\)Jt; 3 a Mo1 Iq96 320 E 1231World Bank Discussion Papers Household and Intrahousehold Impact of the Grameen Bank and Similar Targeted Credit Programs in Bangladesh Mark M. Pitt Shahidur R. Khandker Recent World Bank Discussion Papers No. 252 Projectiing the Govemance Approach to Civil Service RefoTn: An InstitutionalEnvironrmentAssessmentforPreparinga Sectoral Adjustment Loan in the Gambia. Rogerio F. Pinto with assistance from AngelousJ. Mrope No. 253 Smal Firms Informafly Financed: Studies from Bangladesh. Edited by, Reazul Islam,J. D. Von Pischke, andJ. M. de Waard No. 254 Indicators forMonitoringPoverty Reduction. Soniya Carvalho and Howard White No. 255 Violence Against Women: The Hidden Health Burden. Lori L. Heise withJacqueline Pitanguy andAdrienne Germain No. 256 Women's Health and Nutrition: Making a Difference. Anne Tinker, Patricia Daly, Cynthia Green, Helen Saxenian, Rama Lakshminarayanan, and Kirrin Gill No. 257 Improving tde Quality of PrimaryEducation in Latin America: Towards the 21st Century. Lawrence Wolff, Emesto Schiefelbein, and Jorge Valenzuela No. 258 How Fastis FertiZity Declining in Botswana and Zimbabwe? Duncan Thomas and Ityai Muvandi No. 259 Policies Affecting Fertility and Contraceptive Use: An Assessment ofTwelve Sub-Saharan Countries. Susan Scribner No. 260 Financial Systems in Sub-Saharan Africa: A Comparative Study. Paul A. Popiel No. 261 Poverty Aleviation andSocial Investment Funds: The Latin American Experience. PhilipJ. Glaessner,Kye Woo Lee, Anna Maria Sant'Anna, and Jean-Jacques de St. Antoine No. 262 Public Policy for the Promotion ofFamilyFarmsin Italy: The Experience ofthe Fundfor the Formation of PeasantProperty. EricB. Shearer and Giuseppe Barbero No. 263 Self-Employment for the Unemployed: Experience in OECD and Transitional Economies. Sandra Wilsonand Arvil V. Adams No. 264 Schoolingand Cognitive Achievements ofChildren in Morocco: Can the GovemmentImprove Outcomes?ShahidurR. Khandker, Victor Lavy, and Deon Filmer No. 265 WorldBankAFinanced Projects with Commurnity Participation: Procurementand Disbursement Issues. Gita Gopal and Alexandre Marc No. 266 Seed Systems in Sub-Saharan Africa: Issues and Options. V. Venkatesan No. 267 Trade Policy Reform in Developing Coun tries since 1985: A Review ofthe Evidence.Judith M. Dean, Seema Desai, and James Riedel No. 268 Farm RestructuringandLand Tenure in RefonningSocialistEconomies: A ComparativeAnalysis ofEastem and CentralEurope. Euroconsult and Centre for World Food Studies No. 269 The Evolution ofthe WorldBank's Railway Lending Alice Galenson and Louis S. Thompson No. 270 Land Reform and Farm Restructuring in Ukraine. Zvi Lerman, Karen Brooks, and Csaba Csaki No. 271 Small Enterprises Adjusting to Liberalization in Five African Countries. Ron Parker, Randall Riopelle, and William F. Steel No. 272 Adolescent Health: Reassessing the Passage to AdulthoodJudith Senderowitz No. 273 Measurement of-Welfare Changes Caused by Large Price Shifts: An Issue in the PowerSector. Robert Bacon No. 274 SocialAction Progriams andSocial Funds: A Review ofDesign and Implementation in Sub-Saharan Africa. Alexandre Marc, Carol Graham, Mark Schacter, and Mary Schmidt No. 275 Investing in Young Children. Mary Eming Young No. 276 ManagingPrimary Health Care: Implications ofthe Health Transition. Richard Heaver No. 277 Energy Demand in Fivc MajorAsian Developing Countries: Structure and Prospects. Masayasu lshiguro and Takamasa Akiyama No. 278 Preshipment Inspection Services. Patrick Low No. 279 RestructuringBanks andEnterprises: Recent Lessons from Transition Countries. Michael S. Borish,Millard F. Long, and Michel Noel No. 280 Agriculture, Poverty, and Policy Reform in Sub-Saharan Africa. Kevin M. Cleaver and W. Graeme Donovan No. 281 The Diffusion ofInformation Technology: Experience oftndustrial Countries and Lessons forDeveloping Countries. Nagy Hanna, Ken Guy, and Erik Arnold No. 282 Trade Laws andInstitutions: Good Practices and the World Trade Organization. Bernard M. Hoekman No. 283 Meeting the Chalenge of Chinese Enterprise Reform. Harry G. Broadman No. 284 Desert Locust Management: A Time for Change. Steen R.Joffe (Continuedon theinside backcover) 32 0 s World Bank Discussion Papers Household and Intrahousehold Impact of the Grameen Bank and Similar Targeted Credit Programs in Bangladesh Mark M. Pitt Shahidur R. Khandker The World Bank Washington, D.C. Copyright C 1996 The International Bank for Reconstruction and Development/THE WORLD BANK 1818 H Street, N.W. Washington, D.C. 20433, U.S.A. All rights reserved Manufactured in the United States of America First printing May 1996 Discussion Papers present results of country analysis or research that are circulated to encourage discussion and comment within the development community. To present these results with the least possible delay, the typescript of this paper has not been prepared in accordance with the procedures appropriate to formal printed texts, and the World Bank accepts no responsibility for errors. Some sources cited in this paper may be informal documents that are not readily available. The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they represent. The World Bank does not guarantee the accuracy of the data included in this publication and accepts no responsibility whatsoever for any conse- quence of their use. The boundaries, colors, denominations, and other information shown on any map in this volume do not imply on the part of the World Bank Group any judgrnent on the legal status of any territory or the endorsement or acceptance of such boundaries. The material in this publication is copyrighted. Requests for permission to reproduce portions of it should be sent to the Office of the Publisher at the address shown in the copyright notice above. The World Bank encourages dissemination of its work and will normally give permission promptly and, when the reproduc- tion is for noncornmercial purposes, without asking a fee. Permission to copy portions for classroom use is granted through the Copyright Clearance Center, Inc., Suite 910, 222 Rosewood Drive, Danvers, Massachu- setts 01923, U.S.A. The complete backlist of publications from the World Bank is shown in the annual Index of Publications, which contains an alphabetical title list (with full ordering information) and indexes of subjects, authors, and countries and regions. The latest edition is available free of charge from the Distribution Unit, Office of the Publisher, The World Bank, 1818 H Street, N.W., Washington, D.C. 20433, U.S.A., or from Publications, The World Bank, 66, avenue d'Ina, 75116 Paris, France. ISSN: 0259-21 OX Mark M. Pitt is a professor in the Department of Economics, Brown University, Providence, Rhode Island. Shahidur R. Khandker is an economist in the World Bank's Poverty and Social Policy Department. Iibrary of Congress Cataloging-in-Publication Data Pitt, Mark Martin, 1949 - Household and intrahousehold impact of the Grameen Bank and similar targeted credit programs in Bangladesh / Mark M. Pitt, Shahidur R. Khandker. p. cm. - (World Bank discussion papers ; 320) Includes bibliographical references. ISBN 0-8213-3594-4 1. Grameen Bank. 2. Rural credit-Bangladesh. 3. Bank loans- Bangladesh. 4. Rural poor-Bangladesh. I. Khandker, Shahidur R. I. Title. Ill. Series. HG3090.6.P57 1996 332.1'095492-dc2O 96-6828 CIP CONTENTS Foreword ................................................. v Abstract . ................................................ vi Acknowledgments .............................................. vii 1. Introduction .............................................. I 2. Evaluating program impact: a framework ............................... 6 The empirical model ............................................. 11 Estimation strategy ............................................. 12 Why might credit program participation be endogenous? ...................... 14 Econometric approach ............................................ 15 Identification of the impact of gender-specific credit ......................... 21 3. Survey design .24 Data description .26 4. Results .28 5. Summary and conclusions .40 Tables ................ 45 Appendix A ...... 63 Appendix B ...... 95 Appendix C ...... 105 References ....... 107 . . I FOREWORD Providing credit to the rural poor and developing viable credit institutions within the broader objectives of poverty alleviation is a well established development policy, but there are few good studies of effects and sustainability. The research project RPO 676-59 "Credit Programs for the Poor: Household and Intrahousehold Impacts and Program Sustainability" was designed with appropriate research methods to examine these important issues. Bangladesh was selected as a suitable location to apply such methods because it has a number of targeted programs with varying designs, including the Grameen Bank, the BRAC and the BRDB's RD-12 operated by the government and non-govermnent organizations. One objective of this research was to develop a methodology to estimate the costs and benefits of group- based credit programs. It included the identification of program effects on household and individual outcomes as well as the analysis of the participation of women in these credit programs and the ensuing effects on household and intrahousehold outcomes by gender. Another objective was to analyze the financial and economic efficiency of the credit programns, which depend on resource-intensive group formation and monitoring. While peer monitoring reduces the transaction costs of lending to the poor, group formation and monitoring is costly and group members may not be able to bear the full costs of a program. The aim was to estimate the cost structures of the programs and examine how the programs operate and whether and under what conditions such group- based credit programs are sustainable. This paper is one of several papers produced as a research output under this research project. It estimates the influence of borrowing by both men and women for each of three programs (GB, BRDB, BRAC) under the study on a variety of household and intrahousehold outcomes. These outcomes include the school enrollment of boys and girls, the labor supply of women and men, the asset holdings of women, recent fertility and contraceptive use, consumption, and the anthropometric status of children. Estimates show that credit is a significant determinant of manyof these outcomes. However, credit provided to women was found more likely to influence these behaviors than credit provided to men. In short, targeted credit to women has a significant effect on the well-being of poor household and the effect is greater when women are the program participants. Ishrat Husain Director Poverty and Social Policy Department Human Capital Development v ABSTRACT Group-based lending programs for the poor have become a focus of attention in the development community over the last several years. To date, there has been no comprehensive investigation of their impact on household behavior that has been sufficiently attentive to issues of endogeneity and self-selection. Perhaps one reason for this is the absence of any data generated from social experiments associated with these credit programs, and from the difficulty in finding valid instrumental variables (exclusion restrictions) to deal with the endogeneity bias in non- experimental data. This paper surmounts these issues by treating the choice of participating in credit programs in a sample of Bangladeshi households and villages as corresponding to a "quasi-experiment" conditional on all observed (in the data) and unobserved village characteristics. It uses the same approach to help identify the separate effects of lending to female and male household members, making use of the fact that credit groups are single-sex and groups for both sexes are not available in all villages. The data were collected in a special survey carried out in 87 rural Bangladeshi villages during 1991-92. A comparison of our econometric method with more naive approaches clearly indicates the importance of our attentiveness to endogeneity in evaluating these credit programs. The paper provides separate estimates of the influence of borrowing by both men and women for each of three credit programs (the Grameen Bank, the Bangladesh Rural Advancement Committee (BRAC), and the Bangladesh Rural Development Board's RD-12 program (BRDB) on a variety of household and individual outcomes. These outcomes include the school enrollment of boys and girls, the labor supply of women and men, the asset holdings of women, recent fertility and contraceptive use, consumption, and the anthropometric status of children. We find that credit is a significant determinant of many of these outcomes. Furthermore, credit provided to women was more likely to influence these behaviors than credit provided to men, and had the greatest impact on variables associated with women's power and independence. In short, program credit has a significant effect on the well-being of poor households in Bangladesh and this effect is greater when women are the program participants. vi ACKNOWLEDGMENTS This paper is one of several outputs of a joint World Bank-BIDS study financed by the World Bank under a research project, "Credit Programs for the Poor: Household and Intrahousehold Impacts and Program Sustainability" (RPO# 676-59). We benefited from the comments of participants at a seminar held at the World Bank. We acknowledge the excellent research assistance provided by Signe-Mary McKernan, Deon Filmer, and Hussain Samad. We also acknowledge with thanks the help received from Stella David and Carrie Palma in the production of this paper. vii 1. Introduction This paper evaluates the effects of three group-based credit programs (the Grameen Bank, the Bangladesh Rural Advancement Committee (BRAC), and the Bangladesh Rural Development Board's (BRDB) Rural Development RD- 12 program) on measures of household welfare and on the intrahousehold distribution of resources. These programs are the major small-scale credit programs in Bangladesh that provide credit and other services to the poor, who are otherwise excluded from formal credit institutions because they lack material collateral. While the BRAC is an NGO, the BRDB's RD-12 is a government project, and the Grameen Bank is a rural bank with only about 10 percent of equity owned by the government (the rural poor owning the remainder), all three programs work exclusively with and for the rural poor. Although the sequence of delivery and the provision of inputs vary from program to program, all three programs essentially offer credit to the poor (defined as those who own less than 50 decimals of land, the poor are henceforth referred to as "target" households) with group collateral where group responsibility and loan repayment are tied to lending.' Unlike formal financial institutions, these targeted programs mobilize the poor into groups, give them training, ask them to regularly save a small amount of money, and help them identify a source of employment for generating income. The self-employment activity is, of course, selected by the individual member, but with group approval. The group's incentive to monitor the behavior of individual members is its collective future ability to borrow. Although some have identified an inadequate credit supply as a constraint on production, and hence channeling credit to the rural poor for productive purposes has been emphasized in many developing countries, including Bangladesh, formal financial institutions have hardly succeeded in reaching the poor.2 ' The landholding ceiling of not more than 50 decimals is the general criterion of participation for all three programs. However, for the Grameen Bank, household assets (both land and non-land) must not exceed the value of an acre of land in areas of its operation. The BRAC and BRDB emphasize that in addition to the ownership of less than 50 decimals of land, at least one family member of the participating households should be selling labor to the local wage market prior to program participation. 2 Several types of credit institutions (such as commercial banks, specialized agricultural credit agencies, rural banks, cooperatives and government-supported projects) have been widely used to deliver rural credit. Because of deliberate policy and for other reasons the interest rates were held below the market-clearing rates and credit was thus rationed. Evaluations have found that the rich rural elite have been the principal beneficiaries of these credit 1 This is partly because of the formal institutions' stringent asset-based collateral requirements and partly because of inherent weaknesses in program design.3 Although informal credit markets operate in rural areas, moneylenders usually charge very high rates of interest (for varying reasons), preventing the rual poor from making any sustained gains in income through productive investments. Affordable credit for productive activities would lead, if the effects are sustainable, to improvements in income, welfare and asset positions. Among the poor, this may have a significantly greater impact on women than men, since in many societies the former are burdened by socio-cultural as well as financial constraints.4 The failure of formal institutions to reach the rural poor led to the evolution of credit cooperatives and lending groups as alternative vehicles of rural financial intermediation. Both group-based organizations and credit cooperatives were seen as ways of reaching those who did not otherwise have access to the formal financial system. The risk of default and transaction costs were also expected to decrease as these groups incorporated some form of joint liability and monitoring (for theoretical issues, see Varian, 1990; Stiglitz, 1990). In practice, there have been problems with credit cooperatives and group lending in India, Egypt, Venezuela, Kenya and Lesotho, but examples from Cameroon, Malawi, South Korea, Malaysia, and Bangladesh highlight their successes. The small-scale credit programs, such as the Grameen Bank, BRAC and BRDB RD-12 of Bangladesh, seem to have promoted targeted credit as a means of enabling the poor to break out of the programs and, thus, the major portion of the credit did not reach the intended beneficiaries -- the poorest rural households (World Bank 1975). 3 Inadequate emphasis is placed on the mobilization of rural savings, which has weakened the formal sector institutions. Also, the role of interest rates in stimulating rural financial markets is ignored in program design (Adams and Von Pischke 1984). Since credit is sometimes seen as a process of intermediation (rather than as an input for production), the critical issue is improving this intermediation process through market forces. This involves reducing the costs of intermediation, increasing the dependability of the lender, providing appropriate services to the borrower and enhancing savings mobilization. However, viewed from the framework of imperfect information, financial intermediation does not resolve the problems of screening, incentives and enforcement in the rural credit market (Hoff and Stiglitz 1990). Nor does it ensure that important groups, such as the landless or poor women, gain access to credit. It follows that providing credit and other financial services, especially to the poor and women, requires innovative program design. 4By expanding opportunities for women (relative to men) to undertake productive income-earning activities that affect their status, the welfare of their families may be positively and more than proportionately affected. This is, of course, a testable proposition that will be addressed in this paper. For discussion on the plight of poor women in rural Bangladesh, see World Bank, 1989. 2 vicious cycle of low capital, low productivity, low income, low savings, and consequent low capital. The Grameen Bank, for example, provides credit to members in self-selected groups of five persons, who are collectively responsible for each member's repayment. Members are required to make weekly repayments and minimum weekly savings as well as mandatory contributions to group savings and insurance funds (for details, see Hossain, 1988; Khandker and others, 1994a; Wahid, 1992). Loan recovery rates have been consistently above 90 percent. By the end of 1993, this program had served 1.8 million borrowers of whom 94 percent were women, disbursing the equivalent of $311 million and mobilizing $218 million in savirgs and deposits (Khandker and others, 1994a). Program evaluations suggest that the Grameen Bank's success rests with its creation of a market niche and its outreach to poor rural women (Khandker and others, 1994a; Von Pischke, 1991; Yaron, 1992). Although the committed leadership of founder Professor Muhammad Yunus and the availability of foreign subsidized funds and grants were instrumental in its inception and institutional development, the Grameen Bank has institutionalized a highly decentralized management structure with the potential capacity to operate on market-based resources and without the continued leadership of Professor Yunus. Over time the Grameen Bank has reduced its reliance on foreign funds for on-lending: the foreign proportion of total funding was 58 percent in 1993 compared to 98 percent in 1987 (Khandker and others, 1994a). About 54 percent of the Grameen Bank's 1,040 branches recorded profits in 1993. Similar analyses of the BRAC and the BRDB's RD-12 program suggest that although there is scope for improving cost efficiency, these targeted credit programs have the potential to become viable, given their program design, leadership, and institutional development (Khandker and Khalily, 1994; Khandker and others, 1994b). However, the long-run sustainability of these programs depends to a large extent on the viability of the borrowers that they serve. Since these programs are organizations for the poor and their objective is to alleviate poverty, they cannot sustain their operations unless the accrued benefits to the poor from program participation are sustainable. As such, the critical issues are what these programs have accomplished and for whom, whether their impacts are quantifiable and sustainable and, if so, what policy implications may result. 3 Participation in a targeted credit program such as the Grameen Bank is self-selective; an individual member of a target household is free to choose whether to participate. The decision to participate is based on her/his expected costs and benefits from program participation. Although membership is free, program participation is costly, since group formation, training, and other group activities are time consuming and involve opportunity costs of time spent in group-based activities. But program participation (joining the group) provides access to institutional credit and other organizational inputs that are often inaccessible to many rural households. Once a household decides to participate, it is important to identify the effects of program participation on household and individual outcomes, such as assets, consumption, employment, time allocation and investment in children. This is cruicial in order to quantify whether a credit program achieves its stated goal of reducing poverty. The fragmented literature on credit programs suggests that participants do benefit from the programs, as reflected in higher income and employment among participants (e.g., Hossain, 1988; Wahid, 1993; Amin and others, 1994). However, there are serious weaknesses in the methodologies used in the pre-existing literature to study the impact of credit programs on household outcomes. More rigorous research is needed to fully identify and quantify this impact. A related task is to analyze women's participation in these credit programs and measure the impact on the productivities of women and men and any induced effects on household and intrahousehold consumption and investment. As noted earlier, the major beneficiaries of these group-based credit programs are women who, independently of their husbands, earn cash income from investments made as a result of their access to credit and related inputs. In Bangladeshi society, where the mobility of women is restricted and they are traditionally not allowed to participate in income-earning activities outside the home, direct access to credit and other inputs can significantly influence women's cash earnings. This raises two important questions: (i) Does increased personal income enhance women's influence in household decision- making, and, if so, what are the results on intrahousehold resource allocation? (ii) Do the induced effects of credit programs differ by the gender of the program participants? The third aspect of household and intrahousehold impacts of credit programs is to distinguish credit effects from non-credit effects. Programs such as the Grameen Bank and the BRAC also provide non-credit 4 services to the poor, such as consciousness-raising and skill development training. Such social intermediation is often seen as a complement to financial intermediation for the poor. Since program participation thus provides access to both financial and non-financial services, their relative importance cannot be discerned by examining the total impact of program participation. For policy purposes it is necessary to document the relative importance of these financial and non-financial services in the household or individual behavioral outcomes, in particular to ascertain whether non-financial services are a major factor limiting effective poverty alleviation.5 Very few studies have attempted to identify the causal effects of program participation, let alone credit versus non-credit effects or gender effects of credit and non-credit services or program participation.6 The studies that attempted to evaluate programn impact did so by comparing the outcomes between participating and non-participating households. To the extent that prograrn participation is self-selective, it is not clear whether measured program effects reflect, in part, unobserved attributes of households that affect both the probability they will participate in the programs (and the extent of that participation) and the relevant household outcomes (schooling of children, fertility, asset accumulation). These unobserved factors include such things as unmeasured ability, health and preferences. Moreover, because of the fungibility of credit, it is very difficult to identify the independent effect of credit on household and individual outcomes. Unlike other studies, this one takes into account the endogeneity of program participation and the amount borrowed while assessing their impacts on household and individual behavioral outcomes. The study uses a quasi-experimental survey design to solve the identification problem plaguing earlier attempts to document the program or credit effects. The survey design covers one group of households with the choice to enter a credit program that may alter their behavior and a "control" group which is not given that choice but still allows monitoring of their behavior. Similarly, the identification of program or credit 5This is also important for the program design and placement. Since the major cost of such a program is the administrative cost (see Khandker and others, 1 994a) necessary for group mobilization and training, it is imperative to know what the contribution of the non-financial services of the Grameen Bank and similar programs is for the poor. Evaluation of programs such as the Grameen Bank is extensive in Hossain (1988). There are other studies such as the one carried out by the BIDS (1990) that have also looked at the program effects on a set of household-level outcomes. 5 impact by gender is done based on the comparison between a group of each gender which has a choice to participate and a group which does not have that choice. However, analyzing the program impacts by comparing program-participating households or individuals with control groups may be erroneous because of the possibility that program placement is endogenous. Thus, it will not be clear whether the measured program impact is due to the credit program itself or due to unobservable village characteristics that influence program placement. To avoid such problems, we will use a village-level fixed-effects method to estimate the impact of targeted credit programs on various household and individual outcomes, including differential effects within the household attributable to the gender of the borrower, identified through a quasi-experimental survey design. The remaining portion of the paper is organized as follows. Section 2 discusses a household model framework to motivate the specification of conditional demand equations that provide estimates of the impact of credit program participation by gender on a set of household- and individual-level outcomes. Section 3 presents the quasi-experimental survey design of household and community surveys that were conducted in Bangladesh during 1991-92 and presents the descriptive statistics of major variables identified for model estimation. Section 4 presents the results of the determinants of program participation and credit and the impact by gender on household- and individual-level outcomes. The concluding section summarizes the results and provides policy conclusions. 2. Evaluating program impact: a framework To motivate the evaluation of the effects of group-based credit program participation on household behavior and intrahousehold resource allocation, consider a simple model that generates an efficiency argument for targeted credit for the rural poor. Assume that households of size n, consisting of two working age adults (the male head and his wife) plus n-2 dependents, maximize a lifetime utility function containing time-specific utility functions of the form Ut = U(QI H... Hz,11* (1) 6 where Qi is a set of market goods consumed by household member i, the set of non-market household- produced goods allocated to member i is Hi, and Ii is leisure time consumed by household member i. As a generalization of (1), each of the two adult household members, denoted byf and m, wishes to maximizes his (if m) or her (ifj) own utility ui, Uii = Ui (Q.. Q, Hi... H., i *l)i= m (2) where household social welfare is some function of the individual utility functions U, U(uy u,,, a simple form of which is U,, = XUft+(1-X)Ur,, 0X'1 (3) in which x is the weight given to women's preferences in the household's social welfare function. The parameter X can be thought of as representing the bargaining power of female household members relative to males in determining the intrahousehold allocation of resources. When X=O, female preferences are given no weight and the household's social welfare function is identically that of the males.7 The household-produced goods H include "household care" activities such as food preparation, child care, and the gathering of fuel.8 H= H(Lth,LJh,G;F) (4) where L,,h, and LAf are time devoted to the production of H by males and females, respectively, G is a vector of market goods used as inputs in the production of H, and F is a vector of technology parameters that affect efficiency in H good production. 7 The reader is referred to McElroy (1990), McElroy and Homey (1981), and Manser and Brown (1989) for a formal exposition of game theoretic approaches to household decision making. 8 Some of these household goods, such as food preparation and child care, cannot be stored for consumption in later periods. 7 Due to socio-cultural factors, relatively few poor women work in the wage labor market. The reservation wvage for market work is, therefore, relatively high.9 In addition to this preference effect on female wage employment, workers typically must commit to a full day's employment even in the spot labor market.10 If men's time (or that of other household members) is a poor substitute for women's time, and if important H-good outputs, such as child care and food preparation, must be "produced" daily (cannot be stored), then working a full day may entail foregoing the production and consumption of highly valued H- goods. Thus, the non-storability and time-intensity of production of household goods H, the indivisibility of time allocation in the wage labor market, and high reservation wages due to cultural impediments to wage employment outside the home all result in most women being engaged in the production of household goods H in every period to the exclusion of employment in market activities. These effects are magnified if X is small and male preferences tend to favor certain kinds of H-goods produced on women's time. However, there are also economic activities that produce goods for market sale that are not culturally frowned upon. These activities, producing what we refer to as Z-goods, permit part-day labor and do not require that production occur away from the home. Although many of these production activities can be operated at low levels of capital intensity, for many Z-goods a minimum level of capital is necessary. This minimum is often the result of the indivisibility of capital items. For example, dairy farming requires no less than one cow, and hand-powered looms have a minimum size. For other activities, such as paddy husking, where the indivisibility of physical capital is not an issue, transaction costs (or the high costs of information) place a floor on the minimal level of operations. In many societies these indivisibilities may be inconsequential, but among the rural poor of many developing countries, including Bangladesh, household income and wealth is so low that the costs of initiating production at minimal economic levels are quite high. 9 Poverty alleviation programs, such as the Rural Works Programs, which target households by drawing them into yin-kind) wage labor have a comparatively small direct effect on the time allocation and productivity of women. ° In addition, transportation and other transaction costs in labor markets may be so high as to make part-day labor unremunerative. 8 Formally, we represent the production function for the Z-goods as: Z = Z(K,L,,Lft, A;J) (5) where Lmz and Lft are labor time of head and wife devoted to the production of Z, K is capital in Z production, A is a vector of variable inputs, and J is a vector of technology parameters that affect efficiency in Z-good production (information). Positive production requires a minimal level of capital K K 2 Kmin. The production function (5) can be operated at a non-zero level when L,,.2 or Lf are zero, but not when both are zero. For example, in the case of milk production, although at least one cow is required, any person's labor can be used to obtain the milk. In other cases, K,,i, may represent the minimal information required to produce and market home production. Households maximize lifetime utility subject to a budget constraint that requires that the present discounted value of expenditure on goods and leisure equal the present value of all wealth, defined as assets plus the discounted present value of the time endowments, and the two production function equations (4) and (5). Household ability to borrow has significant influence on the time path of household consumption. Households having very low levels of initial assets as collateral may not be able to borrow to achieve the minimum capital requirements necessary to operate the Z-good activity. At very low levels of income and consumption, reducing current consumption to accumulate assets for this purpose may not be optimal because it may seriously threaten health (and production efficiency) and life expectancy, as shown in Gersovitz (1983). As a result, for many households, the Z-good activity is never carried out (and Lf = 0) and women who do not work in the wage labor market devote all their time to production of the non-market good H and to leisure. This simple model, which has some of the features of the "two-gap" models of aid and development, demonstrates the role of a credit program.1' For the very poor, access to credit may alter the 1I1 In the two-gap model, the effect of foreign aid on the rate of growth of output is high as long as imported capital requirements exceed labor availability. The two-gap model requires that domestic capital cannot substitute for imported capital, and that labor cannot be substituted for imported capital in production. Without sufficient capital, labor is unemployable. In the household model described here, labor is also unemployed, or rather underemployed, for lack of a minimum level of capital in the production of the Z-good. As in the two-gap model, this result requires 9 optimal time allocation for women from home production of H to market production of Z. Conceivably, if household consumption is at or near minimal levels necessary for survival, so that saving is almost infinitely costly, even a small quantity of credit for the purchase of Kmin can have a large impact on household welfare by shifting women's time from the production of H, which may have a low shadow value to the household, to high marginal product Z-good production. In addition, progran participation may alter the technology parameters, F and J, by providing information and training, which may affect efficiency in H- or Z-good production and, hence, income and consumption. It is straightforward to allow for heterogeneity in preferences (including X) and in human capital endowments (including ability) in the model. The introduction of a rural credit program into a poor village economy composed of heterogeneous households may induce some households to participate and borrow to finance K,,i,. Since Z-goods can be produced with part-time and flexible labor and can take place at home where an H-good, such as child care, can be jointly produced, women who undertake Z-good production will allocate time for it by reducing time in one or both other activities (H-good production and leisure). Some households, in which the marginal utility of H-good production is high (perhaps because x is small), or in which wage labor opportunities are superior, may choose not to participate in a credit program. The production of H-goods may rise or fall in households that initiate program borrowing in order to start Z-good production. The direction of change in H-good production depends on the size of the income effects, the substitutability of market inputs G with time inputs, and the degree to which a unit of (women's) time can jointly produce the Z-good and the H-good. Program participation may also affect household allocations by altering the value of X, the weight given women's preferences in the household's social welfare function. The value of x may increase with the greater bargaining power of women, resulting from having additional resources under their control through targeted credit and training and from the "consciousness raising" acquired from group participation (such as the Grameen Bank's Sixteen Decisions). the non-substitutability of other factors (including labor) for capital in the production of Z over some range of the production technology. 10 The empirical model From the model presented above, the reduced-form determinants of credit program participation include the prices of market time, the price of the purchased market good Q, the prices of the market inputs into H-good production including the cost of averting a birth and other determinants of fertility, the prices of variable inputs into Z-good production, the price of the capital good, age and education levels of the borrower and spouse, access to transfers from non-resident relatives, and village-level characteristics (V).12 Whether or not poor households, particularly the women, are credit-constrained is a complex issue. Rashid and Townsend (1993) present an excellent review of this issue in the context of targeted group- based lending. They suggest that risk, private information, communications and enforcement difficulties may result in inefficient consumption and production outcomes. There is substantial evidence of the limited participation of women in the formal credit market due to lack of collateral and education, the health risks and intermittency of employment associated with childbirth, and cultural barriers. Rashid and Townsend note that the evidence does not in itself imply that outcomes are inefficient if, for exarnple, women have access to other sources of finance such as transfers or if male household members obtain funds for female household members. This paper does not test whether credit constraints are binding for women but whether or not access to group-based lending programs alters allocations and whether or not there is a difference if a man or a woman is the participant. It is important to note that the problem of "credit rationing" here is essentially different than that of, say, a fanner who needs to borrow to finance farm inputs (Feder and others 1988). If a farmer is credit-constrained in any season he cannot use inputs at the profit maximizing level during that season (e.g., Feder and others 1988). In the case of group-based lending to the landless, the time path of credit allocated to a member is part of the dynamic optimization problem of a group, and the level of credit provided each individual in the group is tailored to fund a new self-employment project of certain size. 12 The terms of the loan may affect loan demand, but those effects are not statistically identifiable since all Grameen Bank or other credit program loans carry the same terms. Local credit market conditions, including the informal lending market, and the availability of relatives able to transfer funds, will affect the individual demand for credit. 11 Moreover, the cost of credit includes not only the interest rate, but also the timing of repayment and the penalties associated with default. Group-based credit is packaged with both responsibilities (meeting attendance, forced saving, shared default risk) and benefits (training, insurance, consciousness-raising). If there was no monitoring of the use of borrowed funds and no group responsibility and decision-making in the lending program, individuals would likely want to borrow much more than they actually do in order to capture the premiums associated with the soft terms of the loan. In some sense, the monitoring of credit use makes all program participants "credit constrained." Whatever the case, all participating households are presumed to be in the same credit demand regime given the practical impossibility of any other treatment. Estimation strategy A primary focus of this paper is to estimate the impact of credit programs on various household outcomes such as household consumption, time allocation, asset accumulation, contraceptive use, and investments in children. We propose to estimate the conditional demand equation for each outcome to be investigated, conditioned on the household's program participation as measured by the quantity of credit borrowed."3 Consider the reduced fonn equation (6) for the level of participation in one of the credit programs (Cu), where level of participation will be taken to be the value of program credit C(j = X#Pc + Vrc + Zut + £ c (6) where Xij is a vector of household characteristics (e.g., age and education of household head), Vy is a vector of village characteristics (e.g. prices and community infrastructure), Zij is a set of household or village 13 The quantity of credit is, of course, only one measure of the flow of services associated with participation in any one of the group-based lending programs. As the introductory section has made clear, they are much more than just lending institutions. Nevertheless, the quantity of credit is the most obvious and well measured of the services provided. In work in progress, we are attempting to discern the importance of the non-credit services provided group members by estimating conditional demand equations for the same set of outcomes investigated in this paper by conditioning on a variety of measures of non-credit services provided. Since we do not control for these other services in this paper, the estimated credit effects reported below should be interpreted to (imperfectly) include the effects of all aspects of program participation. 12 characteristics distinct from the Xs and Ps in that they affect Cy but not other household behaviors conditional on Cy (see below), j3,yc, and Jc are unknown parameters, and S ' is a random error having three components Y= p. + rl + ec (7) where Pij is an unobserved village-specific effect, Tij is an unobserved household-specific effect, and £ c is a non-systematic error uncorrelated with the other error components or the regressors. The conditional demand for household outcome YU conditional on the level of program participation Cy is Ye/ = Xy By + Vry + CJ8 + (8) where Yc,y,, and 6 are unknown parameters and E Y is comprised of g = (cLp. + p.IY) + ( 07+ + +1 ) + (9) where a and 0 are parameters (corresponding to correlation coefficients), p Y and i Y are additional village- and household-specific errors uncorrelated with ;j and nij, respectively, and 6 Y is a non-systematic error uncorrelated with other error components or with the regressors. If a#O or 00 the errors s" and s are correlated. Econometric estimation that does not take this correlation into account will yield biased estimates of the parameters of equation (8) due to the endogeneity of credit program participation Cy. 13 Why might credit program participation be endogenous? The endogeneity of credit program participation (represented here by the amount of credit borrowed from the targeted credit program) in the household outcome (Y,j) equations may arise from common village-specific unobservable variables, the .z, and from common household-specific unobservables, the Thij. We note the following sources: 1) Non-random placement of credit prograns. It is unlikely that credit programs are randomly allocated across the villages of Bangladesh. Indeed, program officials note that they often place programs in poorer and more flood-prone areas, as well as areas in which villagers have requested program services. Recently, Pitt, Rosenzweig and Gibbons (1993) have shown that treating the timing and placement of programs as random can lead to serious mismeasurement of program effectiveness in Indonesia. Comparison of the two sets of villages as in a treatment/control framework would lead to a downward bias in the estimated effect of the program on household income and wealth (and other outcomes associated with income and wealth) and could even erroneously suggest that credit programs reduced income and wealth if the positive effect of the credit program on the difference between "treatment" and "control" villages did not exceed the negative effect that induced the non-random placement. 2) Unmeasured village attributes affect both program credit demand and household outcones Yj.. Even if credit programs are randomly placed by the agencies involved, village attributes that are not well measured in the data may affect both the demand for program credit and the household outcomes of interest. These attributes (the pe's) include prices, infrastructure, village attitudes and the nature of the environment including climate and propensity to natural disaster. For example, the proximity of villages to urban markets or transport may influence the demand for credit to undertake small-scale activities but may also affect household behavior through altering attitudes and access to urban amenities. 3) Unmeasured household attributes affect both credit demand and household outcomes Ye,. These attributes (the al's) include endowments of innate health, ability, and fecundity, as well as preference heterogeneity. Consider the possibility that households are heterogeneous with respect to the relative treatment of males and females. It seems possible that households that are more egalitarian in their 14 treatment of the sexes are also more likely to have female household members participate in credit programs and are also more likely to provide more resources to females than otherwise identical but less egalitarian households. Ignoring this heterogeneity would wrongly attribute a more egalitarian intra- household resource distribution to the credit program, where it is actually due to the more "egalitarian" preferences of self-selected households themselves. Econometric approach The standard approach to the problem of estimating equations with endogenous regressors, such as equation (8), is to use instrumental variables. In the model set out above, the exogenous regressors Zi, in equation (6) are the identifying instruments. Unfortunately, it is difficult to find any regressors Z4 that can justifiably be used as identifying instrumental variables. The exogenous regressors Z4 must satisfy two conditions: (i) they must affect the decision to participate in a credit program (that is, irO), and (ii) they must not affect the household outcomes of interest YUj conditional on program participation. An approach motivated by demand theory is to use the price of the endogenous variable, conditioned upon as an identifying instrument. The most obvious measure of the "price of credit program participation" is the interest rate charged, but this is ruled out here since it does not vary across the sample.14,15 Using either interest rates or measures of the cost of information as identifying instruments fails for another reason. If households are responsible for repaying the loans made in the name of individual members and jointly make the credit decisions of individual household members, and there is a single price for credit to all members of a household, then gender- or individual-specific allocations of credit to multiple-person households suffer from the classic problem of more goods than prices. An individual- 14 Even if interest rates varied across the sample, it is likely that some of this variation reflects unmeasured household attributes unknown to us but known to the lender and likely to be part of the E Y. efror tern, and hence be an invalid instrument. 15 Another measure of the "price of credit program participation" is some proxy for the information costs associated with learning about these credit programs. To some extent, this depends on the qualities of the credit program organizers and staff. Our survey collected information on the educational background, experience, age and gender of credit program organizers and other staff. There was a substantial number of missing values in these data and these measured attributes tended to vary little across the sample. In any case, the validity of these variables requires that the credit programs allocate program organizers randomly across villages, which is uncertain. 15 specific price of credit (informational or otherwise) to the female adults of a household is likely to be related to the borrowing behavior of male adults and unobserved household attributes. Village fixed-effects estimation, which treats the village-specific error l.4 as a parameter to be estimated, eliminates the endogeneity caused by unmeasured village attributes including non-random program placement. However, fixed-effects estimation raises issues of consistency and computational difficulty. Measured program credit is a limited dependent variable since not all eligible households participate in the credit programs. Some relevant household outcomes -- such as schooling of children, labor supply, and assets -- are also limited dependent variables. As is well known, fixed effects estimation in this case generally yields inconsistent parameter estimates without large numbers of observations on each fixed effects unit. Heckman (1981) provides Monte Carlo evidence that with 8 or more observations per fixed effects unit, the inconsistency problem becomes relatively minor. The average number of target households per village in this study is 20.2. There are 87 village units in the data, 72 with credit programs, and joint estimation of credit use by gender (see below) with each household outcome (such as schooling or labor supply) implies that nearly 200 fixed-effects parameters need to be jointly estimated. Even with village fixed effects, the endogeneity problem still remains if 00; that is, if there are common household-specific unobservables affecting credit demand and household outcomes. Lacking identifying instruments ZQj (exclusion restrictions), another approach is required for identification. Realizing this, the sample survey was constructed so as to provide identification through a quasi- experimental design. To understand the nature of this quasi-experimental design, consider the classic program evaluation problem with non-experimental data. Individuals can elect to receive a treatment offered in their village (or neighborhood). The difference between the outcome (YU) of individuals who chose to receive the treatment and the outcome of those who chose not to is not a valid estimate of the treatment's effect if individuals self- select themselves into the treatment group. Lacking any Z4 (or panel data on individuals before and after treatment availability), one method of identifying the effect of the treatment is based upon (presumed) knowledge of the error distribution. This is the standard sample selection framework of Heckman (1976) and Lee (1976). If the errors are assumed to be normally distributed, as is common, the treatmnent effect is 16 implicit in the deviations from normality within the sample of treatment participants (Moffitt 1991). The nonlinearity of the presumed distribution is crucial. If both the treatment and the outcome are measured as binary indicators, identification of the treatment effect is generally not possible even with the specification of an error distribution. Now consider a "natural experiment" in which the treatment is not available in every village and this availability is not correlated with observables affecting the outcome Yu; that is, treatment availability is randomly placed across villages. In this case, the presence or absence of treatment choice is a legitimate identifying variable, requiring samples of individuals from villages with treatment choice as well as villages without it (Moffitt 1991). What if the availability of treatment were correlated with village-specific unobserved attributes? Then, net of these unobserved attributes, one could identify the parameters of all the observed exogenous household and individual regressors by fixed-effects estimation with the subsample drawn from non-treatment villages only. For example, in equation (8), Cij is identically zero for all households in non-program villages, so that village fixed effects estimation of (8) on that subsample yields consistent parameter estimates of By. The credit-effect parameter 6 and the parameters yy are not identifiable from any part of the sample, since they are "captured" by the village fixed effects. The parameters of interest, 6, the effect of participation in a credit program on the outcome YU, can be identified if the sample includes households in villages with treatment choice (program villages) that are excluded from making a treatment choice by random assignment or some exogenous rule, which would be the exclusion of households owning more than 0.5 acres of land from any of the three credit programs. Data on the behavior of households exogenously denied program choice in this way is sufficient to identify the credit program effect. Thus, rather than relying solely on nonlinearity arising from the specification of an error distribution to identify the program effect 6, another piece of identifying information is available. A comparison of the outcome Yij between households with program choice and those without it, conditioning on all village effects and observed household and individual attributes, is an estimate of the program's effect on that outcome. 17 To illustrate these ideas more formnally, consider a binary treatment (IC=1 if treatment chosen, 0 otherwise) and a binary outcome (4,=1 if outcome is true, 0 otherwise). This is the most difficult model to identify in that nonlinearity is insufficient to identify the credit effect parameter &. The model is C= Xy + (10) c= I if c > 0, Ic = 0 otherwise * ~~~~~~~~~~~~~(11) y = Xy + °lc + Sy Iy = if y > 0, Iy = 0 otherwise where c and y are latent variables associated with, respectively, treatment choice and the outcome, Xc and Xy are vectors of regressors, y, , and 8 are parameters to be estimated, and ec and ey are errors distributed as bivariate normal with unit variances and correlation coefficient p. The parameter 8 represents the treatment effect. The log-likelihood function for this model is logL(y, ,B ,o, p) = E log ®2(Xcydc, (Xyp + 81C)dy, pd4d (12) where 02 is the bivariate standard normal distribution, and dc = 2*Ic - 1 and dy= 2*Iy - 1. If cc and ey are not independent (p#O) and Xy includes all the variables in Xc, the parameters in equation (11) are not identified (Maddala 1983, page 122-123). That is, lacking exclusion restrictions, if the choice into the treatment group is selective, identification of the treatment effect on a binary outcome is not possible with a sample of self-selected individuals. Consider the addition of a subsample of individuals for whom treatment is (exogenously) not available. The log-likelihood becomes 18 log L(y ,0,8, p)= = logO2(Xcydc,(Xy +O Ic)dy, pdcdy) choice (13) + i log 0(Xyj3dy) nochoice where 0 is the univariate standard normal distribution, and "choice" and "no choice" represent those individuals in the sample who have a treatment choice and those for whom no treatment is available. All of the parameters of the model are identifiable even if the errors are not independent and exclusion restrictions do not exist. If program placement is random, all of the households in the second part (no choice) of the likelihood could come from villages without programs. Identification of the credit program effect is then essentially a comparison of outcomes across villages conditioned on village and household/individual observables. If program placement is not random only with respect to village effects, then we can control for village effects by adding a village-specific intercept plk to the vector of regressors. Distinguishing between households with no choice because they reside in a non-program village and households residing in a program village that do not have choice because of the application of an exogenous rule, the likelihood can be written as: logL(y ,B ,o, , p) = L logO2((L + Xcy )dc,(PLk + Xyf + 81c)dy, pdcdy) choice + L lOgO((gk+Xyp)dy)± + log0((. + Xy )dy) (14) no choice no choice progroni non-progranr village village where pk are the village-specific intercepts for program villages and pm are village-specific intercepts for non-program villages. It is the ability to estimate the marginal probability 0(pk + XypB)dy) of the outcome directly from a subsample of households that makes this identification possible.16 16 Implicit in this setup is the assumption that the effect of the treatment (8) is the same for all individuals, an assumption which is common in the program evaluation literature (Moffitt 1991). Furthermore, the model is not 19 Underlying identification in this model is the assumption that land ownership is exogenous in this population. Although it is clearly non-standard to use program eligibility criteria for purposes of identification in most instances of program evaluation, we think its use is well justified here. Unlike the evaluation of job training programs, health/nutrition interventions, and many other types of programs, where lack of job skills, lack of health, or insufficiency in some other behavior are both criteria for eligibility and the behaviors the programs directly act upon, land ownership is used as the primary eligibility criteria for these credit programs only to proxy for unreliable indicators of income, consumption or total asset wealth. Land ownership is simple to quantify, understood within the community and unlikely to change in the medium-term. Market turnover of land is well known to be low in South Asia, and the absence of an active land market is the rationale given for the treatment of land ownership as an exogenous regressor in almost all the empirical work on household behavior in South Asia.'7 A number of theories have been set forth to explain the infrequency of land sales. Binswanger and Rosenzweig (1986) analyzed the set of material and behavioral factors which are important determinants of production relations in land-scarce settings, and concluded that land sales would be few and limited mainly to distress sales, particularly where national credit markets are underdeveloped. Rosenzweig and Wolpin (1985) set out an overlapping generations model incorporating retums on specific experience which uses low land turnover as an implication and, using data from the Additional Rural Incomes Survey of the National Council of Applied Economic Research (NCAER) of India, found a very low incidence of land sales. Even if land ownership is exogenous for the purposes of this analysis, it is necessary that the "landless" and the "landed" can be pooled in the estimation of reduced form equations (6). To enhance the validity of this assumption, we restrict the set of non-target households used in the estimation to those with nonparametrically identified. That is, if the linear indices X.y and (Xyp+8I1) were replaced by nonparametric functions of the Xs and I, the model is not identified. 17 For example, in a classic paper in the field, Rosenzweig (1980) tested the implications of neoclassical theory for the labor market and other behaviors of farm households in India by splitting the sample on the basis of land ownership, treating the sample separation criterion as non-selective. 20 less than 5 acres of owned land. In addition, we include the quantity of land owned as one of the regressors in the vector Xi, and include a dummy variable indicating the target/non-target status of the household. Identification of the impact of gender-specific credit A principal objective of this research is not just to determine whether credit programs for the rural poor affect household behavior in important ways, but whether the sex of the program participant matters. For that reason, the reduced form credit equation is disaggregated by gender Ciw- = Xjffl + Viyqf + S ' (15) 00n = XiYP43. + Vi ym + s i (16) where the additional subscripts f and m refer to females and males respectively. The conditional household outcome equation not only allows for separate female and male credit effects, but also for different effects for each of the three credit programs Yu} = Xuy + V'yy + ; C,ffDy,k8Jk + E CyDyi&k8k + , Y (17) k k where Dk is a dummy value such that Dk=l if the individual participates in credit program k and Dk=O otherwise (kBRDB, BRAC, and Grameen), C,f is the credit participation of females in household i of villagej, Cijm is similarly defined for males, and the 6's are program-specific parameters specific to each sex. Introducing gender-specific credit is not a trivial generalization of the econometric model. First, it is likely that the errors s are correlated with the errors E S; that is, there are common unobservables influencing the credit program behavior of both women and men in the household. Second, additional identification restrictions are required when there are both male and female credit programs with possibly 21 different effects on behavior. The first issue is computational; bivariate probability distributions need to be evaluated when estimating equations (15) and (16). Furthermore, if Yu, is a limited dependent variable and limited information maximum likelihood methods are applied to the full system given by (15), (16) and (17), trivariate probability distributions need to be evaluated. The second issue, that of identification, is handled by an extension of the quasi-experimental setup described above. All of these group-based credit programs have single-sex groups. It was established above that identification could be achieved, even if program placement was non-random, by including in the estimation sample observations for households that are in villages with credit programs but are unable to join because they possess more than the threshold quantity of land, considered an exogenous rule. Similarly, identification of gender-specific credit is achieved by a quasi-experimental survey design that includes some households from villages with only female credit groups, so that even males in landless households are denied the choice of joining a credit program, and some households from villages with only male credit groups, so that even landless females are denied program choice. In particular, of the 87 villages in the sample, 15 had no credit program, 40 had credit groups for both females and males, 22 had female-only groups and 10 had male-only groups. Table 2.1 provides the details by type of credit program. Since each village had only one type of credit program available, there is no need to model which program members of a household join -- the BRDB, BRAC or Grameen.18 While the likelihood given by (14) illustrates the general principle and method used in estimating the effect of credit programs on behavior in Bangladesh, the actual likelihoods maximized are substantially more complex for the following reasons: 1) The likelihood for binary and tobit outcome variables involve trivariate and bivariate normal distribution functions because two credit equations ((15 and (16)) are being estimated simultaneously with the outcome equation. In addition, some of the outcomes are continuous (such as child anthropometry and expenditure) or tobit (such as labor supply). In each case, estimation was done by limited information maximum 18 A small number of individuals belonged to credit programs that met in other villages. For example, there were some women who belonged to Grameen Bank groups even though there was no Grameen Bank group in their village. These participation decisions were treated as exogenous in the analysis. 22 likelihood. For the tobit case, our method is a substantial generalization of the LIML likelihoods presented in Smith and Blundell (1986) and Rivers and Vuong (1988) for limited dependent variables because the endogenous right-hand-side variables are also tobits. 2) Observations on Yy are sometimes for multiple members of the same household, as in child anthropometry and schooling where more than one child per household appears in the sample, or observations on the same individual in different seasons, as in labor supply. Thus, it is unlikely that the errors are independently and identically distributed. Unobserved household attributes that affect one child's schooling or nutrition are likely to also affect the schooling and nutrition of that child's sibling. Not accounting for this lack of independence will yield biased estimates of the parameter covariance matrix (t- ratios). Our approach is to use an asymptotic bootstrap estimator of the covariance matrix, essentially White's (1980) heteroskedasticity-consistent covariance matrix estimator in which the outer-product of the derivatives of the log densities (commonly known as the Berndt-Hall-Hall-Hausman or BHHH estimator) is defined so that the log density contains the full set of observations for any household or household member. The log densities thus defined are independently and identically distributed and the resulting pararneter covariance matrix is consistent. 3) The sample design is choice-based (see Section 3.1 below). In particular, program participants are over- sampled. The use of choice-based sampling somewhat complicates the econometrics but allows researchers to get the most statistical efficiency per dollar spent on data collection. Lancaster and Imbens (1991) have demonstrated the large efficiency gains to be obtained from a well-designed choice-based sampling strategy and Lancaster (1992) has reviewed methods for estimation with choice-based samples. Not correcting for the choice-based nature of the sample would lead to biased parameter estimates. The Weighted Exogenous Sampling Maximum Likelihood (WESML) methods of Coslett (1981) were grafted onto the limited information maximum likelihood (LIML) methods described above in the estimation of both parameters and the parameter covariance matrix. To remind the reader of these crucial aspects of the maximum likelihood approach taken in this paper, the method is referred to as WESML-LIML-FE, which stands for Weighted Exogenous Sampling Maximum Likelihood - Limited Information Maximum Likelihood - Fixed Effects. 23 3. Survey design A multi-purpose quasi-experimental household survey was conducted in 87 villages of 29 thanas in rural Bangladesh during the year 1991-92. The survey's major focus was to analyze the credit and other input effects of three major credit programs and was designed to include both target (qualified to participate) and non-target households from both program and non-prograrn (i.e. control) areas. The sample consists of 29 thanas (subdistricts) randomly drawn from 391 thanas in Bangladesh. Out of the 29 thanas selected for the study, 24 have at least one of the three credit programs in operation, while 5 thanas have none. That is, the proportion of thanas surveyed under each progran coverage is 28 percent, while 16 percent of the 29 thanas do not have any program. The program thanas are distributed among four regions in the following way: 8 thanas in Khulna region, 3 thanas in Chittagong region, 10 thanas in Dhaka region, and 8 thanas in Rajshahi region.19 Three villages in each program thana were then randomly selected from a list, supplied by the program's local office, of villages in which the program had been in operation at least three years. Three villages in each non-program thana were randomly drawn from the village census of the Government of Bangladesh (GOB). For both prograrn and non-program thanas, if a village contained less than 50 and more than 600 households it was dropped from the list and replaced by another randomly selected village in this size class. Furthermore, if the selected village had between 301 and 600 households, the household census (see below) was begun from one randomly selected corner of the village and stopped when some 200 households were covered. A census was conducted in each village selected for the study. The purpose of the village census was to help identify target (i.e., those qualified to join a program) and non-target households, as well as to 19 Note that more than one-third of the Chittagong region was devastated by the 1991 cyclone and dropped from sampling. This is why few thanas are drawn from the Chittagong region. It is also worth noting that there are several thanas where the three credit programs under study overlap. However, although programs may overlap in a thana, they do not overlap the same individual. Because of program design, the program officials ensure that no individual is a member of two or more programs simultaneously. Technically, therefore, a particular thana could have been drawn twice for two different programs. This did not happen in the actual sample selection, but some of the 24 program thanas do have more than one credit program in operation. 24 identify program participating and non-participating households among the target households in any village. From the village census list of households, 20 were drawn from each program and non-program village from both target and non-target households for the in-depth household survey. The distribution of these 20 households by target and non-target groups was 17:3 in each program village and 16:4 in each non-program village. A random sampling technique was used to draw the required sample of 17 target group households from the non-program villages as well as the sample of 3 non-target households from both program and non-program villages. However, a simple random sampling technique could not be applied to draw target households from the program villages; although a good percentage of the target households in program villages did participate in the program, we did not know whether this percentage was above 50 percent. This was significant because the survey design required a sufficient number of program participants among the target households to enable us to analyze the credit or program participation impact on various household and individual outcomes. Instead, a stratified random sampling technique was used to draw households in the ratio of 12:5 (i.e., 12 program participants and 5 non-participants) from the list of target households in the program villages.20 A total of 1,798 households was drawn for the in-depth household survey, where 1,538 were target households and 260 non-target households. Among the target households, 905 were found to be participating in any of the three credit programs, representing 59 percent of the target households sampled for the study. The actual distribution of program participating and non-participating households in the study villages, according to the village census, is 44:66. Therefore, the households were disproportionately drawn for the study and thus the sample ratio needed to be adjusted to make it representative of the actual village distribution. In addition to the general household survey (that collected household- and individual-level information on income, employment, education, health, consumption, borrowing, savings, etc.) and a 20 The sample size and its ratio between participating and non-participating households are different in five program thanas (2 for the Grameen Bank, 2 for the BRAC and I for the BRDB) which were also selected for nutrition surveys. In each nutrition study thana the number of the target households drawn was higher than 17, although the number of non-target households drawn remained the same (i.e., 3). Thus, in the Grameen Bank and BRAC nutrition thana 20 target households were drawn from the target households where the ratio between participating and non-participating households was 16:4. By contrast, for the BRDB nutrition thana 25 target households were drawn for in-depth study at a ratio of 18:7 between participating and non-participating households. 25 nutrition sub-survey (that collected individual dietary intake, weight, and height), a village survey questionnaire was also administered. Note that the general household survey was conducted three times over the crop cycle year 1991-92 to match the three crop seasons, and informnation on village-level prices and wages was collected in the same manner. On the other hand, the nutrition survey was conducted twice over the same year to collect dietary intake information during the peak (December to February) and slack (July to September) seasons in terms of food availability. In addition, data were also collected on village- level infrastructures that tend not to vary seasonally. Data description Table 3.1 presents the weighted mean and standard deviations of all exogenous variables used in the regression. Because the samples drawn are not representative of the village population, the means of the variables are adjusted by appropriate weights based on the actual and sample distribution of the households covered in the study villages. The sample of individuals aged between 15-64 is quite young, since the mean age is only 23 years. Approximately half of the samnple is female. The educational level is very low, averaging only 1.4 years. About 61 percent qualify to join one of the credit programs under study. Those who have joined a credit program have, on average, 3.7 years of membership. The number of potential transferees of the households who own more than 50 decimals of land provides an alternative source of credit. As the table suggests, the average number of such relatives (for example, parents, sons, daughters, brothers, sisters, uncles and aunts) of the household head and his/her spouse is less than 1. Approximately 11 percent of target households are BRAC members, while 6 percent belong to the BRDB and 8 percent to the Grameen Bank. The average household landholding size is only 30 decimals. About 13 percent of households do not have a spouse present; however, 95 percent are headed by men. The average education of the household head is 1.9 years of schooling, the average age is 41 years. The average highest educational level among the adult females in each household is 1.6 years, and the average highest educational level among adult males in each household is 3.1 years. Only 3.5 percent of the households have no adult male, while an even smaller 1.7 percent have no adult female. 26 About 95 percent of participants in all three programs borrow. The average (cumulative) amount borrowed since November 1986 is greater for female than for male borrowers of the BRAC and GB, although it is higher for male than for female borrowers of the BRDB.2' However, the amount borrowed by females from the GB is the largest among the loans received by men or women from any program. Women's credit from the GB is about 8 times larger than that from the BRDB and 3 times larger than that from the BRAC. Women's credit from the GB is also 3 times larger than men's credit from the GB. Since loans from all three programs are annual, the higher loan amounts for female or male borrowers of the GB may represent the longer program participation of GB borrowers relative to borrowers from other programs. The explanatory variables also include availability of a primary school (68 percent of households reported having a primary school in their village), rural health center (30 percent), family planning center (10 percent), and Dai/midwife (67 percent). They also include the village-level prices of major commodities and the wages of male and female labor. Although few women participate in the wage labor market (about 19 percent of the villages have no active wage labor market for women), the female wage is about 40 percent of the male wage. Even if one assumes that participation in any of these targeted credit programs involves foregone wage income, it appears that women have a lower opportunity cost than men in joining the Grameen Bank or another program. Although the availability of a commercial bank in the area does not ensure a large number of targeted households' borrowing from a formal financial institution, its presence may nevertheless increase the availability of credit. The average distance from a study village to a commercial bank is about 3.5 km. Table 3.2 presents some household- and individual-level outcomes that are of particular interest in this paper and disaggregated by various groups -- participants and non-participants of program areas, target households of non-program areas, and aggregates for all households of all areas. There are differences in behavioral outcomes between participating and non-participating households, between men and women and between boys and girls. For example, contraceptive use among married women aged 14-50 is 42 percent for program participants, 37 percent for non-participants in program areas, and 36 percent among target 21 Credit is deflated by regional cost-of-living indices to constant Taka. 27 households in non-program areas. About 68 percent of women had a child in the 3 years prior to the survey among participants, 70 percent among non-participants, and 72 percent among the target households in non- program villages. School enrollment among children aged 5-17 is 54 percent for girls and 57 percent for boys among participants, 43 percent for girls and 41 percent for boys among non-participants, and 54 percent for girls and 48 percent for boys among the target households of non-program villages. The hours worked per month by women for cash-earning activities are 40 for participants, 38 for non-participants and 44 for target households in non-program villages. By contrast, the hours worked by men are 202 among participants and non-participants in program areas and 195 among target households in non-program areas. More interestingly, the non-land assets owned by women are higher among participants (Tk. 2,267) than among non-participants (1,145) and much higher than among target households in non-program areas (Tk. 585). Our objective is to analyze whether program participation has benefited the poor, especially women and children. 4. Results In this section we present and interpret the results of estimating conditional demand equations of the form given by equation (16) for a wide variety of behaviors. All of the parameter estimates are WESML-LIML-FE (Weighted Exogenous Sampling Maximum Likelihood-Limited Information Maximum Likelihood-Fixed Effects) estimates using the quasi-experimental identification restrictions set out in Section 2 above (Appendix B, Tables B 1 -B8 provides WESML-LIML-FE estimates for different outcomes). We also present two "naive" estimates which do not treat credit program placement or participation as endogenous (Appendix A, Tables Al-A15). One set of naive estimates treats the choice- based samnpling nature of the survey appropriately and uses WESML methods, while the other does not. The latter is actually more consistent with the maintained hypothesis of the naive model that choice -- credit program participation -- is exogenous, and thus fully consistent estimates are obtained by ignoring varying sampling proportions.22 Since village fixed effects are not accounted for in the naive estimates, a set of 22 Furthermore, neither naive model deals with the possible nonindependence of the errors. This is not atypical of much of the applied literature in this area. If the exogeneity assumption is valid, ignoring nonindependence 28 village characteristics, consisting of 5 measures of village infrastructure, 6 goods prices and two wage rates, are included as regressors (see Table 3. 1.), as is common in this type of cross-sectional analysis. In a separate table (Table 4.1), we present WESML-LIML-FE estimates side-by-side with WESML-LIML estimates. If program placement is random, the WESML-LIML estimates are efficient and WESML-LIML-FE estimates are consistent but inefficient. If program placement is non-random, the WESML-LIML estimates are inconsistent. Hausman-like tests of the consistency of the WESNvLLIML models were attempted, but the covariance matrix of the differences in the parameter vectors were not positive definite in every case tried. This problem is not uncommon in estimation problems of this kind. The test statistic computed is: (P FE- XFE E ) (DFE-P) (18) where PFE and P (X_FE and E ) refer to the WESML-LIML-FE and WESML-LIMIL parameter (covariance) vectors (matrices) respectively. Typically, the problem is that one or more of the diagonal elements of the covariance matrix (_FE - _ ) is very close to zero, and sometimes negative. Essentially, the implication is that the test statistic is infinitely large, and the null hypothesis that the fixed- effects and non-fixed-effects parameter vectors are the same is thus rejected. This implies that credit programs are not placed randomly across the villages of Bangladesh. The results of Table 4.1 will be addressed as we discuss individual outcomes. Presenting fixed- effects and non-fixed-effects estimates side-by-side but separately is intended to allow the interested reader to eyeball the parameters and their t-ratios, to subjectively judge the importance of the difference between these methods. One important drawback of estimating program impacts from data on two cohorts (those from villages with and without programs available) in which cohort assignment is non-random, meaning provides consistent parameter estimates but inconsistent estimates of the parameter covariance matrix (the t- statistics). 29 deliberate program placement, is the possible misinterpretation of the village fixed effects. The discussion so far has treated the village effects as time-invariant attributes. However, it is possible that credit prograrns can alter village attitudes and other village characteristics, perhaps through demonstration effects, and thus can alter the attitudes of non-participants as well as participants. The full behavioral effect of the program must then include the effect of any such village "externalities" and not just the direct effect on credit participants. As an example, consider the limiting case in which program placement is in fact random but program activities, particularly those aimed at altering attitudes, successfully alter the views of non- participants on the value of contraception and limiting family size. In this case, unobserved village contraception propensities would be correlated with program placement, but the causation would not go from village unobserved effects to program placement, but from program placement to village unobserved effects. In this scenario, programs are not placed in villages because of their relative attitudes on contraception, but rather program placement affects the attitudes of non-participants in villages. Unfortunately, the only way these external effects can be measured is to collect data on villages before and after program introduction. A more formal statement of this measurement problem explicitly allows for the placement of a credit program to cause a village effect (Qj) in addition to a pre-existing village effect p1. Equation (8) is then rewritten as: Yq = X'JPy + VJlY + Cii6 + Dj + £ y (19) where all terms are defined as before except that a new term Qj is added to the conditional demand equation. This term represents the external effects of a program in a village and has the value zero if no program is located in the village. Significantly, the existence of non-zero credit program externalities Qi does not affect the consistency of any estimate of 5, only its interpretation.23 The program effect parameter 8 estimated by WESML-LIML-FE captures all program effects only if Qj=O in all villages; that is, none of 23 This result relies on the linearity of the conditional demand equation. 30 the village-specific heterogeneity in behavior is caused by programs. If village externalities exist (piO), the WESML-LIML-FE estimate of o represents only the effect of credit on program participants above and beyond its effects on non-participants in the village. If program placement is random and q.#O, then WESML-LIML is a more efficient estimator than WESML-LIML-FE and the estimated o has the same interpretation as for WESML-LIML-FE. If program placement is non-random, WESML-LIML is inconsistent. It is generally not possible to estimate the village externality Qj from a single cross-section of data. Before describing those results, we first present the results of estimating the credit equations (14) and (15), which are estimated jointly with equation (16) in every case where WESML-LIML-FE is applied. Table 4.2 presents these estimates. Since there are no endogenous right-hand-side regressors in the credit equations, they can be estimated separately from the conditional demand equation (16) using WESML bivariate tobit with village fixed effects, which was the method used for the estimates presented in Table 4.2. Implicit in these estimates is a set of restrictions on the parameters Jcf and P,, of equations (14) and (15) that can clearly be seen by rewriting these equations as Cy = X,,pfcf + X,/DmcLfi + Jlc1 + E c (20) (G'n = Xj 1cm + X,,D5amI + ptcm + Eim (21) where Dm=l if there is a male credit group in the village, Dm=O otherwise, Df=1 if there is a female credit group in the village, Df=O otherwise, and afm and amf are parameters.24 The set of village-specific regressors Vj of equations (14) and (15) are replaced by W.'s in the equations above, representing the village fixed effects. If the a parameters are non-zero, the determinants of women's (men's) credit participation (the 13's) depends on whether men (women) also have a choice of joining the credit program. The restriction that o4m = anf= 0 was tested with a likelihood ratio test and could not be rejected at common levels of 24 Essentially, the idea is that there may be two regimes each with different parameter vectors for each sex: a regime in which only one sex is able to choose to participate in a credit program and a regime in which both sexes can participate. 31 significance (X2(28)=22.6, p=0.25). Note that this does not necessarily imply that the presence or absence of a credit program for the opposite sex does not matter, only that it does not affect the slope parameters (,). The "demand" curve may be shifted up or down but such shifts are not statistically identifiable in this model, since they are fully captured by the village-specific intercepts w. The other restriction is that the slope parameters P are common to the three credit programs. Again, the credit equations may be shifted up or down but such shifts are not statistically identifiable in this model as they are fully captured by the village-specific intercepts Pi. While individual loans are small by formal credit market standards, they were never less than Taka 1000 in the data. The censoring threshold for the credit equations (1) and (2) was taken to be 1000 in the estimation. Redefining Cijf and Cy,m as the logarithm of program credit provided female and male members of household i in village j, and defining C,- and C, as the latent variables associated with these female and male credit variables, respectively, the model estimated is Cj, = X#Pf + wci + s jf (22) CV = CC,; if C,; ) log(1000) C,;m = X1J0L + Ficm + S. c (23) Cij.= C.* if Ci- ) log(1 000) where latent credit demand of less than Taka 1000 results in censoring of the observed credit variable. The logarithmic form implies that latent credit demand is strictly positive. Latent demand less than the censoring threshold of Tk. 1000 does not result in borrowing. The set of variables describing the availability of potential sources of intra-family transfers was not a significant determinant of credit demand for either gender. The household head's age and sex are apparently important determinants of credit demand for both women and men, but of opposite signs 32 between the sexes. Having a male head reduces the credit received by women, as do increases in the age of the head. A test of the hypothesis that the slope parameters in women's and men's credit demand are equal is strongly rejected (X2( 14)=50.94, p=0.00), reflecting to a large extent the opposite and significant sex and age of household head effects.25 Table 4.3 presents estimates of the effects of credit program participation on the school enrollment status of children aged 5-17 at the time of the survey. Separate sets of estimates were made for girls and for boys. The WESML-LIML-FE estimates demonstrate that the schooling of boys is increasing in all 6 credit variables, and the schooling of girls is increasing in 4 of 6 credit variables, although only a few of the individual parameters have large t-ratios. Tests of the joint significance of the six credit variables find little evidence of joint significance for girls (X 2(6)=4. 11, p=0.66) but much stronger evidence for boys (X2 (6)=20.00, p=O.00).26 It is female program credit that drives the positive credit effect on boys' schooling. The test statistics for women's credit are significant at the 0.01 level (X2(3)=15.18, p=0.00). The largest and most precisely estimated individual credit parameter for both boys' and girls' school enrollment is for credit obtained by women from the Grameen Bank (t=2.36 for boys, t=1.30 for girls).27 25 The variables "No adult females in household" and "No adult males in household" were included as regressors because the adult education variables "Highest grade completed by an adult female in household" and "Highest grade completed by an adult male in household" are undefined when there are no adults (defined as a household member 16 years of age or older) of that sex in the household. Whenever there was no adult member of one sex in the household, the relevant "Highest grade completed..." variable was coded zero. The "No adult..." variable thus picks up the difference between having zero as the highest number of years of schooling of adults of a particular sex and not having any adult of that sex in the household. 26 All of the x2 test statistics from the WESML-LIML-FE estimates are reproduced in tabular form in Appendix C. 27 There are no a priori grounds to expect that the signs of the credit parameters will be positive in this or any other of the conditional demand equations estimated. One might expect school enrollment rates to be increasing in household full and cash income resulting from credit program particaption and borrowing if this kind of human capital formation is a normal good. In addition, to the extent that credit to women increases their bargaining power in the household, and thus their utility weight X, and their preferences for human capital investment in children, girl children in particular, is "greater" than for household males, credit programs will increase school enrollment rates through changes in the households social welfare function (3). Changes in the social welfare function can also come about from the information credit programs provide women (and men) about the returns on schooling or by altering perceived social pressures that act to reduce schooling. On the other hand, if girls' time is a close substitute for the time of their mothers, an increase in the value of mothers' time in self-employement (production of Z-goods) attributable to credit programs may induce a substitution of daughters' time from schooling and into either household goods production or into the self-employment activity, or both. The sign of the sum of the income, substitution and x effects is indeterminant. Similar types of logic, standard in the household production framework, apply to the other conditional demand equations estimated. 33 The WESML-LIML girls' schooling estimates (Table 4.11B) are algebraically larger than the WESML-LIML-FE estimates. Furthermore, for each credit program, the parameters on female participation are larger than those on male participation. In this model, participation in the Grameen Bank has the largest effect of all. The striking difference between the WESML-LIML and WESML-LIML-FE estimates of the effects of women's program participation on girls' schooling is mirrored in the estimates of the women's correlation coefficients p. The WESML-LIML estimates suggest that women in households that are less likely to educate their daughters than observationally equivalent women are also more likely to choose to participate in a credit program. The WESML-LIML-FE estimates suggest that self-selection into the program is of the opposite sort -- women in households that are more likely to educate their daughters, conditional on the observed regressors and all observed and unobserved village characteristics, are more likely to participate in a credit program. A joint test of the exogeneity of credit program participation cannot reject the null hypothesis that individual credit program participation is exogenous in the determination of girls' and boys' schooling conditional on the village fixed effects. Table 4.1 D presents WESML-FE estimates of those conditional demand equations for which the hypothesis of exogeneity could not be rejected, as well as the relevant test statistics. Imposing the statistically valid restriction of exogeneity provides more efficient estimates of program effects. The estimates in the first column of Table 4.ID demonstrate a strong and statistically significant effect of female Grameen Bank credit on girls' schooling (t=2.92). No other credit parameters are statistically significant. The small effect of women's credit on their daughters' schooling for the other credit programs may reflect the close substitution of women's and girls' time in both the production of household goods and in the self-employment activity. If mothers are drawn into self-employment, daughters' time may be used to replace the time mothers formerly spent on household products, such as child care and food preparation. Table 4.1 D (column 2) provides WESML-FE estimates of the determinants of boys' schooling that demonstrate a pattern of statistical significance conforming to that found in the WESML-LIML-FE 28 The test is that the two correlation coefficients p are jointly zero. 34 estimates. The estimated t-ratios are higher for female BRDB and Grameen credit and male Grameen credit, but the size of the women's Grameen effect falls. The two sets of naive parameter estimates presented for the boys' school enrollment equation are quite different from the WESML-LIML-FE. The magnitude and significance of women's BRDB and Grameen credit on boys' school enrollment is strikingly miscalculated by the naive models. For the determinants of girls' schooling, the weighted naive model only finds a significant positive effect for male credit from the BRAC. The point estimate is the same as the WESML-LIML-FE, but the t-ratio for the naive estimate is much larger. Table 4.4 presents estimates of the program credit impact on the market labor supply, including self-employment (log hours in the past week), by gender using all three rounds of the survey. The WESML-LIML-FE estimates for women find no significant credit effects (X 2(6)=1.39, p=0.97). As both labor supply and credit are entered in logged forn, the credit parameters are the elasticities of (latent) hours of market labor supply with respect to credit. The naive estimates (Table 4.4) substantially overestimate the effect of credit provided women on their labor supply. Table 4.1 provides non-fixed-effects WESML- LIML estimates of the determinants of women's labor supply that, except for female credit from the Grameen Bank, are not very different from the fixed-effects estimates. A test of the null hypothesis that credit program participation is exogenous in the determination of women's labor supply could not be rejected; hence the WESML-FE estimates of Table 4.1D are preferred. As in the case of girls' schooling, these estimates find a statistically significant positive effect of women's participation in the Grameen Bank on women's labor supply. In addition, the women's BRAC and BRDB parameters change sign and are marginally statistically significant, with asymptotic t-ratios above 1.8. Both own- and cross-effects are important in the male labor supply (Table 4.4). Both male credit (x2(3)=98.66, p=O.OO) and female credit (X2(3)=53.1 1, p=O.OO) reduce the labor time of adult male household members. Since it seems unlikely that they are substituting home time for market time, the only conclusion to be drawn is that these negative cross-effects reflect income effects. If the market value of men's time is unchanged by women's borrowing, their labor supply should fall if male leisure is a normal 35 good. This is consistent with a variety of scenarios. One is that men already have ready access to non- program credit markets, so that program credit provides men mostly with rents proportional to the difference between the program and next-best-alternative rates of interest. Table 4.5 presents estimates of the impact of credit program participation on the natural logarithm of food, non-food and total expenditure per capita using all three rounds of survey data. All three female credit parameters are positive and statistically significant determinants of total expenditure, with no t- statistic less than 3.8, and are jointly significant (X2(3)=19.03, p=0.00). By contrast, none of the male credit parameters has a t-statistic over 2.0 and the hypothesis that all the male credit parameters are zero cannot be rejected at the 0.05 level of significance (X2(3)=4.1 1, p=0.25). The estimated female credit effects are approximately double the male credit parameters for the same program. The largest elasticity is, with respect to Grameen Bank credit, provided to women (0.043). The WESML-LIML parameter estimates of the determinants of (log) total expenditure in Table 4.IC again show the importance of the village fixed effects in the estimation. Women's credit effects are underestimated by WESML-LIML, and all three male credit parameters are negative and two (BRAC and Grameen) are statistically significant. The naive estimates presented in Table 4.5 enormously underestimate the positive effects of program credit on total household expenditure. Credit provided to women and men increases expenditure on both food and non-food items. These parameters are less precisely estimated than the total expenditure parameters, and because of the logarithmic specification chosen, the adding up property of expenditure equations does not hold.29 Table 4.6 presents estimates of the effects of credit programs on current contraceptive use and the recent (last 36 months) fertility of currently married women aged 15-49 years. The WESML-LIMIL-FE estimates provide mixed statistical evidence of the influence of program credit on both behaviors. Female credit from all three programs apparently reduces the use of contraceptive devices among program participants (X2(3)=6.15, p=0.10), with t-statistics greater than 2.0 (in absolute value) for the BRDB and Grameen. By contrast, male credit from the BRAC and BRDB tends to increase the use of contraceptives 29 WESML-FE estimates of the determinants of food and non-food expenditures are not provided since they are simply disaggregations of the total expenditure, for which exogeneity was firmly rejected. 36 (X2(3)=8.58, p=0.04). The naive weighted contraceptive use equation (Table 4.6) also does not find strong positive effects of credit program participation by women. The WESML-LIML-FE correlation coefficient (p) is positive and fairly large (p=.425) implying that the women who join these credit programs are more likely to already use contraceptives than observationally equivalent women, controlling for village effects. The WESML-LIML estimates (Table 4.1B) paint an opposing picture of the effects of women's programn credit on contraceptive use. Without controlling for village effects, all the female credit parameters are positive and the BRAC and Grameen parameters have t-statistics greater than 2.0. Moreover, all the male credit effects change sign from positive to negative. Contraceptive use is one behavior for which village externalities (as defined above) might be important. Consequently, the total effect of the credit program on a participant Cifi + Q-j may in fact be positive, but we are still left with the implication that the effect of the credit program on women participants is less than its effect on non- participants in the same village, since the estimated o's are negative. The null hypothesis that credit program participation is exogenous in the determination of contraceptive use is only marginally rejected (X2(2)=4.90, p=0.09), and the null hypothesis for women's credit program participation is more firmly rejected (t--2.075). Nonetheless, WESML-FE estimates for contraceptive use are presented in Table 4.1 D because of the marginal significance of the joint test. The WESML-FE estimates find a higher t-ratio for male BRDB credit, and still find negative women's BRDB and Grameen Bank credit effects, although they are no longer statistically significant. There remains a lack of evidence that women's credit program participation increases the use of contraceptives.30 The WESML-LIML-FE fertility estimates (Table 4.6) are mostly consistent with the contraceptive use estimates for women's credit. Fertility is increasing with women's participation in the BRAC and BRDB, although only statistically significantly for the BRAC. The set of three women's credit parameters are jointly different from zero (X2(3)=8.36, p=0.04), as are the men's credit parameters (X2(3)=8.17, 30 Furthermore, the WESML-LIML correlation coefficient is negative and large in absolute value (p=0.325) whereas the WESML-LIML-FE estimate is large and positive (p=0.425). A negative correlation coefficient implies that women who are less likely to use contraception than observationally equivalent women are more likely to join a credit program, which strikes us as less intuitive than the opposite sort of self-selection. 37 p=0.04). However, the male BRDB and Grameen credit effects are negative and have t-statistics near or above 2.0 in absolute value. That is, male participation seemingly reduces fertility while female participation increases it. The null hypothesis that women's and men's credit effects on fertility are the same is rejected (X2(3)=17.85, p=0.00). 3132 Table 4.7 presents the results of estimating WESML-LIML-FE and naive models of the determinants of women's non-land asset value.33 The WESML-LIML-FE estimates are all positive, implying that credit program participation by both sexes increases the value of women's non-land asset holdings, with the female participation parameters for each program larger than the male participation parameters in each case. However, these parameters are not statistically different from zero (X2(6)=4.36, p=0.54), nor are the women's and men's parameters statistically different from each other (X2(3)=2.95, p=0.40). The naive estimates find large positive effects for women's BRAC and Grameen participation on the value of their non-land assets. Women's non-land assets is apparently the behavior for which the difference between the unweighted and weighted naive estimates is the greatest amongst those studied. That is, the choice-based nature of the sample matters most. The WESML-LIML estimates show a statistically significant effect of female participation only for the BRAC, but male participation shows such an effect in all three programs, most strikingly in the BRDB.34 The null hypothesis of the exogeneity of program credit in the determination of women's non-land assets cannot be rejected (X2(2)=1.76, p=0.41). The WESML-FE estimates of Table 4.11D find strong and 31 The naive weighted fertility estimates only find women's participation in the Grameen Bank to have a statistically significant effect in reducing fertility. Furthermore, unlike contraception, the WESML-LIML and WESML-LIML- FE estimates of p are not of opposite sign. Both sets of estimates suggest that individuals with lower recent fertility, conditional on their observed attributes, are more likely to participate in a credit program. 32 In work in progress, Pitt and others (1995) investigate the contraceptive and fertility effects of these credit programs in more detail by estimating the model with age-defined subsamples of the data and by altering the econometric specification in other ways. 33 The asset variables are sex-specific rather than individual-specific in that they are defined as the total value of assets held by all individuals of each sex in the household. Thus, no household contributes more than one observation to each of the sex-specific asset equations estimated. The quality of asset data is typically suspect in household surveys, even more so when there is an attempt to break down assets by sex of ownership. The relative variance of the asset data is very high (see Table 3.2), with many household reporting zero for women's assets. The male asset data was even more troublesome. We were unable to get the any of the likelihoods for the determinants of male assets to converge. 38 statistically significant positive effects of credit program participation on women's asset holdings. The BRDB and Grameen Bank parameters are nearly twice as large as the BRAC parameter, and all are larger than male participation parameters. The last group of reported estimates examines the determinants of the anthropometric status of children aged 0-14 years -- height, weight and body mass index. There are 6 sets of estimates -- the three anthropometric measures for each sex. The high cost of collecting anthropometric data forced us to draw samples of children from only 15 of the 87 villages. All of the 15 villages had credit programs present -- 6 each with the BRAC and Grameen Bank, and 3 with the BRDB -- and all the sampled households were in the target group. As a result, a substantial part of the statistical identification obtained from the quasi- experimental framework was lost. Gender-specific credit effects are still identified from the fact that not all villages in this subsample had credit programs for both sexes. Because the anthropometric dependent variables are strictly continuous, nonlinearity arising from the specification of the errors as having a joint normal distribution is used to identify the model. Body Mass Index (BMI), defined as the ratio of weight to height squared (weight/ (height2), is most often the preferred indicator of anthropometric status. The WESML-LIML-FE estimates in Table 4.8 reveal that all six credit variables negatively affect boys' BMI (X2(6)=4.17, p=0.65) and positively affect girls' BMI (X2(6)=9.82, p=O. 13). Neither the set of three women's credit variables or men's credit variables are significant in the boys' BMI equation (X2(3)=3.32, p=0.34 for women's credit; X2(3)=1.76, p=0.62 for men's credit) or in the girls' BMiI equation (X2(3)=4.14, p=0.25 for women's credit; X2(3)=5.98, p=O. 1 for men's credit). The largest positive effects on girls' BMI came from Grameen Bank credit, the smallest from BRDB. The estimated elasticities are quite small; the largest is .009 for Grameen Bank credit provided to men. The weighted naive girls' BMI equation finds that women's credit has a negative effect on BMI, the opposite sign of the WESML-LIML-FE equation. One reason for the "wrong" sign of the naive model is the negative p for women's credit, implying a negative correlation between the errors of the women's credit equation and girls' BMI. The WESML-LIML estimates without village fixed effects (Table 4.1A) similarly estimate the wrong sign for women's credit in the girls' BMI equation. With village unobservables treated 39 as random effects, the p for women's credit is positive but small, suggesting that credit programs are more likely to be placed where girls' anthropometric status is somewhat higher but that the target households that participate in these credit programs are those with girls' BMI of below village average. Exogeneity could not be rejected for both the girls' and boys' BMi equations, and thus WESML-FE estimates are presented in Table 4.1 D. Like the WESML-LIMI-FE estimates, these estimates continue to find that the only statistically significant effect is of male Grameen Bank credit on girls' BMI. Height and weight estimates are presented for completeness. The pattern of the p's is interesting here. They suggest that women who borrow tend to have children of higher than average weight and height among target households. No similar selection mechanism appears among men. This is consistent with rhe preference heterogeneity explanation suggested above, in which more egalitarian households are both more likely to treat their girl children favorably and to permit their adult females access to program credit and self-employment activities. 5. Summary and conclusions Group-based lending programs for the poor have become a focus of attention in the development community over the last several years. To date, there has been no comprehensive investigation of their impact on household behavior that has been sufficiently attentive to issues of endogeneity and self- selection. Perhaps one reason for this is the absence of any data generated from social experiments associated with these credit programs, and from the difficulty in finding valid instrumental variables (exclusion restrictions) to deal with endogeneity in non-experimental data. This paper surmounts these issues by treating the choice of participating in credit programs in a sample of Bangladeshi households and villages as corresponding to a "quasi-experiment" conditional on all observed (in the data) and unobserved village characteristics. It uses this same approach to help identify the separate effects, if any, of lending to female and male household members, making use of the fact that credit groups are single-sex and groups for both sexes are not available in all villages. The econometric methods used are much more complex than those ordinarily applied in this area. In order to demonstrate the 40 value of resorting to these methods, the paper presents alternative estimates of program impacts using simpler approaches such as ordinary least squares. This simplicity is obtained by ignoring to some extent issues of endogeneity. A comparison of these methods clearly indicates the importance of our attentiveness to endogenity in evaluating these credit programs and the mistaken conclusions that could be drawn from the simple "naive" estimates. The paper provides estimates for a wide variety of household and individual outcomes and separate estimates of the influence of borrowing by both men and women and for each of three credit programs. The results are summarized as follows: A. Joint tests reveal that credit is often a significant determinant of household behavior. Either the set of female credit variables, male credit variables or both are statistically significant at the 0.05 level of significance in all 8 key behaviors studied (excluding anthropometry and disaggregations of total household expenditure).35 B. Joint tests reveal that credit provided to women somewhat more often has a statistically significant effect on these 8 outcomes than credit provided to men. The set of female credit variables is statistically significant in 7 of 8 cases at the 0.05 level. By contrast, the set of male credit variabl.es is significant in 3 of 8 cases. However, the hypothesis that female and male credit parameters are jointly equal for each of the three programs is rejected in only four cases: women's labor supply, women's non-land assets, contraception and fertility. C. Credit provided by the Grameen Bank had the greatest positive impact on variables typically associated with household wealth and women's power and independence than credit from any other 35 Identification of the determinants of anthropometric outcomes is somewhat weaker in that anthopometry is only available from villages in which there is a credit program and only for target households, as discussed above, and as a consequence we treat them separately below. Food and non-food expenditures are not counted separately here since they are encompassed by total household expenditure. The 8 outcomes are: girls' and boys' schooling, women's and men's labor supply, total household expenditure, contraception, fertility, and the value of women's non-land assets. The test statistics referenced here and below are for WESML-LIML-FE estimates unless the joint test of exogeneity could not be rejected, in which case the test statistics are for the WESML-FE estimates of Table 4.1 D. 41 program source.36 Grameen Bank credit to women had the largest impacts on girls' schooling, women's labor supply and total household expenditure, and Grameen Bank credit to men had the largest impact on fertility (tied with male BRDB credit). Women's credit from the BRDB had the largest impact on boys' schooling and the value of women's assets. D. Little evidence is provided of any impact of credit on the anthropometric status of children. However, this might reflect the somewhat weaker statistical identification available in the data when estimating the determinants of anthropometric outcomes. E. Treating the placement of credit programs across villages as non-random, and the decision to join and borrow from one of these programs as endogenous, has an important influence on the estimated program impacts. For example, the WESML-LIML-FE credit parameters in the conditional demand equation for contraceptive use are of the opposite sign of their WESML-LIML (without village fixed effects) counterparts. In addition, the naive estimates, which treat program participation and program placement as exogenous, miscalculate the effects of credit program participation on behavior. For example, they grossly underestimate the effects of the credit programs on increasing total household expenditure. Our results provide evidence that program participation benefits the poor, especially women and children. Furthermore, the magnitude of the benefits accruing to individuals in a participating household depends on whether the participant is a woman or a man. Three important policy conclusions can be drawn from this exercise. First, targeted credit programs such as the Grameen Bank can "empower" women by increasing their contribution to household consumption expenditure, their hours devoted to production for the market, and the value of their assets. Second, targeted credit programs can be seen as anti-poverty schemes. Poverty in rural Bangladesh largely means low levels of consumption, and our results clearly indicate that credit from all three programs increases the total per capita consumption of the poor and the asset holdings of women. Third, group-based 36 These outcomes are girls' and boys' schooling, and women's labor supply, assets, and total expenditure. 42 credit provided to men can also have beneficial effects, particularly on the schooling of children, contraceptive use, fertility and total household expenditure. Further research is, however, needed to broaden our understanding of the influence of these credit programs in altering the lives of participants and their families. We are currently undertaking research using data from the survey described above to study the importance of the non-credit services provided by these group-based programs, the determinants of the choice of self-employment activity, the effect of program borrowing on intrafamily transfers and borrowing from other non-program sources, and the effects of program credit and the self-employment it engenders on seasonal patterns of consumption. 43 Table 2.1 Distribution of villages by credit program and group type Credit programr Group type BRAC BRDB GB None Total Female only 7 3 12 0 22 Male only 0 9 1 0 10 Female and 17 12 11 0 40 male No program 0 0 0 15 15 Total 24 24 24 15 87 Source: BIDS-World Bank household survey data, 1991-92. 45 Table 3.1 Weighted mean and standard deviations of independent variables Independent variables |_No. of observations | Mean J_Standard deviation Age of the individual 9,215 23.00 18.00 Education of individual (years) 7,886 1 .377 2.773 Parents of HH head own land 1,725 0.256 0.564 Brothers of HH head own land 1,725 0.815 1.308 Sisters of HH head own land 1,725 0.755 1.208 Parents of HH head's spouse own land 1,735 0.529 0.784 Brothers of HH head's spouse own land 1,735 0.919 1.427 Sisters of HH head's spouse own land 1,735 0.753 1.202 Household land (in decimals) 1,757 76.142 108.543 Highest grade completed by HH head 1,757 2.486 3.501 Sex of household head (1 =male) 1,757 0.948 0.223 Age of household head (years) 1,757 40.821 12.795 Highest grade completed by an adult 1,757 1.606 2.853 female in HH (in years of education) l Highest grade completed by an adult 1,757 3.082 3.081 male in HH (in years of education) __l No adult male in HH 1,757 0.035 0.185 No adult female in HH 1,757 0.017 0.129 No spouse present in HH 1,757 0.126 0.332 Amount borrowed by female from 1,757 350.345 1573.659 BRAC (Tk.) l Amount borrowed by male from BRAC 1,757 171.993 1565.006 (Tk.) Amount borrowed by female from 1,757 114.348 747.301 BRDB (Tk.) l Amount borrowed by male from BRDB 1,757 203.250 1572.667 (Tk.) . Amount borrowed by female from GB 1,757 956.159 4293.366 (Tk.) 46 Table 3.1 (continued) Weighted mean and standard deviations of independent variables Independent variables No. of observations Mean Standard deviation Amount borrowed by male from GB 1,757 374.383 2922.794 (Tk.) Non-target household 1,757 0.295 0.456 Has any primary school? 1,757 0.686 0.464 Has rural health center? 1,757 0.300 0.458 Has family planning center? 1,757 0.097 0.296 Is Dai/midwife available? 1,757 0.673 0.469 | Price of rice 1,757 11.15 0.85 Price of wheat flour 1,757 9.59 1.00 | Price of mustard oil 1,757 52.65 5.96 Price of hen egg 1,757 2.46 1.81 Price of milk 1,757 12.54 3.04 __Price of potato 1,757 3.74 1.59 Average female wage 1,757 16.154 9.613 No female wage dummy 1,757 0.193 0.395 Average male wage 1,757 37.893 9.400 Distance to bank (km) 1,757 3.49 2.85 Note: Amount borrowed is the cumulative amount of credit (2 Tk.1,000) borrowed since December 1986 from any of these three credit programs. These amounts are then adjusted with proper CPI indices. Source: BIDS-World Bank household survey data, 1991-92. 47 Table 3.2 Weighted Mean and Standard Deviations of Dependent Variables Partici- Obs. Non- Obs. Total Obs. Non- Obs. Aggrcgate Obs. Dependent Variables pants participants programn I______ ______________ __________ areas Sum of program loans of females 5498.854 779 326 2604.454 1105- 2604.454 1105 (Taka) (7229.351) . (5682.398 (5682.398) Sum of program loans by males 3691.993 631 263 1729.631 894 - - 1729.631 895 (Taka) (7081.581) (5184.668) (5184.668) __ Contraceptive use by currently .418 902 .375 546 .389 1448 .322 283 .378 1731 married women aged 14-50 years (.493) (4.84) (.488) (.468) | (.485) Fertility: Number of Children .679 902 .703 546 .695 1448 .712 281 .697 1729 Bom last 3 years to currently (.736) (.717) (.723) (.702) (.719) married women aged 14-50 years (Any child Yes= 1; No=O) Current school enrollment by girls .535 802 .528 434 .531 1236 .552 225 .534 1461 oo aged 5-17 years (Yes=1; No=0) (.499) (.500) (.499) (.498) (.499) Current school enrollment by boys .566 856 .555 468 .558 1324 .560 265 .559 1589 aged 5-17 years (Yes=1; No=0) (.496) (.498) (.497) (.497) (.497) Weight of girls aged 0- 14 years 13.00 263)- 12.00 146 12.00 409 - 12.00 409 (4.00) (4.00) (4.00) (4.00) Weight of boys aged 0-14 years 13.00 287 12.00 91 13.00 378 - - 13.0 378 (kg) (4.00) (4.00) (4.00) (4.0()) Height of girls aged 0- 14 years 96.00 263 94.00 146 94.00 409 - - 94.00 409 (cm) (17.00) (18.00) (18.00) (18.00) Height of boys aged 0-14 year 97.00 287 93.00 91 95 378 - - 95.00 378 (cm) (17.00) (16.00) (17.00) (17.00) Body Mass Index of girls aged 0- .001 263 .001 146 .001 409 - - .001 409 14 years - (.000) (.000) (.000) (.000) Body Mass Index of boys aged 0- .001 287 .001 91 .001 378 .001 378 14 vems I (.000) (.000) t.000) (.000) Table 3.2 (continued) Weighted Mean and Standard Deviations of Dependent Variables Partici- Obs. Non- Obs. Total Obs. Non- Obs. Aggregate Ohs. lYependent Viriables pants participants program areas Sum of program loans of females 5498.854 779 326 2604.454 1105- - 2604.454 1105 (Takar- . - (7229.351) (5682.398 . _ (5682.398) Employ-rnent hours per month by 40.328 3420 37.680 2108 38.905 5528 43.934 1074 39 54() 6602 women aged 16-59 years (70.478) (71.325) (70.934) (74.681) (71.432) Employment hours per month by 202.758 3534 185.858 2254 191.310 5788 180.94 1126 189.477 6914 men aged 16-59_years (100.527) (104.723) (103,678) (98.805) (102.902) Per capita HH food expenditure 59.166 2696 62.265 1650 61.242 4326 61.985 872 61.366 5218 (Taka) (19.865) _ (23.256) (22.239) (23.897) - (22.522) Per capita HH non-food 17.848 2696 23.621 1650 21.716 4346 27.676 872 22.706 5218 4s expenditure (Taka) (31.538) _ (54.791) (48.439) (51.409) _ (48.990) Per capita HH total expenditure 77.014 2696 85.886 1650 82 959 4346 89.661 872 -84.072 5218 (Taka) (41.496) (64.820) __ (58.309) (66.823) (59.851) Female Non-land assets (Taka) 7399.231 899 4716.416 542 5608.033 1441 1801.839 292 4970.67 1733 _ (2930.02) _ _ (19901.035) (23509.09) (6287.491) (21649.42) Male Non-land assets (Taka) 54767.57 873 83116.58 542 73893.11 1415 71858.15 276 73559.46 1691 (73152.98) (94047.46) (88753.85) (76653.98) (86867.58) Note: Standard deviations are in the parentheses. Contraceptive use and fertility variables are based on only round I data. Nutrition variables (weight, height and BMI) are based on round I and round 2 data of the nutrition survey which match up with round I and round 3 of the gencral household survey. All other variables are based on all 3 rounds of the general household survey. Source: BIDS-World Bank household survey data, 1991-92. Table 4. lA Fixed- and Nonfixed-Effects Estimates of the Impact of Credit on Women's Log Labor Supply, Boys Log Body Mass Index, Women's Log Non-land Assets and Men's Log Labor Supply Women's Log Labor Supply Boys Log BM1 Women's Log Non-land Assets Men's Log Labor Supply Explanatory WESML- WESML- WESML-LIML- WESML- WESML- WESML- WESML- WESML- Variables LIML-FEL' LMILY FE'" L1MLb' LIML-FE" LIMLb/ LIMI-FEL' LIMLI Amount borrowed by -.0117 .0096 -.0130 .0020 .0318 .0425 -.1813 -.2008 female from BRAC (-.128) (.144) (-1.248) (2.388) (.356) (2.302) (-5.884) (-8.350) Amount borrowed by -.0448 -.0908 -.0050 -.0139 .1005 .2589 -.1369 .0246 male from BRAC (-.520) (-1.090) (-0.536) (-2.343) (.468) (2.367) (-2.155) (1.036) Amount borrowed -.0139 .2087 -.0110 -.0046 .1257 .0473 -.2308 -.2051 by female from (-.139) (3.185) (-0.827) (-0.529) (1.043) (.300) (-7.066) (-7.635) BRDB Amount borrowed by -.0144 .0281 -.0110 -.0153 .0334 3.8329 -.1440 .0172 male from BRDB (-.181) (.398) (-1.017) (-2.468) (.141) (3.340) (-2.129) (.777) Amount borrowed by .0152 .1449 -.0090 -.0077 .1131 1.3484 -.2189 -.2175 female from GB : t.162) (2.042) - (-0.797) (-0.928)- (1.317) (1.452) (-6.734) (-8.232) Amount borrowcd by -.0570 .0357 -.0060 -.0140 -.0457 .3377 -.1592 .0126 malc lrom GB. (-.677) (.440) (-0.623) (-2.095) (-.200) (2.386) (-2.524) ( .522) Rho(women) .1255 -.0173 .482 .6206 .1136 -.0168 .6564 .7151 (1.062) (-.196) (1.156) -(3 458) (1.325) (-.198) (7.461) ( 11.698) Rho (men) .0560 .0415 .399 .3423 -.0148 -.7656 .4929 -.0481 (.592) (.435) (1.146) (1.070) (-.053) (-36.311) (2.512) ( -.794) Log likelihood -15069.781 -15774.111 -2998.448 -3176.737 -4226.176 -4951.408 -18395.082 -18954.702 No. of observations 6602 378 1757 - 6914 "Wcighted Exogenous Sampling Maximum Likelihood:Limited Information Maximum Likelihood:Fixcd Effects. ' Weighted Exogenous Sampling Maximum Likelihood:Limited Information Maximum Likelihood. Note: Figures in parentheses represent asymptotic t-ratios. Source: BIDS-World Bank household survey data, 1991-92. Table 4.1 B Fixed- and Nonlixed-Effects Estimates of the Impact of Credit on Girls Schooling, Girls Log BMI, Contraceptive Use and Recent Fertility Girls Schooling Girls Lo BM1 Contracetivc Usc Rccent Fertility Explanatory Variables WESML- WESML- WESML- WESML- WESML- WESML- WESNMI.- WESML- LIML-FE' LIML' LJIML-FEs LIMLw IMIML-FE" LlvMLb LIML-FE" IjMLb | Amount borrowed by female from -.0203 .0693 .004 -0.002 -0735 .0745 .0790 0374 BRAC (-.552) ( 1.990) (1.365) (-0.175) (-I 693) (2.095) (2.372) (.933) Amount borrowed by male from .0495 .0612 .006 .008 0395 -.0212 .0543 016t) BRAC (1.152) ( 1.891) (1.070) (1.571) (.745) (-.406) (1.353) (.399) - Amount borrowed by female from -.0099 .0591 .002 -0.003 -.1163 .0443 .0502 0218 BRDB (-.220) ( 1.616) (244) (-0.244) (-2.421) (1.214) (1.312) (.495) Amount borrowed by male from .0321 .0341 .000 .007 .0839 - 0067- -.0744 -.0547 BRDB (.665) ( 1.036) (.145) (1.962) (1.475) (-.132) (-1.976) (-1. 191) Amount borrowed by female from GB .0128 .0853 .005 -0.001 - 0905 .0946 -.0348 -.0160 (.334) (2.289) (1.822) (-0.098) (-2.011) (2.580) (- 951) (-.362) Amount borrowed by male from GB .0582 .0697 .009 .010 .4253 -.0879 -.0743 -.0420 (1.298) ((2.103) (2.293) (2.524) (2.975) (-1.625) (-2.193) (-.851) Rho(women) .1648 -2728 -0.165 136 4253 -.3253 -.4319 -.2635 (1.029) (-1.370) (-1.441) (.242) (2.075) (-1.777) (-2.718) (-1.201) Rho (men) -.1360 -.1409 -0.051 -0.161 -.2032 1991 .3511 .2445 (-.720) (-.922) (-0.534) (-1.184) (-.700) (.643) (2.701) (1.097) Log likelihood -3702.947 -3949.170 -2921.321 -3104.000 -2458.954 -2709.3012 -2444.341 -2657.02067 No. of observations 2885 409 1884 1882 "Weighted Exogenous Sampling Maximum Likelihood:Limited Information Maximum Likelihood:Fixed Effects. ' Weighted Exogenous Sampling Maximum Likelihood:Limited Information Maximum Likelihood. Note: Figures in parentheses represent asymptotic 1-ratios. Source: BIDS-World Bank household survey data, 1991-92. Table 4. IC Fixed- and Nonfixed-Effects Estimates of the Impact of Credit on Boy's Schooling, Log Expenditure per Capita, Log Expenditure Per Capita on Non-Food Goods, and Log Expenditure Per Capita on Foods Boys Schooling Log Total Expenditurc per Capita Log Total Non-Food Expenditure Log Total Food Expcnditurc Pcr Explanatory Per Capita Capita Variables WESML- WESML- WESML-LIML- WESML- WESML- WESML- WESML- WESML- .______________ LIML-FE" LIMLb' FE" LIMLb LIL-FEa LIMIY LIML-FE' LIMI. , Amount borrowed by .0394 .0991 .0394 .0340 .0220 -.0183 .0057 .0094 female from BRAC (.917) (3.196) (4.237) (2.291) (.544) (-.822) (.658) (1.325) Amount borrowcd -.0040 .0113 .0192 -.0161 .0364 -.0150 .0060 -.0075 by male from BRAC (-.107) (.333) (1.593) (-1.658) (1.388) (-.680) (.801) (-.890) Amount borrowed by .1210 .0956 .0402 .0258 .0139 -.0269 .0101 .0044 female from BRDB (2.573) (3.066) (3.813) (1.723) (.320) (-1.197) (1.051) (.601) Amount borrowed .0361 .0370 .0233 -.0155 .0349 -.0246 .0138 -.0055 by male from BRDB (.934) (1.181) (1.936) (-1.788) (1.330) (-1.246) (1.845) (-.707) Amount borrowed by .1025 .1307 .0432 .0371 .0199 -.0184 .0114 .0114 female from GB (2.364) (4.022) (4.249) (2.174) (.467) (-.759) (1.263) (1.435) Amount borrowed by .0736 .0561 .0179 -.0225 .0182 -.0220 .0087 -.0142 male from GB (1.688) (1.607) (1.431) (-2.291) (.665) (-.982) (1.163) (-1.602) Rho (women) -.2192 -.4665 -.4809 -.3897 -.0564 .1824 -.1026 -.1023 (-1.054) (-2.490) (4.657) (-2.056) (-.222) ( 1.357) (-.697) (-.820) Rho (men) -.0284 -.0222 -.2060 .2999 -.1300 .2152 -.1077 .2050 (-.177) (-.144) (-1.432) (2.998) (-.858) ( 1.923) (-.980) ( 1.648) Lo likelihood -3802.873 4141.386 -6633.559 -7281.469 -10620.080 -11259.596 -5311.365 -6024.498 No. of observations 2940 5218 5218 5218 'Weighted Exogenous Sampling Maximum Likelihood:Limited Information Maximum Likclihood:Fixed Effects. v Weighted Exogenous Sampling Maximum Likelihood:Limited Information Maximum Likelihood. d These variables are applied to outcomes specific to individuals. Note: Figures in parentheses represent standard dcviations. Source: BIDS-World Bank household survey data, 1991-92. Table 4.1D WESML-FE Estimates of the Impact of Credit on Boy's and Girl's Schooling, Boy's and Girl's BMI, Women's Labor Supply and Assets, and Contraceptive Usea [ Explanatory variables Girl's | Boy's | Girl's | Boy's | Women's Women's Log | Contraceptive l_________________________ |Schooling | Schooling BMI | BMI | Labor Supply Non-land Assets use Amount borrowed by female 0.0119 -0.0028 0.0012 -0.0043 0.0721 0.1151 0.0081 from BRAC (0.682) (-0.173) (0.516) (-1.783) (1.884) (2.003) (0.433) Amount borrowed by male 0.0242 -0.0076 0.0055 0.0037 -0.0126 0.0878 0.0075 from BRAC (0.897) (-0.341) (1.150) (0.948) (-0.231) (1.007) (0.289) Amount borrowed by female 0.0233 0.0793 -0.0014 0.0019 0.0766 0.2172 -0.0287 from BRDB (0.804) (3.106) (-0.234) (0.301) (1.803) (2.408) (-1.134) Amount borrowed by male 0.0069 0.0293 -0.0001 -0.0012 0.0268 0.0244 0.0524 from BRDB (0.309) (1.475) (-0.041) (-0.420) (0.682) (0.426) (2.663) Amount borrowed by female 0.0469 0.0611 0.0020 0.0009 0.1037 0.1989 -0.0032 from GB (2.919) (3.644) (0.937) (0.317) (3.016) (3.950) (-0.199) Amount borrowed by male 0.0304 0.0720 0.0081 0.0025 -0.0229 -0.0603 -0.0411 from GB (1.376) (2.743) (2.322) (1.127) (-0.506) (-0.878) (-1.631) No. of observations 2,885 2,940 409 378 6,602 1,757 1,882 Joint test both p's=0 in 1.64 1.20 2.33 1.96 1.53 1.76 4.90 1 WESML-LIML-FE X2(2) (p=0.44) j (p=0.55) | (p=0.31)| (p=0.37) J (p=0.47) (p=0.41) (p =0.09) Note: aFigures in parentheses represent asymptotic t-ratios except for x2 statistics. Source: BIDS-World Bank household survey data, 1991-92. Table 4.2 WESML Bivariate Tobit Fixed Effects Estimates of the Demand for Credit by Gender Dependent Variable: Log of cumulative credit (Taka) since 1986 Women Men Explanatory Variables Coef. t-stat Coef. t-stat Parents of HH head own land -0.010 -0.098 .042 .250 Brothers of HH head own land .036 .458 .170 1.622 Sisters of HH head own land .051 .621 -0.034 -0.339 Parents of HH head's spouse own land .005 .049 -0.185 -1.126 Brothers of HH head's spouse own land .002 .034 -0.027 -.295 Sisters of HH head's spouse own land .100 1.196 -0.004 -0.045 Log household land .026 .540 .207 3.154 Highest grade completed by HH head -0.021 -0.352 -0.029 -0.334 Sex of household head -2.068 -3.532 1.399 1.551 Age of household head (years) 0.015 2.089 -0.024 -2.373 Highest grade completed by an adult female -0.074 -1.754 -0.026 -0.458 inHH Highest grade completed by an adult male in .029 .534 0.142 1.802 HH__ _ ___ _ _ No adult male in HH -1.257 -1.923 No adult female in HH -0.850 -0.961 No spouse present in HH -0.831 -2.483 -1.351 -2.951 Sigma women's credit 2.083 33.211 Sigma men's credit 2.312 26.878 Rho - Coef. (t-stat) -0.075 (-1.313) Log likelihood -1424.393 No. of observations 1105 895 Source: BIDS-World Bank household survey data, 1991-92. 54 Table 4.3 Alternative Estimates of the Impact of Credit on School Enrollment of Children Aged 5-17 Boys Girls Explanatory Variables Naive WESML- Naive WESML- Unweighted Weighted LIML-FE Unweighted Weighted LIML-FE (Probit) (Probit) (Probit) (Probit) Amount borrowed by female from BRAC .024 .027 .0394 .020 .015 -.0203 l __________________________________ (1.515) (1.745) (0.917) (1.167) (0.938) (-0.552) Amount borrowed by male from BRAC -0.005 0.002 -.0040 .044 .049 .0495 (-0.231) (.085) (-0.107) (1.986) (2.192) (1.152) Amount borrowed by female from BRDB .042 .035 .1210 .011 .002 -.0099 (2.346) (1.588) ( 2.573) (0.612) (0.082) (-0.220) E_n Amount borrowed by male from BRDB .022 .028 .0361 .005 -0.005 .0321 (1.466) (1.447) ( 0.934) (0.331) (-0.236) (0.665) Amountborrowedby female from GB .053 .062 .1025 .023 .019 .0128 (4.141) (4.461 ( 2.364) (1.785) (1.412) (0.334) Amount borrowed by male from GB .053 .074 .0736 .100 .029 .0582 (2.707) (3.089) ( 1.688) (0.614) (0.532) (1.298) Rho (women) -.2192 .1648 (-1.054) (1.029) Rho (men) -.0284 -.1360 (-0.177) (-0.720) Log likelihood -786.506 -779.369 -3802.873 -728.630 -729.449 -3702.947 No. of observations 1341 1341 2940 1269 1269 2885 Note: Figures in parentheses are t-ratios. Source: BIDS-World Bank household survey data, 1991-92. Table 4.4 Alternative Estimates of the Impact of Credit on Log Labor Supply by Gender Men Women | Explanatory Variables Naive WESML- Naive WESML- Unweighted Weighted LIML-FE Unweighted Weighted LIML-FE |____________________________ (Tobit) (Tobit) (Tobit (Tobit) l Amount borrowed by female from .010 .013 -.1813 .028 .054 -.0117 BRAC (1.290) (1.623) (-5.884) (1.163) (2.106) (-.128) Amount borrowed by male from .007 .002 -.1369 -0.072 -0.042 -.0448 BRAC (0.676) (0.169) (-2.155) (-2.049) (-1.103) (-.520) Amount borrowed by female from -0.002 -0.000 -.2308 .131 .178 -.0139 BRDB (-0.169) (40.020) (-7.066) (4.969) (5.043) (-.139) Amount borrowed by male from .006 .001 -.1440 -0.007 .043 -.0144 BRDB (0.813) (0.072) (-2.129) (-0.303) (1.278) (-.181) Amountborrowedby female from .012 .013 -.2189 .116 .134 .0152 GB (1.910) (1.803) (-6.734) (6.275) (6.236) ( .162) Amount borrowed by male from -0.014 -0.027 -.1592 .081 .084 -.0570 GB (-1.594) (-2.488) (-2.524) (3.012) (2.406) (-.677) Rho (women) .6564 .1255 ( 7.461) (1.062) Rho (men) .4929 .0560 .__________________________ _____________ ( 2.5 12) ( .592) Log likelihood -10401.817 _ -10537.668 -18395.082 -9020.541 -8696.531 -15069.781 j No. of observations 5846 5846 6914 5693 5693 6602 Note: Figures in parentheses are t-ratios. Source: BIDS-World Bank household survey data, 1991-92. Table 4.5 Alternative Estimates of the Impact of Credit on Per Capita Expenditure Food Non-food Total Explanatory Variables Naive WESML- Naive WESML- Naive WESML- Un- Weighted LIML-FE Un- Weighted LIML-FE Un- Weighted LIML-FE weighted (OLS) weighted (OLS) weighted (OLS) (OLS) (OLS) (OLS) Amount borrowed by female .006 .005 .0057 .008 .009 .0220 .007 .007 .0394 from BRAC (3.040) (2.563) (0.658) (1.390) (1.668) (0.544) (3.048) (2.847) (4.237) Amount borrowed by male .007 .007 .0060 .014 .017 .0364 .010 010 .0192 from BRAC (2.544) (2.296) (0.801) (1.966) (2.130) (1.388) (2.906) (2.835) (1.593) Amount borrowed by female .003 .003 .0101 -0.001 .006 .0139 .002 .003 .0402 from BRDB (1.523) (.025) (1.051) (-0.215) (0.826) (0.320) (0.573) (0.906) (3.813) u Amount borrowed by male .008 .007 .0138 .001 .011 .0349 .007 .007 .0233 fromBRDB (4.721) (2.727) (1.845) (0.108) (1.536) (1.330) (3.118) (2.253) (1.936) Amount borrowed by female .005 .005 .0114 .000 .009 .0199 .003 .004 .0432 from GB (3.098) (2.700) (1.263) (0.007) (1.760) (0.467) (1.400) (1.765) (4.249) Amount borrowed by male -0.001 -0.001 .0087 -0.001 .004 .0182 .001 .001 .0179 from GB (-0.252) (-0.471) (1.163) (-0.216) (0.476) (0.665) (0.252) (0.325) (1.431) Rho (women) -.1026 -.0564 -.4809 (-.697) (-0.222) (-4.657) Rho (men) -.1077 -.1300 -.2060 (-.980) (-0.858) (-1.432) Log likelihood -5090.877 -5090.877 -5311.365 -8712.608 -8712.608 -10620.08 -5784.156 -5784.156 -6633.559 No. of observations 4567 4567 5218 4567 4567 5218 4567 4567 5218 Note: Figures in parentheses are t-ratios. Source: BIDS-World Bank household survey data, 1991-92. Table 4.6 Alternative Estimates of the Impact of Credit on Contraceptive Use and Recent Fertility of Currently Married Women Aged 15-49 years Contraceptive Use Recent Fertility i Explanatory Variables Naive WESML- Naive WESML- LIMLFE LIML-FE Unweighted Weighted Unweighted Weighted (Probit) (Probit) (Probit) (Probit) l Amount borrowed by female from BRAC 0.017 0.006 -.0735 0.006 -0.008 .0790 (1.143) (0.374) (-1.693) (0.414) (-0.493) (2.372) Amount borrowed by male from BRAC 0.012 0.004 .0395 -0.012 -0.005 .0543 (0.570) (0.170) (0 .745) (-0.605) (-0.237) (1.353) Amount borrowed by female from BRDB -0.023 -0.032 -.1163 -0.015 -0.106 .0502 (-1.348) (-1.285) (-2.421) (-0.849) (-0.447) (1.312) u Amount borrowed by male from BRDB 0.026 0.034 .0839 0.019 0.026 -.0744 (1.906) (1.610) (1.475) (1.318) (1.215) (-1.976) Amount borrowed by female from GB 0.033 0.021 -.0905 -0.024 -0.035 -.0348 (2.842) (1.469) (-2.011) (-1.991) (-2.534) (-0.951) Amount borrowed by male from GB -0.036 -0.494 .0000 0.013 0.008 -.0743 (-2.084) (-2.059) ( 0.000) (0.735) (0.365) (-2.193) Rho (women) .4253 -.4319 ( 2.075) (-2.718) Rho (men) -.2032 .3511 ____________ (-0.700) (2.701) Log likelihood -2458.954 ___________ __________ -2444.341 No. of observations 1731 1731 1731 1557 1557 1557 Note: Figures in parentheses are t-ratios. Source: BIDS-World Bank household survey data, 1991-92. Table 4.7 Alternative Estimates of the Impact of Credit on Log Women's Non-land Assets Women Explanatory Variables Naive WESML- Unweighted Weighted LIML-FE (Tobit) (Tobit) Amount borrowed by female from BRAC .277 .182 .0318 (4.359) (2.834) (0 .356) Amount borrowed by male from BRAC .141 .110 .1005 (1.615) (1.214) (0.468) Amount borrowed by female from BRDB .078 -0.096 .1257 (1.040) (-0.949) (1.043) Amount borrowed by male from BRDB .234 .138 .0334 (3.934) (1.608) (0.141) Amount borrowed by female from GB .232 .195 .1131 (4.402) (3.318) (1.317) Amount borrowed by male from GB .125 .096 -.0457 (1.676) (1.029) (-0.200) Rho (women) .1136 (1.325) Rho (men) -.0148 (-0.053) Log likelihood -3007.646 -2939.802 -4226.176 No. of observations 1517 1517 1757 Note: Figures in parentheses are t-ratios. Source: BIDS-World Bank household survey data, 1991-92. Table 4.8 Alternative Estimates of the Impact of Credit on Log Body Mass Index (BMI) of Children of Age Less Less than 10 Boys Girls Explanatory Variables Naive WESML- Naive_WESML- LIML-FE LIML-FE Unweighted Weighted Unweighted Weighted (OLS) (OLS) (OLS) (OLS) Amount borrowed by female from BRAC -0.003 -0.003 -0.013 .000 -0.001 .004 (-1.692) (-1.602) (-1.248) (0.105) (-0.299) (1.365) Amount borrowed by male from BRAC -0.000 .000 -0.005 .006 .007 .006 (-0.102) (0.095) (-0.536) (1.256) (1.390) (1.070) Amount borrowed by female from BRDB -0.004 -0.003 -0.011 -0.000 -0.496 .002 (-0.975) (-0.496 (-0.827) (-0.100) (-0.317) (.244) o | Amount borrowed by male from BRDB .002 .002 -0.011 .001 .000 .001 (0.693) (0.519) (-1.017) (0.481) (0.091) (.145) Amount borrowed by female from GB .002 .001 -0.009 -0.000 -0.000 .005 (1.020) (0.790) (-0.797) (-0.173) (-0.179) (1.82) Amount borrowed by male from GB .001 .001 -0.006 .005 .007 .009 (0.461) (0.651) (-0.623) (2.562) (2.917) (2.293) Rho (women) .482 -0.165 l___.___._ (1.156) (-1.441) Rho (men) .399 -0.051 ____________________________________ (1.146) .___________ _ ______1 4 (-0.534) Log likelihood -2998.448 _ -2921.321 No. of observations 378 409 Note: Figures in parentheses are t-ratios. Source: BIDS-World Bank household survey data, 1991-92. Table 4.9 Alternative Estimates of the Impact of Credit on Log Height of Children Boys Girls Explanatory Variables Naive WESML- Naive WESML- LIML-FE LIML-FE Unweighted Weighted Unweighted Weighted (OLS) (OLS) (OLS) (OLS) Amount borrowed by female from BRAC .001 .002 .024 .002 .004 -0.049 (0.465) (1.255) (.449) (1.285) (2.012) (1.401) Amount borrowed by male from BRAC .006 .005 .010 -0.004 -0.004 .023 (1.969) (1.646) (.236) (-0.999 (-0.834) (.598) Amount borrowed by female from BRDB -0.001 -0.001 .102 .002 .001 .071 (-0.266) (-0.223) (1.814) (0.705) (0.121) (1.583) c> Amount borrowed by male from BRDB -0.003 -0.003 .047 -0.002 -0.006 .011 (-1.342) (-1.084) (1.100) (-1.083) (-1.903) (.272) Amount borrowed by female from GB -0.004 -0.003 .091 .001 .001 .085 (-2.820) (-1.923) (1.738) (0.94) (0.528) (2.554) Amount borrowed by male from GB -0.002 -0.002 .068 .000 -0.001 .031 (-1.301) (-1.104) (1.557) (0.076 (-0.261) (.797) Rho (women -0.141 -0.153 l___________ ___________ (-0.578) (-1.036) Rho (men) -0.060 -0.011 l_____________ _____________ (-0.344) (-0.083) Log likelihood -3542.159 . -3497.840 No. of observations 378 1341 409 1269 Note: Figures in parentheses are t-ratios. Source: BIDS-World Bank household survey data, 1991-92. Table 4. 10 Alternative Estimates of the Impact of Credit on Log Weight of Children Boys Girls Explanatory Variables By il Naive WESML- Naive WESML- LIML-FE LIML-FE Unweighted Weighted Unweighted Weighted (OLS) (OLS) (OLS) (OLS) Amount borrowed by female from BRAC -0.002 .001 -0.017 .005 .007 -0.015 (-0.434) (0.229) (-1.308) (1.193) (1.713) (-1.641) Amount borrowed by male from BRAC .011 .010 .018 -0.003 -0.001 .002 (1.702) (1.467) (1.756) (-0.264) (-0.091) (.177) Amount borrowed by female from BRDB -0.005 -0.005 -0.019 .004 -0.000 -0.030 (-0.730) (-0.439) (-1.030) (0.576) (-0.044) (-1.623) Amount borrowed by male from BRDB -0.004 -0.005 -0.002 -0.004 -0.012 -0.009 (-0.855 (-0.669) (-0.199) (-0.723) (-1.714) (-0.820) Amount borrowed by female from GB -0.006 -0.005 -0.027 .002 .001 -0.022 (-2.067) (-1.340) (-1.981) (0.751) (0.401) (-1.987) Amount borrowed by male from GB -0.003 -0.003 -0.001 .006 .006 .009 (-0.953) (-0.637) (-0.090) (1.339) (1.187) (1.246) Rho (women) .560 .655 (2.115) (3.628) Rho (men) -.033 -0.049 .__________ (-0.169) (-0.592 Log likelihood -3159.857 _ -3137.600 No. of observations 378 409 Note: Figures in parentheses are t-ratios. Source: BIDS-World Bank household survey data, 1991-92. APPENDIX A Table Al Unweighted Naive Estimates of Impact of Credit by Gender on Log Labor Supply (Tobit) Men Women Explanatory Variables Coef. asymptotic Coef. asymptotic t-ratio t-ratio Parents of HH head own land '-0.005 -0.113 .574 4.802 Brothers of HH head own land -0.026 -1.247 -0.066 -1.053 Sisters of HH head own land .056 2.886 -0.008 -0.134 Parents of HH head's spouse own land .037 1.169 .183 1.948 Brothers of HH head's spouse own land -0.025 -1.534 -0.011 -0.214 Sisters of HH head's spouse own land .018 0.966 .083 1.423 Log HH land assets in decimal -0.016 -1.358 .011 0.318 Highest grade completed by HH head -0.069 -5.698 .218 5.035 Sex of HH head (1=male) .326 2.484 -0.587 -1.440 Age of HH head (years) -0.007 -3.266 .005 0.924 Highest grade completed by adult female -0.029 -3.115 -0.112 -2.359 in HH ______ Highest grade completed by adult male in -0.101 -7.964 -0.300 -7.457 HH__ _ ___ _ _ No adult male in HH 1.770 3.980 No adult female in HH .130 0.665 No spouse present in HH .059 0.691 .569 2.134 Round 2 dummy -0.023 -0.251 .076 0.279 Round 3 dmmy -0.104 -1.025 -0.111 -0.366 Age in years .128 10.538 .507 14.516 Age in years squared -0.002 -9.575 -0.007 -13.614 Highest grade completed .147 13.193 .013 0.241 Amount borrowed by female from BRAC .010 1.290 .028 1.163 Amount borrowed by male from BRAC .007 0.676 -0.072 -2.049 Amount borrowed by female from BRDB -0.002 -0.169 .131 4.969 63 Table Al (continued) Unweighted Naive Estimates of Impact of Credit by Gender on Log Labor Supply (Tobit) Men Women Explanatory Variables Coef. asymptotic Coef. asymptotic t-ratio t-ratio Amount borrowed by male from BRDB .006 0.813 -0.007 -0.303 Amount borrowed by female from GB - .012 1.910 .116 6.275 Amount borrowed by male from GB -0.014 -1.594 .081 3.012 Participated but did not take credit -0.289 -3.328 -0.065 -0.246 Has any primary school .026 0.557 .532 3.861 Has rural Health center -0.214 4.330 .130 0.866 Has family planning center? .225 2.861 -0.031 -0.135 Is Dal/Midwife available? .015 0.315 -0.594 4.367 Price of rice .016 0.636 -0.323 -4.119 Price of wheat flour -0.040 -2.076 .067 1.115 Price of mustard oil -0.002 -0.623 -0.012 -1.078 Price of hen egg .000 0.009 .028 0.623 Price of milk -0.009 -1.161 -0.072 -3.096 Price of potato -0.005 -0.297 -0.078 -1.654 Average female wage -0.001 -0.377 -0.049 -4.281 No female wage dummy -0.025 -0.262 -1.425 -4.962 Average male wage -0.001 -0.466 -0.007 -0.780 Distance to bank (kin) -0.001 -0.130 .080 3.670 Constant 3.411 7.689 -2.356 -1.840 Log likelihood -10401.817 -9020.541 Pseudo R2 0.037 0.047 No. of observations 5846 5693 Source: BIDS-World Bank household survey data, 1991-92. 64 Table A2 Weighted Naive Estimates of Impact of Credit by Gender on Log Labor Supply (Tobit) Men Women Explanatory Variables Coef. asymptotic Coef. asymptotic t-ratio t-ratio Parents of HH head own land -0.053 -1.213 .559 4.227 Brothers of HH head own land -0.014 -0.634 -0.044 -0.648 Sisters of HH head own land .083 4.008 -0.138 -2.175 Parents of HH head's spouse own land .111 3.269 .099 0.952 Brothers of HH head's spouse own land -0.031 -1.759 .061 1.082 Sisters of HH head's spouse own land .002 0.118 .080 1.204 Log HH land assets in decimal -0.015 -1.177 .034 0.874 Highest grade completed by HU{ head -0.066 -5.205 .205 4.357 Sex of HH head (1 =male) .406 3.085 -1.569 -3.774 Age of HH head (years) -0.004 -2.026 .013 2.243 Highest grade completed by adult female -0.042 -4.419 -0.115 -2.382 in HH l Highest grade completed by adult male in -0.080 -6.231 -0.315 -7.293 HH No adult male in HH 1.803 3.959 No adult female in HH -0.005 -0.027 _ _ No spouse present in HH -0.058 -0.720 .054 0.197 Round 2 dummy -0.088 -0.950 -0.032 -0.112 Round 3 dmmy -0.164 -1.594 -0.146 -0.461 Age in years .161 13.005 .473 12.999 Age in years squared -0.002 -12.506 -0.006 -12.272 Highest grade completed .123 10.861 .044 0.834 Amount borrowed by female from BRAC .013 1.623 .054 2.106 Amount borrowed by male from BRAC .002 0.169 -0.042 -1.103 Amount borrowed by female from BRDB -0.000 -0.020 .178 5.043 Amount borrowed by male from BRDB .001 0.072 .043 1.278 Amount borrowed by female from GB .013 1.803 .134 6.236 65 Table A2 (continued) Weighted Estimates of Impact of Credit by Gender on Labor Supply Men Women Explanatory Variables Coef. asymptotic Coef. asymptotic t-ratio t-ratio Amount borrowed by male from GB -0.027 -2.488 .084 2.406 Participated but did not take credit -0.288 -2.971 .391 1.315 Has any primary school .022 0.476 .641 4.484 Has rural Health center -0.290 -5.988 .024 0.155 Has family planning center? .344 4.182 -0.007 -0.027 Is Dai/Midwife available? .176 3.611 -0.821 -5.648 Price of rice -0.012 -0.460 -0.298 -3.573 Price of wheat flour -0.018 -0.848 -0.045 -0.695 Price of mustard oil -0.001 -0.353 -0.014 -1.252 Price of hen egg -0.009 -0.525 .069 1.514 Price of milk -0.014 -1.703 -0.103 -4.155 Price of potato -0.000 -0.005 -0.077 -1.567 Average female wage .006 1.646 -0.041 -3.432 No female wage dummy .056 0.574 -1.550 -5.142 Average male wage .001 0.335 .003 0.352 Distance to bank (kim) .004 0.452 .090 3.803 Constant 2.569 5.758 -0.456 -0.344 Log likelihood -10537.668 -8696.531 Pseudo R2 0.036 0.051 No. of observations 5846 5693 Source: BIDS-World Bank household survey data, 1991-92. 66 Table A3 Unweighted Naive Estimates of the Impact of Credit on Log Per Capita Expenditure (OLS) Food Non-food Total Explanatory Variables Coef. t-ratio Coef. t-ratio Coef. t-ratio Parents of HH head own land .021 2.270 .065 2.471 .029 2.548 Brothers of HH head own land .003 0.565 .016 1.144 .003 0.559 Sisters of HH head own land -0.008 -1.663 .007 0.526 -0.005 -0.785 Parents of HH head's spouse own land -0.004 -0.539 .049 2.269 .004 0.457 Brothers of HH head's spouse own land -0.006 -1.347 .012 1.043 .000 0.069 Sisters of HH head's spouse own land -0.003 -0.683 .001 0.094 - -0.004 -0.647 Log household land .002 0.880 .042 5.307 .011 3.280 Highest grade completed by HH head -0.005 -1.268 -0.029 -2.885 -0.010 -2.246 Sex of household head (I=male) .058 1.715 .001 0.011 .037 0.891 Age of household head (years) -0.002 -3.801 -0.006 -5.233 -0.003 -4.844 Highest grade completed by an adult female .011 4.574 .045 6.571 .020 6.632 in HH Highest grade completed by an adult male in .015 4.541 .070 7.443 .027 6.580 HH No adult male in HH -0.023 -0.596 -0.220 -2.072 -0.051 -1.098 No adult female in HH .190 4.558 .146 1.250 .209 4.110 No spouse present in HH .078 3.862 .188 3.298 .104 4.182 Round 2 -0.037 -1.678 .159 2.565 -0.018 -0.675 Table A3 (continued) Unweighted Naive Estimates of the Impact of Credit on Log Per Capita Expenditure (OLS) Food Non-food Total Explanatory Variables Coef. t-ratio Coef. t-ratio Coef. t-ratio Round 3 -0.083 -3.391 -0.762 -11.138 -0.204 -6.856 Amount borrowed by female brom BRAC .006 3.040 .008 1.390 .007 3.048 Amount borrowed by male from BRAC .007 2.544 .014 1.966 .010 2.906 Amount borrowed by female from BRDB .003 1.523 -0.001 -0.215 .002 0.573 Amount borrowed by male from BRDB .008 4.721 .001 0.108 .007 3.118 Amount borrowed by female from GB .005 3.098 .000 0.007 .003 1.400 Amount borrowed by male from GB -0.001 -0.252 -0.001 -0.216 .001 0.252 00 Participate but no credit .039 1.872 -0.165 -2.852 .011 0.434 Has any primary school -0.054 -4.962 -0.132 -4.321 -0.074 -5.617 Has rural health center -0.043 -3.585 -0.136 -3.997 -0.059 -4.018 Has family planning center? .068 3.599 .050 0.938 .073 3.178 Is Dal/Midwife available? -0.068 -6.169 -0.001 -0.024 -0.053 -3.927 Price of rice .029 4.567 -0.021 -1.198 .020 2.567 Price of wheat flour .012 2.535 .056 4.123 .020 3.425 Price of mustard oil -0.002 -2.049 -0.006 -2.615 -0.003 -2.90 Price of hen egg .000 0.058 .008 0.695 .002 0.385 Price of milk .009 4.632 .016 3.046 .011 4.573 Table A3 (continued) Unweighted Naive Estimates of the Impact of Credit on Log Per Capita Expenditure (OLS) Food Non-food Total Explanatory Variables Coef. t-ratio Coef. t-ratio Coef. t-ratio Price of potato -0.000 -0.111 .030 2.862 .009 1.952 Average female wage -0.000 -0.517 -0.008 -3.085 -0.002 -1.641 No female wage dummy .007 .307 -0.308 4.775 -0.050 -1.776 Average male wage .002 2.631 .007 3.718 .002 2.905 Distance to Bank (kQm) -0.007 4.008 -0.006 -1.180 -0.006 -2.979 Constant 3.620 38.503 2.141 8.114 3.900 33.929 Adjusted R2 0.131 0.257 0.179 No. of observations 4567 4567 4567 Source: BIDS-World Bank household survey data, 1991-92. Table A4 Weighted Naive Estimates of the Impact of Credit on Log Per Capita Expenditure (OLS) Food Non-food Total Explanatory Variables Coef. t-ratio Coef. t-ratio Coef. t-ratio Parents of HH head own land .021 2.127 .058 2.118 .029 2.390 Brothers of HR head own land -0.000 -0.020 .012 0.832 .000 0.058 Sisters of HH head own land -0.010 -1.964 .003 0.223 -0.006 -0.936 Parents of HH head's spouse own land .001 0.134 .062 2.761 .011 1.124 Brothers of HH head's spouse own land -0.007 -1.649 .005 0.429 -0.001 -0.252 Sisters of HH head's spouse own land -0.004 -0.730 -0.000 -0.000 -0.006 -0.913 Log household land .008 2.738 .048 5.769 .017 4.577 Highest grade completed by HH head -0.001 -0.398 -0.031 -3.036 -0.009 -1.958 Sex of household head ( = male) .019 0.579 -0.043 -0.463 .010 0.242 Age of household head (years) -0.002 -5.262 -0.007 -6.283 -0.003 -6.187 Highest erade completed by an adult female in HH .016 6.208 .066 9.556 .028 9.285 Highest grade completed by an adult male in HH .011 3.248 .069 7.227 .024 5.686 No adult male in HH -0.083 -2.171 -0.258 -2.452 -0.091 -2.003 No adult female in HH .135 3.636 .023 0.228 .123 2.758 No spouse present in HH .069 3.513 .133 2.480 .085 3.612 Round 2 -0.039 -1.741 .148 2.420 -0.036 -1.367 Round 3 -0.089 -3.636 -0.776 -11.438 -0.223 -7.536 Amount borrowed by female brom BRAC .005 2.563 .009 1.668 .007 2.847 Amount borrowed by male from BRAC .007 2.296 .017 2.130 .010 2.835 Amount borrowed by female from BRDB .003 1.025 .006 0.826 .003 0.906 Amount borrowed by male from BRDB .007 2.727 .011 1.536 .007 2.253 Table A4 (continued) Weighted Naive Estimates of the Impact of Credit on Log Per Capita Expenditures (OLS) Food Non-food Total Explanatory Variables Coef. t-ratio Coef. t-ratio Coef. t-ratio Amount borrowed by female from GB .005 2.700 .009 1.760 .004 1.765 Amount borrowed by male from GB -0.001 -0.471 .004 0.476 .001 0.325 Participated but did not take credit .039 1.718 -0.141 -2.224 .018 0.666 |Has any primarv school -0.034 -3.091 -0.095 -3.163 -0.051 -3.840 Has rural health center -0.037 -3.154 -0.074 -2.273 -0.042 -2.973 Has family planning center? .065 3.379 .014 0.264 .059 2.577 Is Dat/Midwife available? -0.045 -3.036 .049 1.556 -0.018 -1.313 | Price of rice .027 4.238 -0.040 -2.286 .014 1.860 Price of wheat flour ;013 2.559 .065 4.674 .023 3.822 Price of mustard oil -0.003 -2.952 -0.006 -2.486 -0.004 -3.740 Price of hen egg -0.005 -1.219 .012 1.138 -0.001 -0.196 Price of milk .002 5.828 .022 4.000 .014 5.906 Price of potato .001 0.350 .033 3.186 .015 2.957 Average female wage -0.001 -1.058 -0.008 -3.040 -0.002 -1.927 No female wage dummy -0.026 -1.093 -0.329 -5.115 -0.073 -2.614 Average male wage .002 2.575 .007 3.533 .002 2.823 Distance to Bank (km) -0.006 -3.463 -0.008 -1.480 -0.006 -2.721 Constant 3.671 38.619 2.097 8.005 3.915 34.269 Adiusted R2 0.135 0.279 0.199 No. of observations __ _ 4567 | 4567 | 4567 Source: BIDS-World Bank household survey data, 1991-92. Table AS Weighted and Unweighted Naive Estimates of the Impact of Credit on Log Non-land Assets by Gender Weighted obit) Unweighted (Tobit) Explanatory Variables Male Female Male Female Coef. asymptotic Coef. asymptotic Coef. asymptotic Coef. asymptotic t-ratio t-ratio t-ratio t-ratio Parents of HH head own land -0.049 -0.689 .582 1.877 -0.080 -1.099 .305 1.026 Brothers of HH head own land -0.013 -0.326 .107 0.644 .030 0.76 .141 0.876 Sisters of HH head own land -0.014 -0.379 -0.099 -0.597 -0.008 -0.205 -0.054 -0.349 Parents of HH head's spouse -0.043 -0.736 .328 1.251 -0.020 -0.338 .305 1.224 own land Brothers of HH head's spouse .022 0.687 -0.311 -2.101 .018 0.560 -0.250 -1.818 own land Sisters of HH head's spouse .033 0.872 -0.321 -1.865 .058 .1567 -0.184 -1.179 own land , Log household land .296 12.980 -0.009 -0.094 .266 11.535 -0.050 -0.536 Highest grade completed by -0.090 -3.264 .066 0.547 -0.045 -1.610 -0.098 -0.868 HH head Sex of household head 7.503 25.454 -7.944 -7.291 7.359 24.403 -7.319 -6.802 (l=male) Age of household head (years) -0.005 -1.522 -0.011 -0.771 .003 0.772 -0.021 -1.474 Highest grade completed by an .070 3.887 .283 3.570 .055 2.877 .289 3.718 adult female in HH ___ Highest grade completed by an .179 7.007 .143 1.281 .132 5.004 .240 2.274 adult male in HH I Non adult male in HH .070 0.058 .513 0.434 No adult female in HH -0.587 -2.209 -0.642 -1.941 No spouse present in HH -0.224 -1.573 .-.817 -1.234 -0.098 -0.608 -1.053 -1.531 Amount borrowed by female -0.015 -1.004 .182 2.834 -0.015 -0.920 .277 4.359 brom BRAC Table A5 (continued) Weighted and Unweighted Naive Estimates of the Imapct of Credit on Log-Non-land Assets by Gender Weight (Tobit) Unweigh ed (Tobit) Explanatory Variables Male Female Male Female Amount borrowed by male .061 2.923 .110 1.214 .071 3.260 .141 1.615 from BRAC Amount borrowed by female .028 1.243 -0.096 -0.949 .006 0.305 .078 1.040 from BRDB _ Amount borrowed by male .062 3.180 .138 1.608 .057 3.877 .234 3.934 from BRDB Amount borrowed by female .003 0.188 .195 3.318 .013 0.969 .232 4.402 from GB Amount borrowed by male .037 1.768 .096 1.029 .044 2.415 .125 1.676 from GB _ Participated but did not take .058 0.354 -0.274 -0.385 .046 0.2?l .576 0.909 credit ; Has any primary school -0.061 -0.706 -0.366 -0.977 -0.012 -0.130 -0.699 -1.861 Has rural health center .326 3.724 1.228 3.197 .331 3.444 1.164 2.987 Has family planning center? -0.226 -1.558 -2.081 -3.159 -0.214 -1.417 -1.065 -1.684 Is Dal/Midwife available? .121 1.366 .127 0.323 .102 1.115 .071 0.187 Price of rice -0.131 -2.416 -0.689 -2.878 -0.047 -0.805 -0.702 -2.969 Price of wheat flour .070 1.524 .867 4.247 .007 0.143 .778 3.822 Price of mustard oil -0.007 -1.063 .096 3.249 -0.007 -0.995 .062 2.111 Price of hen eig .007 0.393 .315 4.257 -0.008 -0.397 .335 4.267 Price of milk .026 1.523 -0.329 -4.312 .022 1.239 -0.355 -4.888 Price of ptato .037 1.452 -0.244 -2.153 .009 0.317 -0.025 -0.223 Average female wage -0.001 -0.070 -0.011 -0.345 .012 1.600 -.000 -0.010 No female wage dummy -0.015 -0.086 1.445 1.832 .252 1.352 1.503 1.936 Average male wage -0.006 -1.097 .037 1.63 1 -0.010 _ -1.908 .051 2.442 Table A5 (continued) Weighted and Unweighted Naive Estimates of the Imapct of Credit on Log-Non-land Assets by Gender Explanatory Variables Weighted (Tobit) Unweigh ed gobit) Explanatory Variables MFemale Male Female Distance to Bank (km) -0.076 -5.663 .117 2.003 -0.061 -4.429 .099 1.772 Constant 1.647 4.896 1.510 .792 0.963 6.259 1.920 Pseudo R2 0.182 0.054 0.158 .051 Log likelihood -2488.602 -2939.802 -2574.151 -3007.646 No. of observations 1475 1517 1475 1517 Source: BIDS-World Bank household survey data, 1991-92. Table A6 Unweighted Naive Estimates of Impact of Credit on Children's School Enrollment by Gender (Probit) Explanatory Variables Bo rs Girls Coef. asymptotic Coef. asymptotic t-ratio t-ratio Parents of HH head own land .248 2.775 .094 1.083 Brothers of HH head own land -0.075 -1.860 .049 1.229 Sisters of HH head own land -0.050 -1.240 -0.043 -1.062 Parents of HH head's spouse own land -0.098 -1.573 -0.012 -0.188 Brothers of HH head's spouse own land .045 1.373 .006 0.175 Sisters of HH head's spouse own land .035 0.949 -0.018 -0.477 Log household land .079 3.454 .034 1.392 Highest grade completed by HH head .060 2.301 .027 1.010 Sex of household head (1 =male) .462 1.590 .109 0.293 Age of household head (years) -0.011 -2.574 -0.008 -1.818 Highest grade completed by an adult .051 2.399 .019 0.871 female in HH I Hi hest grade completed by an adult male .027 1.110 .072 2.998 in SHI No adult male in HH .128 0.368 .221 0.539 No adult female in HH -0.454 -0.838 -0.800 -1.791 No spouse present in HH -0.075 -0.369 -0.005 -0.022 Age in years .646 9.178 .836 11.151 Age in years squared -0.031 -9.607 -0.039 -11.213 Amount borrowed by female brom BRAC .024 1.515 .020 1.167 Amount borrowed by male from BRAC -0.005 -0.231 .044 1.986 Amount borrowed by female from BRDB .042 2.346 .011 0.612 Amount borrowed by male from BRDB .022 1.466 .005 0.331 Amount borrowed by female from GB .053 4.141 .023 1.785 Amount borrowed by male from GB .053 2.707 .031 1.585 Participate but no credit .243 1.463 .100 0.614 Has any primary school .098 1.020 .171 1.702 Has rural health center .129 1.321 -0.049 -0.456 Has family planning center? -0.245 -1.601 -0.566 -3.305 Is Dai/Midwife available? -0.095 -1.029 .064 0.655 Price of rice .029 0.485 .139 2.217 Price of wheat flour .059 1.204 -0.056 -'.139 75 Table A6 (continued) Unweighted Naive Estimates of Impact of Credit on Children's School Enrollment by Gender (Probit) Explanatory Variables Bo rs Girls Coef. asymptotic Coef. asymptotic I___________ t-ratio t-ratio Price of mustard oil -0.010 -1.395 -0.011 -1.383 Price of hen egg -0.003 -0.128 -0.011 -0.487 Price of milk -0.010 -0.545 .037 2.108 Price of potato -0.023 -0.723 .015 0.505 Average female wage .008 1.115 .015 1.883 No female wage dummy .276 1.473 .369 1.879 Average male wage .012 2.262 .005 0.876 Distance to Bank (km) .006 0.479 -0.022 -1.537 Constant 4.284 4.637 -5.650 -5.755 Pseudo R2 0.151 0.169 Log likelihood -786.506 -728.630 No. of observations 1341 1269 Source: BIDS-World Bank household survey data, 1991-92. 76 Table A7 Weighted Naive Estimates of Impact of Credit on Children's School Enrollment by Gender (Probit) Explanatory Variables Bo s Girls Coef. asymptotic Coef. asymptotic t-ratio t-ratio Parents of HH head own land .263 2.774 .164 1.743 Brothers of HH head own land -0.125 -3.002 .070 1.745 Sisters of HH head own land -0.037 -0.876 -0.094 -2.332 Parents of HH head's spouse own land , -0.037 -0.568 -0.040 -0.591 Brothers of HH head's spouse own land .006 0.169 -0.044 -1.241 Sisters of HH head's spouse own land .049 1.244 .041 1.021 Log household land .076 3.106 .013 0.526 Highest grade completed by HH head .044 1.696 .011 0.393 Sex of household head (1 =male) .352 1.243 .209 0.534 Age of household head (years) -0.015 -3.688 -0.010 -2.293 Highest grade completed by an adult .040 1.822 .036 1.669 female in HH Highest grade completed by an adult male .052 2.114 .087 3.608 No adult male in HH .196 0.551 .165 0.386 No adult female in HH .141 0.329 -0.796 -2.012 No spouse present in HH -0.080 -0.421 -0.006 -0.029 Age in years .673 9.410 .730 9.711 Age in years squared -0.033 -9.749 -0.034 -9.675 Amount borrowed by female brom BRAC .027 1.745 .015 0.938 Amount borrowed by male from BRAC .002 0.085 .049 2.192 Amount borrowed by female from BRDB .035 1.588 .002 0.082 Amount borrowed by male from BRDB .028 1.447 -0.005 -0.236 Amount borrowed bv female from GB .062 4.461 .019 1.412 Amount borrowed by male from GB .074 3.089 .029 1.222 Participate but no credit .355 1.964 .094 0.532 Has any primary school .136 1.438 .235 2.395 Has rural health center .222 2.321 -0.076 -0.734 Has family plannin center? -0.545 -3.370 -0.551 -3.225 Is Dai/Midwife available? -0.152 -1.596 .102 1.038 Price of rice -0.015 -0.260 .089 1.440 Price of wheat flour .081 1.661 .010 0.203 77 Table A7 (continued) Weighted Naive Estimates of Impact of Credit on Children's School Enrollment by Gender (Probit) Explanatory Variables Bo rs Girls Coef. asymptotic Coef. asymptotic _______________________________________________ t-ratio t-ratio Price of mustard oil -0.014 1.947 .007 0.963 Price of hen egg .003 0.136 .015 0.729 Price of milk -0.003 -0.183 .035 1.938 Price of potato -0.013 -0.410 .027 0.953 Average female wage .015 1.884 .015 1.885 No female wage dummy .396 2.070 .316 1.597 Average male wage .010 1.808 .003 0.526 Distance to Bank (kIm) -0.005 -0.325 -0.030 -1.932 Constant -3.860 -4.298 -6.267 -6.342 Pseudo R2 0.161 0.1704 Log likelihood -779.369 -729.449 No. of observations 1341 1269 Source: BIDS-World Bank household survey data, 1991-92. 78 Table A8 Unweighted Naive Estimates of the impact of Credit on Children's Log Height by Gender (OLS) Explanatory Variables Boys Girls Coef. t-ratio Coef. t-ratio Parents of HH head own land .011 1.531 .012 1.462 Brothers of HH head own land .008 2.277 -0.009 -2.041 Sisters of HH head own land .003 0.748 .006 1.300 Parents of HH head's spouse own land -0.016 -2.152 -0.012 -1.458 Brothers of HH head's spouse own land -0.001 -0.379 .003 0.779 Sisters of HH head's spouse own land .008 2.109 .003 0.817 Log household land .002 0.749 -0.006 -2.079 Highest grade completed by HH head -0.006 -1.880 -0.005 -1.607 Sex of household head (1 =male) -0.040 -1.109 .023 0.847 Age of household head (years) -0.001 -1.219 .000 0.419 Highest grade completed by an adult 0.000 0.003 .008 3.360 female in HH__ _ _ __ _ __ _ _ _ Hi hest grade completed by an adult male .006 2.037 .005 2.200 No adult male in HH .080 1.691 .053 1.416 No adult female in HH -0.071 -0.785 -0.057 -0.861 No spouse present in HH -0.057 -1.478 .029 1.036 Round 3 -0.007 -0.188 .008 0.231 Age in years .110 20.385 .104 23.880 Age in years squared -0.005 -9.783 -0.005 -11.398 Amount borrowed by female brom BRAC .001 0.465 .002 1.285 Amount borrowed by male from BRAC .006 1.969 -0.004 -0.999 Amount borrowed by female from BRDB -0.001 -0.266 .002 0.705 Amount borrowed by male from BRDB -0.003 -1.342 -0.002 -1.083 Amount borrowed by female from GB -0.004 -2.820 .001 0.943 Amount borrowed by male from GB -0.002 -1.301 .000 0.076 Participate but no credit .044 2.534 .009 0.585 Has any primary school .011 0.745 -0.049 -3.139 Has rural health center -0.014 -0.913 -0.003 -0.184 Has family planning center? .001 0.023 .031 1.359 Is Dai/Midwife available? -0.035 -1.431 .036 1.337 Price of rice .008 0.965 .001 0.193 Price of wheat flour -0.012 -1.247 .011 1.190 79 Table A8 (continued) Unweighted Naive Estimates of the Impact of Credit on Children's Height by Gender (OLS) Explanatory Variables Bov Girls Coef. t-ratio Coef. t-ratio Price of mustard oil .001 0.907 -0.000 -0.201 Price of hen eer -0.006 -0.289 -0.033 -1.421 Price of milk .003 1.486 -0.003 -1.404 Price of potato .005 0.792 .001 0.265 Average female wage .000 0.102 .001 1.198 No female wage dummy -0.015 -0.615 -0.006 -0.223 Average male wage -0.001 -0.572 -0.003 -1.609 Distance to Bank (krm) .003 0.729 -0.006 -1.420 Constant 4.210 34.155 4.234 34.487 Adjusted R2 0.843 0.834 No. of observations 378 409 Source: BIDS-World Bank household survey data, 1991-92. 80 Table A9 Weighted Estimates of Impact of Credit on Children's Log Height by Gender Explanatory Variables Bovs l Girls Coef. t-ratio Coef. t-ratio Parents of HH head own land .015 2.096 .019 2.336 Brothers of HH head own land .005 1.521 -0.011 -2.839 Sisters of HH head own land -0.001 -0.297 .010 2.174 Parents of HH head's spouse own land -0.030 4.166 -0.015 -1.862 Brothers of HH head's spouse own land .007 1.716 .002 0.430 Sisters of HH head's spouse own land .008 2.160 .005 1.138 Log household land .002 0.743 -0.001 -0.225 Highest grade completed by HH head -0.004 -1.540 -0.002 -0.682 Sex of household head (1 =male) -0.008 -0.227 .019 0.692 Age of household head (years) -0.001 -1.488 .000 0.170 Highest grade completed by an adult -0.002 -1.019 .004 1.717 female in HH__ _ _ __ _ __ _ _ _ Highest grade completed by an adult male .007 2.555 .005 2.128 No adult male in HH .125 2.346 .045 1.117 No adult female in HH -0.105 -1.062 -0.077 -1.036 No spouse present in HH -0.046 -1.062 .037 1.415 Round 3 -0.015 -0.390 -0.028 -0.790 Age in years .114 20.022 .108 25.221 Age in years squared -0.006 -10.225 -0.005 -12.525 Amount borrowed by female brom BRAC .002 1.255 .004 2.012 Amount borrowed by male from BRAC .005 1.646 -0.004 -0.834 Amount borrowed by female from BRDB -0.001 -0.223 .001 0.121 Amount borrowed by male from BRDB -0.003 -1.084 -0.006 -1.903 Amount borrowed by female from GB -0.003 -1.923 .001 0.528 Amount borrowed by male from GB -0.002 -1.104 -0.001 -0.261 Participate but no credit .059 3.070 .024 1.273 Has any primary school .029 1.971 -0.017 -1.149 Has rural health center -0.005 -0.327 -0.007 -0.383 Has family planning center? .000 0.017 -0.029 -1.141 Is Dai/Midwife available? -0.043 -1.737 .024 0.981 Price of rice .007 0.981 -0.005 -0.732 Price of wheat flour -0.006 -0.605 .010 1.015 Price of mustard oil .000 0.277 .001 0.343 Table A9 (continued) Weighted Naive Estimates of Impact of Credit on Children's Height by Gender (OLS) Explanatory Variables YS Girls Coef. t-ratio Coef. t-atio Price of hen e_g -0.017 -0.757 -0.024 -1.027 Price of milk .003 1.387 -0.001 -0.270 Price of potato .008 1.307 .008 1.544 Average female wage .001 0.875 .001 0.904 No female wage dummy .014 0.575 -0.001 -0.019 Average male wage -0.001 -0.383 -0.003 -1.639 Distance to Bank (klm) .002 0.439 -0.001 -0.288 Constant 4.142 34.849 4.189 37.585 Adjusted R2 0.837 0.850 No. of observations 378 409 Source: BIDS-World Bank household survey data, 1991-92. 82 Table A1O Unweighted Naive Estimates of Impact of Credit on Children's Log Weight by Gender (OLS) Explanatory Variables Boys Girls Coef. t-ratio Coef. t-ratio Parents of HH head own land .002 0.124 .048 2.536 Brothers of HH head own land .012 1.496 -0.022 -2.349 Sisters of HH head own land .007 0.761 .011 1.114 Parents of HH head's spouse own land -0.035 -2.123 -0.035 -1.947 Brothers of HH head's spouse own land -0.008 -0.950 .000 0.036 Sisters of HH head's spouse own land .019 2.240 -0.000 -0.054 Log household land -0.002 -0.272 -0.011 -1.685 Highest -grade completed by HH head -0.009 -1.271 -0.013 -1.926 Sex of household head (1 =male) -0.006 -0.073 .069 1.133 Age of household head (years) -0.001 -0.805 .003 2.148 Highest grade completed by an adult .002 0.479 .020 3.827 female in HH Hiahest grade completed by an adult male .014 1.967 .013 2.391 No adult male in HH .110 1.026 .170 2.022 No adult female in HH -0.286 -1.390 -0.324 -2.177 No spouse present in HH -0.031 -0.355 .050 0.775 Round 3 .000 0.002 .050 0.620 Age in years .188 15.382 .169 17.314 Age in years squared -0.008 -6.714 -0.007 -7.299 Amount borrowed by female brom BRAC -0.002 -0.434 .005 1.193 Amount borrowed by male from BRAC .011 1.702 -0.003 -0.264 Amount borrowed by female from BRDB -0.005 -0.730 .004 0.576 Amount borrowed by male from BRDB -0.004 -0.855 -0.004 -0.723 Amount borrowed by female from GB -0.006 -2.067 .002 0.751 Amount borrowed by male from GB -0.003 -0.953 .006 1.339 Participate but no credit .080 2.041 .019 0.532 Has any primary school .000 0.000 -0.105 -2.999 Has rural health center -0.036 -1.043 -0.060 -1.538 Has family planning center? -0.014 -0.273 .034 0.654 Is Dai/Midwife available? -0.037 -0.668 .102 1.695 Price of rice .018 1.016 .012 0.735 Price of wheat flour -0.036 -1.713 .001 0.036 83 Table A10 (continued) Unweighted Naive Estimates of Impact of Credit on Children's Log Weight by Gender (OLS) Explanatory Variables Bc a Girls Coef. t-ratio Coef. t-ratio Price of mustard oil .004 1.148 .000 0.045 Price of hen egg .036 0.741 -0.030 -o.565 Price of milk .006 1.214 -0.009 -1.720 Price of potato -0.004 -0.309 -0.009 -0.740 Average female wage .001 0.488 .003 1.072 No female waLe dummy -0.012 -0.219 -0.019 -0.329 Average male wage -0.003 -0.950 -0.003 -0701 Distance to Bank (Ian) .014 1.613 -0.005 -0.541 Constant 1.859 6.636 1.830 6.617 Adjusted R2 0.780 0.759 No. of observations 378 409 Source: BIDS-World Bank household survey data, 1991-92. 84 Table A 1 Weighted Naive Estimates of Impact of Credit on Children's Log Weight by Gender (OLS) Explanatory Variables B ys Girls Coef. t-ratio Coef. t-ratio Parents of HH head own land .010 0.621 .062 2.580 Brothers of HH head own land .003 0.398 -0.034 -3.999 Sisters of HH head own land -0.002 -0.180 .027 2.590 Parents of HH head's spouse own land -0.049 -2.894 -0.035 -2.037 Brothers of HH head's spouse own land .003 0.297 -0.006 -0.729 Sisters of HH head's spouse own land .015 1.672 -0.003 -0.321 Log household land .002 0.321 -0.001 -0.140 Highest grade completed by HH head -0.006 -0.826 -0.012 -1.986 Sex of household head (1 =male) .090 1.093 .069 1.179 Age of household head (years) -0.002 -1.255 .002 2.175 Highest grade completed by an adult -0.004 -0.781 .014 2.524 female in HH Highest grade completed by an adult male .016 2.486 .014 2.711 in ~~tH No adult male in HH .203 1.613 .155 1.757 No adult female in HH -0.397 -1.703 -0.367 -2.291 No spouse present in HH .022 0.212 .055 0.976 Round 3 .013 0.147 -0.060 -0.772 Age in vears .199 14.769 .182 19.651 Age in years squared -0.009 -6.968 -0.008 -8.928 Amount borr- ..ed by female brom BRAC .001 0.229 .007 1.713 Amount borrowed by male from BRAC .010 1.467 -0.001 -0.091 Amount borrowed by female from BRDB -0.005 -0.439 -0.000 -0.044 Amount borrowed by male from BRDB -0.005 -0.669 -0.012 -1.714 Amount borrowed by female from GB -0.005 -1.340 .001 0.401 Amount borrowed by male from GB -0.003 -0.637 .006 1.187 Participate but no credit .114 2.507 .046 1.139 Has any primary school .018 0.525 -0.032 -1.008 Has rural health center -0.027 -0.809 -0.061 -1.611 Has family planning center? .007 -0.122 -0.117 -2.137 Is Dai/Midwife available? -0.003 -0.044 .084 1.558 Price of rice .028 1.628 -0.014 -0.942 Price of wheat flour -0.029 -1.262 .002 0.118 85 Table A1l (continued) Weighted Naive Estimates of Impact of Credit on Children's Log Weight by Gender (OLS) Explanatory Variables B vys Girls Coef. t-ratio Coef. t-ratio Price of mustard oil .001 0.313 .000 0.073 Price of hen egg .014 0.270 -0.022 -0.429 Price of milk .005 0.870 -0.004 -.864 Price of potato -0.002 -0.153 .011 0.965 Average female wage .003 1.159 .003 1.167 No female wage dummy .063 1.091 .004 0.075 Average male wage -0.002 -0.477 -0.001 -0.383 Distance to Bank (kIn) .014 1.509 .005 0.535 Constant 1.604 5.723 1.842 7.637 Adiusted R2 0.757 0.800 No. of observations 378 409 Source: BIDS-World Bank household survey data, 1991-92. 86 Table A12 Unweighted Naive Estimates of Impact of Credit by Gender on Log Body Mass Index (BAO) af Children Under Age 10 Explanatory Variables B S Girls Coef. t-ratio Coef. t-ratio Parents of HH head own land -0.020 -2.381 .023 2.495 Brothers of HH head own land -0.004 -1.041 -0.005 -1.081 Sisters of HH head own land .001 0.307 -0.000 -0.082 Parents of HH head's spouse own land . -0.003 -0.393 -0.012 -1.315 Brothers of HH head's spouse own land -0.006 -1.232 -0.005 -1.321 Sisters of HH head's spouse own land .003 0.750 -0.007 -1.569 Lo-z household land -0.006 -1.959 .001 0.323 Highest grade completed by HH head .002 0.596 -0.003 -1.007 Sex of household head ( =male) .057 1.395 .023 0.766 Age of household head (years) .000 0.402 .002 3.579 Highest grade completed by an adult .003 0.968 .004 1.699 female in HH__ _ _ _ _ _ _ Hi h t grade completed by an adult male .002 0.548 .002 0.880 No adult male in HH -0.038 -0.696 .064 1.540 No adult female in HH -0.210 -2.846 No spouse present in HH .055 1.346 -0.009 -0.291 Round 3 .024 0.543 .034 0.836 Age in years -0.032 4.973 -0.038 -7.841 Age in years squared .002 3.678 .003 5.684 Amount borrowed by female brom BRAC -0.003 -1.692 .000 0.105 Amount borrowed by male from BRAC -0.000 -0.102 .006 1.256 Amount borrowed by female from BRDB -0.004 -0.975 -0.000 -0.100 Amount borrowed by male from BRDB .002 0.693 .001 0.481 Amount borrowed by female from GB .002 1.020 -0.000 -0.173 Amount borrowed by male from GB .001 0.461 .005 2.562 Participated but did not take credit -0.013 -0.655 .000 0.027 Has any primary school -0.022 -1.256 -0.007 -0.427 Has rural health center -0.008 -0.423 -0.054 -2.770 Has family planning center? -0.016 -0.574 -0.029 -1.113 Is Dai/Midwife available? .032 1.103 .031 1.023 Price of rice .004 0.411 .010 1.136 87 Table A12 (continued) Unweighted Naive Estimat of Impact of Credit by Gender on Log Body Mass Index (BMI) of Children Under Age 10 (OLS) Explanatory Variables Girls ._________________ __ Coef. t-ratio Coef. t-ratio Price of wheat flour -0.011 -1.023 -0.022 -2.057 Price of mustard oil .001 0.704 .001 0.451 Price of hen ez .049 1.941 .037 1.405 Price of milk -0.001 -0.320 -0.002 -0.953 Price of potato -0.016 -2.251 -0.012 -1.965 Average female wage .001 0.808 .000 0.016 No female wage dummy .021 0.704 -0.008 -0.262 Average male wage -0.002 -0.959 .003 1.466 Distance to Banlc (lam) .008 1.855 .007 1.451 Constant -6.554 -44.940 -6.638 -4.370 Ad justed_ R 0.131 0.288 No. of observations 378 409 J Source: BIDS-World Bank household survey data, 1991-92. 88 Table 13 Weighted Naive Estimates of Impact of Credit by Gender on Log Body Mass Index (BMI) of Children Under Age 10 (OLS) Explanatory Variables Bos Girls Coef. t-ratio Coef. t-ratio Parents of HH head own land -0.020 -2.212 .025 2.902 Brothers of HH head own land -0.008 -1.776 -0.012 -2.809 Sisters of HH head own land .001 0.300 .006 1.187 Parents of HH head's spouse own land .011 1.292 -0.005 -0.645 Brothers of HH head's spouse own land -0.012 -2.277 -0.009 -2.302 Sisters of HH head's spouse own land -0.001 -0.289 -0.012 -2.804 Log household land -0.002 -0.655 .000 0.139 Highest grade completed by HH head .003 0.750 -0.008 -2.770 Sex of household head (1=male) .087 2.077 .032 1.101 Age of household head (years) -0.000 -0.126 .002 4.124 Highest grade completed by an adult .001 0.199 .005 1.915 female in HH _ Hi h t grade completed by an adult male .003 0.840 .004 1.520 No adult male in HH -0.029 -0.441 .064 1.482 No adult female in HH -0.214 -2.725 No spouse present in HH .076 1.578 -0.019 -0.677 Round 3 .053 1.148 -0.003 -0.086 Age in years -0.030 4.216 -0.034 -7.471 Age in years squared .002 3.222 .002 5.408 Amount borrowed by female brom BRAC -0.003 -1.602 -0.001 -0.299 Amount borrowed by male from BRAC .000 0.095 .007 1.390 Amount borrowed by female from BRDB -0.003 -0.496 -0.002 -0.317 Amount borrowed by male from BRDB .002 0.519 .000 0.091 Amount borrowed by female from GB .001 0.790 -0.000 -0.179 Amount borrowed by male from GB .001 0.651 .007 2.917 Participated but did not take credit -0.012 -0.515 -0.002 -0.078 Has any primary school -0.038 -2.125 .002 0.110 Has rural health center -0.017 -0.981 -0.047 -2.568 Has family planning center? -0.009 -0.304 -0.059 -2.212 Is Dai/Midwife available? .082 2.666 .034 1.331 Price of rice .015 1.610 -0.004 -0.543 89 Table A13 (continued) Weighted Naive Estimates of Impact of Credit by Gender on Log Body Mass Index (BMI) of Children Under Age 10 (OLS) Explanatory Variables _ _ s G irls Coef. t-ratio Coef. t-ratio Price of wheat flour -0.015 -1.251 -0.017 -1.674 Pnce of mustard oil .000 0.169 -0.001 -0.499 Price of hen egg .049 1.744 .026 1.063 Price of milk -0.002 -0.727 -0.003 -1.256 Price of potato -0.020 -2.743 -0.005 -0.942 Average female wage .001 0.868 .001 0.677 No female wage dummy .039 1.279 .005 0.189 Average male wage -0.001 -0.377 .004 2.312 Distance to Bank (kn) .010 2.144 .007 1.637 Constant -6.671 -45.063 -6.536 -55.356 Adiusted R2 0.140 0.316 No. of observations 378 409 Source: BIDS-World Bank household survey data, 1991-92. 90 Table A14 Unweighted and Weighted Naive Estimates of the Impact of Credit on Contraceptive Use of Currently Married Women Aged 15-49 Years (Pobir) Explanatory Variables Wei hted Unw ighted Coef. asymptotic Coef. asymptotic t-ratio t-ratio Parents of HH head own land -0.026 -0.369 .007 0.108 Brothers of HH head own land .008 0.200 .023 0.617 Sisters of HH head own land -0.063 -1.672 -0.041 -1.139 Parents of HH head's spouse own land -0.015 -0.265 .003 0.059 Brothers of HH head's spouse own land .002 0.069 -0.000 -0.009 Sisters of HH head's spouse own land .026 0.720 .008 0.249 Log household land -0.030 -1.323 -0.078 -3.615 Highest grade completed by HH head .023 0.856 -0.002 -0.095 Sex of household head (1 =male) 1.016 2.342 1.147 2.751 Age of household head (years) -0.007 -1.727 -0.007 1.536 Highest grade completed by an adult .023 1.270 .041 2.282 female in HH . Highest grade completed by an adult male .016 0.655 .026 1.088 in H No spounse present in HH .258 1.012 .448 1.540 Age in vears .319 10.790 .291 9.891 Age in years squared -0.005 -10.614 -0.004 -9.626 Amount borrowed by female brom BRAC .013 0.868 .019 1.218 Amount borrowed by male from BRAC .003 0.139 .012 0.573 Amount borrowed by female from BRDB -0.031 -1.390 -0.024 -1.385 Amount borrowed by male from BRDB .030 1.546 .023 1.690 Amount borrowed by female from GB .026 1.942 .035 2.884 Amount borrowed by male from GB -0.050 -2.310 -0.035 -2.020 Participate but no credit .138 0.860 -0.032 -0.217 Has any primary school .094 1.083 .044 0.501 Has rural health center .148 1.697 .141 1.567 Has family planning center? .156 1.077 .189 1.348 Is Dai/Midwife available? .162 1.791 .171 1.942 Price of rice .033 0.598 .065 1.186 Price of wheat flour -0.140 -3.019 -0.097 -2.076 Price of mustard oil -0.022 -3.272 -0.022 -3.246 Price of hen egg .043 2.276 .030 1.598 91 Table A14 (continued) Unweighted and Weighted Naive Estimates of the Impact of Credit on Contraceptive Use of Currently Married Women Aged 15-49 Years (Probit) Explanatory Variables Wei hted Unw ighted Coef. asymptotic Coef. asymptotic l ______________________________________ ___________ t-ratio t-ratio Price of milk -0.023 -1.349 -0.050 -2.945 Price of potato .034 1.313 .048 1.752 Average female wage -0.005 -0.629 .002 0.300 No female wage dummy -0.010 -0.055 .090 0.503 Average male wage -0.002 -0.394 -0.003 -0.538 Distance to Bank (Ian) .004 0.267 -0.024 -1.817 Constant -3.760 -4.124 -3.954 -4.266 Pseudo R2 0.113 0.113 No. of observations 1498 1498 Source: BIDS-World Bank household survey data, 1991-92. 92 Table A15 Unweighted and Weighted Naive Estimates of the Impact of Credit on Fertility of Currently Married Women Aged 15-49 (Probit) Explanatory Variables Wei hted Unw ighted Coef. aLsymptotic Coef. asymptotic t-ratio Coef. t-ratio Parents of HH head own land .235 3.330 .168 2.466 Brothers of HH head own land -0.042 -0.999 -0.017 -0.426 Sisters of HH head own land -0.053 -1.339 -0.035 -0.928 Parents of HH head's spouse own land .024 0.405 .008 0.133 Brothers of HH head's spouse own land -0.012 -0.343 -0.001 -0.019 Sisters of HH head's spouse own land -0.012 -0.308 -0.004 -0.108 Log household land .035 1.485 .054 2.355 Highest zrade completed by HH head -0.070 -2.386 -0.059 -2.030 Sex of household head -0.761 -2.181 -0.570 -1.584 Age of household head (years) -0.004 -0.847 -0.004 -0.934 Highest grade completed by an adult -0.034 -1.725 -0.022 -1.141 female in HHR ______ Highest grade completed by an adult male .037 1.407 .024 0.911 No spouse present in HH -0.639 -2.116 -0.516 -1.613 Age in years .192 5.089 .200 5.054 Age in vears squared -0.004 -6.207 -0.004 -6.214 Amount borrowed by female brom BRAC -0.029 -1.752 -0.014 -0.819 Amount borrowed by male from BRAC .001 0.053 -0.011 -0.538 Amount borrowed by female from BRDB .008 0.321 .004 0.200 Amount borrowed by male from BRDB .042 2.129 .031 2.193 Amount borrowed by female from GB -0.022 -1.558 -0.021 -1.587 Amount borrowed by male from GB -0.000 -0.018 -0.000 -0.025 Participate but no credit .066 0.375 -0.009 -0.058 Has any primary school -0.022 -0.243 -0.005 -0.049 Has rural health center -0.055 -0.581 .023 0.231 Has family planning center? -0.206 -1.302 -0.124 -0.825 Is Dai/Midwife available? .047 0.489 .022 0.239 Price of rice -0.110 -1.931 -0.089 -1.513 Price of wheat flour .108 2.224 .104 2.083 Price of mustard oil .020 2.718 .018 2.425 Price of hen egg -0.011 -0.545 -0.002 -0.115 93 Table A15 (continued) Unweighted and Weighted Naive Estimates of the Impact of Credit on Fertility of Currently Married Women Aged 15 49 (Probit) Explanatory Variables Weitbted Unweighted Coef. asymptotic Coef. asymptotic t-ratio t-ratio Price of milk .040 2.267 .034 1.940 Price of potato -0.046 -1.595 -0.059 -1.964 Average female wage . -0.001 -0.106 -0.014 -1.837 No female wage dummy .175 0.932 -0.175 -0.929 Average male wage .008 1.445 .008 1.447 Distance to Bank (lam) .020 1.422 .025 1.787 Constant -3.113 -3.287 -3.109 -3.111 Pseudo R2 0.150 0.148 No. of observations 1496 1496 Source: BIDS-World Bank household survey data, 1991-92. 94 APPENDIX B Table BI WESIML-LIML-FE Estimates of the Impact of Credit on Log Labor Supply by Gender Men Women Explanatory Variables Coef. asymptotic Coef. asympototic t-ratio t-ratio Parents of HH head own land .065 .895 .437 2.262 Brothers of HH head own land .008 .232 -0.189 -1.672 Sisters of HH head own land .108 2.797 -0.242 -2.104 Parents of HH head's spouse own land .114 1.672 .038 .238 Brothers of HH head's spouse own land -0.067 -1.922 .048 .514 Sisters of HH head's spouse own land -0.016 -0.399 .001 .045 Log HH land assets in decimal -0.009 -0.356 -0.025 -0.412 Highest grade completed by HH head -0.079 -2.770 .228 2.612 Sex of HH head (1 =male) -0.214 -0.352 -1.085 -1.629 Age of HH head (years) .000 .045 -0.002 -0.254 Highest grade completed by adult female -0.044 -2.056 -0.048 -0.623 inHH Highest grade completed by adult male in -0.053 -1.858 -0.302 -3.636 HH No adult male in HH 2.305 3.167 No adult female in HH -0.070 -0.272 No spouse present in HH -0.232 -1.006 -0.186 -0.430 Round 2 dummy -0.032 -0.678 -0.268 -1.810 Round 3 dmmy -0.101 -1.755 -0.075 -0.512 Age in years .119 6.684 .470 8.453 Age in years squared -0.002 -6.361 -0.006 -7.822 Highest grade completed .116 4.688 .057 .756 Amount borrowed by female from BRAC -0.212 -8.470 .234 2.302 Amount borrowed by male from BRAC -0.157 -3.688 .058 .492 Amount borrowed by female from BRDB -0.249 -9.149 .233 2.275 95 Table BI (continued) WESML-LIML-FE Estimates of the Impact of Credit on Log Labor Supply by Gender Men Women Explanatory Variables Coef. asymptotic Coef. asympototic t-ratio t-ratio Amount borrowed by male from BRDB -0.163 -3.521 .060 .562 Amount borrowed by female from GB -0.255 -9.619 .320 3.153 Amount borrowed by male from GB -0.167 -3.604 .027 .231 Participated but did not take credit -0.303 -2.515 .191 .435 Sigma 1.746 23.982 3.782 41.498 Rho (women) .697 10.608 -0.206 -1.850 Rho (men) .503 3.826 -0.034 -0.264 Log likelihood -13778.692 -12300.124 No. of observations 5846 5693 Source: BIDS-World Bank household survey data, 1991-92. 96 Table B2 WESML-LIML-FE Estimates of the Impact of Credit by Gender on Log Per Capita Expenditure Food Non-food Total Explanatory Variables Coef. asymptotic Coef. asymptotic Coef. asymptotic t-ratio t-ratio t-ratio Parents of HH head .016 1.074 .039 .991 .020 .991 own land Brothers of HH head .002 .227 -0.000 -0.016 -0.000 -0.002 own land Sisters of HH head -0.000 -0.039 .016 .802 .004 .349 own land Parents of HH head's .015 1.278 .048 1.624 .021 1.400 spouse own land Brothers of HH head's -0.006 -0.823 .008 .440 -0.000 -0.006 spouse own land Sisters of HH head's .005 .676 .003 .150 .002 .053 spouse own land Log HH land assets .005 1.026 .055 4.528 .015 2.431 in decimal Highest grade -0.002 -0.426 -0.024 -1.746 -0.007 -0.853 completed by HH head Sex of HH head .096 2.164 .070 .530 .110 1.856 (1=male) Age of HH head -0.002 -2.926 .-0.007 -3.642 -0.003 -3.657 (years) Highest grade .015 3.736 .065 5.996 .029 5.149 completed by adult female in HHR Highest grade .009 1.773 .060 4.528 .019 2.437 completed by adult male in HH No adult male in HH -0.020 -0.293 -0.176 -1.117 -0.014 -0.167 No adult female in .158 2.090 .132 .910 .159 2.038 HH No spouse present in .122 4.195 .188 2.483 .141 4.283 HH 97 Table B2 (continued) WESML-LIML-FE Estimates of the Impact of Credit by Gender on Log Per Capita Expenditure Food Non-food Total lExplanatory Variables Coef. asymptotic Coef. asymptotic Coef. asymptotic t-ratio t-ratio t-ratio Round 2 dummy -0.069 -6.284 .230 7.822 -0.021 -1.586 Round 3 dmmy -0.148 -13.266 -0.657 -19.182 -0.222 -17.302 Amount borrowed by .026 4.032 .019 .471 .038 3.702 female from BRAC I Amount borrowed by .012 1.343 .041 1.680 .018 1.615 male from BRAC Amount borrowed by .032 4.491 .017 .395 .041 3.620 female from BRDB Amount borrowed by .021 2.531 .050 2.228 .024 2.341 male from BRDB || Amount borrowed by .032 4.926 .022 .518 .044 3.899 female from GB Amount borrowed by .016 1.752 .029 1.231 .018 1.660 male from GB Participated but did .056 1.868 .015 .196 .059 1.714 not take credit Sigma .312 37.113 .820 52.067 .383 25.371 Rho (women) 0.409 4.917 -0.055 -0.224 -0.464 -3.940 Rho (men) -0.205 -1.705 -0.187 -1.505 -0.191 -1.633 Log likelihood -5090.877 -8712.608 -5784.156 No. of observations 4567 4567 4567 Souice: BIDS-World Bank household survey data, 1991-92. 98 Table B3 WESML-LIML-FE Estimates of the Impact of Credit by Gender on Log Non-land Assets Male Female Explanatory Variables Coef. asymptotic Coef. asymptotic t-ratio t-ratio Parents of HH head own land _ 121 .963 .361 1.346 Brothers of HH head own land .042 .723 .086 .613 Sisters of HH head own land .026 .408 .190 1.305 Parents of HH head's spouse own land -0.053 -0.498 .167 .688 Brothers of HH head's spouse own land .014 .227 -0.041 -0.315 Sisters of HH head's spouse own land .102 1.509 -0.251 -1.718 Log HH land assets in decimal .342 9.825 .055 .576 Highest grade completed by HH head -0.101 -2.311 -0.023 -0.209l Sex of HH head (1=male) 7.007 23.409 -6.823 -7.003 Age of HH head (years) -0.012 -2.389 -0.013 -1.113 Highest grade completed by adult female .049 1.782 .167 2.440 in HH Highest grade completed by adult male in .198 4.691 .159 1.516 HH No adult male in HH -0.518 -1.305 .549 .556 No adult female in HH H No spouse present in HH -0.624 -3.131 .375 .554 Amount borrowed by female from BRAC -0.007 -0.137 .070 .869 Amount borrowed by male from BRAC -0.156 -4.656 .328 1.733 Amount borrowed by female from BRDB .003 .042 .189 1.745 Amount borrowed by male from BRDB -0.169 -5.012 .332 1.906 Amount borrowed by female from GB .001 .023 .219 2.920 Amount borrowed by male from GB -0.218 -5.956 .240 1.293 Participated but did not take credit .037 .140 -0.116 -0.188 Sigma non-land assets 1.361 45.865 3.990 26.992 Rho (women) .023 .122 .027 .405 Rho (men) .830 28.686 -0.328 -1.745 Log likelihood -3245.862 -3403.751 No. of observations 1475 1517 Source: BIDS-World Bank household survey data, 1991-92. 99 Table B4 WESML-LIML-FE Estimates of the Impact of Credit by Gender on School Enrollment of Children Aged 5-17 Boys Girls Explanatory VariablesCef EpeCoef. asymptotic Coef. asymptotic t-ratio t-ratio Parents of HH head own land .225 2.172 .260 2.609 Brothers of HH head own land -0.038 -0.854 .078 1.513 Sisters of HH head own land -0.053 -1.147 -0.094 -1.802 Parents of HH head's spouse own land -0.061 -0.783 -0.056 -0.778 Brothers of HH head's spouse own land .045 1.114 .027 .633 Sisters of HR head's spouse own land .042 .849 .019 .410 Log HH land assets in decimal .041 1.400 .056 1.942 Highest grade completed by HH head .042 1.311 .043 1.551 Sex of HH head (1=male) .616 1.812 -0.005 -0.032 Age of HH head (years) -0.015 -2.805 -0.011 -2.419 Hihest grade completed by adult female .054 2.260 .008 .322 HiWet grade completed by adult male .044 1.439 .073 3.013 No adult male in HH .453 1.267 -0.099 -0.279 No adult female in HH .137 .222 -0.658 -1.344 No spouse present in HH -0.122 -0.513 -0.536 .260 Age in years .669 9.084 .752 9.194 Age in years squared -0.032 -9.539 -0.034 -9.058 Amount borrowed by female from BRAC .024 .449 -0.049 1.401 Amount borrowed by male from BRAC .010 .236 .023 .598 Amount borrowed by female from BRDB .102 1.814 .071 1.583 Amount borrowed by male from BRDB .047 1.100 .011 .272 Amount borrowed by female from GB .091 1.738 .085 2.554 Amount borrowed by male from GB .068 1.557 .031 .797 Participated but did not take credit . .257 1.375 .249 1.349 Rho (women) -0.141 -0.578 -0.153 -1.036 Rho (men) -0.060 -0.344 -0.011 -0.083 Log likelihood -3542.159 -3497.840 No. of observations 1341 1269 Source: BIDS-World Bank household survey data, 1991-92. 100 Table B5 WESML-LIML-FE Estimates of the Impact of Credit by Gender on Log Height of Children Aged 0-14 years B ys Girls Explanatory Variables Coef. asymptotic Coef. asymptotic l_____________ . t-ratio _ t-ratio Parents of HH head own land .014 1.526 .010 .939 Brothers of HH head own land .008 1.857 -0.011 -1.990 Sisters of HH head own land -0.002 -0.282 .007 1.185 Parents of HH head's spouse own land -0.030 -2.618 -0.012 -1.153 Brothers of HH head's spouse own land .004 .725 .002 .417 Sisters of HH head's spouse own land .007 1.547 .004 .742 Log HH land assets in decimal .002 .531 .002 .416 Highest grade completed by HH head -0.008 -2.007 -0.003 -.045 Sex of HH head (1=male) -0.007 -0.075 -0.013 -0.467 Age of HH head (years) -0.001 -1.580 .000 .579 Highest grade completed by adult female -0.002 -0.612 .005 1.526 in HH Highest grade completed by adult male in .009 2.541 .006 2.034 HH No adult male in HI .108 2.347 .024 .617 No adult female in HH -0:.073 -0.811 -0.038 -0.855 No spouse present in HH -0.047 -0.562 .016 .451 Round 3 dummy .021 3.877 .023 4.587 Age in years .112 15.199 .107 18.661 Age in years squared -0.006 -7.184 -0.005 -10.079 Amount borrowed by female from BRAC -0.004 -0.670 -0.007 -1.800 Amount borrowed by male from BRAC .007 2.223 -0.001 -0.192 Amount borrowed by female from BRDB -0.006 -0.818 -0.012 -1.793 Amount borrowed by male from BRDB -0.000 -0.071 -0.004 -0.704 Amount borrowed by female from GB -0.011 -1.854 -0.010 -2.282 Amount borrowed by male from GB -0.002 -0.485 .001 .439 Participated but did not credit .061 3.654 .023 .940 Sigma .073 7.885 .083 10.914 Rho (women) .478 1.546 .634 3.903 Rho (men) -0.042 -0.357 -0.092 -0.790 Log likelihood -2930.308 -2859.767 No. of observations 378 409 Source: BIDS-World Bank household survey data, 1991-92. 101 Table B6 WESML-LIML-FE Estimates of the Impact of Credit by Gender on Log Weight of Children Aged 0-14 Years Bo s Girls Explanatory Variables Bv l ____il Coef. asymptotic Coef. asymptotic t-ratio t-ratio Parents of HH head own land .011 .479 .047 1.769 Brothers of HH head owD land .012 1.186 -0.033 -2.433 Sisters of HH head own land -0.001 -0.055 .027 1.779 Parents of HH head's spouse own land -0.051 -1.962 -0.025 -0.994 Brothers of HH head's spouse own land -0.003 -0.211 -0.007 -0.649 Sisters of HH head's spouse own land .009 .813 -0.004 -0.312 Log HH land assets in decimal .002 .233 .007 .655 Highest grade completed by HH head -0.014 -1.790 -0.014 -2.090 Sex of HH head ( =male) .041 .204 .003 .048 Age of HH head (years) -0.001 -0.874 .004 2.215 Hi%hest grade completed by adult female -0.003 -0.294 .012 1.768 Hihest grade completed by adult male .018 2.293 .014 2.5133 No adult male in HH .130 1.052 .121 1.386 No adult female in HH -0.418 -2.477 -0.325 -3.532 No spouse present in HH -0.003 -0.019 .014 .209 Round 3 dummy .019 1.227 .039 3.184 Age in years .198 10.515 .184 13.127 Age in years squared -0.009 4.781 -0.008 -6.287 Amount borrowed by female from BRAC -0.017 -1.308 -0.015 -1.641 Amount borrowed by male from BRAC .018 1.756 .002 .177 Amount borrowed by female from BRDB -0.019 -1.030 -0.030 -1.623 Amount borrowed by male from BRDB -0.002 -0.199 -0.009 -0.820 Amount borrowed by female from GB -0.027 -1.981 -0.022 -1.987 Amount borrowed by male from GB -0.001 -0.090 .009 1.246 Participated but did not take credit .125 3.234 .060 1.140 Sigma .179 6.881 .189 9.038 Rho (women) .560 2.115 .655 3.628 Rho (men) -0.033 -0.169 -0.049 -0.592 Log likelihood -3159.857 -3137.600 No. of observations 378 409 Source: BIDS-World Bank household survey data, 1991-92. 102 Table B7 WESML-LIML-FE Estimates of the Impact of Credit by Gender on Body Mass Index (BMI) of Children of Age Less than 10 Explanatory Variables _ I Coef. asymptotic Coef. asymptotic Parents of HH head own land -0016 -1 325 .028 2.299 rothers of H head wn land 002 -0325 . 014 -2.39 Sisters of HH head own land 002 .344 .013 1.534 Parents of HH head's spouse own land .006 .568 -0.001 -0.083 Brothers of HH head's spouse own land . O 09 -1.674 4.Q10 -2.09 Sisters of HH head's spouse own land 0 004 -0 595 0 013 -2.174 L-og HH land assets in decimal -,001 -0250 .002 .5 Hiehest -rade completed by HH head .oo .044 -0.00.7 -2.044 Sex of HH head (I =male) .06i0 1.085 .051 1.787 Age of HH head (vears) Q .127 _ 003 3.328 Highest grade completed by adult female .001 .333 .004 1.221 in -i H H§hest grade completed by adult male in .002 .499 .002 .600 H. No adult male in HH 40095 -1.558 086 2.690 No adult female in HH _,I288 -3.968 -0.238 -4.756 No spouse present in HH .092 _1890 -0.014 -0.619 Round 3 dumrmy 40024 -2.435 -0.007 -0 720 Aze in years _0,025 -2.846 -0 028 -3.972 Age in vears sauared .002 2.797 .002 3.094 Amount borrowed by female from BRAC -0,013 -1-248 .004 1.365 Amount borrowed by male from BRAC O O05 O0536 .006 1.070 Amount borrowed by female from BRDnR 0.011 0827 .002 .244 Amount borrowed by male from BRDB -0.01 -1,017 .001 .145 Amount borrowed by female from GB -0 009 -0 797 .005 1.822 Amount borrowed by male from GB -.6 -0.623 2.2 Participated but did not take credit .007 .254 .011 .513 SigDma .097 3.900 .088 20.395 Rhg (woMen 482 1156 -0.165 -1 41 Rho (men) .394 1-146 -0051 0534 Log likelihood -2998 448 -2921.321 No. of observations 378 409 Source: BIDS-World Bank household survey data, 1991-92. 103 Table B8 WESML-LIML-FE Estimates of the Impact of Credit on Contraceptive Use by and Fertility of Currently Married Women Aged 15-49 years Contrace tive Use Recent Fertility E:xplanatory Variables Coef. asymptotic Coef. asymptotic t-ratio t-ratio Parents of HH head own land -0.002 -0.019 .258 3.019 Brothers of HH head own land .016 .363 -0.080 -1.719 Sisters of HH head own land -0.032 -0.714 4.039 -0.818 Parents of HH head's spouse own -0.019 -0.267 .029 .413 land ___________ ___________ Brothers of HH head's spouse own -0.003 -0.073 -0.029 -0.707 land__ _ _ _ _ Sisters of HH head's spouse own land .001 .016 -0.032 -0.672 Log HH land assets in decimal -0.056 -1.508 .049 1.538 Highest grade completed by HH head .019 .580 -0.062 -1.767 Sex of HH head (I=male) .912 1.931 -1.043 -2.671 Age of HH head (years) -0.002 -0.447 -0.003 -0.579 Highest grade completed by adult .025 1.135 -0.052 -2.024 female in HH _ ___ Highest grade completed by adult .021 .685 .030 .901 mae in HH No spouse present in HH .307 .903 -0.666 -1.902 Age in Years .344 6.448 .214 3.445 Age in years squared -0.005 -6.420 -0.004 -4.176 Amount borrowed by female from -0.023 -0.444 -0.042 -0.732 BRAG Amount borrowed by male from .092 1.831 -0.043 -0.689 BRAC ________ Amount borrowed by female from -0.086 -1.549 -0.046 -0.781 BRDB__ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Amount borrowed by male from .146 3.002 .040 .665 BRDB . - Amount borrowed by female from -0.051 -0.933 .051 .884 GB . - Amount borrowed by male from GB .040 .687 .033 .533 Participated but did not take credit .288 1.421 .040 .191 Rho (women) .226 .932 .097 .364 Rho (men) -0.464 -1.910 -0.082 -0.323 Log likelihood -2181.475 -2140.187 No. of observations 1498 1496 Source: BIDS-World Bank household survey data, 1991-92. 104 APPENDIX C Table C: Wald Test (X2) Statistics a Joint Significance ofb-c Outcome Variables Credit Female Male Transfer Equality of variables credit credit variables gender credit (6) variables variables (6) variables (3) (3) (3) Girl's schooling 4.11 2.16 2.34 7.62 1.64 Boy's schooling 20.10 15.18 5.54 10.00 3.03 Women's labor supply 1.39 0.44 0.79 15.84 1.00 Men's labor supply 98.66 53.11 7.65 23.27 2.26 Per capita food 7.97 3.41 4.37 10.40 1.06 expenditure . - Per capita non-food 5.05 0.81 4.08 14.23 2.31 expenditure Per capita total 22.69 19.03 4.11 13.16 3.39 expenditure Contraception 16.90 6.15 8.58 4.53 12.42 Fertility 13.87 8.36 8.17 14.20 9.20 Women's non-land 4.36 2.42 1.91 2.55 2.95 assets Girls BMI 9.82 4.14 5.98 26.63 0.92 Boys BMI 4.17 3.32 1.76 6.88 1.77 Girls' Height 9.35 5.78 1.28 6.92 5.50 Boys' Height 14.00 7.89 9.78 17.05 2.54 Girls' Weight 9.34 4.12 2.94 9.77 6.64 Boys' Height 10.39 4.88 7.89 9.06 4.70 *Based on WESML-LIML-FE estimates. bdegress of freedom in parenthesis 'critical values are: X 2(3).1 = 6.25 X2(6).1o = 10.64 X2(3).,5 = 7.82 X2(6).o5 = 12.59 X2(3).o1 = 11.34 X2(6).o, = 16.81 105 REFERENCES Adams, Dale W., Douglas H. Graham, and J. D. Von Pischke. 1984. Undermining Rural Development with Cheap Credit. Boulder, Co., Westview Press. Amin, R., M. Kabir, J. Chowdhury, A. U. Ahmed and R.B. Hill. 1994. "Impact of Poor Women's Participation in Credit-based Self-employment on Their Empowerment, Fertility and Contraceptive Use, and Fertility Desire in Rural Bangladesh." (Mimeo). BIDS. 1990. "Evaluation of Poverty Evaluation Programmes." Various volumes (Draft). Dhaka: Bangladesh Institute of Development Studies. Binswanger, Hans and Mark Rosenzweig. 1986. "The Behavioural and Material Determinants of Production Relations in Agriculture," The Journal of Development Studies 32, 503-539. Coslett, S. R. 1981. "Maximum Likelihood Estimation for Choice-Based Samples," Econometrica 49:1289-1316. Feder, Gershon. 1988. Land Policies and Farm Productivity in Thailand, John Hopkins University Press. Baltimore, MD. Gersovitz, Mark, "Savinas and Nutrition at Low Incomes, " Journal of Political Economy, October 1983, 841-855. Hoff, Karla and Joseph E. Stiglitz. 1990. "Introduction: Imperfect Information and Rural Credit Markets - Puzzles and Policy Perspectives." The World Bank Economic Review, vol. 4 no. 3:235-251. Heckman, James J. 1981. "The Incidental Parameters Problem and the Problem of Initial Conditions in Estimating a Discrete Time-Discrete Data Stochastic Process," in C. F. Manski and D. MdFadden, eds., Structural Analysis of Discrete Data with Econometric Applications, Cambridge, Mass: MIT Press, 178-195. Heckman, James J. 1976. "The Common Structure of Models with Continuous and Discrete Endogenous Variables and a Simple Estimator for Such Models," Annuals of Economic and Social Measurement 5:475-492. Hossain, Mahabub. 1988. Credit for Alleviation of Rural Poverty: The Grameen Bank in Bangladesh. Research Report 65. International Food Policy Research Institute, Washington, DC. Khandker, Shahidur R., Zahed Khan and Baqui Khalily. 1994b. "Sustainability of a Government Targeted Credit Program: The BRDB RD-12 Project in Bangladesh." Mimeo, The World Bank, Washington, DC. Khandker, Shahidur R., Baqui Khalily and Zahed Khan. 1994a. "Sustainability of Grameen Bank: What Do We Know?" Mimeo, The World Bank, Washington, DC. 107 Khandker, Shahidur R. and Baqui Khalily. 1994. "Designing A Sustainable Poverty Alleviation Program: The BRAC Strategy in Bangladesh." Mimeo, The World Bank, Washington, DC. Lancaster, T., 1992. "The Theory of Choice-Based Sampling: A Review," manuscript, Brown University, Department of Economics, March 1992. Lancaster, T. and G. Imbens, 1991. "Choice-Based Sampling - Inference and Optimality," Brown University, Department of Economics Working Paper No. 91-17. Lee, L. F., 1976. "Estimation of Limited Dependent Variable Model by Two-Stage Methods," Ph.D. Dissertation, University of Rochester. Maddala, G. S., 1983. Limited Dependent and Qualitative Variables in Econometrics, New York: Cambridge University Press. Manser, M. and M. Brown, 1980. "Marriage and Household Decision-making: A Bargaining Analysis," International Economic Review, 21, 31-44. McElroy, Marjorie. 1990. "The Empirical Content of Nash-Bargained Household Behavior". The Journal of Human Resources. vol. 25 no. 4:559-583. McElroy, Marjorie and Mary Jean Horney. 1981. "Nash-Bargained Household Decisions: Towards a Generalization of the Theory of Demand." International Economic Review, 22, June. pp.333-49. Moffitt, R., 1991. "Program Evaluation with Nonexperimental Data," Evaluation Review 15, 291-314. Pitt, Mark M., Shahidur R. Khandker, Signe-Mary McKernan and M. A. Latif. 1995. "Credit Programs for the Poor and Reproductive Behavior in Low Income Countries: Are the Reported Causal Relationships the Result of Heterogeneity Bias?" Prepared for presentation at the annual meeting of the Population Association of America, San Francisco, April 1995. Pitt, Mark, Mark Rosenzweig and D. M. Gibbons, 1993. "The Determinants and Consequences of the Placement of Government Programs in Indonesia," World Bank Economic Review, September, 319-348. Rashid, Mansoora and Robert M. Townsend. 1994. "Targeting Credit and Insurance: Efficiency, Mechanism Design and Program Evaluation," ESP Discussion Paper No. 47, The World Bank, Washington, DC. Rivers, Douglas and Quang Vuong. 1988. "Limited Information Estimators and Exogeneity Tests for Simultaneous Probit Models," Journal of Econometrics 39, pp.347-366. Rosenzweig, Mark. 1980. "Neoclassical Theory and the Optimizing Peasant: An Econometric Analysis of Family Farm Labor Supply in a Developing Country," Ouarterly Journal of Economics XCIV, pp.31-55. 108 Rosenzweig, Mark and Kenneth Wolpin. 1985. "Specific Experience, Household Structure and Intergenerational Transfers: Farm Family Land and Labor Arrangements in Developing Countries," Ouarterly Journal of Economics, 100:961-987. Smith, R. J. and R. W. Blundell, 1986. "An Exogeneity Test for a Simultaneous Equation Tobit Model with an Application to Labor Supply," Econometrica 54, pp.679-685. Stiglitz, Joseph E. 1990. "Peer Monitoring and Credit Markets". The World Bank Economic Review. vol. 4 no. 3:351-366. Varian, Hal R. 1990. "Monitoring Agents with Other Agents". Journal of Institutional and Theoretical Economics. 146:153-74. Von Pischke, J. D. 1991. Finance at the Frontier: Debt Capacitv and the Role of Credit in the Private Economy. EDI Development Studies. The World Bank, Washington, DC. Wahid, Abu (ed.) 1993. The Grameen Bank: Poverty Relief in Baneladesh. Boulder, CO, Westview Press. White, H., 1980. "A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity," Econometrica 48, 817-838. World Bank. 1975. Agricultural Credit: Sector Policy Paper. Washington, D.C. World Bank. 1989. Bangladesh: Strategies for Enhancing the Role of Women in Economic Development. A World Bank Country Study, Washington, DC. Yaron, Jacob. 1992. Successful Rural Finance Institutions. World Bank Discussion Paper 150. Washington, DC. 109 CHfIA FRANCE ISRAEL NEW ZEALAND RUSSIAN FEDERATION Wanernrerp-tlianniAS Distributors O Chna Fmana & Eonoiric Wald Bank Piicatons Yozmnd Librare Ltd. EBSCO NZ Ltd. Isdaattslalo