’ calculation from RIC data. Note: a. “smu� is the standard deviation of the community random effect. b. The standard ordered probit and random effect ordered probit models use a different standardization, which is immaterial to the estimated values of slope parameters. Annex I. Econometric Analysis of RIC Survey Data 225 APPENDIXES TO ANNEX I. STATA PROCEDURES Appendix A1. OLS Estimation Let the dependent variable of the empirical model be called Y. Let X1 and X2 be the explanatory variables at the individual level and let Z1 and Z2 be the explanatory variables at the community level; of course, there may be more variables at each level, or there may be just one. Let WGT be the weight of an observation and let GRP indicate the community. Stata estimates this model with the statement: regress Y X1 X2 Z1 Z2 [pweight=WGT],cluster(GRP) Appendix A2. Fixed Effects Estimation In the following program statements, the weight WGT is standardized such that the sum of WGT over the sample equals the number of observations in the sample. This standardized weight is called wgt; note that Stata is case-sensitive and that WGT and wgt are distinct variables. quietly su WGT,meanonly g wgt=WGT/r(mean) The �xed effects estimates of b are found with the following few statements. g swgt=sqrt(wgt) egen wgt_g=total(wgt),by(GRP) foreach x of varlist Y X1 X2 { egen m`x'=total(`x'*wgt),by(GRP) quietly replace m`x'=m`x'/wgt_g g d`x'=swgt*(`x' - m`x') } reg dY dX1 dX2,noconst The estimate of the variance of the individual-speci�c disturbance n, namely s2, is found as follows. The n result is stored in a scalar named s2v. predict vtilhat,resid g vtilhat2=vtilhat*vtilhat g wgt2=wgt*wgt g wgt2stan=total(wgt*wgt/wgt_g) quietly su vtilhat2 scalar den=r(sum) quietly su wgt scalar den1=r(sum) quietly su wgt2stan scalar den2=r(sum) scalar s2v=num/(den1-den2) display "Estimated sigv2 = " s2v; The �xed effects m are found with the following statements. They appear as a new variable Muhat in the N database, having the same value for all observations in a given community. matrix b=e(b) g Muhat = mY - b[1,1]*mX1 - b[1,2]*mX2 In order to estimate the effect of Z, one must �rst estimate s2 : m matrix V=e(V) foreach x of varlist Y X1 X2 { quietly replace d`x'=d`x'*swgt 226 The Rural Investment Climate } matrix accum XWXstar = dX1 dX2,noconstant egen totwgt2j=total(wgt*wgt),by(GRP); g wgtterm = (wgt*wgt - (wgt*totwgt2j/totwgtj))/totwgtj; foreach x of varlist X1 X2 { quietly replace d`x'=d`x'/swgt quietly replace d`x'=d`x'*sqrt(s2v)*wgtterm egen t`x'=total(d`x'),by(GRP) } keep if GRP ~= GRP[_n-1] mkmat mX1 mX2,matrix(Xbar) mkmat tX1 tX2,matrix(Xt) g varmu=s2v*totwgt2j/(totwgtj*totwgtj) mkmat varmu,matrix(varmuv) matrix varmu=diag(varmuv)-Xbar*invsym(XWXstar)*Xt' - /* */ Xt*invsym(XWXstar)*Xbar'+Xbar*V*Xbar' scalar J_com = rowsof(varmu) g iota=1 mkmat Z1 Z2 iota,matrix(Z) mkmat Muhat,matrix(muhat) matrix lambda=muhat-Z*syminv(Z'*Z)*(Z'*muhat) matrix sll=lambda'*lambda scalar tvarmu=trace(varmu) scalar sigmu2=(sll[1,1]-tvarmu)/J_com display "Estimated sigmu2 = " sigmu2 The parameter g that measures the effect of Z on y is now quickly found: matrix SIG = sigmu2*I(J_occ) + varmu matrix invSIG=syminv(SIG) matrix bgls = syminv(Z'*invSIG*Z)*(Z'*invSIG*muhat) matrix vgls = syminv(Z'*invSIG*Z) matrix stdgls=vecdiag(vgls)' matrix tgls = bgls local j=1 while `j' <= rowsof(stdgls) { matrix stdgls[`j',1]=sqrt(stdgls[`j',1]) matrix tgls[`j',1]=tgls[`j',1]/stdgls[`j',1] local j=`j'+1 } display "Impact of Z:" matrix result=(bgls,stdgls,tgls) matrix colnames result = bAMEMIYA stdev tstat matrix list result Appendix A3. Random Effects Estimation In the following program code, the weight WGT is standardized such that the sum of WGT over the sample equals the number of observations in the sample. This standardized weight is called wgt; note that Stata is case-sensitive and that WGT and wgt are distinct variables. quietly su WGT,meanonly g wgt=WGT/r(mean) Annex I. Econometric Analysis of RIC Survey Data 227 The �rst task is to compute the variance s2 of the individual idiosyncratic disturbance nji. This estimator n derives from a �xed effects regression and so is similar to what was provided in Appendix A.2, with one improvement that helps to get an estimator of s2 more likely to be positive. The �rst step is to compute a m matrix labeled XQWQX, of which the trace is needed later on. During the process of the computation, all but the �rst observation in a group (or community) are removed. Therefore, the existing dataset is stored and subsequently retrieved when the computation is �nished. quietly save temptrash,replace foreach x of varlist X1 X2 { g w05`x'=`x'*sqrt(`wgt') g w`x'=`x'*`wgt' egen tw`x'=total(w`x'),by(`GRP') egen tww`x'=total(w`x'*`wgt'),by(`GRP') local w05x = "`w05x' w05`x'" local wx = "`wx' w`x'" local twx = "`twx' tw`x'" local twwx = "`twwx' tww`x'" } matrix accum XW1X = w05X1 w05X2,noconst matrix accum XW2X = wX1 wX2,noconst quietly drop if `GRP' == `GRP'[_n-1] mkmat twX1 twX2,matrix (ZWX) mkmat twwX1 twwX2,matrix(ZW2X) mkmat totwgt_g,matrix(ZWZ) matrix ZWZ = diag(ZWZ) matrix iZWZ = invsym(ZWZ) mkmat totwgt2_g,matrix(ZW2Z) matrix ZW2Z = diag(ZW2Z) matrix XQWQX = XW2X - (ZW2X')*iZWZ*ZWX - (ZWX')*iZWZ*ZW2X + /* */ (ZWX')*iZWZ*ZW2Z*iZWZ*ZWX use temptrash,clear At this point, the program proceeds with a regression on variables in deviation that generates residuals from which the estimate of s2 is computed. n g swgt=sqrt(wgt) egen wgt_g=total(wgt),by(GRP) foreach x of varlist Y X1 X2 { egen m`x'=total(`x'*wgt),by(GRP) quietly replace m`x'=m`x'/wgt_g g d`x'=swgt*(`x' - m`x') } reg dY dX1 dX2,noconst predict vtilhat,resid g vtilhat2=vtilhat*vtilhat g wgt2=wgt*wgt g wgt2stan=total(wgt*wgt/wgt_g) quietly su vtilhat2 scalar den=r(sum) quietly su wgt scalar den1=r(sum) quietly su wgt2stan 228 The Rural Investment Climate scalar den2=r(sum) scalar s2v=num/(den1-den2) display "Estimated sigv2 = " s2v; The second task is to compute the variance s2 of the community disturbance mj. In this program, we �rst m run an OLS regression to retrieve the residuals that are the basis of the estimator. regress Y X1 X2 Z1 Z2 [pw=WGT],cluster(GRP) predict uhat,resid egen sumwgtu=total(wgt*uhat),by(GRP) g sumwgtu2=sumwgtu*sumwgtu quietly su sumwgtu2,meanonly scalar num1=r(mean) g wgtu2=wgt*wgt*uyr*uyr quietly su wgtu2,meanonly scalar num2=r(mean) egen sumwgt=total(wgt),by(GRP) g sumwgt2=sumwgt*sumwgt quietly su sumwgt2,meanonly scalar den1=r(mean) g wgt2=wgt*wgt quietly su wgt2,meanonly scalar den2=r(mean) scalar sigmu2 = (num1-num2)/(den1-den2) display "Estimated sigmu2 = " sigmu2 The third step in the program is to compute the estimated parameters d. by GRP: g Nj = _N g tau2j = sumwgt*sigmu2 + s2v foreach x of varlist Y X1 X2 Z1 Z2 { egen tm`x'=total(`x'*wgt),by(GRP) g m`x'=tm`x'/sumwgt g s`x'=(`x' - (1-sqrt(s2v/tau2j)) * m`x')/sqrt(s2v) quietly replace tm`x'=0 if GRP == GRP[_n-1] } g intercpt = (1 - (1-sqrt(s2v/tau2j)) * 1)/sqrt(s2v) reg sY sX1 sX2 sZ1 sZ2 intercpt [pw=WGT],noconst drop tmlnwage - intercpt **Now do by hand, in order to get correct standard errors** display "** estimate deltaGLS **" g twgt1=sqrt(wgt/s2v) g twgt2=sqrt(sigmu2/(tau2j*s2v)) g intercpt = 1 foreach x of varlist Y X1 X2 Z1 Z2 intercpt { g s`x'=`x'*twgt1 egen tm`x'=total(`x'*wgt),by(GRP) quietly replace tm`x'=0 if GRP == GRP[_n-1] quietly replace tm`x'=tm`x'*twgt2 } matrix accum tbatb1 = sY sX1 sX2 sZ1 sZ2 sintercpt,noconst Annex I. Econometric Analysis of RIC Survey Data 229 matrix accum tbatb2 = tmY tmX1 tmX2 tmZ1 tmZ2 tmintercpt,noconst scalar tK=rowsof(tbatb1) matrix tat=tbatb1[2..tK,2..tK] - tbatb2[2..tK,2..tK] matrix tay=tbatb1[2..tK,1] - tbatb2[2..tK,1] matrix itat = invsym(tat) matrix deltaGLS = itat*tay drop twgt1 twgt2 slnwage - tmintercpt The fourth step in the program secures the estimated covariance matrix of the estimated parameters d. display "** estimate Var(deltaGLS) **" g twgt1=wgt/sqrt(s2v) g twgt2=sqrt(sigmu2/(tau2j*s2v)) foreach x of varlist X1 X2 Z1 Z2 intercpt { g s`x'=`x'*twgt1 egen t1`x'=total(`x'*wgt),by(GRP) quietly replace t1`x'=0 if GRP == GRP[_n-1] quietly replace t1`x'=t1`x'*twgt2 egen t2`x'=total(`x'*wgt*wgt),by(GRP) quietly replace t2`x'=0 if GRP == GRP[_n-1] quietly replace t2`x'=t2`x'*twgt2 } matrix accum tbatb1 = sX1 sX2 sZ1 sZ2 sintercpt,noconst matrix accum tbatb2 = t1X1 t1X2 t1Z1 t1Z2 t1intercpt /* */ t2X1 t2X2 t2Z1 t2Z2 t2intercpt,noconst scalar tK=rowsof(tbatb1) matrix tavat=tbatb1 - tbatb2[(tK+1)..(2*tK),1..tK] /* */ - tbatb2[1..tK,(tK+1)..(2*tK)] egen twgt3a=total(wgt*wgt),by(GRP) g twgt3=sqrt(sigmu2 + (sigmu2*sigmu2*twgt3a/s2v))/tau2j foreach x of varlist X1 X2 Z1 Z2 intercpt { quietly replace t1`x'=t1`x'*twgt3/twgt2 } matrix accum tbatb3 = t1X1 t1X2 t1Z1 t1Z2 t1intercpt,noconst matrix tavat = tavat + tbatb3 matrix VardeltaGLS = itat*tavat*itat Finally, report the results: the estimated parameters, their standard errors, and the t-statistics. matrix stdev=vecdiag(VardeltaGLS)' matrix tstat=J(tK,1,0) local j=1 while `j' <= tK { matrix stdev[`j',1]=sqrt(stdev[`j',1]) matrix tstat[`j',1]=deltaGLS[`j',1]/stdev[`j',1] local j=`j'+1 } matrix result=(deltaGLS,stdev,tstat) matrix colnames result = deltaGLS stdev tstat matrix list result drop sumwgtu - twgt3 230 The Rural Investment Climate Appendix A4. Weighted Random Effect Probit Estimation The weighted random effect probit model is estimated by maximum likelihood. In Stata, this means the re- searcher must provide a program called by the Stata’s built-in ml procedure. In the following Stata code, this program is called wrep. This ml procedure requires some data set-up, which is provided below. The program wrep de�nes a temporary variable `mu’ which is initialized at 0 but subsequently �lled with random draws of a standard normal distribution. These draws must be reused for every iteration of the maximum likelihood search routine. For that reason, the random draws are inserted into a matrix ZZ prior to the start of the maximum likelihood estimation. Furthermore, the program uses an antithetical sim- ulation method, in which random draws of `mu' are reused in a different form to raise the precision of the simulated probability: here, the sign of the random draws are switched. At the end of the program, the vari- ables gradient1 and gradient2 are computed for use in computations when the estimation is completed. *** De�ne the weighted RE probit likelihood function ***; capture program drop wrep; program de�ne wrep; version 9.0; args todo b lnf g; tempvar xb lntheta theta wgtmean wgt mu xbmu p; tempvar lnLgi lnLg elnLg Lg lnfgi arg g1 g2; mleval `xb' = `b',eq(1); mleval `theta' = `b',eq(2); /******************************************************* * ML_y1 is the categorical variable, with two outcomes * ML_y2 is the group indicator * ML_y3 is the weight of each individual * lntheta is the log-stdev of mu * theta will be the standard deviation of mu *******************************************************/ quietly su `theta'; if (r(min) <= 0) {; scalar `lnf' = .; exit; }; egen `wgtmean'=mean($ML_y3),by($ML_y2); g `wgt'=$ML_y3/`wgtmean'; quietly g double `xbmu'=0; quietly g double `mu'=0; quietly g double `p'=0; quietly g double `Lg'=0; quietly g double `lnLgi'=0; quietly g double `elnLg'=0; quietly g double `arg'=0; quietly g double `g1'=0; quietly g double `g2'=0; local jr=1; local JR = colsof(ZZ); local Ngroups = rowsof(ZZ); while `jr' <= `JR' {; *display "jr = `jr' "; local jg=1; Annex I. Econometric Analysis of RIC Survey Data 231 quietly replace `mu'=0; while `jg' <= `Ngroups' {; quietly replace `mu' = `mu' + ($ML_y2 == `jg')*ZZ[`jg',`jr']; local jg=`jg'+1; }; **Work the regular draw**; quietly replace `xbmu' = `xb'+`theta'*`mu'; quietly replace `p' = normal(`xbmu'); quietly replace `p' = 1-`p' if $ML_y1 == 0; quietly su `p'; if (r(min) == 0) {; scalar `lnf' = .; exit; }; quietly replace `lnLgi' = `wgt'*ln(`p'); quietly replace `lnLgi' = -999 if `lnLgi' == .; quietly egen `lnLg' = total(`lnLgi'),by($ML_y2); quietly replace `elnLg' = exp(`lnLg'); quietly su `elnLg'; if (r(min) == 0) {; scalar `lnf' = .; exit; }; quietly replace `Lg' = `Lg' + `elnLg'; quietly replace `arg' = (2*$ML_y1 - 1) * normalden(`xbmu') / `p'; quietly replace `g1' = `g1' + `elnLg'*`arg'; quietly replace `g2' = `g2' + `elnLg'*`arg'*`mu'; drop `lnLg'; **Work the antithetical draw**; quietly replace `xbmu' = `xb'-`theta'*`mu'; quietly replace `p' = normal(`xbmu'); quietly replace `p' = 1-`p' if $ML_y1 == 0; quietly su `p'; if (r(min) == 0) {; scalar `lnf' = .; exit; }; quietly replace `lnLgi' = `wgt'*ln(`p'); quietly replace `lnLgi' = -999 if `lnLgi' == .; quietly egen `lnLg' = total(`lnLgi'),by($ML_y2); quietly replace `elnLg' = exp(`lnLg'); quietly su `elnLg'; if (r(min) == 0) {; scalar `lnf' = .; exit; }; quietly replace `Lg' = `Lg' + `elnLg'; quietly replace `arg' = (2*$ML_y1 - 1) * normalden(`xbmu') / `p'; quietly replace `g1' = `g1' + `elnLg'*`arg'; quietly replace `g2' = `g2' - `elnLg'*`arg'*`mu'; drop `lnLg'; local jr=`jr'+1; }; quietly replace `Lg' = `Lg'/(2*`JR'); quietly g double `lnfgi' = `wgtmean'*ln(`Lg') * ($ML_y2 ~= $ML_y2[_n-1]); mlsum `lnf' = `lnfgi'; if (`todo'==0 | `lnf' >= .) exit; tempname d1 d2; quietly replace gradient1 = $ML_y3*`g1'/(`Lg'*(2*`JR')); quietly replace gradient2 = $ML_y3*`g2'/(`Lg'*(2*`JR')); mlvecsum `lnf' `d1' = gradient1, eq(1); mlvecsum `lnf' `d2' = gradient2, eq(2); matrix `g' = (`d1',`d2'); end ; 232 The Rural Investment Climate The following program wreprobit is a driver for weighted random effect probit estimation. After some initializations, it creates sequential group (which, in the case of RIC data, are community) numbers. It per- mits the user to specify a few comparison models such as ordinary probit, weighted probit (without ran- dom effects), and unweighted random effect probit. The latter is used to generate starting values for the ML estimation of the weighted random effect probit model. The next steps are: generating a matrix ZZ of standard normal random draws, executing the ml proce- dure with a specific reference to the wrep program provided above, and computing corrected standard errors and t-statistics. Relative to the discussion in Sections 3.3 and 3.4, Stata’s matrix V stands for the vari- ance matrix V1; Stata’s matrix G stands for V2 1; and Stata’s matrix Varb stands for V3. The program only reports standard errors based on V3, but alternatives are easily inserted. At the very bottom of the program, commented code briefly explains the manner in which wreprobit should be called by the user’s do-�le. *** De�ne weighted RE probit as a regression command ***; capture program drop wreprobit; program de�ne wreprobit; version 9.0; syntax varlist(min=3)[,trace simnum(integer 25) compareUOP compareWOP compareURP trialrun]; gettoken y rest : 0 /*,parse(" ,")*/; gettoken g rest : rest ,parse(" "); gettoken wgt rest : rest ,parse(" "); gettoken xzvar rest : rest ,parse(","); display "Dependent variable is `y'"; display "Group variable is `g'"; display "Weight variable is `wgt'"; display "Explanatory variables are `xzvar'"; display "Number of draws used by simulator = `simnum' (used twice: +/-)"; display " "; ** rescale weight to an average of 1 **; quietly su `wgt',meanonly; g `wgt'_stan=`wgt'/r(mean); label variable `wgt'_stan "Scaled weight at individual level"; display "Variation in weights rescaled to an average of 1, at individual level"; su `wgt'_stan,detail; scalar stdvwgt = r(sd); ** Determine the number of groups **; sort `g'; quietly g `g'_nummer=1 if _n==1; quietly replace `g'_nummer=`g'_nummer[_n-1] + (`g' ~= `g'[_n-1]) if _n > 1; label variable `g'_nummer "Numeric group number"; di " "; quietly su `g'_nummer; local Ngroups = r(max); display "Number of groups = `Ngroups'"; tempvar gsize; Annex I. Econometric Analysis of RIC Survey Data 233 sort `g'_nummer; by `g'_nummer: g `gsize' = _N; quietly su `gsize'; display "Smallest group size = " r(min); display "Average group size = " r(mean); display "Largest group size = " r(max); di " "; *di "Tabulation of group sizes for each group"; *tab `g'_nummer; di " "; ** Estimate weighted ordinary probit as a trial run for the speci�cation **; if "`trialrun'" ~= "" {; display " "; display " Estimate weighted ordinary probit as a trial run for the speci�- cation "; probit `y' `xzvar' [pw=`wgt'_stan]; }; else {; /*begin of nontrial segment*/ ** Estimate unweighted models for comparison, as requested through options **; if "`compareUOP'" ~= "" | "`trialrun'" ~= "" {; display " "; display ">> Unweighted Ordinary Probit Model, for comparison only <<"; probit `y' `xzvar'; display ">> Note: without applying weights or random effects, estimates may be biased and inconsistent"; display ">> End of comparison <<"; display " "; }; ** Estimate weighted models for comparison, as requested through options **; if "`compareWOP'" ~= "" {; display ">> Weighted Ordinary Probit Model, for comparison only <<"; probit `y' `xzvar' [pw=`wgt'_stan]; display ">> End of comparison <<"; display " "; }; else {; quietly probit `y' `xzvar' [pw=`wgt'_stan]; }; matrix b=get(_b)'; ** Estimate weighted models for comparison, as requested through options **; if "`compareURP'" ~= "" {; display ">> Unweighted Random Effects Probit Model, for comparison only <<"; xtprobit `y' `xzvar',re i(`g'_nummer); display ">> End of comparison <<"; display " "; }; else {; quietly xtprobit `y' `xzvar',re i(`g'_nummer); 234 The Rural Investment Climate }; matrix bre=get(_b)'; scalar thet = bre[rowsof(bre),1; if thet < 0.05 {; matrix theta=0.05; }; else {; matrix theta=thet; }; matrix bb=(b*0.95)\theta; matrix rownames bb = `xzvar' intercept stdevmu; matrix list bb; preserve; ** Compute a matrix of random draws for simulation **; quietly drop if _n > `Ngroups'; g id=_n; set seed 19901; matrix ZZ=J(`Ngroups',1,0); local jr=1; while `jr' <= `simnum' {; g z=invnorm(uniform()); mkmat z; matrix ZZ=ZZ,z; drop z; local jr=`jr'+1; }; matrix ZZ=ZZ[1...,2...]; *matrix list ZZ; restore; if stdvwgt > 0 {; ** Start ML estimation of random effect probit model **; di "** Start ML estimation of weighted random effect probit model **"; g gradient1=0; g gradient2=0; ml model d1 wrep (eq1:`y' `g'_nummer `wgt'_stan=`xzvar') (eq2:); ml init bb,copy; ml search; if "`trace'" ~= "" {; ml max,trace grad; }; else {; ml max; }; ** organize the outerproduct of the gradient for computation of correct stdev of b **; local ggvar = ""; local j = 0; foreach aa of varlist `xzvar' {; Annex I. Econometric Analysis of RIC Survey Data 235 local j = `j' + 1; g ggg`j' = gradient1*`aa'; local ggvar = "`ggvar' ggg`j'"; display "Adding to local ggvar: `j' equals `aa'"; }; g gggintercept = gradient1; g gggtheta = gradient2; local ggvar = "`ggvar' gggint gggtheta"; display "Complete local ggvar equals `ggvar'"; matrix accum G = `ggvar',noconst; matrix b_est = e(b)'; matrix V = e(V); matrix list G; matrix list V; matrix VARb = V * G * V; matrix SDb = cholesky(diag(vecdiag(VARb))); matrix tstatb = syminv(SDb)*b_est; matrix SDb = vecdiag(SDb)'; matrix result = b_est,SDb,tstatb; matrix colnames result = b stdev tstat; matrix list result; drop ggg*; }; else {; di "** Weights are constant across observations"; di "** Use random effect probit without weights"; xtprobit `y' `xzvar',re i(`g'_nummer); }; drop `g'_nummer `wgt'_stan gradient*; }; /*end of nontrial segment*/ end; /*** Information on wreprobit routine ************************************************* * Depvar: y * Group indicator: group * Weight: weight * Explanatory variables, individual level: x1a x1b x1c * Explanatory variables, group level: x2a x2b * * The call to the weighted �xed effect regression command must �rst list * the depvar groupvar and weightvar, and then the explanatory variables. * The explanatory variables may be in any order, mixing individual level and * group level variables in random order. The program will sort them into * a set of individual level variables and a set of group level variables. * The option 'simnum' must specify an integer, indicating the number * of random draws that the simulated random probit estimation uses. A higher * number yields greater precision but lengthens the computation time more or * less proportionately. Default is 25. * The option 'trace' asks for more detail during the iterative search. * The option 'compareUOP' will run a unweighted ordinary probit model for comparison. * The option 'compareWOP' will run a weighted ordinary probit model for comparison. * The option 'compareURP' will run a unweighted random effects probit model for comparison. * The option 'trialrun' will only estimate a weighted ordinary probit model to * test the speci�cation of a preliminary model. 236 The Rural Investment Climate * * The output will appear to yield estimates of two equations. The �rst equation * represents the parameters of the explanatory variables (and intercept). The * second equation shows the estimate of the standard deviation of the random effect * (under the label of '_cons'). * The code of the program is simply modi�ed to estimate the natural log of the * standard deviation of the random effect (log-stdev) rather than the standard * deviation itself. An estimate of the standard deviation (stdevmu) * is obtained by taking the antilog of the log-stdev. The standard * error of stdevmu is found by multiplying the standard error of the * estimate of the log-stdevmu with the estimated value of stdevmu. * * Examples of how to call the program: * * wreprobit y group weight x1a x2a,trace simnum(6); * wreprobit y group weight x1a x2a x1b x1c x2b,compareUOP compareWOP; * wreprobit y group weight x1a x2a x1b x1c x2b,trialrun; * ***************************************************************************/; It should be noted that wreprobit does not tolerate missing values. Observations with missing values must be removed before calling wreprobit. Moreover, whereas many Stata regression routines will auto- matically remove variables that are perfectly collinear, such incidences will generate a fatal error in wrepro- bit. For this reason, the user ought to inspect his model speci�cation with the trialrun option, which will only estimate an ordinary probit model. If a variable is omitted in this estimation, the user must examine the speci�cation of the regression model more carefully. Appendix A5. Weighted Random Effect Ordered Probit Estimation The estimation of the random effect ordered probit model proceeds along a line similar to that of the ran- dom effect probit model. The program wreop de�nes the likelihood function for the random effect ordered probit model for use by Stata’s ml routine. The program wreoprobit is the driver program that sets up the estimation and delivers the estimates. For information on how to use this program in a Stata do-�le, see the instructions at the end of the text. It should be emphasized that this version of the wreoprobit routine estimates an ordered probit where the dependent variable takes on �ve responses. The program does not automatically adjust to the number of distinct responses in the dependent variable. If the number of responses is four instead of �ve, for exam- ple, the program must be adjusted accordingly by omitting the estimation of alpha4 and adjusting com- parisons of $ML_y1 with potential response values. *** De�ne the weighted RE ordered probit likelihood function ***; capture program drop wreop; program de�ne wreop; version 9.0; args todo b lnf g; tempvar xb lntheta theta alpha2 alpha3 alpha4 wgtmean wgt mu xbmu p; tempvar lnLgi lnLg elnLg Lg lnfgi arghi arglo g1 g2 g3 g4 g5; mleval `xb' = `b',eq(1); mleval `theta' = `b',eq(2); mleval `alpha2' = `b',eq(3); mleval `alpha3' = `b',eq(4); Annex I. Econometric Analysis of RIC Survey Data 237 mleval `alpha4' = `b',eq(5); /******************************************************* * ML_y1 is the categorical variable, with two outcomes * ML_y2 is the group indicator * ML_y3 is the weight of each individual * lntheta is the log-stdev of mu * theta will be the standard deviation of mu * alpha_j are cutpoints * The program is designed for number of alternatives equal to 5, * ranging from 0,...,4 * By constraint, alpha_0 = -inf * alpha_1 = 0 * alpha_2 .. alpha_4 are estimated parameters * alpha_5 = +inf * If y == j, then alpha_j < ystar <= alpha_{j+1} *******************************************************/ ** Check consistency of parameters provided **; quietly su `theta'; if (r(min) <= 0) {; scalar `lnf' = .; exit; }; quietly su `alpha2'; scalar m2=r(min); if m2 <= 0 {; scalar `lnf' = .; exit; }; quietly su `alpha3'; scalar m3=r(min); if m3 <= m2 {; scalar `lnf' = .; exit; }; quietly su `alpha4'; scalar m4=r(min); if m4 <= m3 {; scalar `lnf' = .; exit; }; egen `wgtmean'=mean($ML_y3),by($ML_y2); g `wgt'=$ML_y3/`wgtmean'; quietly g double `xbmu'=0; quietly g double `mu'=0; quietly g double `arglo'=0; quietly g double `arghi'=0; quietly g double `p'=0; quietly g double `Lg'=0; quietly g double `lnLgi'=0; quietly g double `elnLg'=0; quietly g double `g1'=0; quietly g double `g2'=0; quietly g double `g3'=0; quietly g double `g4'=0; quietly g double `g5'=0; local jr=1; local JR = colsof(ZZ); local Ngroups = rowsof(ZZ); while `jr' <= `JR' {; *display "jr = `jr' "; local jg=1; quietly replace `mu'=0; while `jg' <= `Ngroups' {; 238 The Rural Investment Climate quietly replace `mu' = `mu' + ($ML_y2 == `jg')*ZZ[`jg',`jr']; local jg=`jg'+1; }; **Work the regular draw**; quietly replace `xbmu' = `xb'+`theta'*`mu'; quietly replace `arglo' = -`xbmu' + ($ML_y1 == 2)*`alpha2' + ($ML_y1 == 3)*`alpha3' + ($ML_y1 == 4)*`alpha4'; quietly replace `arghi' = -`xbmu' + ($ML_y1 == 1)*`alpha2' + ($ML_y1 == 2)*`alpha3' + ($ML_y1 == 3)*`alpha4'; quietly replace `p' = ($ML_y1 == 4) + ($ML_y1 < 4) * normal(`arghi') - ($ML_y1 > 0) * normal(`arglo'); quietly su `p'; if (r(min) <= 0) {; scalar `lnf' = .; exit; }; quietly replace `lnLgi' = `wgt'*ln(`p'); quietly replace `lnLgi' = -999 if `lnLgi' == .; quietly egen `lnLg' = total(`lnLgi'),by($ML_y2); quietly replace `elnLg' = exp(`lnLg'); quietly su `elnLg'; if (r(min) == 0) {; scalar `lnf' = .; exit; }; quietly replace `Lg' = `Lg' + `elnLg'; if (`todo'~= 0) {; quietly replace `arglo' = (($ML_y1 > 0) * normalden(`arglo')) / `p'; quietly replace `arghi' = (($ML_y1 < 4) * normalden(`arghi')) / `p'; quietly replace `g1' = `g1' - `elnLg'*(($ML_y1 < 4)*`arghi' - ($ML_y1 > 0)*`arglo'); quietly replace `g2' = `g2' - `elnLg'*(($ML_y1 < 4)*`arghi' - ($ML_y1 > 0)*`arglo')*`mu'; quietly replace `g3' = `g3' + `elnLg'*(($ML_y1 == 1)*`arghi' - ($ML_y1 == 2)*`arglo'); quietly replace `g4' = `g4' + `elnLg'*(($ML_y1 == 2)*`arghi' - ($ML_y1 == 3)*`arglo'); quietly replace `g5' = `g5' + `elnLg'*(($ML_y1 == 3)*`arghi' - ($ML_y1 == 4)*`arglo'); }; drop `lnLg'; **Work the antithetical draw**; quietly replace `xbmu' = `xb'-`theta'*`mu'; quietly replace `arglo' = -`xbmu' + ($ML_y1 == 2)*`alpha2' + ($ML_y1 == 3)*`alpha3' + ($ML_y1 == 4)*`alpha4'; quietly replace `arghi' = -`xbmu' + ($ML_y1 == 1)*`alpha2' + ($ML_y1 == 2)*`alpha3' + ($ML_y1 == 3)*`alpha4'; quietly replace `p' = ($ML_y1 == 4) + ($ML_y1 < 4) * normal(`arghi') - ($ML_y1 > 0) * normal(`arglo'); quietly su `p'; if (r(min) == 0) {; scalar `lnf' = .; exit; }; quietly replace `lnLgi' = `wgt'*ln(`p'); quietly replace `lnLgi' = -999 if `lnLgi' == .; quietly egen `lnLg' = total(`lnLgi'),by($ML_y2); quietly replace `elnLg' = exp(`lnLg'); quietly su `elnLg'; if (r(min) == 0) {; scalar `lnf' = .; exit; }; quietly replace `Lg' = `Lg' + `elnLg'; if (`todo'~= 0) {; quietly replace `arglo' = (($ML_y1 > 0) * normalden(`arglo')) / `p'; quietly replace `arghi' = (($ML_y1 < 4) * normalden(`arghi')) / `p'; Annex I. Econometric Analysis of RIC Survey Data 239 quietly replace `g1' = `g1' - `elnLg'*(($ML_y1 < 4)*`arghi' - ($ML_y1 > 0)*`arglo'); quietly replace `g2' = `g2' + `elnLg'*(($ML_y1 < 4)*`arghi' - ($ML_y1 > 0)*`arglo')*`mu'; quietly replace `g3' = `g3' + `elnLg'*(($ML_y1 == 1)*`arghi' - ($ML_y1 == 2)*`arglo'); quietly replace `g4' = `g4' + `elnLg'*(($ML_y1 == 2)*`arghi' - ($ML_y1 == 3)*`arglo'); quietly replace `g5' = `g5' + `elnLg'*(($ML_y1 == 3)*`arghi' - ($ML_y1 == 4)*`arglo'); }; drop `lnLg'; local jr=`jr'+1; }; quietly replace `Lg' = `Lg'/(2*`JR'); quietly g double `lnfgi' = `wgtmean'*ln(`Lg') * ($ML_y2 ~= $ML_y2[_n-1]); *su `lnfgi' `g1' `g2' `g3' `g4' `g5'; mlsum `lnf' = `lnfgi'; if (`todo'==0 | `lnf' >= .) exit; tempname d1 d2 d3 d4 d5; **Note: $ML_y3 contains the full weight**; **Note: gradient1 and gradient2 are variables made available in wreprobit, to help compute the correct stdev**; quietly replace gradient1 = $ML_y3*`g1'/(`Lg'*(2*`JR')); quietly replace gradient2 = $ML_y3*`g2'/(`Lg'*(2*`JR')); quietly replace gradient3 = $ML_y3*`g3'/(`Lg'*(2*`JR')); quietly replace gradient4 = $ML_y3*`g4'/(`Lg'*(2*`JR')); quietly replace gradient5 = $ML_y3*`g5'/(`Lg'*(2*`JR')); mlvecsum `lnf' `d1' = gradient1, eq(1); mlvecsum `lnf' `d2' = gradient2, eq(2); mlvecsum `lnf' `d3' = gradient3, eq(3); mlvecsum `lnf' `d4' = gradient4, eq(4); mlvecsum `lnf' `d5' = gradient5, eq(5); matrix `g' = (`d1',`d2',`d3',`d4',`d5'); end ; *** De�ne weighted RE ordered probit as a regression command ***; capture program drop wreoprobit; program de�ne wreoprobit; version 9.0; syntax varlist(min=3)[,trace simnum(integer 25) compareUOP compareWOP]; gettoken y rest : 0 /*,parse(" ,")*/; gettoken g rest : rest ,parse(" "); gettoken wgt rest : rest ,parse(" "); 240 The Rural Investment Climate gettoken xzvar rest : rest ,parse(","); display "Dependent variable is `y'"; display "Group variable is `g'"; display "Weight variable is `wgt'"; display "Explanatory variables are `xzvar'"; display "Number of draws used by simulator = `simnum' (used twice: +/-)"; display " "; ** rescale weight to an average of 1 **; quietly su `wgt',meanonly; g `wgt'_stan=`wgt'/r(mean); label variable `wgt'_stan "Scaled weight at individual level"; display "Variation in weights rescaled to an average of 1, at individual level"; su `wgt'_stan,detail; ** Estimate unweighted models for comparison, as requested through options **; if "`compareUOP'" ~= "" {; display " "; display ">> Unweighted Ordinary Probit Model, for comparison only <<"; oprobit `y' `xzvar'; display ">> Note: without applying weights or random effects, estimates may be biased and inconsistent"; display ">> End of comparison <<"; display " "; }; ** get starting values from a weighted ordinary probit **; if "`compareWOP'" ~= "" {; display ">> Weighted Ordinary Probit Model, for comparison only <<"; oprobit `y' `xzvar' [pw=`wgt'_stan]; display ">> End of comparison <<"; display " "; }; else {; quietly oprobit `y' `xzvar' [pw=`wgt'_stan]; }; matrix b=get(_b)'; scalar bK=rowsof(b); ** oprobit sets the intercept to zero and estimates all cutpoints **; ** this program estimates the intercept and sets the �rst cutpoint to zero **; matrix b_b = b[1..(bK-3),1]; matrix b_alpha = b[(bK-2)..bK,1] - (b_b[bK-3,1]*J(3,1,1)); matrix theta=0.1; matrix bb=(b_b*0.95)\theta\b_alpha; matrix rownames bb = `xzvar' intercept stdevmu a2 a3 a4; matrix list bb; ** Determine the number of groups **; sort `g'; Annex I. Econometric Analysis of RIC Survey Data 241 quietly g `g'_nummer=1 if _n==1; quietly replace `g'_nummer=`g'_nummer[_n-1] + (`g' ~= `g'[_n-1]) if _n > 1; label variable `g'_nummer "Numeric group number"; di " "; quietly su `g'_nummer; local Ngroups = r(max); display "Number of groups = `Ngroups'"; di " "; *di "Tabulation of group sizes for each group"; *tab `g'_nummer; di " "; preserve; ** Compute a matrix of random draws for simulation **; quietly drop if _n > `Ngroups'; g id=_n; set seed 19901; matrix ZZ=J(`Ngroups',1,0); local jr=1; while `jr' <= `simnum' {; g z=invnorm(uniform()); mkmat z; matrix ZZ=ZZ,z; drop z; local jr=`jr'+1; }; matrix ZZ=ZZ[1...,2...]; *matrix list ZZ; restore; ** Start ML estimation of random effect probit model **; di "** Start ML estimation of weighted random effect probit model **"; g gradient1=0; g gradient2=0; g gradient3=0; g gradient4=0; g gradient5=0; ml model d1 wreop (eq1:`y' `g'_nummer `wgt'_stan=`xzvar') (eq2:) (eq3:) (eq4:) (eq5:); ml init bb,copy; ml search; if "`trace'" ~= "" {; ml max,trace grad; }; else {; ml max; }; ** organize the outerproduct of the gradient for computation of correct stdev of b **; 242 The Rural Investment Climate local ggvar = ""; local j = 0; foreach aa of varlist `xzvar' {; local j = `j' + 1; g ggg`j' = gradient1*`aa'; local ggvar = "`ggvar' ggg`j'"; display "Adding to local ggvar: `j' equals `aa'"; }; g gggintercept = gradient1; g gggtheta = gradient2; g gggcut2 = gradient3; g gggcut3 = gradient4; g gggcut4 = gradient5; local ggvar = "`ggvar' gggint gggtheta gggcut2 gggcut3 gggcut4"; display "Complete local ggvar equals `ggvar'"; matrix accum G = `ggvar',noconst; matrix b_est = e(b)'; matrix V = e(V); matrix list G; matrix list V; matrix VARb = V * G * V; matrix SDb = cholesky(diag(vecdiag(VARb))); matrix tstatb = syminv(SDb)*b_est; matrix SDb = vecdiag(SDb)'; matrix result = b_est,SDb,tstatb; matrix colnames result = b stdev tstat; matrix list result; drop ggg*; drop `g'_nummer `wgt'_stan gradient*; end; /** Information on wreoprobit routine ********************************************** * Depvar: y * Group indicator: group * Weight: weight * Explanatory variables, individual level: x1a x1b x1c * Explanatory variables, group level: x2a x2b * * The call to the weighted random effect ordered probit command wreoprobit must �rst * listthe depvar groupvar and weightvar, and then the explanatory variables. * The explanatory variables may be in any order, mixing individual level and * group level variables in random order. The program will sort them into * a set of individual level variables and a set of group level variables. * The option 'simnum' must specify an integer, indicating the number * of random draws that the simulated random probit estimation uses. A higher * number yields greater precision but lengthens the computation time more or * less proportionately. Default is 25. * The option 'trace' asks for more detail during the iterative search. * The option 'compareUOP' will run a unweighted ordered probit model for comparison. * The option 'compareWOP' will run a weighted ordered probit model for comparison. * * NOTE: the program is set up for a y variable with FIVE categories. It is * easily modi�ed to handle a different number of categories, but at the moment Annex I. Econometric Analysis of RIC Survey Data 243 * the program does not automatically accommodate an arbitrary number of * categories. * * The output will appear to yield estimates of �ve equations. The �rst equation * represents the parameters of the explanatory variables (and intercept). The * second equation shows the estimate of the standard deviation of the random effect * (under the label of '_cons'). The third, fourth, and �fth equation shows the * estimated second, third, and fourth cutpoints. The �rst cutpoint is standardized * to be equal to 0 a priori. * * The code of the program is simply modi�ed to estimate the natural log of the * standard deviation of the random effect (log-stdev) rather than the standard * deviation itself. An estimate of the standard deviation (stdevmu) * is obtained by taking the antilog of the log-stdev. The standard * error of stdevmu is found by multiplying the standard error of the * estimate of the log-stdevmu with the estimated value of stdevmu. * * Examples of how to call the program: * * wreoprobit y group weight x1a x2a,trace simnum(6); * wreoprobit y group weight x1a x2a x1b x1c x2b,compareUOP compareWOP; * *************************************************************************************/ Annex J. Employment and Income Estimates in the Surveyed Communities Table J.1 Nicaraguan Communities Unweighted Communities Entire Sample Smaller Larger (n 98 communitites) (n 46) (n 48) 1. Employment Shares Total rural nonfarm 89% 85% 92% Total rural nonfarm self-employed 55% 53% 58% __ Total wage workers l1 41% 42% 42% __ Agriculture l2 30% 40% 19% Labor force participation ratios 74% 76% 72% 2. Income Shares __ Rural nonfarm l3 89% 89% 92% __ Agriculture l4 11% 11% 8% 3. Average Per Capita and Per Worker Incomes ($) __ Gross household income/capita l5 490 388 571 __ Household “earned� income/capita l6 395 317 482 __ Household earnings/cap. fm RNF activity l7 449 354 534 Average per rural nonfarm worker 1048 839 1272 Source: RIC Survey, Nicaragua. __ l1 includes part-time and seasonal farm labor. __ l2 includes all household members engaged in agricultural operations; viz. full-time self-employed, part-time, and seasonal workers. l3 ∑ (enterprise income wage earnings/HH “earned� income [cf. ff 6 below]) * 100; N.B., a small portion of wage earnings likely derives __ from agriculture, thus the estimate contains moderate upwards bias. __ l4 possible moderate downwards bias — cf. ff 3 above. l5 = ∑ (enterprise & farm incomes wage earnings other cash inflows [e.g. interest, remittances, etc.])/HH size. __ l6 = ∑ (enterprise & farm incomes wage earnings)/HH size. __ l7 ∑ (enterprise income and wage earnings); n.b. a small portion of wage earnings likely derives from agriculture, thus the estimate contains __ moderate upwards bias. 245 246 The Rural Investment Climate Table J.2 Sri Lankan Communities Unweighted Communities Entire Sample Smaller Larger (n 147 communitites) (n 73) (n 73) 1. Employment Shares Total rural nonfarm 86% 81% 92% Total rural nonfarm self-employed 42% 40% 44% __ Total wage workers l1 49% 46% 52% __ Agriculture l2 35% 42% 27% Labor force participation ratios 62% 63% 61% 2. Income Shares __ Rural nonfarm l3 74% 69% 79% __ Agriculture l4 26% 31% 21% 3. Average Per Capita and Per Worker Incomes ($) __ Gross household income/capita l5 406 369 448 __ Household “earned� income/capita l6 190 172 210 __ Household earnings/cap. fm RNF Activity l7 259 252 268 Average per rural nonfarm worker 518 484 552 Source: RIC Survey, Sri Lanka. __ l1 includes part-time and seasonal farm labor. __ l2 includes all household members engaged in agricultural operations; viz. full-time self-employed, part-time, and seasonal workers. l3 ∑ (enterprise income wage earnings/HH “earned� income [cf. ff 6 below]) * 100 ; N.B., a small portion of wage earnings likely __ derives from agriculture, thus the estimate contains moderate upwards bias. __ l4 possible moderate downwards bias — cf. ff 3 above. l5 ∑ (enterprise & farm incomes wage earnings other cash inflows [e.g. interest, remittances, etc.])/HH size. __ l6 __ ∑ (enterprise & farm incomes wage earnings)/HH size. l7 __ ∑ (enterprise income and wage earnings); n.b. a small portion of wage earnings likely derives from agriculture, thus the estimate contains moderate upwards bias. Annex J. Employment and Income Estimates in the Surveyed Communities 247 Table J.3 Tanzanian Communities Unweighted Communities Entire Sample Smaller Larger (n 154 communitites) (n 71) (n 73) 1. Employment Shares Total rural nonfarm 42% 42% 43% Total rural nonfarm self-employed 38% 38% 38% __ Total wage workers l1 15% 16% 17% __ Agriculture l2 89% 90% 90% Labor force participation ratios 90% 90% 90% 2. Income Shares __ Rural nonfarm l3 65% 63% 69% __ Agriculture l4 35% 37% 31% 3. Average Per Capita and Per Worker Incomes ($) __ Gross household income/capita l5 149 126 157 __ Household “earned� income/capita l6 124 102 128 __ Household earnings/cap. fm RNF Activity l7 77 59 85 Average per rural nonfarm worker 391 301 453 Source: Calculated from RIC household surveys. __ l1 includes part-time and seasonal farm labor. __ l2 includes all household members engaged in agricultural operations; viz. full-time self-employed, part-time and seasonal workers. l3 ∑ (enterprise income wage earnings/HH “earned� income [cf. ff 6 below] ) * 100; N.B., a small portion of wage earnings likely derives __ from agriculture, thus the estimate contains moderate upwards bias. __ l4 possible moderate downwards bias — cf. ff 3 above. l5 ∑ (enterprise & farm incomes wage earnings other cash inflows [e.g. interest, remittances, etc.])/HH size. __ l6 __ ∑ (enterprise & farm incomes wage earnings)/HH size. l7 __ ∑ (enterprise income and wage earnings); n.b. a small portion of wage earnings likely derives from agriculture, thus the estimate contains moderate upwards bias. Annex K. International Benchmarking of the Nonmetropolitan/Rural Investment Climate See Table K.1 on the following pages. 249 250 Table K.1 International Benchmarking of the Nonmetropolitan/Rural Investment Climate Bangladesh: by geographic segment Bangladesh Nicaragua Sri Lanka Pakistan Tanzania Peri-urban Town Village Mean Mean Mean Mean Mean Mean Mean Mean Index 1: Connectivity 0.45 0.34 0.40 0.38 0.20 0.55 0.50 0.43 Subindex 1 Time taken by main means of transportation 0.44 0.32 0.47 0.32 0.26 0.57 0.55 0.40 to the nearest major city Subindex 2 Cost of transportation to the nearest major city 0.18 0.09 0.66 0.25 0.08 0.18 0.14 0.19 (by public transportation) Subindex 3 Time taken by main means of transportation 0.78 0.33 0.49 0.51 0.52 0.86 0.86 0.74 to the main market Subindex 4 Cost of transportation to the main market 0.52 0.07 0.61 0.39 0.08 0.58 0.52 0.51 (by bus) Subindex 5 Distance to the post of�ce 0.67 0.47 0.69 0.67 0.26 0.82 0.68 0.63 Subindex 6 Rail stop in walking distance 0.57 0.12 0.34 0.23 0.64 0.60 0.55 Subindex 7 Percentage of households with �xed 0.04 0.59 0.08 0.38 0.03 0.08 0.10 0.04 telephone lines Subindex 8 Percentage of households with cell phones 0.42 0.51 0.10 0.20 0.13 0.68 0.54 0.39 Index 2: Infrastructure Services 0.44 0.51 0.35 0.58 0.19 0.54 0.50 0.40 Subindex 1 Percentage of households that use electricity 0.70 0.67 0.69 0.87 0.12 0.88 0.80 0.71 Subindex 2 Availability of electricity index 0.00 0.62 0.55 0.46 0.31 0.00 0.00 0.00 Subindex 3 Percentage of households with access to 0.98 0.62 0.80 0.72 0.47 1.00 0.99 0.97 protected water source Subindex 4 Percentage of households with �xed 0.04 0.59 0.08 0.38 0.03 0.08 0.10 0.04 telephone lines Subindex 5 Percentage of households with cell phones 0.42 0.51 0.10 0.20 0.13 0.68 0.54 0.39 Subindex 6 Sewage channels in the community n.a. 0.08 0.11 0.57 0.07 n.a. n.a. n.a. Subindex 7 Garbage collection or disposal service in the n.a. 0.49 0.12 0.34 0.13 n.a. n.a. n.a. community Subindex 8 Most common road surface (internal road) is 0.39 n.a. 0.33 0.88 0.22 0.60 0.56 0.31 concrete or asphalt Index 3: Business Services 0.05 0.24 0.15 0.40 0.07 0.03 0.21 0.02 Subindex 1 Engineering services available for businesses 0.05 0.23 0.06 0.42 0.04 0.04 0.20 0.02 in the community Subindex 2 Management consulting services available for 0.04 0.22 0.07 0.26 0.15 0.04 0.20 0.01 businesses in the community Subindex 3 Marketing services available for businesses in 0.06 0.19 0.08 0.37 0.13 0.04 0.32 0.02 the community Subindex 4 Accounting services available for businesses 0.02 0.27 0.11 0.29 0.02 0.00 0.16 0.00 in the community Subindex 5 Legal services available for businesses in the 0.06 0.31 0.26 0.55 0.07 0.08 0.20 0.03 community Subindex 6 Insurance services available for businesses in 0.09 0.21 0.39 0.46 0.03 0.04 0.28 0.07 the community Subindex 7 Information services available for businesses 0.02 0.24 0.10 0.40 0.03 0.00 0.12 0.01 in the community Index 4: Governance 0.62 0.68 0.67 0.72 0.50 0.68 0.67 0.60 Subindex 1 General policy and institutional constraints 0.83 0.79 0.75 0.94 0.85 0.85 0.78 0.84 Subindex 2 Infrastructure and services 0.21 0.50 0.49 0.74 0.01 0.27 0.29 0.19 Subindex 3 Dealing with government agencies 0.91 0.94 1.00 0.86 0.89 0.95 0.90 Subindex 4 Rule of law . 0.13 0.57 0.49 . . . Subindex 5 Conflict resolution and contract enforcement 0.77 0.59 0.98 0.62 0.66 0.78 0.82 0.76 Index 5: Human Capital 0.24 0.21 0.33 0.21 0.21 0.29 0.28 0.22 Index 6: Finance Services* 0.19 0.17 0.50 0.39 0.18 0.27 0.33 0.15 Subindex 1 Number of �nancial and insurance institutions that offer services in the community, weighted 0.24 0.20 0.56 0.73 0.24 0.34 0.42 0.19 by distance from the community Subindex 2 Number of �nancial and insurance services 0.23 0.15 0.47 n.a. 0.12 0.34 0.41 0.16 provided in the community by formal institutions, weighted by distance from the community Subindex 3 Access to loans 0.11 0.47 0.05 0.19 0.12 0.16 0.10 Source: Data for Nicaragua, Sri Lanka, and Tanzania, from RIC2; for Bangladesh and Pakistan, from World Bank (2008a; forthcoming). Note: * See note to Table E.6. 251 Notes techniques; these subsequently serve as dependent vari- EXECUTIVE SUMMARY ables in a statistical model designed to get at the underlying 1 The survey methodology has been addressed in “Rural determinants of ef�ciency. The authors conclude that Investment Climate Assessment: Implementation Manual� household incomes could be improved in The Gambia if ad- (World Bank 2007c). ditional household resources were allocated to nonfarm ac- 2 For Nicaragua and Sri Lanka, gaps in weights required that tivities. They suspect that imperfections in local labor and some observations be discarded as well. capital markets are the chief constraints preventing house- 3 Related to this, consistent database design should be an- holds from doing so. In a study based on data from China other objective. The RIC databases for Nicaragua, Sri Lanka, taken during a period of factor-market liberalization, Yang and Tanzania followed entirely different formats. Consi- (2004) �nds that education is a key determinant of the rate derable effort was necessary to reconcile the analysis of one at which households reallocated assets to more pro�table country with data from another. Among other undesirable nonfarm activities. consequences, lack of consistency and even of common 10 These �ndings apply to other regions as well. Similar results de�nitions both muddled and discouraged the prospects for were observed in notes prepared for The Rural Investment robust cross-country analysis. Climate: It Differs and It Matters (World Bank 2006), using survey data from the three pilots under review in this report (Nicaragua, Sri Lanka, and Tanzania). A main problem for investment is that �rms with pro�table investment projects INTRODUCTION often cannot use external funds to �nance them. 11 Noting that the LSS approach captures in the main sole pro- 4 Perhaps, more accurately, we should say that the growth of prietorships and partnerships, it was evident that even small informal enterprises—which predominate in rural some of these enterprises had not been captured in the LSS areas—has lagged behind growth of the larger “formals,� databases. Meanwhile, incorporated enterprises, joint which are largely urban-based. ventures and cooperatives, and parastatals remained com- 5 Studies offering some relevant information were based on pletely outside the LSS sampling design. Consequently, household-based surveys without particular focus on the both a broader enterprise and community survey and com- conditions of nonfarm enterprises and thus provided lim- plementary surveys of household-owned enterprises were ited evidence on investment climate constraints. Probably recommended as essential for more rigorous assessments of the most elaborate study to date to use household survey the rural investment climate. These �ndings guided the data is that by Vijverberg et al. (2006), compiled using the scope and design of the prototype questionnaire employed Vietnam Household Living Standards Survey data of 2006. for the RIC pilot surveys. 6 The RIC Implementation Manual developed by the Bank’s 12 An additional resource, not available when the pilot surveys Agriculture and Rural Development Department is being were being designed, is the draft WDR (2008b), which doc- used by task teams throughout the regions, as are the pro- uments and emphasizes the importance of spatial consider- totype questionnaires. The manual backstopped a training ations. The RIC Implementation Manual (World Bank course delivered in 2007 to participants of a multinational 2007c) suggests ways to address this issue in the context of workshop convened in Tanzania. the connectivity of rural-urban RNFE markets, although, 7 Haggblade et al. 2007, p. 400. unfortunately, the surveys did not really address spatial 8 Though an intersect exists, the orientation of the PICS has considerations, except for a parallel sample of larger and largely been toward IC constraints endured by registered, better-connected rural enterprises in the Sri Lanka RIC as- larger, mainly urban-based enterprises. The rural nonfarm sessment. This led to indirect inferences, as the larger enter- enterprises (RNFEs), in contrast, are relatively smaller and, prises tend to serve markets beyond the borders of their for the most part, informally organized. communities. 13 It also contradicts the traditional assumption that earnings from labor migration exceed those from local nonfarm CHAPTER 2 14 activities (Haggblade 2007, p. 122). Ibid., p. 123. 9 Using household data from The Gambia, Chavas, Petrie, 15 Ibid., p. 138–39. and Roth (2005) examine the determinants of ef�ciency in 16 Examples are indicators for governance (World Bank households that generate farm and nonfarm income. They 2007a), investment climate (World Bank 2004), cost of doing take a two-stage approach that �rst generates ef�ciency business (World Bank 2007b), and political rights and civil measures from a joint farm-nonfarm household production liberties (Freedom House 2008), http:/ /www.freedomhouse model based on nonparametric linear programming .org/template.cfm?page=15 . 253 254 Notes 17 See the RIC Implementation Manual for details (World Bank 28 A precise de�nition of entrepreneurship appears in the next 2007c). section. The de�nitions for Tanzania and Nicaragua are the 18 To illustrate, suppose that a community has 185 small en- same; that for Sri Lanka differs. terprises typically employing 1.5 workers and 15 large ones 29 These variables relate to the productivity of household employing 10 workers on average. Thus, the population members in their own enterprises and to the attractiveness proportion of large enterprises is 7.5 percent (15 of 200). A of alternative employment and therefore determine the sample of 10 enterprises is drawn, consisting of 5 large ones choice between them. (a sampling rate of 33.3 percent) and 5 small ones (a sam- 30 Among the explanatory variables, household assets and pling rate of 2.7 percent). The simple sample proportion of community income levels might have been endogenous. large enterprises is 50 percent, which of course greatly ex- Fortunately, a test of endogeneity of these two variables ceeds the actual population proportion it seeks to estimate. gave no evidence that this was the case (Annex F). Sample weights would equal 3 ( 15/5) for large enterprises 31 The distribution of assets is highly skewed. For both coun- and 37 ( 185/5) for small enterprises. A weighted sample tries, the change in assets is also equivalent to the difference proportion of large enterprises equals 3 5 / (3 5 37 between the seventy-�fth and about the ninety-second 5) 0.075, which is the correct estimate. The unweighted es- percentiles. timate of the number of employees equals 5.75 per enter- 32 It is noted that this estimated effect is causal and not merely prise, which is completely unrealistic since the population a reflection of reverse causation where prosperous entre- includes so many small enterprises; the weighted mean preneurs are accumulating household wealth: as mentioned equals (3 5 10 37 5 1.5) / (3 5 37 5) 2.14. above, assets were found to be exogenous relative to entre- 19 The dependent variable is continuous in Chapter 3, binary preneurship choice. in Chapter 4, and ordered-categorical in Chapter 5. Each im- 33 A portion of the difference between Nicaragua and Sri plies its own variation of the same econometric structure. Lanka derives from the difference in de�nition: Nicaragua 20 This is not to say that a successful investment climate nec- uses the income-based de�nition and Sri Lanka a work- essarily raises entrepreneurship rates. It may equally well based de�nition. Thus, the percentage actually engaged in stimulate enterprise growth, which soaks up labor in the farm work rises to 35 percent in Nicaragua. By comparison, community and gives some entrepreneurs an incentive to the percentage of rural households in the 2001 Living abandon their own enterprise and seek employment in a Standards Measurement Study data that reported any agri- larger, more successful business—thus driving the entre- cultural activity was 35.6 percent. Additionally, it may be preneurship rate down. the case that a signi�cant number of wage workers are ac- tually agricultural laborers; the questionnaires did not dif- ferentiate wage employment by sector. 34 The choices are probably interdependent, and some ef�- CHAPTER 3 ciency gain might be derived from joint estimation. A multinomial logit model incorporates all alternatives simul- 21 The questionnaires did not formally de�ne enterprise or self- taneously but requires that individuals or households make employment nor did it set a clear minimum threshold for en- exclusive choices. As seen in Table 4.1, households fre- terprise size. With a sharper de�nition of enterprise some of quently participate in several activities, contrary to the logic these cases would not have been included, although the var- of a multinomial logit application. ious sources of income would still have been recorded as 35 These conditions might provide incentives to start one’s household income. own business instead, but column 1 of the table reveals no 22 This is because much more variation relates to enterprise such effect. size than to community variables. 36 The effect of management consulting and marketing ser- 23 This �nding mirrors the conclusion in Haggblade et al. vices is strong and statistically signi�cant and in a plausible (2007) reported above in Chapter 2. direction—but may have to be taken with a grain of salt, as the number of communities with such services (9 and 8 re- spectively, with an overlap of 6 between them, out of 117) is small. CHAPTER 4 37 The precise timing of the RIC survey matters as well, since 24 Haggblade et al. (2007) mention a range of 20 to 50 percent. the period of start-up measurement covers calendar years 25 This suggests—or, rather, reaf�rms from the stratum of only. This recognition leads to an upward revision of the es- rural households—that as GDP/capita increases (Table 3.1) timated start-up rate in Nicaragua to 4.0 to 4.5 percent (see the signi�cance of agricultural employment declines some- Annex F). what monotonically. 38 If p measures the annual proportion of enterprises that exit 26 When describing household incomes it is important to note and are replaced by new start-ups, the two-year ratio of that the Sri Lanka questionnaire contained a flaw that start-up enterprises relative to the total number of enter- caused understating of income components other than prises equals p(1 p) p, which then may be equated to the wages and salaries and farming. In addition, the Tanzania two-year percentage q of start-up enterprises in the sample. statistics did not use sampling weights, and the strati�cation The survival rate equals 1 p and solves to (1 q)0.5. The as- is explicitly based on a feature (entrepreneurship) related to sumption of a stationary enterprise population is crucial in income. identifying the survival rate. 27 In addition to farm income, income from wages and salaries are lumped together, and expenses related to wage earning are deducted from the gross wage income to estimate net wage income. Households also receive other income in the CHAPTER 5 form of remittances and gifts from friends and relatives, in- 39 Entrepreneurs’ responses on the importance of governance terest income from savings and dividends, retirement ben- and corruption as possible IC constraints are addressed in e�ts, and lottery winnings. the �rst section of Annex H and recorded in Table H.10. Notes 255 40 In Nicaragua, different enterprise questionnaires were ad- some communities have a very low proportion of house- ministered to household-based and stand-alone enterprises. holds with this amenity. Enumerators apparently skipped questions about labor 53 For Tanzania, this EICO did not score in the top ten, and it EICOs for household-based enterprises. Another difference is therefore not included in the table. In a regression run for among the countries is that the Sri Lanka and Tanzania Tanzania not reported here, marketing services reduce the questionnaires use screen questions to uncover general obstacle of low market demand, as does the human capital problems with, for example, �nance or infrastructure; if benchmark. Connectivity, however, had no effect. af�rmative, they proceed with the detailed EICOs. If the 54 Annex L contains a broader summary of governance and respondent reports no general problems, the researcher is corruption as perceived investment climate constraints. forced to conclude there are also no problems with detailed 55 There is no dummy for traders; the parameters compare en- EICO aspects. This may have led to some underreporting trepreneurs with traders. of EICO barriers in Sri Lanka and Tanzania relative to 56 Unfortunately, standard estimation routines in statistical Nicaragua. software such as Stata and SAS allow either sampling 41 A detailed check of data by enterprise suggests that this is weights or random effects, but not both. Annex I describes not due to a “halo� effect caused by entrepreneurs who, dis- Stata programs that allow both features to be included in the gruntled about a few barriers, flagged all EICOs across the estimation. board: their responses varied by EICO. 57 As described in the analysis above, including enterprise per- 42 The precise wording is “In many countries, companies give formance in the EICO equations creates biased estimates. To informal payments (kickbacks, bribes, or other exemptions) the degree that these biased parameter estimates are still to government of�cials to gain advantages in the writing of reliable, it appears that enterprise size raises perceptions of laws, decrees and regulations, etc. In this country, to what investment climate constraints, and enterprise pro�tability extent do these practices have a direct impact on your reduces perceived constraints. But one may question the di- establishment?� rection of causality, since the EICO models in this chapter do 43 In Sri Lanka, the questions about influencing of�cials or not confront this issue: are managers of pro�table enterprises drafting legislation are answered only by entrepreneurs em- better able to circumvent investment climate barriers, or are ploying at least three workers. such enterprises more pro�table because barriers are lower? 44 The remaining benchmark indicator, governance, is not used since it is constructed out of community-averaged en- trepreneur responses, some of which are among the EICOs and EICIs these models seek to explain. CHAPTER 6 45 When studying whether entrepreneurs list postal service as 58 Community size varies greatly, however, especially in an obstacle, for example, it is useful to know the distance to Nicaragua, but also in Tanzania, such that the smallest com- the nearest post of�ce. This is merely one of eight elements munity surveyed in Nicaragua has only slightly more than of the connectivity benchmark. one-half the population of the smallest Sri Lankan village, 46 In the case of variables entered in logarithmic form, the though the size of its largest community exceeded Sri mean and standard deviation of the regular variable are Lanka’s by more than 16 times. considered in determining the magnitude of the standard- 59 The percentage with debt (formal and informal sources to- ized effect. gether) is 21, 42 and 23 percent respectively in Nicaragua, 47 These estimates are presented in Table H.3. Sri Lanka and Tanzania. 48 Ibid. Table H.3, section C. 60 Moreover, the speci�cation of the outcome variables is not 49 Ibid. Table H.3 section B. Note that the dummy for trade is always optimal from a theoretical point of view because of omitted in the model. limitations imposed by the database. 50 A mixed enterprise is active in more than one sector. 61 Measured by the average tendency of enterprises to interact 51 Table H.5 reports the estimates of the impact of size, as mea- with suppliers and clients outside the community. sured by the log of sales, and productivity, as measured by 62 Despite a number of relevant variables in the regression the log of the ratio of net value added over total factor cost model, the F-statistic is extremely low. This may occur in an (V/C). The reported values are not scaled estimates as in the equation suffering from substantial heteroskedasticity. The other tables; rather, they are unscaled estimates of the para- standard errors of the parameter estimates are computed to meters of the weighted random effect ordered probit model. allow for heteroskedasticity, but the computed p-value of In Nicaragua, many of the two effects are statistically signif- the F-statistic could be seriously biased. The ratio variable icant; in Sri Lanka, the effect is statistically relevant for only here may be subject to wild swings because of measurement one EICO, that is, telecommunication; and in Tanzania, �ve error in costs of production. of the twenty parameter estimates are statistically signi�cant. It is notable that the size effect is positive for 23 of the 30 es- timates, that the productivity effect is negative for 24 of the 30 estimates, and that the effects have the same sign only �ve CHAPTER 7 times. Taken at face value, this implies that operators of 63 Thus, comparisons of benchmarks between countries in this larger enterprises tend to complain more often about invest- study are illustrative but tentative. ment climate conditions and that entrepreneurs overseeing 64 With the loss of weights, as in the case of Tanzania, much of more productive businesses tend to view the environment in the value of the RICS is lost. For Nicaragua and Sri Lanka less problematic terms. But, as mentioned, these effects are some observations had to be discarded because of gaps in probably biased downward: the size effect is probably more information on weights. strongly positive, and the productivity effect less negative. 65 The bene�ts from utilizing the pilot assessments in Nicaragua 52 The average percentage of households in a community with and Tanzania is less clear, owing to delays in completing the access to protected water sources is higher in Sri Lanka assessments, lukewarm government response (in Tanzania), (80%) than Nicaragua (62%) and Tanzania (48%), suggest- and recurrent staf�ng changes both on the country teams ing the possibility of a threshold effect. But even in Sri Lanka and among counterparts. 256 Notes entails complicates the Heckit method greatly. Moreover, ANNEX A sampling weights for enterprises differ from those for 66 Incomplete data on benchmark indicators causes a loss of households, even if a given enterprise belongs to a particu- 167 enterprises. Unavailability of other community vari- lar household. Sorting out these econometric issues is left for ables reduces the sample size by 53 observations. Many of the future. these enterprises are located in the war-torn northeast re- 77 The prices of some items, in particular Coca-Cola, ap- gion of the country. peared to be signi�cant in several of the draft regression 67 The questionnaires included no formal de�nition of enter- speci�cations. prise and set no minimum threshold size. 78 This is statistically relevant, but it may not be unique to rural 68 Other common reasons might be added, such as response enterprises. No comparable data for PICS are available, but error and data entry mistakes. it is not unlikely that for urban enterprises as well enterprise 69 Determination of the threshold was arbitrary. Taking characteristics dominate other variables. 25 percent of an adult’s employment per year and assuming 79 In simpli�ed models similar to variant (1), population size 240 working days per year at a net income of US$0.25 per and enterprise density have signi�cant positive impacts, but day gives a threshold of US$60 per year. In the analysis of these effects vanish in the more elaborate speci�cations of household income and employment in Annex D no enter- variants (4) and (5). prises were excluded. 80 In variants (3) and (4) without quadratic terms (not reported in the table), the parameter estimates of log-labor and log- capital are 0.25 and 0.09, respectively, both highly signi�- cant; in variant (5) without quadratic terms, the estimates ANNEX C equal –0.04 (imputed labor, not signi�cant), 0.11 (paid 70 The Cobb-Douglas model augmented with quadratic and labor), 0.04 (imputed capital), and 0.07 (paid capital). Thus, interactive labor and capital variables is a partial imple- a positive relation exists between input use and net value mentation of the translog model, which would add qua- added, but these parameter estimates add up to much less dratic interactive terms involving nonfactor inputs and than 1, which suggests decreasing returns to scale. depreciation. As the NVA model omits the latter, it may be 81 That is, 100(e1.34 – 1) percent. seen as a complete translog speci�cation. 71 Following common practice, there is no depreciation charge for capital. 72 Although variables for effective access to bank loans and ANNEX D maximum amount of informal credit obtained are available, 82 The exchanges rates used in this paper are 15.9372 cordobas they were not selected because of endogeneity. Banks and per dollar for Nicaragua, 1089.33 shillings per dollar for informal lenders presumably extend credit based on their Tanzania, and 96.7699 rupees per dollar for Sri Lanka. evaluations of enterprise performance. 83 According to human capital theory, the wage rate repre- 73 As a result, the registration dummy variable may behave as sents payment for the rental of a worker’s human capital. a dummy for larger enterprises. Size is already measured Thus, a wage w equals the product of the rental rate r and elaborately, however, with the input variables of labor, cap- the human capital stock h. Since r is unobservable but the ital, nonfactor costs, and depreciation. The measured impact same for all participants in the labor market, h is determined of the registration dummy is therefore likely to measure the up to a multiplicative scalar constant, found as an intercept pure effect of registration, rather than the effect of enterprise in the ln(h) equation. For now, this intercept in the ln(h) size differential. equation is set to 0. Later on, a further scaling is introduced 74 Note that these dummy variables do not reflect differences that effectively modi�es the scaling applied here. in technology, which would imply a different structural 84 As a concept, Bils and Klenow (2000) also include a more in- relationship between inputs and output by sector. volved version of the same model, which allows for nonlin- Theoretically, differences in sectoral technologies are en- ear effects of education as well as intergenerational transfers tirely plausible, but the sectoral subsamples are too small to from parents and teachers to children. Nonlinear effects permit estimation of relationships by sector. Moreover, one proved to be less important empirically, and intergenera- subsample consists of enterprises that cannot be clearly lo- tional transfers are dif�cult to establish with the data at cated in a single sector, causing analytical problems as well. hand. It should also be noted, however, that the (semi)translog 85 These assumptions are inspired by the evidence in the data, speci�cation of the input-output speci�cation alleviates the as the completion of a given level of schooling is typically problem caused by sample pooling. achieved after a set number of years, which is frequently 75 Among the benchmark indicators in the regression model is reported by respondents. a human capital index that summarizes the community’s 86 Households reported on nonwage, nonfarm income (in- skills that are accumulated through education and experi- cluding enterprise income) only if they held any wage job. ence. Illiteracy is therefore more speci�c than having little A later section, provides more information. human capital: there is room for both variables in the re- 87 For example, if the respondent offered information about gression model. The two variables also describe different the plot size in both acres and perches, the ratio of reported parts of the labor force, and it is an empirical question which perches over reported acres should vary around 65. parts the rural non-farm enterprises draw from. Alternatively, if the total plot size should be the sum of the 76 Self-selection effects may pertain to many other variables, two reported values, one would expect that the number of including the use of household labor and enterprise capital, perches would be consistently below 65. Moreover, what- and the sector of activity. Correcting for self-selection is ever interpretation is followed, the price per unit of land often done by means of the so-called Heckit method, which should also correspond roughly to that recorded for house- is appropriate when disturbances are identically and inde- holds using only one measure. pendently distributed among observations. The clustering 88 According to information provided at http://www by community and the community random effect this .sizes.com/units/manzana.htm. Notes 257 similar evidence in the context of Living Standards Survey ANNEX F data in Ghana, Guatemala, Kyrgyz Republic, and Vietnam, 89 A logical alternative to these de�nitions is to rely on the see Vijverberg (2005). The lesson from this is that when col- RICS enterprise sample, which is linked to the household lecting survey data such as RICS, survey managers must be sample: a household would be entrepreneurial if it is linked on guard against common inconsistencies of this nature. to an enterprise in the enterprise sample. Due to a design Proper training of enumerators is a �rst step. feature of the early RICS data, however, if the household owned and operated a stand-alone enterprise, it was not coded as operating a household enterprise. Thus, the rate of entrepreneurship based on the link in the RICS data reflects ANNEX H only household-based entrepreneurship and would under- 91 The �rst insertion makes the grammar appropriate for the state the overall rate of entrepreneurship among rural context of the text here; the second insertion is implied by households. Examining income receipt and work activity is the context of the screening question, but the actual text in a way around this design flaw. loco does not speci�cally refer to the operation and growth 90 Note that these are unweighted counts. Thus, Tanzania’s of the enterprise. sample evidence is valid here too. In Tanzania, 16.7 percent of the households received enterprise income but reported no “work on own account,� and 10.2 percent had household members working in a household enterprise but reported ANNEX I no income from a nonfarm enterprise. In other words, 92 This follows from deriving the (pseudo-)likelihood function Tanzanian data suffer from a more serious consistency and retaining the portion that contains the parameters of the problem. In this regard, RICS data are not unique: for regression equation. References Amemiya, Takeshi. 1978. “A Note on a Random Coef�cients Haggblade, Steven, Peter B. R. Hazell, and Thomas Reardon, Model.� International Economic Review, Vol. 19, No. 3 (Oct., eds. 2007. Transforming the Rural Nonfarm Economy: Opportu- 1978), pp. 793–796. nities and Threats in the Developing World. Baltimore: The Johns Baltagi, Badi H. 2005. Econometric Analysis of Panel Data, Third Hopkins University Press for the International Food Policy edition. Chichester: Wiley. Research Institute. Barrett, C. B., T. Reardon, and P. Webb. 2001. “Nonfarm Income Hayami, Yujiro, ed. 1998. Towards a Rural-Based Development of Diversi�cation and Household Livelihood Strategies in Rural Commerce and Industry: Selected Experiences from East Asia. Africa: Concepts, Dynamics, and Policy Implications.� Food Washington, DC: Economic Development Institute of the Policy 26 (4): 315–31. World Bank. Batra, G., D. Kaufmann, and A. H. W. Stone. 2003. Investment Lanjouw, Peter. 2001. “Nonfarm Employment and Poverty in Climate Around the World: Voices of the Firms from the World Rural El Salvador.� World Development 29 (3): 529–47. Business Environment Survey. Washington, DC: IBRD. Lanjouw, Jean O., and Peter Lanjouw. 2001. “The Rural Non- Bigsten, A., P. Collier, S. Dercon, M. Fafchamps, B. Gauthier, Farm Sector: Issues and Evidence from Developing J. W. Gunning, A. Oduro, R. Oostendorp, C. Pattillo, Countries.� Agricultural Economics 26 (1): 1–23. M. Söderbom, F. Teal, and A. Zeufack. 2003. “Credit Larson, Donald F., and Yyannu Cruz-Aguayo. 2008. “On the Constraints in Manufacturing Enterprises in Africa.� Journal Role of Self-Employment and Specialization in Family Farms of African Economies 12 (1): 104–25. and Businesses: Evidence from Nicaragua.� Unpublished Bigsten, A., P. Collier, S. Dercon, B. Gauthier, J. W. Gunning, manuscript. A. Isaksson, A. Oduro, R. Oostendorp, C. Pattillo, Maloney, W. F. 2004. “Informality Revisited.� World Development M. Söderbom, M. Sylvain, F. Teal, and A. Zeufack. 1999. 32 (7): 1159–78. “Investment in Africa’s Manufacturing Sector: A Four Moock, Peter, Philip Musgrove, and Morton Stelcner. 1990. Country Panel Data Analysis.� Oxford Bulletin of Economics “Education and Earnings in Peru’s Informal Nonfarm Family and Statistics 61 (4): 489–512. Enterprises.� Living Standard Measurement Study Working Bigsten, A., P. Kimuyu, and K. Lundvall. 2004. “What to Do with Paper 64, World Bank, Washington, DC. the Informal Sector?� Development Policy Review 22 (6): Perry, Guillermo, W.F. Maloney, O.S. Arias, P. Fajnzylber, A.D. 701–15. Mason, and J. Saavedra-Chanduvi. 2007. Informality: Exit and Bigsten, A., and M. Soderbom. 2006. “What Have We Learned Exclusion. Washington, DC: World Bank Latin American and from a Decade of Manufacturing Enterprise Surveys in Caribbean Studies and IBRD. Africa?� World Bank Research Observer 21 (2): 241–65. Reardon, T., J. A. Berdegué, and G. Escobar. 2001. “Rural Bils, Mark, and Peter J. Klenow. 2000. “Does Schooling Cause Nonfarm Employment and Incomes in Latin America: Growth?,� American Economic Review, vol. 90(5), pages 1160– Overview and Policy Implications.� World Development 29 (3): 1183, December. 395–409. Chavas, Jean-Paul, Ragan Petrie, and Michael Roth. 2005. “Farm Sundaram-Stukel, Reka, Klaus Deininger, and Songqing Jin. Household Production Ef�ciency: Evidence from The 2006. “Fostering Growth of the Rural Non-Farm Sector in Gambia.� American Journal of Agricultural Economics 87 (1): Africa: The Case of Tanzania.� Paper prepared for presenta- 160–179. tion at the American Agricultural Economics Association Deininger, Klaus, Songqing Jin, and Mona Sur. 2007. “Sri Annual Meeting, Long Beach, California, July 23–26. Lanka’s Rural Non-Farm Economy: Removing Constraints to Washington, DC: IBRD-DECRG. Pro-Poor Growth.� World Development 35 (12): 2056-2078. Van der Sluis, Justin, Mirjam van Praag, and Wim Vijverberg. De Soto, Hernando. 1989, reissued 2002. The Other Path. New 2005. “Entrepreneurship Selection and Performance: A York: Basic Books. Meta-analysis of the Impact of Education in Developing Eifert, B., A. Gelb, and V. Ramachandran. 2005. “Business Countries.� World Bank Economic Review 19 (2): 225–61. Environment and Comparative Advantage in Africa: Vijverberg, Wim P.M. 1990. “Non-Farm Self-Employment and Evidence from the Investment Climate Data.� Policy the Informal Sector in Côte d’Ivoire: A Test of Categorical Research Working Paper, World Bank, Washington, DC. Identity.� Journal of Developing Areas 24 (4): 523–42. Freedom House. 2008. Freedom in the World 2007, http://www ———. 1997. “Monte Carlo Evaluation of Multivariate Normal .freedomhouse.org/template.cfm?page=15. Probabilities.� In: Journal of Econometrics, January 1997, 76: Greene, William H. 2008. Econometric Analysis. Upper Saddle 1–2, pp. 281–307. River, New Jersey: Pearson Prentice Hall. ———. 1998. “Nonfarm Household Enterprises in Vietnam.� In Haggblade, Steven, Peter Hazell, and James Brown. 1989. Household Welfare and Vietnam’s Transition, ed. David Dollar, “Farm-Nonfarm Linkages in Rural Sub-Saharan Africa.� Paul Glewwe, and Jennie Litvack, 137–78. Washington, DC: World Development 17 (8): 1173–1201. World Bank. 259 260 References ———. 1999. “The Impact of Schooling and Cognitive Skills on ———. 2007c. “Rural Investment Climate Assessments: Income from Non-Farm Self-Employment.� In The Economics Implementation Manual� (draft). Washington, DC: World of School Quality Investments in Developing Countries: An Bank. Empirical Study of Ghana, ed. Paul Glewwe, 206–52. New ———. 2007d. World Development Indicator Database. York: St. Martin’s Press. Washington, DC: World Bank. www.worldbank.org/data/ ———. 2005. “Rural Investment Climate and Non-Farm onlinedatabases/onlinedatabases.html Entrepreneurship in Living Standards Surveys.� Un- ———. 2007e. Tanzania: Pilot Rural Investment Climate published manuscript, ARD, World Bank, Washington, DC. Assessment: Stimulating Non-farm Microenterprise Growth. Vijverberg, Wim, and Jonathan Haughton. 2004. “Household Report No. 40108-TZ. Washington, DC: World Bank. Enterprises in Vietnam: Survival, Growth, and Living ———. 2008a (forthcoming). Bangladesh Second Investment Standards.� In Economic Growth, Poverty and Household Climate Assessment: Harnessing Competitiveness for Stronger Welfare: Policy Lessons from Vietnam, ed. David Dollar and Growth. Washington, DC: World Bank. Paul Glewwe, 95–132. Washington, DC: The World Bank. ———. 2008b. Agriculture for Development. World Development Vijverberg, Wim, with Hoang Thi Thanh Huong, Nguyen Chien Report. Washington, DC: IBRD. Tang, Nguyen Ngoc Que, Nguyen The Quan, Phung Duc ———. 2009 (forthcoming). Reshaping Economic Geography. Tung, and Vu Thi Kim Mao. 2006. “Non-Farm Household World Development Report. Washington, DC: IBRD. Enterprises in Vietnam: A Research Project using Data from World Bank and Asian Development Bank. 2005. Sri Lanka – VHLSS 2004, VHLSS 2002 and AHBS 2003.� The World Bank, Improving the Rural and Urban Investment Climate. Washington, Hanoi Of�ce. DC: World Bank; Manila: Asian Development Bank. http:/ / World Bank. 2003. Reaching the Rural Poor: A Renewed Strategy for www.worldbank.lk/WBSITE/EXTERNAL/COUNTRIES/ Rural Developmemt. Washington, DC: World Bank. SOUTHASIAEXT/SRILANKAEXTN/0,,contentMDK: ———. 2005. A Better Investment Climate for Everyone. World 20569317~menuPK:287057~pagePK:141137~piPK:217854~ Development Report 2005. Washington, DC: World Bank theSitePK:233047,00.html. and IBRD. World Bank/RUTA. 2008 (forthcoming). Nicaragua: Rural ———. 2006. The Rural Investment Climate: It Differs and It Investment Climate. Washington, DC: World Bank. Matters. Washington, DC: IBRD. Yang, Dennis Tao. 2004. “Education and Allocative Ef�ciency: ———. 2007a. Governance Matters 2007: Worldwide Governance Household Income Growth during Rural Reforms in China.� Indicators, 1996–2006. Washington, DC: World Bank. Journal of Development Economics 74: 137–62. ———. 2007b. Doing Business 2008. Washington, DC: World Bank.