LSM- 92 NOV.1992 Living Standards Measurement Study Working Paper No. 92 Economic Aspects of Child Fostering in Cote d'Ivoire LSMS Working Papers No. 20 Statistical Experimentation for Household Surveys: Two Case Studies of Hong Kong No. 21 The Collection of Price Data for the Measurement of Living Standards No. 22 Household Expenditure Surveys: Some Methodological Issues No. 23 Collecting Panel Data in Developing Countries: Does It Make Sense? No. 24 Measuring and Analyzing Levels of Living in Developing Countries: An Annotated Questionnaire No. 25 The Demand for Urban Housing in the Ivory Coast No. 26 The C6te d'Ivoire Living Standards Survey: Design and Implementation No. 27 The Role of Employment and Earnings in Analyzing Levels of Living: A General Methodology with Applications to Malaysia and Thailand No. 28 Analysis of Household Expenditures No. 29 The Distribution of Welfare in C6te d'Ivoire in 1985 No. 30 Quality, Quantity, and Spatial Variation of Price: Estimating Price Elasticities from Cross-Sectional Data No.31 Financing the Health Sector in Peru No. 32 Informal Sector, Labor Markets, and Returns to Education in Peru No. 33 Wage Determinants in C6te d'Ivoire No.34 Guidelines for Adapting the LSMS Living Standards Questionnaires to Local Conditions No. 35 The Demand for Medical Care in Developing Countries: Quantity Rationing in Rural C6te d'Ivoire No. 36 Labor Market Activity in C6te d'Ivoire and Peru No.37 Health Care Financing and the Demand for Medical Care No.38 Wage Determinants and School Attainment among Men in Peru No. 39 The Allocation of Goods within the Household: Adults, Children, and Gender No. 40 The Effects of Household and Community Characteristics on the Nutrition of Preschool Children: Evidence from Rural C6te d'Ivoire No. 41 Public-Private Sector Wage Differentials in Peru, 1985-86 No. 42 The Distribution of Welfare in Peru in 1985-86 No. 43 Profits from Self-Employment: A Case Study of C6te d'Ivoire No. 44 The Living Standards Survey and Price Policy Reform: A Study of Cocoa and Coffee Production in C6te d'Ivoire No. 45 Measuring the Willingness to Pay for Social Services in Developing Countries No. 46 Nonagricultural Family Enterprises in C6te d'Ivoire: A Descriptive Analysis No. 47 The Poor during Adjustment: A Case Study of C6te d'Ivoire No. 48 Confronting Poverty in Developing Countries: Definitions, Information, and Policies No. 49 Sample Designs for the Living Standards Surveys in Ghana and Mauritania/Plans de sondage pour les enqutes sur le niveau de vie au Ghana et en Mauritanie No. 50 Food Subsidies: A Case Study of Price Reform in Morocco (also in French, 50F) No. 51 Child Anthropometry in C6te d'Ivoire: Estimates from Two Surveys, 1985 and 1986 No. 52 Public-Private Sector Wage Comparisons and Moonlighting in Developing Countries: Evidence from C6te d'Ivoire and Peru No. 53 Socioeconomic Determinants of Fertility in C6te d'Ivoire No. 54 The Willingness to Pay for Education in Developing Countries: Evidence from Rural Peru (List continues on the inside back cover) Economic Aspects of Child Fostering in C6te d'Ivoire The Living Standards Measurement Study The Living Standards Measurement Study (LSMs) was established by the World Bank in 1980 to explore ways of improving the type and quality of house- hold data collected by statistical offices in developing countries. Its goal is to foster increased use of household data as a basis for policy decisionmaking. Specifically, the ISus is working to develop new methods to monitor progress in raising levels of living, to identify the consequences for households of past and proposed gov- ernment policies, and to improve communications between survey statisticians, an- alysts, and policymakers. The Laus Working Paper series was started to disseminate intermediate prod- ucts from the Lsus. Publications in the series include critical surveys covering dif- ferent aspects of the Lss data collection program and reports on improved methodologies for using Living Standards Survey (Wss) data. More recent publica- tions recommend specific survey, questionnaire, and data processing designs, and demonstrate the breadth of policy analysis that can be carried out using Lss data. LSMS Working Paper Number 92 Economic Aspects of Child Fostering in Cote d'Ivoire Martha Ainsworth The World Bank Washington, D.C. Copyright 0 1992 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 November 1992 To present the results of the Living Standards Measurement Study 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. 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 consequence of their use. Any maps that accompany the text have been prepared solely for the convenience of readers; the designations and presentation of material in them do not imply the expression of any opinion whatsoever on the part of the World Bank, its affiliates, or its Board or member countries concerning the legal status of any country, territory, city, or area or of the authorities thereof or concerning the delimitation of its boundaries or its national affiliation. 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 reproduction is for noncommercial purposes, without asking a fee. Permission to copy portions for classroom use is granted through the Copyright Clearance Center, 27 Congress Street, Salem, Massachusetts 01970, 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, Department F, The World Bank, 1818 H Street, N.W., Washington, D.C. 20433, U.S.A., or from Publications, The World Bank, 66, avenue d'I&6na, 75116 Paris, France. ISSN: 0253-4517 Martha Ainsworth is an economist in the Population, Health, and Nutrition Division of the Africa Technical Department at the World Bank. Library of Congress Cataloging-in-Publication Data Ainsworth, Martha, 1955- Economic aspects of child fostering in COte d'Ivoire / Martha Ainsworth. p. cm. - (LSMS working paper, ISSN 0253-4517 ; no. 92) Includes bibliographical references. ISBN 0-8213-2280-X 1. Foster home care-Economic aspects-Ivory Coast. 2. Children- Employment-Ivory Coast. I. Title. II. Series. HV887.19A56 1992 338.4'7362733'096668-dc2O 92-33538 CIP - ABSTRACT Throughout West Africa, it is a common and socially accepted practice for parents to send their children to live with other households. Several economic motives have been suggested for these "fostering" arrangements: the demand for child labor; investments in human capital; child care; and income insurance. Child fostering, like fertility, is a measure of the demand for children. Evidence of economic factors underlying fostering decisions would demonstrate that the demand for children in Africa is potentially susceptible to policy intervention. This paper examines the economic determinants of child fostering decisions in Cote d'lvoire, where, in 1985, one fifth of non-orphaned children age 7-14 were living away from both natural parents. The economic determinants of both sending and receiving decisions are examined separately and evaluated with respect to their support for child labor and human capital explanations. The determinants for both sides of the fostering market are then estimated simultaneously, using a model of friction developed by Rosett (1959), so that the symmetry of fostering determinants can be formally tested. The results indicate that important economic factors affect fostering decisions on both sides of the fostering market; however, different factors explain sending and receiving decisions. The economic determinants also vary by the sex of the child and whether the household is in an urban or rural area. The findings are consistent with a child labor explanation and inconclusive with respect to schooling investments as a motive. Although the number of children fostered out increases with family size, the paper finds no evidence that children are fostered out because parents cannot afford them. The symmetry of fostering decisions could not be rejected; nevertheless, in all regressions the significant determinants of the sending and receiving decisions are different. -Vi - ACKNOWLEDGMENTS Comments from T. Paul Schultz, John Strauss, Robert Evenson, T.N. Srinivasan, Mark Rosenzweig and participants in Yale and World Bank Seminars are gratefully acknowledged. I am indebted to Dorothy Seavey for providing programming of the Rosett model.  Vii - FOREWORD Population is growing more rapidly in Sub-Saharan Africa than in any other developing region, the result of continued high fertility and substantial mortality decline since the second World War. Although many African couples would have fewer children with better access to family planning methods, most analysts agree that the "demand* for children -- as reflected in actual fertility and statements about desired family size - remains high over much of the continent. This paper analyzes the economic factors underlying the widespread practice of child fostering in C8te d'Ivoire - a practice often subjected to sociological analysis but not previously analyzed by economists. The paper provides convincing evidence that: (a) child fostering can be partly explained by the demand for child labor; and (b) the economic benefits of children remain strong in C6te d'Ivoire, and must be taken into consideration in any strategy to reduce fertility and rapid population growth. This paper is part of a broader program of research in the Population and Human Resources (PHR) Department on the extent of poverty in developing countries and on policies to reduce poverty. This research program is located in the Poverty Analysis and Policy Division. Martha Ainsworth is a human resources economist in the Africa Technical Department who worked extensively in the design of the Living Standards Measurement Study (LSMS) survey methodology. This paper, which uses LSMS data from CMte d'Ivoire, is an extract from her doctoral dissertation. The extensive linking between individuals within households and between households and extended families in the LSMS surveys is quite a rarity among household surveys. This linking makes the LSMS data sets a rare and valuable source of information for empirical analysis of child fostering. Ann 0. Hamilton Director Population and Human Resources Department 4 - lx - TABLE OF CONTENTS 1. Introduction............................................. . . . . . . . . . . 1 2. Economic Correlates of Child Fostering ....... ............................... 4 3. A Model of the Determinants of Child Fostering ............................... 12 4. The Economic Determinants of Sending and Receiving Decisions ..................... 23 5. Testing the Symmetry of Fostering Decisions ................................. 32 6. Conclusions ...................................................... 36 Appendix . ...................................................... 39 References ....................................................... 41 Fieure 1 Percent of children living away from parents, by age and sex of child .................... 5 2A Percent of girls who worked in the last 7 days................................... 7 2B Percent of boys who worked in the last 7 days................................... 7 3A Percent of children in Abidjan who worked in the last 7 days ........................ 8 3B Percent of children in urban areas who worked in the last 7 days ....................... 8 3C Percent of children in rural areas who worked in the last 7 days ........................ 8 4 Percent of children enrolled in the last 12 months ............................... 10 5 Enrollment of foster children 7-14 and their siblings ............................. 11 6 The Rosett model of friction with differential slopes and intercepts ...................... 34 Tables 1 Description of variables............................................... 17 2 Variable means and standard deviations ..................................... 20 3 Tobit estimates, fostering of girls and boys ................................... 24 4 Variable means and standard deviations, urban and rural samples ..................... 27 5 Tobit estimates, fostering by rural households ................................. 29 6 Tobit estimates, fostering by urban households . ................................ 31 7 Wald tests for equality of intercepts and coefficients on opposite sides of the fostering market ...................................... 35 Al Rosett model estimates, fostering of children 7-14 .............................. 39 A2 Rosett model estimates, fostering of boys and girls 7-14 ............................ 40 1. INTRODUCTION Throughout West Africa, it is a common and socially accepted practice for parents to send their children to live with other households. Anthropological studies in C8te d'Ivoire, Ghana, and Sierra Leone have described these "fostering" arrangements (Antoine and Guillaume, 1986; Bledsoe, 1989; Bledsoe and Isiugo- Abanihe, 1989; Brydon, 1979, 1985; Etienne, 1979a, 1979b; Fiawoo, 1978; Goody, 1982; Issac and Conrad, 1982; Saint-Vil, 1981; Schildkrout, 1973). In certain instances, fostering arises out of family crises, such as the death of one or both parents, parental separation, illegitimacy, or infertility of a relative. In these instances, parental obligations may be permanently transferred to the foster parents, much like an adoption. Far more prevalent, however, is the practice of voluntarily sending children on a temporary basis to live in another household; the foster child retains all obligations to his or her natural parents and remains a member of the parents' lineage. A child may be fostered for short or long periods, more than once, and to different households. Children of all ages are fostered and most are sent to the households of relatives. In some African societies fosterage has been practiced traditionally, while in others it is a more recent phenomenon arising from economic development. This paper analyzes the economic determinants of child fostering in C6te d'Ivoire, where, in 1985, one fifth of non-orphaned children age 7-14 were living away from both parents. Fostering, like fertility, is a measure of the demand for children. While fertility decisions are irreversible (short of infanticide or abortion), fostering is a temporary and reversible way of adjusting family size post-natally. If couples make fertility decisions based on the costs and benefits of children, the price of children relative to other investment and consumption goods, and a budget constraint, then a similar calculation should apply to fostering decisions. Evidence of economic determinants of fostering would demonstrate that the demand for children in Africa is potentially susceptible to policy intervention.' Several economic motives for child fostering have been suggested: child labor; investment in human capital; child care; and insurance.2 1 Fostering also has demographic implications (Isiugo-Abanihe, 1985; Page, 1986). Households with low fertility may have large family size due to the presence of foster children. Fostering spreads the costs of childbearing over the extended family, minimizing the costs of bearing children to the parents. It has been argued that fostering is an institution for "managing" high fertility in Africa and that it divorces child costs from fertility decisions (Isiugo-Abanihe, 1983, 1985; Bledsoe, 1989; Bledsoe and Isiugo-Abanihe, 1989; Frank, 1984; Oppong and Bleek, 1982). Parental characteristics may have little effect on the nutrition, morbidity, or mortality outcomes for children, if children are living away from their parents; the relevant variable might be characteristics of the receiving household. Finally, widespread fostering means that there is substantial migration of children in West Africa, independent of migration of parents. 2 The predominance of fostering to the households of relatives can be explained in terms of minimizing transactions costs for both the sending and receiving households, ensuring that the child will not be mistreated, and minimizing supervision costs of the receiving household (Ben-Porath, 1986; Pollack, 1985). -2- * The Demand for Child Labor. The role of foster children in home production is often cited as a motivation for receiving older foster children, particularly girls (Antoine and Guillaume, 1986; Bledsoe and Isiugo-Abanihe, 1989; Etienne, 1979b; Fiawoo, 1978; Goody, 1982; Issac and Conrad, 1982; Oppong and Bleek, 1982; Schildkrout, 1973).' Foster children help out with chores in the host household in return for all or part of their upkeep, including school expenses. Foster girls perform domestic work, substituting for the time inputs of working mothers and of the host household's own children who may be enrolled in school. A life-cycle pattern of fostering for child labor has also been suggested; children may be fostered out of households with a temporary surplus of child labor at mid-life and into households with child labor deficits at either extreme of the life cycle (Frank, 1984; Isiugo-Abanihe, 1985; Page, 1986). * Investments in Human Capital. Children are often fostered out to learn a trade or to attend school (Brydon, 1979; Chernichovsky, 1985b; Goody, 1982; Gould, 1985; Saint-Vil, 1981). School-related fostering is prevalent among children of secondary-school age, since middle and secondary schools are 6ften located in urban areas. It may also occur at the primary level if local primary schools are inaccessible, crowded, or of poor quality. The costs of schooling and upkeep may be paid entirely by the sending household, entirely by the host household, or shared (Goody, 1982; Brydon, 1979; Schildkrout, 1973). Other types of child investments associated with fostering are apprenticeships and "alliance" fostering, in which a child is fostered to a household of higher social standing in hope of eventual social advancement (Goody, 1982). * Child Care. There is a growing tendency for urban working mothers to foster out their young children (under 6) to grandmothers and other relatives in rural areas for child care (Bledsoe and Isiugo- Abanihe, 1989; Brydon, 1979; Fiawoo, 1978). Children can be raised more cheaply in villages and their urban mothers can continue working. These arrangements generally involve some type of compensation to the caretaker in the host household.s The value of child labor in home production is thought to be an important reason for the high demand for children in many parts of the developing world (Chemichovsky, 1985; Levy, 1985; Rosenzweig and Evenson, 1977). Numerous studies have documented the contribution of children to home production (Cain, 1977; Mueller, 1976,1985; White, 1975). Children as young as 7 years of age fetch drinking water and firewood, shop, clean, mind younger siblings, and tend livestock, freeing older children and adults to engage in more productive activities. Beginning at age 11-13, children often participate directly in agricultural production. DaVanzo (1972), Levy (1985), and Rosenweig and Evenson (1977) found positive relationships between the child wage, child labor force participation, and fertility in developing countries. ' Gould (1985) notes that in East Africa children do not necessarily attend the nearest school; they try to enroll in the schools with the best pass rate. Antoine and Guillaume (1986) note that parents in Abidjan often send their children to schools in the interior because they are more "developed." ' In Sierra Leone, grandmothers actually compete for the care of their infant grandchildren to insure a flow of resources from the parents. "Parents know that one of the main reasons why grannies lobby for children is that they will soon begin asking the children's parents for things they would ask for anyway-food, clothes, and money-but would ordinarily have little leverage to demand" (Bledsoe and Isiugo-Abanihe, 1989, p.21). .3- * Income Insurance. Bledsoe and Gage (1987), Bledsoe and Isiugo-Abanihe (1989), and Frank (1984) have suggested that foster children are received to augment their foster parents' potential sources of support in old age.' Fostering may also enforce implicit insurance contracts between urban and rural households: rural farm households may accept an urban child to have a claim on resources from the sending household in bad times, while urban households may accept a rural child to obtain transfers from farm households in times of high unemployment (Bledsoe and Isiugo-Abanihe, 1989; Brydon, 1979; Etienne, 1979b; J. Goody, 1987). Finally, fostering may represent an ex post attempt to smooth consumption for remaining household members in response to exogenous income shocks. This paper examines the economic determinants of fostering children age 7-14 using data from the 1985 C6te d'Ivoire Living Standards Survey (CILSS), a national sample of 1599 African households.' Foster children are defined as non-orphaned children living away from both biological parents. The paper will focus on evidence that fostering is due to demand for child labor and, to a more limited extent, to schooling investments. Insurance explanations cannot be explored with this data set. The next section describes the prevalence of fostering in C8te d'Ivoire, the characteristics of households that participate in fostering arrangements, and the activities of foster children compared to those not fostered. In Section 3 a short-run model of household demand and supply of child labor is posited in which the stock of children is adjusted in response to short-run child labor needs by fostering them into or out of the household. This model motivates a reduced form equation of the determinants of the number of children fostered on both sides of the fostering "market", which is estimated empirically in section 4. In Section 5 the two sides of the fostering market are estimated jointly using a model of friction, and the "symmetry" of fostering decisions is formally tested. The final section summarizes the results. * Whether foster parents actually receive extra compensation later in life has not been shown. Since most children are fostered to relatives, it would have to be shown that the income flows from foster children exceed the flows that foster parents would have received anyway by virtue of being related (Schildkrout, 1973). 7 The survey is documented in Ainsworth and Mufioz (1986) and Grootacrt (1986). -4- 2. ECONOMIC CORRELATES OF CHILD FOSTERING A foster child can be defined in terms of the absence of the mother, the father, or both parents. To focus on the economic reasons for fostering, a foster child is defined as a child who is living away from both parents, at least one of whom is alive. Orphans who have lost both parents and children living with one parent are not considered foster children and will be classified as belonging to the household in which they reside. The paper will concentrate on fostering of children age 7-14, since these children are old enough to contribute to home production but too young to be forming households of their own. Households that have sent a child away to be fostered will be referred to as sending households; those that have accepted a foster child are receiving, or host households. Prevalence of Fostering in C6te d'Ivoire Among the 3110 children age 7-14 residing in CILSS households, 20.2 percent were living away from both parents. The percent of children fostered increases with age, from about 1 percent of infants to one- quarter of 14 year olds (see Figure 1). Ninety-three percent of foster children were related to the head of the host household or his/her spouse. A significantly higher proportion of girls 7-14 was fostered than boys (22.4 vs. 18.0 percent, respectively). If schooling investments were a major reason for fostering children, greater fostering of boys might be expected: the 1984 gross primary enrollment ratio for boys in C6te d'Ivoire was 91 percent, compared to only 63 percent for girls (World Bank, 1987). MarriageTdoes not account for the higher fostering of girls: all 11 cases of ever-married children under the age of 15 have been dropped from the analysis. Overall, 24.3 percent of households had a foster child 7-14 living with them and 18.6 percent had fostered a child out. The imbalance between the proportion of households fostering in and out is most likely due to an undercount of children fostered out." Included in these percentages are 52 households (3.3 percent) that sent and received foster children simultaneously; these cases generally involved fostering in and out children of different ages and sexes. Sixty percent of households had neither sent nor received foster children age 7-14. Rural households were the most likely to foster out children, while urban households were the most likely to foster them in. In fact, farm households were more likely to foster out a child and were less likely to foster in a child than were non-farm households: 21.4 percent of farm households fostered out a child 7-14 compared to only 13.1 percent of non-farm household. For fostering in, the figures are 22.5 and 27.9 1 The number of children fostered out was obtained from a list of non-resident children of household members. In keeping with the definition of a foster child as one living away from both parents, for a nonresident child to be considered fostered out, both of his/her parents had to be residing in the household. This definition automatically leads to under- reporting of children fostered out, since a child could have been sent away from one parent and be living with someone other than the second parent. The discrepancy between household fostering in and out is not due to net fostering across international boundaries. FIGURE 1. Percent of children living away from parents, by age and sex of child (Sample size: 6338) 35 30- /1 / 25 00 0 15 101 0 2 4 6 8 10 12 14 Ag In yeas -6- percent of farm and non-farm households, respectively. This suggests that the demand for farm labor is not the driving force behind fostering. There are no meaningful differentials in fostering between households with and without a family business. Economic activities of own and foster children compared The main economic activities of the children 7-14 were farming and housework. In the seven days preceding the survey, 60 percent of all children performed housework, which includes cleaning, meal preparation, shopping, laundry, and fetching of firewood and drinking water. Among children living in households with a family farm, 23 percent did farmwork. Among children living in households with a family business, 2.5 percent worked in a business belonging to the household. Only 29 children (0.9 percent) had ever been apprentices, and 20 children (0.6 percent) had been or were going to be paid for the work they performed the 7 days before the survey. Figures 2 and 3 compare the activities of foster children and the resident own children in host households. Foster children of both sexes are more likely to have farmed or performed housework in the past 7 days than are resident own children. Fostered girls have the highest participation rate in housework, but the differential between the participation rates in housework of foster and own boys is even greater than that for girls. Girls seem to specialize in domestic tasks, although their participation rates in farming are not much different than those of boys. The difference in housework participation rates between fosters and own children persists across all three major locations - Abidjan (the largest city, with one fifth of the population), other urban areas (with one-fifth of the population), and rural areas (with three-fifths of the population). The differential for housework is greatest in Abidj an, however: 59 percent of foster children performed housework compared to only 26 percent of resident own children. The greater role for foster children in housework in urban areas is consistent with the hypothesis that children are fostered in to substitute for the time in home production of working adults or own children attending school. In rural areas, foster children are also more likely to work on the host household's farm than are own children. Schooling of own and foster children compared Sixty-three percent of the children 7-14 were enrolled in school in the past 12 months. Because of late enrollment and high repeater rates, almost all enrolled children (97 percent) were in primary school. Foster children of both sexes have substantially lower enrollment rates than the host household's own resident children (see Figure 4). The difference in enrollment rates between fosters and own children is particularly great for girls and, not surprisingly, in urban areas (including Abidjan), where many girls are fostered. Note that the enrollment rate for foster children in urban areas is similar to that of own children in rural areas and 9 percentage points greater than fosters in rural areas. -7- FIGURE 2A: Percent of girls who worked In the last 7 days Own and foster children in the same households compared go 80 70 e 60 * OfOwn 5o s000 Fosters 40 (Sample size: 617) * For children 20 in farm households 10 (n=375) 0 Any job Housework Family farm* FIGURE 2B: Percent of boys who worked In the last 7 days Own and foster children in the same households compared 60 -O 50 Y a___ _gfOwn 0 % 40 - Fosters 30 - 0' c- -------- - (Sample size: 626) For children 20- "in farm households 10 - (nz394) 0 A Any job Housework Family farm* -8- FIGURE 3A: Percent of children In Abidjan who worked In the last 7 days Own and foster children in the saMe households compared 70 W! 60- 150.0 40 so -0 (Sample size: 252) 20 - Any job Housework FIGURE 30: Percent of children In urban areas who worked In the last 7 days Own and foaster children in the same households compared 90Y 80 70 60 _ - For chidren 20 in farm - 1-0o- households 10 (nS7 (nei87) 0 Any job Hollnework Family farm* FIGURE SC: Percent of children In rural areas who worked In the last 7 days Own and foster children in the same households compared 70.00 60 50a0 W own 40 Fosler 30 (Sample size: SS) 20 * For chidren In farm 10 households 0 Any job Housework Family farm* -9- The enrollment of children 7-14 fostered out and of their siblings of the same age who remained at home is compared in Figure 5. Children fostered out had a -lower enrollment rate (48.4 percent) than their siblings who remained at home (58.4 percent). When the sample of children is broken down by sex, foster boys and their brothers have almost the same enrollment rates, while foster girls have a lower rate then their sisters who were not fostered. Children fostered from urban areas have similar enrollment rates to their siblings, while foster children from rural areas are considerably less likely to enroll than their siblings.' The fact that foster children get the least schooling does not necessarily contradict the hypothesis that they are fostered to obtain schooling to which they otherwise would not have had access. There are several reasons to believe that school-related fostering is minimal for the group of children analyzed here, however. Most children 7-14 are of primary school age. Among the 57 rural communities surveyed, all but 9 had their own village primary school, and the most distant school was 8 km away. Further, by the definition of household membership--persons residing 3 months or more with the household--children away at school for 9 months of the year would still be classified as members of their parents' household and not fostered. Some school-related fostering at the primary level may nonetheless occur to offset the direct costs of schooling or to obtain better quality instruction. Fostering might also be necessary to attend secondary school. The vast majority of foster children enrolled in school (91 percent) received funding for school expenses from the host household.o Foster children were about four times more likely than own children to receive financial help for schooling from persons outside the household, however--39.7 percent of fosters, compared with 9.9 percent of own children--and foster children in rural areas were more likely to get outside financial help for schooling and less likely to be financed by the host household than were foster children in urban areas. This suggests that fostering into rural areas and to some extent towns may be motivated by schooling, while in Abidjan outlays for the schooling of foster children may represent a form of compensation to the children's parents for the contributions of foster children to housework. Host households made substantial outlays for the schooling of foster children ($59 annually per enrolled foster child) even though more was spent annually on the schooling of their own children ($82). Host households spent about the same amount annually on the schooling of their daughters and of foster girls receiving no outside support ($67 and $64, respectively), while they spent far more on their sons ($91) than on foster boys receiving no outside support ($54). Foster girls enrolled in school received more support from host households than did foster boys. A higher proportion of enrolled fosters are in secondary school (12.3 percent) than are their enrolled siblings (2.2 percent), however. 10 Virtually all own children (98 percent) received funding of school expenses from the host household. FIGURE 4. Percent of children enrolled in the last 12 months Own and foster children in the same households compared 100 60IR CC C)) CC O. 40 0 otr 20 0 All Boys Girls Abidjan Urban Rural n=1243 n=626 n=617 n=252 n=395 n=595 FIGURE 5. Enrollment of Foster Children 7-14 and their Siblings Percent of children currently enrolled 90 80 S50 .4 V _ 01- 40 I Fser 30 20 10 0 Total Boys Girls Total Boys Girls Total Boys Girls -NATIONAL- -URBAN- -RURAL- - 12 - 3. A MODEL OF THE DETERMINANTS OF CHILD FOSTERING The choice of variables for the reduced form equations of the determinants of child fostering is guided by a short-run household production model in which children are a variable input into production of a non-marketed home-produced good." Consider a household in which men specialize in market production, women engage in market and home production, and children specialize in home production. Home production is consumed solely by the household and there is no joint production. In the short run, adult household members maximize their utility over market goods (M), home goods (H), and women's leisure (L), given the number of their own children (C) and the number of adults in the household (A): (3.1) U=U(M,H,L;C,A), U' >0,U" <0 where U.' and U." are the first and second partial derivatives of the utility function with respect to argument i (i = M,H,L). Adult household composition is assumed fixed and, to focus on interactions between women and children in home production, the labor supply of adult males in the household is also assumed fixed. 12 Adult household members derive utility from their children, but in the time-frame over which fostering decisions are made, the number of surviving own children, C, is given by past fertility and child mortality.'3 The utility that parents receive from their children does not depend on the children's physical presence in the household.14 n This model is an adaptation. of Gronau's (1973) model of the intrafamily allocation of time in home production between husbands and wives. Rosenzweig (1977) and Rosenzweig and Evenson (1977) incorporated child time in home production into models of fertility. In the Reosenzweig (1977) model, the contribution of mature children to farm production is included in a multiperiod household model of fertility that is tested on a US farm population. Rosenzweig and Evenson (1977) estimated a household model of fertility and child schooling in which the pecuniary returns from child labor are explicitly taken into account. Strauss (1986) presents the general recursive agricultural household model, for which these are all special cases. 12 This model assumes that child household composition adjusts in the short run but that adult household composition does not. It may be less costly for households to adjust the number of children than to add and subtract adults. Children bring fewer complementary inputs into the household than do adults. Adjustment in adult composition (via marriage, divorce, or production arrangements with relatives) affects non-human assets available to the household-breaking up or consolidating farms, renting or building new housing, and so forth. Further, adults are more likely than children to have accumulated human capital that is specific to productive activities of the household. 's Fertility is considered exogenous to fostering decisions in this model. While both fertility and fostering reflect the demand for children, fostering is a temporary decision with a much shorter time frame than fertility, a life-cycle decision. The number of children fostered, conditional on fertility, can adjust upward or downward quickly, while cumulative fertility cannot be adjusted downward once a child is born. For the group of children analyzed here, fertility decisions preceded current fostering decisions by 7-14 years and the number of own 7-14 year olds cannot be adjusted upward. The exogeneity of fertility to fostering is perhaps more difficult to maintain for the fostering of infants. '4 If parents do get utility from the presence of their own children, this would introduce an additional cost to fostering out that does not occur on the fostering in side-a source of asymmetry in fostering decisions. - 13 - The production of home goods is described by a linear homogeneous production function with market goods (X), women's time (7,.), and child time (N) as inputs:'s (3.2) H=f(X,TH ,N), f,'> 0, f," <0, i= X,T.,N. The number of children used in home production in the short run, N, may be less than, equal to, or greater than the number of children belonging to household members. Own children and foster children are assumed to be substitutes in home production." An exogenous market wage for children is not observed'7 it is assumed that additional child labor can be obtained only by bringing children to live in the household, and by feeding and clothing them at a cost per child of P. It is also assumed that children devote all of their time to home production."s Under the assumptions of constant returns to scale and nonjoint production, the marginal cost 7 of producing a unit of home goods defines an implicit commodity price of home goods that is independent of the household's consumption bundle (Deaton and Muellbauer, 1980; Pollak and Wachter, 1975; Singh e al., 1986):" (3.3) T = %P. + Y.w. + 'YP, where P. is the price of market inputs, w, is the woman's market wage, and % , and -, are input coefficients that are invariant with respect to the amount of H produced. Input demands are proportional to the level of home goods produced, but the price of home goods is not influenced by the production level. " Girls' and boys' labor might be entered separately to allow for imperfect substitutability of children by sex in home production. One might also include some factor, such as land or productive assets, that is fixed in the short run. I The predominance of kin fostering should make foster and own children close substitutes, as it minimises transactions costs for both the sending and receiving households. One could also argue that household chores require little or no household-specific human capital (in contrast to Rosenzweig's (1977) model in which farm-specific human capital makes own children and hired labor imperfect substitutes). Imperfect substitutability between foster and own children in home production is an additional potential source of asymmetry in sending and receiving decisions. " The assumption that a market wage for children is not observed is not meant to imply that the market for child labor does not exist. Children may be working for payment in kind, a plausible explanation if their primary contribution is to non-marketed production and if credit markets are imperfect. Mutual exchanges of labor are also common in rural areas, and were reported in 70 percent of the 57 villages in the survey used here. " When children are allowed to work less than full time, the marginal price of child time and commodity prices become endogenous and depend on demand. Pollak and Wachter (1975) compare the situation when commodity price. depend on demand to that of a monopsonistic consumer, in which case there are "virtually no substantive results that can be predicted. " When assumptions of constant returns to scale technology and nonjoint production are relaxed, implicit commodity prices depend on the amount of home goods produced and thus tastes (Singh et aL, 1986). The budget constraint is no longer linear. - 14 - The woman's total time (T) is allocated among market production (r,) , production of home goods (THw), and leisure (L): (3.4) To=T+TH+L. The full-income budget constraint sets the value of consumption "expenditure" equal to the value of the woman's time endowment and non-labor income: (3.5) irH + P.M + wL = w,To + V , where P. is the price of market goods and V is non-labor income. If it is assumed that male household members spend all available time in market production, then their income can be included in non-labor income. Maximizing equation (3.1) subject to constraint (3.5), given production relations, yields four first order conditions, three of which equate the marginal utility of each consumption good to the product of its price and the marginal utility of income, and the fourth of which is the budget constraint. The system can be solved for reduced form equations expressing the demand for market goods, home goods, and leisure as functions of commodity prices, female wages, input prices, child costs, and unearned income, given the number of own children and the number of adults in the household: (3.6) Z = Z (P., w, Px, P,, V; C, A) , Z = M, H, L . Input demands depend linearly on H and, given H , on exogenous technology and market prices. The reduced form input demands have as arguments exogenous input and commodity prices, women's wages, child costs, and unearned income: (3.7) J=J(P,,w,Px, P., V; C,A), J =X,T,,N. The equation of key interest here is the reduced form equation for the demand for child labor. The signs of the coefficients are implied by theory, given some additional assumptions. If home goods are assumed normal, the impact of an increase in exogenous income on the demand for child labor is positive. The effect of an increase in child costs on the demand for children is unambiguously negative. An increase in the price of market inputs reduces the demand for children, assuming that market inputs and child time are substitutes in production and that the negative effect of a rise in market input prices on the demand for home - 15 - goods and thus on demand for child labor is greater than the positive effect it would have on substitution of children for market inputs in home production. If market and home goods are substitutes in consumption, an increase in the price of market goods raises or lowers the demand for children, depending on whether the positive substitution effect of market goods and prices on the demand for home goods exceeds or falls short of the negative income effect. A rise in the woman's wage affects the demand for children through three channels: it raises demand for children as a substitute for women's time in home production; if home goods and leisure are substitutes in consumption, it raises demand for home goods and thus for child time inputs; and it raises the price of home goods relative to other consumption goods, reducing the demand for home goods and child labor. If the first two effects dominate the negative own price effect, then an increase in women's wages raises the demand for children.2 Finally, an increase in the number of adults is equivalent to an exogenous increase in demand for home goods which also unambiguously raises the demand for child (and female) labor inputs. The hypothesized signs on the reduced form demand for children are as follows: + - - - (3.8) N = N ( P,,, w., P., P., V; C, A) and where an increase in the number of adults in the household would raise the demand for child labor through its effect on the demand for home goods. Child fostering occurs because of the difference between the optimal number of children needed for home production (N) and the number of own children of household members. Factors that raise the "net" demand for child labor, N'-C, will lead the household to foster in more children and foster out fewer. The signs on the demand for foster children are therefore identical to equation (3.8) and those on the supply of foster children are the reverse of equation (3.8). An important prediction of this model of fostering as child-labor adjustments is that the coefficients on reduced form equations for fostering in and out should be "symmetric"--opposite in sign and equal in magnitude. Asymmetry would arise if the motivation for sending and receiving foster children is different--that is, if the demand for foster children arises from the need for child labor but the supply arises from parents' investments in the human capital of their children, such as schooling or other training. There are many reasons why symmetry may be violated, however, even if both sides of the fostering market are motivated by child-labor considerations. Differential transactions costs on the two sides of the market may make fostering on one side appear less sensitive to small changes in exogenous variables. For example, if parents get utility from the presence of their children in the household, then there will be an additional cost of 2 When women are not engaged in market production their value of time exceeds the market wage. In that case both the shadow price of women's time and the shadow price of home goods are endogenously determined. - 16 - fostering out children not present on the fostering in side. The symmetry will also be violated if foster and own children are not perfect substitutes in home production. Specification of the Variables for the Reduced Form The dependent variable could be measured as a "net" demand for child labor - the difference between the number of children fostered in and fostered out - and estimated using ordinary least squares. This specification would be closest to the tradition of the theoretical model, but it would also constrain the coefficients on the sending and receiving sides of the market to be equal and of opposite sign. Since the differences in the determinants on the two sides of the market are of analytic interest, equations for fostering out and fostering in will be estimated separately so that the determinants of the sending and receiving decisions can be compared. The dependent variables for the four equations are the number of girls 7-14 fostered out, the number of girls 7-14 fostered in, the number of boys 7-14 fostered out, and the number of boys 7-14 fostered in (see Table 1). Adult household composition variables include the number of male and the number of female adult household members. Adult household members are defined as persons 15 years or older who ate and slept with the household for at least 3 of the past 12 months. A dummy variable is also included for female-headed households, which may have fewer sources of other income, be more likely to participate in the market, and thus have greater need for child labor as a substitute in home production. Past fertility and child mortality of the household is captured by variables that count the number of daughters 7-14 and sons 7-14 of household members, including children who are living with their parents and who are living away. Average completed years of schooling for women is used as a proxy for women's opportunity cost of time.21 Wages are observed only for a small number of women. As a proxy for wages, an increase in women's schooling should raise the demand for child labor in performing household chores, raising the number of children fostered in and reducing the number fostered out. This effect will be magnified if women with schooling also prefer that their own children enroll: in that case, the household's own children have less time to devote to household chores and the probability of fostering in should increase. If educated women prefer that their children attend school (at home), they should be more reluctant to foster them out. Female schooling is not observed for households with no women; the value for schooling in womanless households has been set equal to zero and a dummy variable denotes households with no women. If children are a substitute for 2 The schooling of the head of household was included in other regressions, with a significant positive sign in the fostering in regression and a negative sign in the fostering out regression. Inclusion of this variable, a proxy for household wealth, rendered female schooling and household income per adult insignificant, however. - 17 - women in home production, then the absence of a woman might e expected to raise the number of children fostered in. On the other hand, the reverse might be true if children require some female supervision in home production. The dummy variable for no woman is included only in the fostering in regressions since, by the definition of fostering used here a child cannot be fostered out from a womanless household. TABLE 1: Description of variables Variable Description Dependent Variables GIRLSOUTI Number of girls 7-14 fostered out BOYSOUT" Number of boys 7-14 fostered out GIRLSIN Number of girls 7-14 fostered in BOYSIN Number of boys 7-14 fostered in Independent Variables MEN Number of male household members >14 years WOMEN Number of female household members >14 years GIRLS Number of daughters 7-14 living at home + those fostered out BOYS Number of sons 7-14 living at home + those fostered out NOWOMANb 1 if household has no adult female, 0 otherwise SCHFEM Mean years of schooling of women in household PINCOME Log of permanent income (consumption) per adult HEADFEM 1 if household head is female, 0 otherwise COTTON 1 if household grew at least one hectare of cotton, 0 otherwise ABIDJAN 1 if household is in Abidjan, 0 otherwise TOWNS 1 if household is in urban area other than Abidjan, 0 otherwise SAVANNA 1 if household is in rural savanna, 0 otherwise WFOREST 1 if household is in rural west forest, 0 otherwise EFOREST 1 if household is in rural east forest, 0 otherwise (left out) CLASS/SCI' Number of classes at the nearest primary school DPRIMc Distance in ilometers to the nearest primary school KIDWORK" 1 if children under 12 in village work for a wage, 0 otherwise DefNed only for households with own children in the relevant age group, of the relevant sex. 'Included only in fostering in regressions since, by the definition of fostering used here, a child cannot be fostered out from a womanless household. 'Variables available for rural regressions only. Annual consumption expenditure per adult is used as a proxy for household permanent income. The sum of annual consumption expenditure, the value of home production consumed, and an imputed value of services from durable goods was divided by the number of adult household members age 15 or older and - 18 - expressed in logarithmic form.' The level of household consumption expenditure is potentially endogenous to fostering decisions directly, through the labor supply of foster children in home production, and indirectly, as children substitute for adults in home production, allowing greater adult labor force participation. The consumption variable has been instrumented using variables representing productive assets and human capital of the household. ' A dummy variable is included for cultivation of cotton, because cotton harvesting is highly intensive in child labor in other countries (Levy, 1985; Mueller, 1976).21 Regional dummy variables are included for households living in Abidjan (the largest city, with one fifth of the population), in urban areas other than Abidjan, and in the rural savanna and west forest regions. The left-out dummy variable is for households in the rural east forest. The coefficients on these variables can be interpreted as reflecting regional differences in the cost of living (including child maintenance and schooling costs) relative to the opportunity cost of time. * The household consumption variable was computed by Kozel (1987). It includes: food expenditure and the value of home production consumed by the household; expenditure on transportation , fuel, services, leisure activities, gifts and ceremonies, clothing, water and electricity, education , medical care, and transfers to other households; imputed value of services for durable goods, including automobiles; rent; and imputed rent for owner-occupied housing in urban areas. 23 The exogeneity of consumption per adult was rejected in two of the four regressions, using a test suggested by Smith and Blundell (1986) for models with limited dependent variables. The four dependent variables were regressed on the actual value of annual household consumption per adult (INCOME), the residuals from a least-squares prediction of income, and the other independent regressors in Table 1. A sufficient condition for the value of annual household consumption per adult to be weakly exogenous is that the regression coefficient on the residuals (RESID) from the prediction of income equal to zero. The exogeneity of consumption was rejected for both fostering in regressions. Instruments for income included: schooling of the household head; the head's age and age squared; hectares of land available; dummy variables for households that cultivate at least one hectare of coffee and cocoa; the value of cattle owned by the household; the value of sheep and goats owned by the household; dummy variables for households that possess a sewing machine, transform cassava into attieke for sale, and transform peanuts or palm nuts into oil for sale; and living area in square meters. The R2 for the OLS regression of the prediction of income was .56. *' We assume that household decisions to engage in agricultural production and the choice of crops are made over a longer time horizon, requiring substantial investments and gestation periods, and are thus not jointly endogenous with short- run child fostering decisions. The amount of land cultivated was considered endogenous to child labor decisions, however. When total land available was entered, it was never significant in any regression. Other agricultural variables were either insignificant or uninterpretable. For example, the coefficient on the value of cattle owned by the household was significant in three of the regressions, but always positive. Since income was already controlled for and tending cattle is not generally a task allocated to children, the coefficient was difficult to interpret in terms of the model. Similar problems arose with dummy variables for coffee and cocoa cultivation and for the value of sheep and goats. ' Fostering differences by ethnic group have received a great deal of attention by anthropologists. Surprisingly, when dummy variables for the five main ethnic groups were entered in these regressions with region also controlled for, none of the ethnic coefficients were significant in the fostering in or out of girls or the fostering in of boys. (Girls were less likely to be fostered in by non-Ivorians, however.) North and South Mande groups were significantly less likely than the Akan to foster out boys. Inclusion of ethnic dummy variables had virtually no effect on the magnitude or significance of other coefficients in the regressions. It is possible that ethnic groups were insufficiently disaggregated to observe differences; alternatively, it may be the case that ethnic differences in other studies can be accounted for almost entirely by differences in socioeconomic status or that fostering decisions are based almost entirely on short-run economic conditions. - 19 - Two variables representing the availability and quality of schooling for rural communities are included; they were collected in rural areas only and are identical for all households in the same community. The number of classes at the nearest primary school is a proxy for both availability and quality. The standard number of grades in primary school is six; a school with fewer than six classes cannot have the full primary curriculum. In addition, such schools are typically small, community-built, and may have difficulty attracting good teachers and financing recurrent inputs.2 The second schooling variable is the distance to the nearest primary school, in kilometers. Children living too far from a school would have to be fostered out to attend. If school attendance is a reason for fostering children 7-14, one would expect the number of classes to be negatively correlated and the distance to the nearest school to be positively correlated with fostering out, and the reverse for fostering in. Finally, although very few instances of child wage labor were observed in the household data (and virtually none in rural areas), in 8 rural communities leaders indicated that children under 12 sometimes work for a wage. A dummy variable for the presence of a child labor market is included in regressions for rural areas. If fostering is due to demand for child labor, then the presence of a local child labor market should reduce fostering in. The effect on fostering out is ambiguous: on the one hand, households with more child labor than they need can participate in the local market, reducing out-fostering; on the other hand, the availability of local substitutes for the labor of own children make it possible to foster out more children for other reasons (such as schooling or insurance arrangements). The means and standard deviations of variables are reported in Table 2. The households in the sample had an average of 2.0 men and 2.4 women with 0.88 daughters and 0.93 sons age 7-14 living either at home or away. The subsamples of households with at least one son or daughter 7-14 had more adults--2.3 men and 2.7 women. Only 6.3 percent of households had no woman. Average completed schooling of females is less than two years. The income variable (mean predicted logarithm of consumption expenditure per adult) translates into about $680 per adult at the average 1985 exchange rate of 450 CFA francs per U.S. dollar. The subsamples of households with own children 7-14 have slightly lower income. Only 7.8 percent of all households were headed by women, and only 3-4 percent of households with own children 7-14 were female- headed. Forty-Three percent of the households are in urban areas (Abidjan or towns); the subsamples excluding childless households are less urban. alternatively, it may be the case that ethnic differences in other studies can be accounted for almost entirely by differences in socioeconomic status or that fostering decisions are based almost entirely on short-run economic conditions. I A survey of communities and schools in rural Mali, which borders CMte d'Ivoire to the north, found that parents perceived these schools to be of lower quality than larger schools with at least one class for each grade (Ainsworth, 1983). - 20 - TABLE 2: Variable means and standard deviations Sample with Sample with Sample Entire GIRLS > 0 BOYS > 0 Variables Mean SD Mean SD Mean SD Dependent GIRLSOUT 0.304 0.601 GIRLSIN 0.217 0.557 0.194 0.516 BOYSOUT 0.217 0.515 BOYSIN 0.180 0.543 0.182 0.578 Regressors MEN 2.012 1.457 2.255 1.526 2.300 1.555 WOMEN 2.362 1.809 2.728 1.880 2.795 1.900 GIRLS 0.882 1.122 1.727 1.003 1.191 1.197 BOYS 0.929 1.147 1.248 1.223 1.757 1.015 NOWOMAN 0.063 0.243 SCHFEM 1.658 2.960 1.685 2.764 1.625 2.676 PINCOME 12.631 0.627 12.609 0.588 12.571 0.608 HEADFEM 0.078 0.269 0.034 0.182 0.038 0.191 COTTON 0.071 0.257 0.080 0.271 0.077 0.267 ABIDJAN 0.211 0.408 0.203 0.403 0.185 0.388 TOWNS 0.220 0.415 0.199 0.399 0.201 0.401 SAVANNA 0.190 0.392 0.176 0.381 0.190 0.392 WFOREST 0.200 0.400 0.212 0.409 0.212 0.409 N 1597 816 844 Notes: 1. Statistics for the three variables available only for rural households (CLASS/SCH, DPRIM, KIDWORK) are reported in Table 5, in the discussion of fostering in urban and rural areas. 2. The entire sample is used for fostering in regressions, the sample of households with at least one girl is used for regressions of girls fostered out, and the sample with at least one boy is used for regressions of boys fostered out. CILSS interviewed 1599 households; one household had to be dropped because of missing values for regressors in the prediction of income and a second household with over 100 members was dropped because of extreme values. -21- Econometric Models A Tobit model is used for estimating fostering in regressions, since the dependent variables-the number of boys and girls fostered in-are censored from below at zero.' Let y* be the true demand for foster children and y the observed number of children fostered in. When the true demand y* is greater than zero, it is observed; when y* is less than zero, zero is observed. The model is: yT = 'xi + k, u. - N(O, a) (3.9) yi = y+ if yt > 0 yi = 0 if y# < 0 , where x. is vector of independent regressors for observation i. The Tobit model assumes that the error is normally distributed with mean zero and constant variance a2. Fostering in equations are estimated for the entire sample. For fostering out regressions, the dependent variables-the number of boys and girls fostered out- are censored from below at zero and from above at the number of own children of the household.' A two-limit Tobit model is used: yt B'.x. + u. , u. ~ N(0, o2) y= 0 if y? < 0 (3.10) y y f< fc (3.10) yi, = yIP if 0 < y+ < ci y, = c. if yT > ci where ci is the upper censoring point (the number of sons or daughters 7-14 in household i), which varies across households. The equations for fostering out are estimated only for households that had at least one own child 7-14, resident or nonresident, of the appropriate sex, since households cannot foster out children they " An alternative, the Poisson count model, takes into consideration both the censored and integer aspect of the dependent variable (Maddala, 1983; Hausman, Hall and Griliches 1984; Cameron and Trivedi 1986). It does not allow for the upper censoring of the fostering out variables, however. These same regressions have been estimated with the Poisson count model (not taking the upper censoring into consideration) in Ainsworth (1989). The major conclusions are not affected. 2 Thirteen percent of the households with at least one daughter 7-14 fostered out all of them and 9.5 percent of the households with at least one son 7-14 fostered out all of them. When the dependent variable is censored, least squares estimates are inconsistent because the error term is not independent of the regressors (Amemiya, 1984; Maddala, 1983). - 22 - don't have. Both Tobit models are estimated using maximum likelihood methods.' Standard errors have been corrected to account for the fact that one of the regressors (PINCOME) is a predicted variable." 2 The likelihood function for the single-censored Tobit model is: L(,3, al y, x) = IIo[l(-a'xf)] II,(1/)[(yi - O'xY/o where the first product is over all yj = 0 , the second product is over all positive values of y, , and 4' and 40 are the normal cumulative distribution and probability density functions, respectively. The likelihood function for the two-limit Tobit model is: L(0, aI y, x) = Il[l(-'x 1aO)] II,(1/o)4[(yi - P'x)/o] II[1 - ((c - P'*/o)] where the first product is over all yi = 0 , the second product over all 0 < yi < ej , and the third product over all yi= ci " The formula used to estimate the consistent second-stage variance-covariance matrix follows from Muphy and Topel (1985). The consistent second-stage variance-covariance matrix when the first stage is linear and the second stage is maximum likelihood is given by: V2 = 1- + LI,[Le'x(x,x)-'L,2'+ L,2(x.X'x'eL, + Li2'1L,2lL , where L, is the derivative of the second-stage likelihood function with respect to the parameters of stage two, -L is the inverse of the second derivative of the second-stage likelihood with respect to stage-two parameters, L, (= L) is the deri- vative of L, with respect to stage-one parameters, e is the vector of first-stage residuals, x is the matrix of first- stage regressors, and V1 is the consistent first-stage variance-covariance matrix. See Ainsworth (1989). - 23 - 4. THE ECONOMIC DETERMINANTS OF SENDING AND RECEIVING DECISIONS Determinants of Fostering for the Entire Sample Tobit estimates of the determinants of fostering of girls and boys are presented in Table 3. The coefficients on adult household composition variables are of the expected signs and symmetric: additional adults of both sexes raise the demand for home production, raising the demand for both boys' and girls' labor and the number of children fostered in. The number of men and women have similar effects on the demand for foster children within each regression." Adult household composition variables are significant in all regressions but one: the number of men has no effect on fostering out girls, while the number of women has a weak negative effect. As expected, own children and foster children are substitutes: the more daughters household members have, the more girls are fostered out and the fewer are fostered in. The same holds true for sons and boys fostered out. There are interesting asymmetries, however. If sons and daughters are perfect substitutes in home production, the effect of an additional son should be the same as the effect of an additional daughter on the number of children fostered. The results show that sons and daughters are substitutes for each other and for foster children of both sexes in fostering in regressions (2) and (4). The coefficients on sons and daughters in the regression for foster boys (4) are quite similar, while in the foster girls regression (2) the coefficient on daughters is about twice the size of that for sons. Sons and daughters are not substitutes in fostering out decisions, but these results may be affected by the fact that many households in regressions (1) and (3) do not have a child of the opposite sex.' The absence of an adult woman significantly and greatly reduces the number of children of both sexes fostered in. This is contrary to what might be expected if children are substitutes for women in home production. Perhaps some minimum amount of women's supervision is necessary to use child labor effectively. Alternatively, for child welfare reasons households may be reluctant to send their adolescent children to live in a household of men. Since men generally do not shop for food, cook, or do laundry, children may not be 31 Wald tests failed to reject the hypothesis that the coefficients on men and women are identical in each of the four regressions in Table 3. n Wald tests failed to reject the hypothesis that the coefficients on sons and daughters are identical in each of regressions (2) and (4), however. " Two other specifications of GIRLS and BOYS were tried and rejected. First, children of each sex were broken into age groups 7-10 and 11-14. The equality of coefficients for children of the same sex and different age groups could be rejected in only one case-for the coefficients on girls 7-10 and 11-14 in the fostering out regression for girls. In a second specification, quadratic terms for girls and boys and an interaction between girls and boys were entered. The three added quadratic terms were jointly significant only in the regression for girls fostered in. - 24 - TABLE 3: Tobit estimates, fostering of girls and boys Dependent GIRLSOUT GIRLSIN BOYSOUT BOYSIN Variable: (1) (2) (3) (4) Regressore 0 T aE(y)/8x 0 T aE(y)/8x 0 T aE(y)/8x 0 T 8E(y)/8x MEN -0.218 (-1.36) -.0262 0.293' (3.84) .0435 -0.411 (-1.96) -.0360 0.396t (4.55) .0489 WOMEN -0.265^ (-1.77)-0.0319 0.298t (4.74) .0442 -0.397$ (-2.32) -.0348 0.241t (3.27) .0297 GIRLS 0.665t (2.71) .0801 -0.309t (-3.63) -.0458 -0.121 (-0.68) -.0106 -0.23Y (-2.24) -.0290 BOYS 0.0109 (0.07) .00131 -0.137^ (-1.71) -.0203 0.577t (2.06) .0506 -0.270' (-2.80) -.0333 NOWOMAN' -1.638t (-2.99) -.243 -1.239$ (-2.23) -.153 SCHFEM -0.237$ (-2.26) -.0285 -0.00904(-0.26) -.00134 -0.253$ (-2.05) -.0222 -0.0117 (-0.26) -.00144 PINCOME 0.474 (0.66) .0571 1.416t (4.02) .210 -0.0423 (-0.05) -.00371 0.903$ (2.33) .1114 HEADFEM -2.452^ (-1.81) -.295 0.827t (3.11) .123 -3.930' (-2.25) -.345 0.0523 (0.15) .00645 COTTON -1.477A (-1.89) -.178 0.206 (0.51) .0305 -1.526 (-1.59) -.134 -0.767 (-0.18) -.0946 ABIDJAN -0.338 (-0.42) -.0407 -1.602t (-3.97) -.238 0.882 (0.89) .0774 -1.477' (-3.33) -.182 TOWNS -0.537 (-0.82) -.0647 -0.749* (-2.45) -.111 0.232 (0.28) .0203 -0.278 (-0.85) -.0343 SAVANNA 0.321 (0.52) .0386 -0.500 (-1.52) -.0742 0.319 (0.44) .0280 -0.0323 (-0.10) -.00399 WFOREST -0.194 (-0.37) -.0234 -1.144t (-3.57) -.170 -0.759 (-1.16) -.0666 -0.783$ (-2.51) -.0966 Constant -7.768 (-0.85) -20.093 (-4.42) -1.464 (-0.13) -14.345 (-2.87) a 3.473 (8.21) 2.084 (12.02) 3.734 (6.88) 2.302 (17.45) LOGL -595.75 -954.52 497.60 -851.89 ?b 52.30 124.36 52.72 84.62 N 816 1597 844 1597 Mean of .304 .217 .217 .180 dep. var. tSignificant at .01 or less. $Significant at .05. ^Significant at .10. 'NOWOMAN is excluded fron fostering out regressions because, by the definition of fostering used here, a child cannot be fostered out from a womanless household. bLikelihood ratio test for joint significance of all coefficients, distributed as X where r = 12 in regressions (1) and (3) and r -13 in (2) and (4). -25- adequately cared for in an all-male household. Further, it may not be culturally acceptable to send a young unmarried girl to such a household, even if it is a household of relatives. Among households with women, higher average female schooling reduces the number of girls and boys fostered out but does not seem to affect fostering in decisions. The coefficients on female schooling imply that a one-year increase in the average schooling of women in the household reduces the number of girls fostered out by 0.029 and the number of boys fostered out by 0.022.' The negative effect of female schooling on fostering out is consistent with a child labor explanation when schooling is a proxy for female wages; however the lack of significance of female schooling on fostering in casts doubt on this interpretation. It may be that women with schooling prefer that their children attend school (at home) instead of being fostered out to work. In that case, the lack of significance of female schooling in the fostering in regressions would imply that schooling of own children does not generate additional demand for foster children as a substitute for own children in home production. If home production is normal, higher income will lead to greater demand for child labor to engage in it. Again, asymmetry in fostering in and out decisions is observed: higher income households are more likely to foster in children of both sexes, but no less likely to foster out. The insignificance of income in fostering out is strong evidence against the hypothesis that fostering is a way for parents to raise children they cannot afford--a way of "managing" high fertility. Were that the case, one would expect income to have a strong negative effect on fostering out. The income coefficients translate into relatively high elasticities of 0.97 for girls fostered in and 0.62 for boys fostered in.' The coefficients for female-headed households display symmetry in sign across all four regressions and are significant in all but the regression for boys fostered in. Female-headed households are less likely to foster out children of either sex and more likely to foster in girls than are male-headed households. The insignificance of this variable in the regression for boys fostered in indicates that other people's sons are not good substitutes for women's time, compared with own sons, own daughters, or daughters of other households. I The coefficients of the Tobit model represent Oy*/Ox . The coefficients 8E(y)/8x for the one-limit Tobit with lower bound zero are given by 4 , where 4 is the normal cumulative distribution function of B'x/a evaluated at the mean and the subscript i denotes an explanatory variable. For the two-limit Tobit with lower bound zero and upper limit ci the expression for 8E(y)/ax is if(4 2- 41) , where the density and distribution functions with the first subscript are evaluated at -B'x/o and those with the second subscript are evaluated at (c - B'x)/o . s Income elasticities are calculated from the coefficient of the expected value locus, OE(y)/8x. These elasticities are larger than the long-run income elasticity of demand for children of 0.10 found in Ainsworth's (1988) study of fertility in C8te d'Ivoire. - 26 - The coefficient on cotton is significant and negative in the regression for girls fostered out. With income and geographic region controlled for, it would be difficult to give this result any other than a child labor interpretation.3 The insignificance of cotton in the regression for boys fostered out or for fostering in regressions is perplexing, however. Possibly, the cotton dummy variable is not measuring demand for child labor, but picking up some other dimension of rural-urban differences not accounted for by the regional dummy variables. Finally, results for regional dummy variables representing relative price differences again reveal asymmetry: households in Abidjan and the rural west forest foster in significantly fewer children of both sexes than households in the rural east forest, but do not foster out more children. Abidjan households are the least likely to foster in children among all regions, when income and adult household composition are controlled for. Child costs may be sufficiently high relative to the price of market inputs as to encourage substitution of other inputs for children in home production. None of the regional variables is significant in fostering out regressions. Rural vs. Urban Fostering and the Effect of Schooling Variables Descriptive statistics for rural and urban subsamples are presented in Table 4. Rural households foster out more children and foster in fewer children than do urban households on average. They have slightly fewer men and more women than do urban households. Rural households also have more sons and daughters 7- 14 than do urban households, but this is evidently due to a larger proportion of households with children in rural areas; in subsamples of households in each area with at least one son or daughter 7-14, the number of sons and daughters in rural and urban households is similar. Urban households are twice as likely to be womanless as rural households but are also twice as likely to be headed by a woman. The difference in average women's schooling between regions is remarkable: the women in rural households have an average of only 0.6 years of formal schooling, compared to 3.1 years for women in urban areas. Mean annual income per adult in rural areas is less than half that in urban areas--$466 and $1116 per adult in the unrestricted samples of rural and urban households, respectively. Regional dummy variables for the savanna and west forest are included in rural regressions (the east forest is the excluded region) and a dummy variable for households in Abidjan is included in the urban regressions. Rural households are distributed roughly one third in each of the rural zones, while almost half of the urban sample lives in Abidjan. The mean distance of rural households to the nearest primary school was 0.6 kilometers and the nearest school had about six grades. Almost 18 percent of rural households lived in a village where children under 12 reportedly work for a wage. 36 The cotton variable remains significant when dummy variables controlling for the ethnic group of the household are also included. - 27 - TABLE 4: Variable means and standard deviations, urban and rural samples Sample with Sample with Sample: Entire GIRLS > 0 BOYS > 0 Variables Mean SD Mean SD Mean SD Rural Sample Dependent GIRLSOUT 0.344 0.651 GIRLSIN 0.172 0.505 BOYSOUT 0.224 0.531 BOYSIN 0.164 0.556 Regressors MEN 1.915 1.328 2.092 1.360 2.166 1.430 WOMEN 2.413 1.778 2.703 1.888 2.799 1.883 GIRLS 0.925 1.134 1.721 1.011 1.185 1.213 BOYS 1.007 1.180 1.273 1.238 1.765 1.409 NOWOMAN 0.0419 0.200 SCHFEM 0.572 1.317 0.644 1.347 0.612 1.278 HEADFEM 0.0518 0.222 0.0266 0.161 PINCOME 12.254 0.380 12.289 0.372 12.257 0.400 COTTON 0.117 0.321 0.125 0.331 0.120 0.325 SAVANNA 0.334 0.472 0.295 0.457 0.309 0.462 WFOREST 0.351 0.478 0.355 0.479 0.346 0.476 CLASS/SCH 5.991 2.633 6.057 2.618 6.103 2.722 DPRIM 0.564 1.514 0.523 1.474 0.537 1.515 KIDWORK 0.176 0.381 0.160 0.367 0.153 0.360 N 908 488 581 Urban sample Dependent GIRLSOUT 0.244 0.514 GIRLSIN 0.277 0.615 BOYSOUT 0.206 0.487 BOYSIN 0.200 0.526 Regressors MEN 2.139 1.604 2.497 1.713 2.512 1.715 WOMEN 2.295 1.849 2.765 1.870 2.788 1.930 GIRLS 0.826 1.104 1.735 0.992 1.199 1.174 BOYS 0.826 1.094 1.210 1.200 1.745 0.961 NOWOMAN 0.0914 0.288 SCHFEM 3.088 3.800 3.233 3.509 3.233 3.427 HEADFEM 0.113 0.317 0.0457 0.209 0.0552 0.229 PINCOME 13.127 0.535 13.086 0.523 13.070 0.546 ABIDJAN 0.489 0.500 0.506 0.501 0.479 0.500 N 689 328 326 - 28 - Tables 5 and 6 present Tobit estimates of fostering by households in rural and urban areas, respectively. Since fostering can occur within and between urban and rural areas, there is no particular reason to expect symmetric results within one area. Turning first to rural households (Table 5), there is a marked asymmetry in the determinants of fostering in and out. The only significant household-level variables in fostering out regressions are the number of women and the number of children of the same sex: the more women there are relative to own children of both sexes, the less likely are children to be fostered out. Male household composition shows no significance in fostering out regressions. Two explanations come to mind: (1) children 7-14 substitute for the time of rural women in home production, but they are not good substitutes for men's tasks; or (2) raising children 7-14 requires women's time (not men's), so when the number of women is insufficient to properly care for them, the children are fostered out. Anthropological evidence and the data on the economic activities of children cited earlier suggest that children 7-14 make important contributions to home production and require minimal child- care time inputs from the mother. It seems more plausible that children are substituting for women's time. The coefficient on women's schooling-a proxy for the value of women's time--should be negative if this is the case, but it is insignificant in both fostering out regressions. The mean level of schooling was very low however, and there was not much variation in rural areas. It may also be he case that schooling is a poor proxy for the value of women's time in rural areas where there are few opportunities for wage employment. The results for fostering in regressions are supportive of demand for child labor explanations, but strongly suggest that foster boys and girls perform different tasks in rural areas. Both adult household composition coefficients are significant and of the predicted signs in the girls in regression, while only the coefficient on male adults is significant in the boys in regression. Adults of both sexes thus generate demand for foster girls' labor, while only men generate demand for foster boys' labor. Note also that sons and daughters are substitutes for girls fostered in, while only sons are substitutes for boys fostered in.' Household income per adult is a particularly important determinant of the number of girls fostered in, corresponding to an income per adult elasticity of 1.11, but it apparently does not affect demand for the services performed by foster boys in rural areas. This is what one would expect, for example, if own children and foster girls perform housework, a non-marketed home-produced good that is normal, while boys are engaged in production of marketed agricultural goods. In the latter case, household demand would not affect the level of production, while in the former it would. Female-headed households demand significantly more foster girls, but not foster boys. Further evidence of the different types of work of boys and girls is that the coefficient on KIDWORK is significant, negative and quite large in the boys in regression, but not in the girls in regression, indicating that child wage labor is a substitute for fostered-in boys, but not for girls. n Wald tests failed to reject the hypothesis that the coefficients on BOYS and GIRLS in regressions (2) and (4) are identical. - 29 - TABLE 5: Tobit estimates, fostering by rural households Dependent GIRLSOUT GIRLSIN BOYSOUT BOYSIN Variable: (1) (2) (3) (4) Regressors 0 T aE(y)/ax f T DE(y)/x $ T 8E(y)/8x P T OE(y)/ox MEN -0.131 (-0.64) -.0170 0.369t (2.86) .0411 -0.233 (-1.00) -.0227 0.339t (2.44) .0365 WOMEN -0.366^ (-1.72) -.0475 0.276t (3.18) .0308 -0.650t (-2.61) -.0633 0.177 (1.52) .0191 GIRLS 0.809' (2.36) .105 -0.317$ (-2.48) -.0353 0.00980 (0.05) .000955 -0.148 (-0.86) -.0160 BOYS 0.0354 (0.17) .00459 -0.247t (-2.18) -.0275 0.840' (2.40) .0818 -0.411' (-2.77) -.0443 NOWOMAN' -1.227 (-1.34) -.137 -1.694^ (-1.66) -.183 SCHFEM -0.284 (-1.29) -.0369 0.127 (1.61) .0142 -0.217 (-1.08) -.0211 0.143 (1.41) .0154 PINCOME 1.097 (1.01) .142 1.727t (3.27) .193 -0.340 (-0.30) -.0331 0.919 (1.43) .0991 HEADFEM -2.118 (-1.11) -.275 1.762' (3.42) .196 -0.281 (-0.44) -.0303 COTTON -1.207 (-1.44) -.157 0.752 (1.56) .0838 -1.143 (-1.31) -.111 0.160 (0.30) .0172 SAVANNA 0.147 (0.23) .0109 -0.499 (-1.25) -.0556 0.230 (0.35) .0224 0.145 (0.34) .0156 WFOREST -0.176 (-0.32) -.0228 -1.214t (-2.83) -.135 -0.456 (-0.75) -.0444 -1.032' (-2.61) -.111 CLASSISCH -0.0455 (-0.43) -.00591 0.0448 (0.79) .00500 -0.00989(-0.09) -.000963 0.0555 (0.93) .00598 DPRIM 0.247^ (1.80) .0321 0.0932 (1.14) .0104 0.106 (0.80) .0103 -0.0203 (-0.17) -.00219 KIDWORK 0.635 (1.02) .0824 -0.0887 (-0.24) -.00989 0.372 (0.60) .0362 -1.009* (-2.12) -.109 Constant -15.535 (-1.13) -24.523 (-3.59) 2.064 (0.14) -14.765 (-1.80) o 3.372 (6.76) 2.148 (7.76) 3.231 (5.63) 2.530 (12.18) LOGL -370.33 -451.57 -313.41 -449.52 ?b 45.38 76.68 35.28 39.14 N 488 908 518 908 Mean of 0.344 0.173 0.224 0.164 dep. var. tSignificant at .01 or less. *Significant at .05. ^Significant at .10. "NOWOMAN is excluded from fostering out regressions because, by the definition used here, a child cannot be fostered out from a womanless household. HEADFEM is excluded from regression (3) because of insufficient variation. bLl.io d ratio test of joint significance of all coefficients, distributed X , where r = 13 in regrssion (1), 14 in regressions (2) and (4), and 12 in regression (3). - 30 - The cotton dummy variable is insignificant in all of the rural regressions. The lack of response of fostering to this and other agricultural variables comes as some surprise, since farming is the major activity in rural areas and presumably an activity in which foster children could be of assistance.3' Possibly foster children (as well as hired children, where a market exists) are poor substitutes for own children in agriculture. There is also a pool of adult migrant labor and sharecroppers from neighboring countries available year round that is perhaps more productive in agricultural tasks relative to cost than foster, hired, or own children." These results confirm our earlier suspicion that in the larger sample the cotton dummy variable may well have been measuring some dimension of urban-rural differences not reflected in regional dummy variables. Alternatively, the cotton variable may not be significant because of the small sample, particularly in fostering out regressions. Finally, the evidence that fostering responds to the availability and quality of schooling in rural areas is weak. Across all four regressions in Table 5 coefficient signs on schooling variables are as expected, but generally insignificant. As was pointed out earlier, the vast majority of villages had their own primary schools and the variability of the distance to a primary school is quite low. Parents do not foster their children to go to primary school in rural areas because the children can attend school while living at home. The results in Table 6 show no significant economic determinants of fostering out in urban households. The sample size for fostering out children in urban areas is small--fewer than 330 households-and may account for the poor results. The absence of variables measuring the cost and quality of schooling in urban areas is regrettable; this conceivably could be an important reason for fostering between urban areas and from urban to rural areas. There is ample evidence of fostering in for child labor in urban areas, however. In regressions (2) and (4), daughters are substitutes for foster children of both sexes, while sons are substitutes for neither. It would seem that daughters and foster children perform similar types of work in urban areas, which is different than the work (if any) performed by sons. This is confirmed by the fact that, unlike in rural areas, income is a significant determinant for fostering in both girls and boys. The coefficients correspond to per adult income elasticities of 0.90 for girls and 0.68 for boys. The income elasticity for girls in urban areas is thus lower than in rural areas (1.11). The highly significant and large negative coefficient on the Abidjan dummy variable indicates that the higher cost of maintaining foster children in Abidjan exceeds the higher value of foster child time in housework. Two variables representing the opportunity cost of women's time-female schooling and female headship--are insignificant on both sides of the fostering market in urban areas. This ' When the larger set of agricultural variables (LAND, COFFEE, COCOA), was added to rural regressions LAND was significant (positive) only in the girls in regression and it rendered insignificant the income coefficient. * Two-thirds of farm households reported outlays in cash or in kind for farm labor. Of the 57 rural communities in the survey, 88 percent reported that people migrate into the villages temporarily to work; these migrant workers are overwhelmingly from Mali and Burkina Faso. In addition, 70 percent of villages reported mutual exchanges of labor within the community. -31- comes as some surprise, since female heads in urban areas are more likely to be engaged in market work than female heads in rural areas and one would expect the need for foster children as substitutes for women's time in home production to be greater. It is also surprising that women's schooling has no effect, since it presumably reflects opportunity costs better in urban areas, where the opportunities for market work are greater. Again, however, this result may reflect a problem of insufficient variation within urban areas; recall that in the regressions for the national sample the coefficient on women's schooling was significant in both fostering out regressions. TABLE 6: Tobit estimates, fostering by urban households Dependent GIRLSOUT GIRLSIN BOYSOUT BOYSIN Variable: (1) (2) (3) (4) Regressors T E(y)/ax # T aE(y)/ox # T aE(y)/8x 0 T aE(y)/ax MEN -0.322 (-1.25) -.0381 0.235 (2.41) .0458 -0.632 (-1.34) -.0432 0.445t (3.66) .0608 WOMEN -0.0649 (-0.29) -.00768 0.325t (3.30) .0633 -0.0477 (-0.16) -.00326 0.311t (3.21) .0425 GIRLS 0.298 (0.85) .0353 -0.334t (-2.93) -.0650 -0.323 (-0.80) -.0221 -0.365t (-2.88) -.0499 BOYS -0.0167 (-0.06) -.00198 -0.00186(-0.02) -.00362 0.166 (0.32) .0114 -0.134 (-1.11) -.0183 NOWOMAN -1.755* (-2.48) -.342 -0.979 (-1.57) -.134 SCHFEM -0.196 (-1.46) -.0232 -0.0202 (-0.45) -.00393 -0.410 (-1.58) -.0280 -0.0529 (-1.03) -.00723 PINCOME 0.0183 (0.02) .00217 1.275* (2.56) .248 0.791 (0.41) .0541 0.998$ (1.99) .136 HEADFEM -2.562 (-1.25) -.303 0.238 (0.74) .0463 -3.711 (-1.35) -.254 0.201 (0.56) .0275 ABIDJAN 0.329 (0.49) .0390 -0.722$ (-2.53) -.141 0.622 (0.61) .0425 -1.025' (-3.31) -.140 Constant -2.058 (-0.14) -18.865 (-2.82) -12.385 (-0.48) -15.745 (-2.32) a 3.489 (4.63) 1.982 (8.52) 4.996 (3.35) 1.960 (8.13) LOGL -219.63 492.59 -181.18 -389.11 ;b 13.90 50.82 20.78 65.72 N 328 689 326 689 Mean of 0.244 0.277 0.206 0.200 dep. var. tSignificant at .01 or less. $Significant at .05. ^Significant at .10. 'NOWOMAN is excluded from fostering out regressions because, by the definition used here, children cannot be fostered out from womanless households. 'Likelihood ratio test for joint significance of all coefficients, distributed i , where r = 8 in regressions (1) and (3) and r = 9 in regressions (2) and (4). - 32 - 5. TESTING THE SYMMETRY OF FOSTERING DECISIONS The Tobit models have allowed us to examine separately sending and receiving decisions. In this section, the model of friction first proposed by Rosett (1959) is adapted to jointly estimate both sides of the fostering market without restricting slope coefficients or intercepts to be the same. Rosett's model is a generalization of the basic Tobit model. Whereas the Tobit model allows for a mass point at one extreme of the conditional cumulative distribution of the dependent variable, Rosett allows the mass point to occur anywhere in the distribution. Rosett's application of the model was in analyzing the effect of changes in yield on asset holdings. He speculated that because of transactions costs, asset holdings might not respond to small changes in yield. There would thus be a range over which the dependent variable is insensitive to changes in the independent variables. Once this "friction" has been overcome, asset holdings respond to changes in yield, the direction of response depending on the sign of the change in yield. Rosett's original model allowed for different intercepts (representing differential friction) on the buying and selling sides of the asset market, but he assumed that the relation between yields and assets was the same on both sides once the friction is overcome. In the application to fostering, the Rosett model assumes that there is a range over which fostering in and out does not respond to small changes in the independent variables. The existence of friction in the fostering market can be explained by transactions costs, which may differ in magnitude between sending and receiving decisions. Further, if parents derive utility from the presence of their children, there will be addi- tional costs to cause friction on the sending side. Here, Rosett's model is adapted to allow both the intercepts and the slope coefficients to vary on the two sides of the market. The symmetry of coefficient estimates for fostering in and out can then be formally tested. Let yT be the desired net number of children fostered in, which can be positive or negative, and yi the observed number. When y# is in the negative range, the relation yt = a,+ Bi'xi + u., holds, and when yt is in the positive range, y* = a + B'jx, + u, holds, where a < 0 and a2 > 0 . When y+ lies between a, and a2 , no fostering is observed (yi = 0) ; when it lies below a, or above a2 we observe negative or positive net fostering, respectively. The random error i is normally distributed as N(0, o0) ; it is assumed to be the same on both sides of the market. (4.1) yi = y - a if y+ < at yi = 0 if al < Y1 < a2 Yi = yt - a2 if a2 < Y - 33 - The Rosett model with differential slopes and intercepts is illustrated in Figure 6 and is estimated using maximum likelihood methods." The dependent variables are the net number of children (NETKIDS), girls (NETGIRLS), and boys (NETBOYS), age 7-14 fostered in. Standard errors have been corrected to account for the fact that one of the regressors (PINCOME) is a predicted variable. Note that the model does not define an upper limit on the number of children fostered out. In order not to violate the constraint that households cannot foster out more children than they have, the model is estimated only for the sample of households with own children, sons, or daughters age 7-14. Estimates of the Rosett model for fostering of all children 7-14, girls 7-14, and boys 7-14 are qualitatively similar to single-censored Tobit estimates for the restricted sample of households with children in each category (see Appendix A). Our main interest in the Rosett model is to test the symmetry of coefficients on both sides of the market. If symmetry holds, the two sides of the fostering market should be exact images of each other: fostering in coefficients should be equal in sign and magnitude to fostering out coefficients. If transactions costs and other sources of friction are the same, then the intercepts on the two sides of the market should also be equal. The results of Wald tests for the equality of coefficients on the opposite sides of the market are presented in Table 7. The equality of the intercepts and coefficients on the two sides of the market is strongly rejected for all three regressions (test 1). The failure to reject the symmetry of coefficients for the net girls and net boys regressions separately may be due to sample size problems. The equality of all coefficients (test 2) is rejected for the net fostering of children 7-14; the Wald statistics for net fostering of boys and girls separately are not statistically significant, however. The equality of the intercepts cannot be rejected for any of the three regressions (test 3). The coefficients on the number of men and the number of women in the household are equal on the opposite sides of the market (test 4 and 5). The coefficients on the number of girls and the number of boys are not equal, however (tests 6 and 7): holding other variables constant, an additional child of either sex generates twice as much out-fostering as it reduces the demand for children fostered in. There are only a few additional cases of asymmetry among the remaining coefficients-the coefficient on female headship in the net boys regression and the west forest coefficient in the net boys and net children regressions. a The likelihood function has three parts (Maddala, 1983; Rosett, 1959): L(aq, a2, 0,, #2, a|y, x) = IIo[1/oj0[(yi + ah - P'x)/o] I(0[a2 - - 4l(a' - oA'x.)/orl II2[(yi + a2 - where the first product is over observations for which yi < 0 , the second for which yi = 0 , and the third for which ye > 0 . Analytical first derivatives were provided to the minimization program (see Appendix 5). Single-censored Tobit estimates were used as staring values for the regressions for boys and girls. For the combined sample of all children 7-14, coefficients were assigned starting values of 0 and the intercepts fl, and 0, were assigned values of-1 and 1, respectively. The stopping criterion was set at a change in function value of <.00000001. The functions for all children, girls, and boys converged in 35, 33, and 33 iterations, respectively. I am indebted to Dorothy Seavey for providing the FORTRAN program with analytical first derivatives. - 34 - FIGURE 6. The Rosett model of friction with differential slopes and intercepts y1 yi = -a2 2A + Ui E(y C1 a2 y =-al + PIN + u, - 35 - These results do not detract from the fact that certain variables--such as household income and regional dummy variables as proxies for relative price differences--have a significant impact on one side of the fostering market but not the other. The strong significance of 'both adult household composition variables and their equality across both sides of the market is supportive of the hypothesis that child fostering in both directions is in part motivated by the demand by adults for home services, services which children of this age group typically perform. The asymmetry in coefficients for the number of own girls and own boys suggests that other factors may also play into fostering decisions, however. It is also conceivable that the undercount of children fostered out is producing the asymmetric results: The number of own children (GIRLS, BOYS) is the sum of the number of own children at home and those fostered out. An undercount of the children fostered out will result in an undercount of own children for households on the fostering out side of the market. On the fostering in side, however, the count of the number of own children will be unaffected by underestimation of the value of the dependent variable, the number fostered in. TABLE 7: Wald tests for equality of intercepts and coefficients on opposite sides of the fostering market (r = number of restrictions) Wald Statistic Test NETKIDS NETGIRLS NETBOYS 1. Equality of all coefficients 289.83t 63.85t 209.01t and intercepts (r = 13) 2. Equality of all coefficients 31.33t 13.66 16.77 (r = 12) 3. Equality of intercepts (r = 1) 2.44 2.16 0.42 4. B (r = 1) 0.12 0.04 0.01 5. BWOMEN (r = 1) 0.03 0.21 0.51 6. B0u (r = 1) 5.84* 5.19* 0.90 7. fBovs (r = 1) 3.79A 0.10 3.07A 8. BSCHFm (r = 1) 0.89 1.42 0.64 9. BPINCOME (r = 1) 1.04 0.89 0.02 10. BHEDFEM (r = 1) 1.97 0.24 3.03A 11. BCrrON (r = 1) 1.30 0.89 2.44 12. flBsjAN (r = 1) 0.44 0.91 0.46 13. BrwKs (r = 1) 0.06 0.20 0.12 14. 8SAVANA (r = 1) 1.30 1.54 0.04 15. BwFoREST (r = 1) 3.34A 1.50 3.95* tSignificant at .01 or less. $Significant at .05. 'Significant at .10. *Wald test statistics are distributed ? , where r is the number of restrictions. - 36 - 6. CONCLUSIONS Household decisions on child fostering in C6te d'Ivoire are sensitive to economic factors on both sides of the fostering market. Different economic factors determine sending and receiving decisions, however, and there are important differences in the determinants of fostering for girls and boys, and in urban and rural areas. Multivariate analysis found significant economic determinants of fostering in children of both sexes that were consistent with a child labor explanation; there were also indications that excess child labor plays some role in fostering out decisions. Separate regressions for urban and rural households revealed that the determinants of fostering in decisions consistently support a child labor explanation in both areas, while there are no significant economic determinants of fostering out in urban areas and very few in rural areas. These results also suggested that foster girls and boys perform different tasks in rural areas and that sons and daughters in urban areas have different roles. Based on analysis of the activities of children, it was hypothesized that housework was the main activity provided by foster children. Wald tests rejected the symmetry of fostering decisions for children 7-14, implying unequal transactions costs and/or possibly different motivations behind sending and receiving decisions. Symmetry could not be rejected for separate regressions of girls and boys, but this result should not detract from the fact that in all regressions the significant determinants of the two sides of the market are different. The support for schooling investments as an important motivation for fostering out children 7-14 was inconclusive, due to the paucity of variables on the quality and availability of primary schooling. There probably is a group of households for which fostering out reflects schooling investments--sending households of foster children enrolled in secondary school and of those receiving outside support for schooling from their parents. There is a second group of foster children receiving no support from their parents but who are attending school with the support of the host household. The host is paying out about the same in school fees for these children as for their daughters, but far less than is spent on their sons. This would appear not to be an investment by the host household, but rather some type of compensation. It may still be considered an investment by the sending household, which is forgoing the value of the child's labor in the short run. Finally, there is a third group comprising 50-60 percent of foster girls and 30-40 percent of foster boys who are not attending school and for whom fostering decisions are clearly not linked to schooling investments. The parents of these children evidently felt that fostering their children was a better investment than enrolling them in local primary schools. - 37 - In a more general sense, the results indicate far more economic determinants of fostering on the demand side (fostering in) than on the supply side (fostering out). This is in sharp contrast to the emphasis in the demographic literature on fostering as a way for households to "manage" the costs of high fertility. Fostering out decisions are insensitive to household income and to the regional dummy variables that serve as proxies for price variations. There is no evidence, therefore, that children are fostered out because parents cannot afford them. On the other hand, there is evidence that fostering in decisions have an economic basis. In addition to the many significant economic determinants of fostering in, foster children perform more housework than own children and host households are making considerable outlays for the schooling and upkeep of foster children.  - 39 - APPENDIX: Rosett Model Estimates This Appendix presents Rosett model estimates of the fostering of all children 7-14, girls 7-14, and boys 7-14. For ease of comparison with earlier Tobit.results, the Rosett coefficients on the fostering out side of the market have been multiplied by -1 and are presented next to the coefficients for fostering in. Most of the qualitative differences between these results and the Tobit estimate. earlier can be explained by the fact that the Roset sample was restricted to households with own children in each category on both sides of the market. TABLE A. 1: Rosett model estimates, fostering of children 7-14* Dependent NETKIDS Variable: Fostering outh Fostering in Regressors 0 T $ T MEN -0.379t (-4.09) 0.418 (4.20) WOMEN -0.321t (-3.83) 0.339' (3.33) GIRLS 0.639t (6.86) 4.364t (-3.65) BOYS 0.453t (5.21) 0.247t (-2.62) SCHFEM -0.0895 (-1.58) 0.0281 (0.57) PINCOME -0.555 (-1.14) 1.135* (2.25) HEADFEM -2.005t (-2.90) 0.961t (2.81) COTTON -0.855* (-2.02) 0.234 (0.53) ABIDJAN 0.783 (1.53) -1.178* (-2.31) TOWNS 0.209 (0.52) -0.326 (-0.82) SAVANNA 0.243 (0.70) 4.739* (-2.10) WFOREST 0.185 (0.60) .890t (-2.73) Constant 5.467 (0.90) -16.705 (-2.59) a 2.191 (24.48) N" 1101 Mean of -0.00636 dep. var S.D. 1.182 'Significant at .01 or less. *Significant at .05. 'Unconditional expected value coefficients are given by: SB(y)/I - pli, + AN(l - 4) , whoe AV and A are coefficients on regressor iS for fostering out and fostering in, respectively, and 01 and 2 ae the cumulative normal distribution functions of [(al - Olx)/a] and [(a% - Ax)/ol , respectively, evaluated at sample means and coefficient estimates. The value of 4, is 0.204 and of (1 - #2) is 0.1927. 'IFor case of comparison with Tobit results, coefficients on the fostering out side have been multiplied by -1. 'Value of function at minimum. 'Likelihood ratio test for joint significance of all coefficients, distributed xL. OThe regression sample is all households with at least one child age 7-14 (i.e., GIRLS + BOYS > 0). -40- TABLE A.2: Rosett model estimates, fostering of boys and girls 7-14" Dependent NETGIRLS NETBOYS Variable: Fostering outh Fostering in Fostering oute Fostering in Regressors & T 4 T 4 T T MEN -0.216t (-2.33) 0.192t (1.99) -0.332t (-2.65) 0.317t (2.85) WOMEN -0.223$ (-2.56) 0.268t (3.31) -0.303t (-2.81) 0.216* (2.38) GIRLS 0.659t (5.54) -0.317* (-2.43) -0.0174 (-0.17) -0.124 (-1.02) BOYS 0.0571 (0.61) -0.0209 (-0.23) 0.596t (4.64) -0.296* (-2.20) SCHFEM -0.0899 (-1.56) 0.00832 (0.18) -0.106 (-1.38) 0.0314 (0.46) PINCOME -0.441 (-0.98) 0.975* (2.16) -0.365 (-0.59) 0.467 (0.93) HEADFEM -1.846^ (-1.91) 1.353t (3.40) -2.105* (-2.16) 0.161 (0.27) COTTON -0.992* (-2.32) 0.372 (0.66) -0.950^ (-1.71) -0.229 (-0.39) ABIDJAN 0.498 (1.00) -1.088* (-2.24) 0.632 (0.97) -1.165* (-2.05) TOWNS 0.0449 (0.12) -0.265 (-0.67) 0.0561 (0.11) 0.147 (0.36) SAVANNA 0.245 (0.71) -0.941^ (-1.90) 0.173 (0.41) -0.277 (-0.65) WFOREST 0.192 (0.64) -0.729^ (-1.88) -0.394 (-1.01) -0.620 (-1.57) Constant 3.948 (0.70) -14.635 (-2.50) 3.050 (0.39) -8.819 (-1.36) a 1.886 (9.31) 2.145 (16.33) LOGL 967.57 893.18 X2 d 129.66 113.32 N* 816 844 Mean of -0.110 -0.034 dep. var. S.D. 0.814 0.799 'Significant at .01 or less. tSignificant at .05. ^Significant at .10. aUnconditional expected value coefficients are given by: 8E(y)/ax, = 4P, + %(1 - 4) where 61 and fl are coefficients on regressor x, for fostering out and fostering in, respectively, and It and 12 are the cumulative normal distribution functions of [(a, - O,x)/a and [(a2 - 02x)/o] , respectively, evaluated at sample means and coefficient estimates. The value of 4W, is 0.1994 and of (1 - i2 is 0.1154 in the regression for girls; the value of 4I1 is 0.1357 and (1 - 42) is 0.1115 in the regression for boys. "For case of comparison with Tobit results, coefficients on the fostering-out side have been multiplied by -1. 'Value of function at minimum. "Likelihood ratio test for joint significance of all coefficients, distributed as X2 . The sample for the net girls regression is all households with at least one daughter age 7-14 (i.e., GIRLS > 0) and the sample for the net boys regression is all households with at least one son age 7-14 (BOYS > 0). -41 - REFERENCES Ainsworth, Martha (1983). "The Demand for Health and Schooling in Mali: Results of the Community and Service Provider Survey." Country Policy Department Discussion Paper, No. 1983-7. Washington, D.C.: The World Bank. (1989a). "Socioeconomic Determinants of Fertility in Cte d'livoire." Livin Standards Measurementu Working Paper, No. 53. Washington D.C.: The World Bank (1989b). "The Demand for Children in C6te d'Ivoire: Economic Aspects of Fertility and Child Fostering." Ph.D. Dissertation in Economics, Yale University. and Juan Muftoz (1986). "The C6te d'Ivoire Living Standards Survey: Design and Implementation." Living Standards Measurement Study Working Paper, No. 26. Washington, D.C.: The World Bank. Amemiya, Takeshi (1984). "Tobit Models: A Survey." Journal of Econometrics 24: 3-61. Antoine, Philippe and Agnes Guillaume (1986). "Une Expression de Ia Solidarite Familiale & Abidjan: Enfmnts du Couple et Enfants Confies" in Association Internationale des Demographes do Langue Prangaise (AIDELF). Les Famllies d'aujourd'hui: Demographie et evolution recente des comportementfsimillams. Colloque de Genbve (17-20 septembre 1984). Becker, Gary S. (1965). "A Theory of the Allocation of Time." Economic Journal 75: 493-517. Ben-Porath, Yoram (1986). "The F-Connection: Families, Friends, and Firms and the Organization of ExchAnge." Population and Development Review 6: 1-30. Bledsoe, Caroline H. (1989). "The Social Management of Fertility: Child Fosterage among the Mends of Sierra Leon" in W. Penn Handwerker, ed., Births and Power. The Politics of Reproduction. Boulder, Colo.: Westview Press. and Uche C. Isiugo-Abanihe (1989). "Strategies of Child Fosterage among Mende Grannies in Sierra Leone" in Ronald Lesthaeghe, ed., African Reproduction and Social Organization. Berkeley: University of California Press. Brydon, Lynne (1979). "Women at Work: Some Changes in Family Structure in Amedzofe-Avatime, Ghana." 4fa 49(2): 97-111. (1985). "The Avatime Family and Circulation, 1900-1977" in Prothero and Chapman (ods), Circlatilon In Third World Countries. London: Routledge and Kegan Paul. Cain, Mead T. (1977). "The Economic Activities of Children in a Village in Bangladesh." Population and Development Review 3: 201-227. Cameron, A. C. and P. K. Trivedi (1986). "Econometric Models Based on Count Data: Comparisons and Application of Some Estimators and Tests." Journal of Applied Econometrics 1: 29-53. Chernichovsky, Dov (1985). "Socioeconomic and Demographic Aspects of School Enrollment and Attendance in Rural Botswana." Economic Development and Cultural Change 33(2). , Robert E. B. Lucas, and Eva Mueller (1985). "The Household Economy of Rural Botswana: An African Case." World Bank Staff Working Paper, No. 715. Washington, D.C.: The World Bank. Da Vanzo, Julie (1972). "The Determinants of Family Formation in Chile, 1960." Santa Monica: Rand Corporation. Report #R-830-AID. August. Deaton, Angus and John Muellbauer (1980). Economics and Consumer Behavior. Cambridge: Cambridge University Press. - 42 - Etienne, Mona (1979a). "Maternite Sociale, Rapports d'Adoption et Pouvoir des Femmes chez le Baoule (C6te d'Ivoire)." L'Homme 19, nos. 3-4: 63-107. (1979b). "The Case for Social Maternity: Adoption of Children by Urban Baule Women." Dialectical Anthropology 4(3). Fiawoo, D.R. (1978). "Some Patterns of Foster Care in Ghana" in Oppong et al (eds), Marriage, Fertility and Parenthood in West Africa. Canberra: Australian University Press. Findley, Sally E. and Assitan Diallo (1988). "Foster Children: Links Between Urban and Rural Families?" in International Union for the Scientific Study of Population (IUSSP), African Population Conference, Vol. 2, Dakar, November 7-12. Frank, Odile (1984). "Child Fostering in Sub-Saharan Africa." Paper presented at the Annual Meeting of the Population Association of America, Minneapolis, Minn. Goody, Esther N. (1982). Parenthood and Social Reproduction: Fostering and Occupational Roles in West Africa. Cambridge, England: Cambridge University Press. Gould, W.T.S. (1985). "Circulation and Schooling in East Africa" in Prothero and Chapman (eds), Circulation in Third World Countries. London: Routledge and Paul. Greene, William H. (1985). LIMDEP7m. Department of Economics, Graduate School of Business Administration, New York University. Gronau, Reuben (1973). "The Intrafamily Allocation of Time: The Value of the Housewives' Time." American Economic Review 63(4): 634-651. Grootaert, Christiaan (1986). "Measuring and Analyzing Levels of Living in Developing Countries: An Annotated Questionnaire." Living Standards Measurement Study Working Paper, No. 24. Washington, D.C.: The World Bank. Hausman, J. A., B. Hall, and Z. Griliches (1984). "Econometric Models for Count Data with an Application to the Patent-R & D Relationship." Econometrica 52: 909-938. Isiugo-Abanihe, Uche, C. (1983). "Child Fostering in West Africa: Prevalence, Determinants and Demographic Consequences." Ph.D. Dissertation in Demography, University of Pennsylvania. (1985). "Child Fosterage in West Africa." Population and Development Review 11(1): 53-73. Issac, Barry L. and Shelby R. Conrad (1982). "Child Fosterage among the Mende of Upper Bambara Chiefdom, Sierra Leone: Rural-Urban and Occupational Comparisons." Ethnology 21: 243-257. Kozel, Valerie (1987). "Income, Consumption and Savings Behavior in the C6te d'lvoire." Ph.D. Thesis, Department of Urban Studies and Planning, Massachusetts Institute of Technology. Levy, Victor (1985). "Cropping Pattern, Mechanization, Child Labor, and Fertility Behavior in a Farming Economy: Rural Egypt." Economic Development and Cultural Change 33(4): 777-791. Maddala, G. S. (1983). Limited Dependent and Qualitative Variables in Econometrics. Cambridge: Cambridge University Press. Mueller, Eva (1976). "The Economic Value of Children in Peasant Agriculture" in Ronald G. Ridker, ed., Population and Development. Baltimore: Johns Hopkins University Press. (1985). "The Value and Allocation of Time in Rural Botswana" in D. Chernichovsky, Robert E. B. Lucas and Eva Mueller, "The Household Economy of Rural Botswana." World Bank Staff Working Paper, No. 715. Washington, D.C.: The World Bank. - 43 - Murphy, Kevin M. and Robert H. Topel (1985). "Estimation and Inference in Two-Step Econometric Models." Journal of Business and Economic Statistics 3(4): 370-379. Oppong, Christine and Wolf Bleek (1982). "Economic Models and Having Children: Some Evidence from Kwahu, Ghana." Africa 52(4). Page, Hilary (1986). "Child-bearing versus Child-rearing: Co-residence of Mothers and Children in Sub-Saharan Africa." Interuniversity Programme in Demography, IPD Working Paper No. 1986-2. Paper presented at the Annual Meeting of the Population Association of America, April 1986. Pollak, Robert A. (1985). "A Transaction Cost Approach to Families and Households." Journal of Economic Literature 23: 581-608. and Michael L. Wachter (1975). "The Relevance of the Household Production Function and Its Implications for the Allocation of Time." Journal of Political Economy 83 (Vol. 2): 255-277. Rosenzweig, Mark R. (1977). "The Demand for Children in Farm Households." Journal of Political Economy 85(1): 123-146. and Robert Evenson (1977). "Fertility, Schooling and the Economic Contribution of Children in Rural India: An Econometric Analysis." Econometrica 45(5): 1065-1079. and Oded Stark (1988). "Consumption Smoothing, Migration and Marriage: Evidence from Rural India." University of Minnesota and Harvard University. Mimeo. May. Rosett, Richard N. (1959). "A Statistical Model of Friction in Economics." Econoinetrica 27: 263-267. Saint-Vil, J. (1981). "Migrations Scolaires et Urbanisation en C6te d'Ivoire." Cahiers d'Outre Mer 34(133): 25-41. Schildkrout, Enid (1973). "The Fostering of Children in Urban Ghana: Problems of Ethnographic Analysis in a Multi-cultural Context." Urban Anthropology 2(1). Singh, Inderjit, Lyn Squire, and John Strauss, eds. (1986). Agricultural Household Models: Extensions, Applications and Policy. Baltimore: Johns Hopkins University Press. Smith, Richard J. and Richard W. Blundell (1986). "An Exogeneity Test for a Simultaneous Equation Tobit Model with an Application to Labor Supply." Econometrica 54(3): 679-685. Strauss, John (1986). "The Theory and Comparative Statics of Agricultural Household Models: A General Approach" in Singh, Squire and Strauss, eds., Agricultural Household Models: Extensions, Applications and Policy. Baltimore: Johns Hopkins University Press. White, Benjamin (1975). "The Economic Importance of Children in a Javanese Village" in Moni Nag, ed., Population and Social Organization. The Hague: Moulton. World Bank (1987). World Development Report 1987. New York: Oxford University Press. Distributors of World Bank Publications ARGENTINA EL SALVADOR JAPAN SOUTH AFRICA, BOTSWANA Carlos Hirsch, SRL Fusades Eastern Book Service Forsingle tiw Galeria Guemes Alam Dr. Manuel Erique Araujo #3530 Hongo 3-Chome, Bunkyo-ku 113 Oxford University Pross Florida 165,4th Floor-Ofe. 453/465 EdifOic SISA, ler. Piso Tokyo - Southern Africs 1333 Buenos Aires San Salvador 011 P.O. Box 1141 KENYA Cape Town 8000 AUSTRALIA, PAPUA NEW GUINEA, FINLAND Africa Book Service (F.A.) Ltd. FIJI, SOLOMON ISLANDS, Akateeminenr Kirjakauppa Qmaran House, Mfangano Street For suiscrptimw arders: VANUATU, AND WESTERN SAMOA P.O. Box 128 P.O. Box 45245 International Subscription Service D.A. Books & Journals SF-00101 Helsinki 10 Nairobi P.O. Box 41095 648 Whitehorse Road Craighall Mitcham3132 FRANCE KOREA, REPUBLIC OF Johannesburg2024 Victoria World Bank Publications Pan Korea Book Corporation 66, avenue d4na P.O. Box 101, Kwangwhamun SPAIN AUSTRIA 75116 Paris Seoul Mundi-Prenam Ubros, SA. Gerold and Co. Castello37 Graben 31 GERMANY MALAYSIA 2O001Madrid A-1011 Wien . UNO-Verlag University of Malaya Cooperative Peppius.ra Bookshop,umited Lbredia InternacionalAEDOS BANGLADESH D-5300 Bonn 1 P.O. Box 1127, Jalan Pantal Baru Consell de Cent, 391 Micro Industries Development 59700 Kuala Lumpur 08009 Barcelona Assistance Society (MIDAS) GUATEMALA HouseS, Road 16 Librerias PiedraSanta MEXICO SRI LANKA AND THE MALDIVES Dhanmondi R/Area Sa. Calle 7-55 INPOTEC Lake House Bookshop Dhaka 1209 Zona 1 Aparlado Postal 224160 P.O. Box 244 Guatemala City 14060 Tlalpan, Mexico D.. 100, SIr ChittampalamA. Breec rcarn Gardiner Mawatha 156, Nur Ahmed Sarak HONG KONG, MACAO NETHERLANDS Colombo2 Chittagong4000 Asia 2000 Ltd. De Lindeboom/InOr-Publikaties 46-48 Wyndham Street P.O. Box 202 SWEDEN 76, ID.A. Avenue Winning Centre 7480 AE Haaksbergen Forsingle titie Kulna 9100 2nd Floor Fritzes Fackbokstoretlaget Central long Kong NEW ZEALAND Regeringsgatan 12, Box 1356 BELGIUM - EASCONZ Ltd. S-10327Stockholm Jean De Lannoy . INDIA Private Mail Bag 99914 Av. du Roi 202 Allied Publishers Private Ltd. New Market For sucrption orders: 1060 Brusses 751 Mount Road Auckland Wennergren-Willms AB Madras- 600 002 Box 30004 CANADA NIGERIA S-104 25 Stockholm La Diffuseur Branch Offin* University Pres Limited C.P. 85, 15018 rue Amphre i J.N. Heredia Mug Three Crowns BuildingJericho SWITZERLAND Boucherville, Qudbec Ballard Estate Private Mail Bag 5095 orosingle fisr J4B5E6 Bombay -400038 lbadan Librairie Payot 1, rue de Bourg CHINA 13/14 Asaf Ali Road NORWAY CH 1002 Lausanne China Financial & Economic New Delhi- 110 002 Narvesen Information Center Publishing House Book Department Forsukcriptionodesw 8, Da Fo Si Dong Jie 17 Chittaranjan Avenue P.O. Box 6125 Etterstad Libraiwe Payot Beijing Calcutta -700 072 N-060 Oslo 6 Service des Abonemente Case posiale 3312 COLOMBIA Jayadeva Hostel Building PAKISTAN CH 100M Lausanne Infoenlace Ltda. 5th Main Road Candhinagar Mirza Book Agency Apartado Aereo 34270 Bangalore -560009 65, Shahrah-eQuaid-*-AzaM TANZANIA Bogota D.E. P.O. Box No. 729 Oxford University Pres 3-5-1129 Kachiguda Cross Road Lahore 54000 P.O. Box 5299 COTE D'IVOIRE Hyderabad - 500 027 MaktAba Road Centre d'Edition et de Diffusion PERU Dar ts Salaam Africalnes (CEDA) Prarthana Plats, 2nd Floor Editorial Desarrollo SA 04 B.P. 541 Near Thakore Baug, Navrangpurs Apartado3824 THAILAND Abidjan 04 Plateau Ahmedabad -380 009 Lima 1 Central Department Store 306 Silom Road CYPRUS Patiala House PHILIPPINES Bangkok Cyprus College Bookstore 16-A Ashok Mar; International Book Center 6, Diogenes Street, Engomi Lucknow- 226 001 PIfth Floor, Filipinas Life Building TRINIDAD & TOBAGO, ANTIGUA P.O. Box 2006 Ayala Avenue, Makati BARBUDA, BARBADOS, Nicosia Central Bazaar Road Metro Manis DOMINICA, GRENADA, GUYANA, 60 BajajNagar JAMAICA, MONTSERRAT, ST. DENMARK Nagpur 4400 POLAND KITTS & NEVIS, ST. LUCIA, SamfundsUtteratur ORPAN ST. VINCENT & GRENADINES Rosenoerns Al16 11 INDONESIA Palac Kultury I Nauli Systematic$ Studies Unit DK*1970 FrederiksbergC Pt Indira Limited 00-901 Warzawa 49 Watts Street JL Sam Ratulangi 37 Curepe DOMINICAN REPUBLIC P.O. Box 181 PORTUGAL Trinidad, West Indies Editors Taller, C. por A. Jakarta Pusat Livraria Portugal Restauraci6n a Isabel la Cat6lica 309 Rua Do Carmo 70.74 UNITED KINGDOM Apartado de Coreas 2190 Z-1 ISRAEL 1200 Usbon Microinfo Ltd. Santo Domingo Yozmot Literature Ltd. P.O. Box 3 P.O. Box 56055 SAUDI ARABIA, QATAR Alton, Hampshire CU34 2PG EGYPT, ARAB REPUBLIC OF Tel Aviv 61560 Jarir Book Store England Al Ahram Israel P.O. Box 3196 Al Galas Street Riyadh 11471 VENEZUELA Cairo ITALY Libreria del Este Licosa Commissionaria Sansont SPA SINGAPORE, TAIWAN, Aptdo. 60.337 The Middle EastObeerver Via Duca Di Calabria, 1/1 MYANMAR,BRUNEI Caracas 1060-A 41, Sherif Street Casella Postale 552 Information Publications Cairo 50125 Firenze Private, Ltd. 02-06 1st Fl., Pei-Fu Industrial Bldg. 24 New Industrial Road Singapore 1953 LSMS Working Papers (continued) No. 55 Rigidt des salaires: Donnies microionomiques et macro6onomiques sur I'ajustement du marchg du travail dans le secteur moderne (in French only) No. 56 The Poor in Latin America during Adjustment: A Case Study of Peru No. 57 The Substitutability of Public and Private Health Care for the Treatment of Children in Pakistan No. 58 Identifying the Poor Is "Headship" a Useful Concept? No. 59 Labor Market Perfornance as a Determinant ofMigration No. 60 The Relative Effectiveness of Private and Public Schools; Evidencefrom Two Developing Countries No. 61 Large Sample Distribution of Several Inequality Measures: With Application to C6te d7voire No. 62 Testingfor Significance of Poverty Differences: With Application to Cbte d7voire No. 63 Poverty and Economic Growth: With Application to C6te d7voire No. 64 Education and Earnings in Peru's Informal Nonfarm Family Enterprises No. 65 Formal and Informal Sector Wage Determination in Urban Low-Income Neighborhoods in Pakistan No. 66 Testingfor Labor Market Duality: The Private Wage Sector in C6te d'voire No. 67 Does Education Pay in the Labor Market? The Labor Force Participation, Occupation, and Earnings of Peruvian Women No. 68 The Composition and Distribution ofIncome in C6te d'voire No. 69 Price Elasticities from Survey Data: Extensions and Indonesian Results No. 70 Efficient Allocation of Transfers to the Poor The Problem of Unobserved Household Income No. 71 Investigating the Determinants of Household Welfare in Cbte d7voire No. 72 The Selectivity of Fertility and the Determinants of Human Capital Investments: Parametric and Semiparametric Estimates No. 73 Shadow Wages and Peasant Family Labor Supply: An Econometric Application to the Peruvian Sierra No. 74 The Action of Human Resources and Poverty on One Another: What We Have Yet to Learn No. 75 The Distribution of Welfare in Ghana, 1987-88 No. 76 Schooling, Skills, and the Returns to Government Investment in Education: An Exploration Using Data from Ghana No. 77 Workers'Benefits from Bolivia's Emergency Social Fund No. 78 Dual Selection Criteria with Multiple Alternatives: Migration, Work Status, and Wages No. 79 Gender Differences in Household Resource Allocations No. 80 The Household Survey as a Tool for Policy Change: Lessons from the Jamaican Survey of Living Conditions No. 81 Patterns of Aging in Thailand and C6te d7voire No. 82 Does Undernutrition Respond to Incomes and Prices? Dominance Tests for Indonesia No. 83 Growth and Redistribution Components of Changes in Poverty Measure: A Decomposition with Applications to Brazil and India in the 1980s No. 84 Measuring Income from Family Enterprises with Household Surveys No. 85 Demand Analysis and Tax Reform in Pakistan No. 86 Poverty and Inequality during Unorthodox Adjustment: The Case of Peru, 1985-90 No. 87 Family Productivity, Labor Supply, and Welfare in a Low-Income Country No. 88 Poverty Comparisons: A Guide to Concepts and Methods No. 89 Public Policy and Anthropometric Outcomes in C6te d7voire No. 90 Measuring the Impact of Fatal Adult Illness in Sub-Saharan Africa: An Annotated Household Questionnaire No. 91 Estimating the Determinants of Cognitive Achievement in Low-Income Countries: The Case of Ghana The World Bank Headquarters European Office Tokyo Office 1818 H Street, N.W. 66, avenue d'I6na Kokusai Building o Washington, D.C. 20433, U.S.A. 75116 Paris, France 1-1, Marunouchi 3-chome Chiyoda-ku, Tokyo 100, Japan Telephone: (202) 477-1234 Telephone: (1) 40.69.30.00 Facsimile: (202) 477-6391 Facsimile: (1) 40.69.30.66 Telephone: (3) 3214-5001 Telex: wul 64145 WORLDBANK Telex: 640651 Facsimile: (3) 3214-3657 RCA 248423 WORLDBK Telex: 26838 Cable Address: INTBAFRAD WASHINGTONDC