LSM- 53 P. JULY 1989 Living Standards Measurement Study Working Paper No. 53 Socioeconomic Determinants of Fertility in CBte dIvoire LSMS Working Papers No. 1 Living Standards Surveys in Developing Countries No. 2 Poverty and Living Standards in Asia: An Overview of the Main Results and Lessons of Selected Household Surveys No. 3 Measuring Levels of Living in Latin America: An Overview of Main Problems No. 4 Towards More Effective Measurement of Levels of Living, and Review of Work of the United Nations Statistical Office (UNSO) Related to Statistics of Levels of Living No. 5 Conducting Surveys in Developing Countries: Practical Problems and Experience in Brazil, Malaysia,and the Philippines No. 6 Household Survey Experience in Africa No. 7 Measurement of Welfare: Theory and Practical Guidelines No. 8 Employment Data for the Measurement of Living Standards No. 9 Income and Expenditure Surveys in Developing Countries: Sample Design and Execution No. 10 Reflections on the LSMS Group Meeting No. 11 Three Essays on a Sri Lanka Household Survey No. 12 The ECIEL Study of Household Income and Consumption in Urban Latin America: An Analytical History No. 13 Nutrition and Health Status Indicators: Suggestions for Surveys of the Standard of Living in Developing Countries No. 14 Child Schooling and the Measurement of Living Standards No. 15 Measuring Health as a Component of Living Standards No. 16 Procedures for Collecting and Analyzing Mortality Data in LSMS No. 17 The Labor Market and Social Accounting: A Framework of Data Presentation No. 18 Time Use Data and the Living Standards Measurement Study No. 19 The Conceptual Basis of Measures of Household Welfare and Their Implied Survey Data Requirements 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 Ulrban Housing in the Ivory Coast No. 26 The Cote d'lvoire 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 Cote 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 Cote 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 Cote d'Ivoire (List continues on the inside back cover) Socioeconomic Determinants of Fertility in Cote 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 household 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 LSMS is working to develop new methods to monitor progress in raising levels of living, to identify the consequences for households of past and proposed government policies, and to improve communications between survey statisticians, analysts, and policymakers. The LSMS Working Paper series was started to disseminate intermediate products from the LSMS. Publications in the series include critical surveys covering different aspects of the LSMS data collection program and reports on improved methodologies for using Living Standards Survey (LSS) data. More recent publications 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 53 Socioeconomic Determinants of Fertility in Cote d'Ivoire Martha Ainsworth The World Bank Washington, D.C. Copyright © 1989 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 July 1989 This is a working paper published informally by the World Bank. To present the results of research with the least possible delay, the typescript 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. 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 Director, Publications Department, 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 photocopy portions for classroom use is not required, though notification of such use having been made will be appreciated. The complete backlist of publications from the World Bank is shown in the annual Index of Publications, which contains an alphabetical title list and indexes of subjects, authors, and countries and regions; it is of value principally to libraries and institutional purchasers. The latest edition is available free of charge from the Publications Sales Unit, 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'1ena, 75116 Paris, France. Martha Ainsworth is a consultant to the Welfare and Human Resources Division of the World Bank's Population and Human Resources Department. Library of Congress Cataloging-in-Publication Data Ainsworth, Martha, 1955- Socioeconomic determinants of fertility in Cote d'Ivoire. (LSMS working paper, ISSN 0253-4517 ; no. 53) Bibliography: p. 1. Fertility, Human--Ivory Coast. 2. Income--Ivory Coast. 3. Women--Education--Ivory Coast. I. Title. II. Series. HB1076.A3A45 1989 304.6'32'096668 89-16412 ISBN 0-8213-1256-1 v ABSTRACT This paper examines the Impact of schooling and income on fertility in Cote d'Ivoire using data from the 1985 C6te d'Ivoire Living Standards Survey. The first part presents graphically the correlations between fertility and area of residence, female schooling and household income. The second part estimates a reduced form equation in which the number of children ever born is regressed on the mother's age and schooling, the location of the household and household income variables. This equation is estimated using ordinary least squares (OLS), maximum likelihood Tobit and a Poisson count model. For the entire sample, female schooling lowers fertility, while household income raises it. Among the subsample of urban women, only the negative effect of schooling is observed; among the subsample of rural women only the positive effect of household income is observed. The absence of a schooling effect among rural women is attributed in part to the low proportion of women with any schooling. When the sample is broken into three age cohorts, the negative effect of schooling on fertility is observed for the youngest and middle cohorts (ages 15-24 and 25-34, respectively), while the positive effect of income is observed for the middle and oldest cohorts (25-34 and 35+, respectively). This suggests that a fertility decline may be underway among young educated women. Experimentation with different specifications of the schooling variable shows that schooling has a negative effect on fertility even during the early primary years, although the negative effect of secondary schooling is even greater. - vi - The robustness of the results to the choice of Income variable Is also examined. Three income measures are used: the value of household consumption per adult (a proxy for permanent income); household income per adult; and household nonlabor income per adult. Results were most robust for the permanent income measure, less so for current income and insig- nificant for nonlabor income. Exclusion of all income variables from the fertility regression lowers the coefficient on schooling and, for rural women, renders it insignificant. - vii - ACKNOWLEDGM1ENTS This research was supported by the Welfare and Human Resources Division of the Population and Human Resources Department of the World Bank. The opinions expressed are those of the author and do not reflect policy of the World Bank or its members. Comments on an earlier draft from T. Paul Schultz, John Strauss, Duncan Thomas and Vassilis Hajivassillou of the Economic Growth Center, Yale University, and Dennis DeTray and John Newman of the World Bank are gratefully acknowledged. 11 ~~~~~~~ I .. I I X1 1 '' .. I I h ~ l - ix - TABLE OF CONTENTS Introduction .1.......................................... I. Socioeconomic correlates of fertility .............. 3 Fertility and area of residence ................ 3 Fertility and schooling ........................ 6 Fertility and income ........................... 8 II. Multivariate analysis of fertility ................ 13 A model of the determinants of fertility ... 13 The reduced form .............................. 16 Estimation techniques . . 21 Estimation results . . 24 1. Determinants of fertility ................. 24 2. Choice of income variable ................. 30 3. Specification of schooling ................ 33 Conclusions ........................................... 35 Footnotes ............................................. 36 Annex A: Data underlying figures 1-5 ................. 40 Annex B: Comparison of CILSS fertility results and other demographic surveys ............... 44 Sources ............................................... 50 x T A B L E S 1 Distribution of women by marital status ......... 4 2 Sample means and standard deviations ........... 20 3 Estimates by location .......................... 25 4 Estimates by age group ....................... 27 5 Point elasticities ....................... 29 6 Sensitivity of results to income specification ............... 31 7 Specifications with and without income ......... 32 8 Various schooling specifications, all women .... 34 Annex A: Data underlying figures 1 - 5 Al Mean children ever born by woman's age, location and schooling .................... 40 A2 Mean children ever born by woman's age and expenditure per adult tercile ......... 41 A3 Proportion of women never married, by woman's age, location and schooling ....... 42 A4 Mean age at first cohabitation, ever married or cohabited women, by woman's age, location and schooling .................... 43 Annex B: Comparison of CILSS fertility results with other demographic surveys BI Distribution of samples by age group, all women 15-50 ............................... 46 B2 Proportion of women never married ............. 47 B3 Distribution of samples by schooling .......... 48 B4 Mean children ever born ....................... 49 FIGURES 1 Children ever born by age and location .......... 5 2 Children ever born by age and schooling .........7 3 Proportion of women never married by age and level of schooling ..................... 9 4 Mean age at first cohabitation by current age and schooling ......................... 10 5 Children ever born by age and per adult expenditure tercile ....................... 12 INTRODUCTION Subsaharan Africa is the poorest region of the world and the region with the highest birth rates. Whereas fertility is declining in every other developing region, there are no documented cases of national fertility decline in Subsaharan Africa, with the possible exception of Zimbabwe (World Bank 1986). Studies of the determinants of fertility in other developing regions have generally found that female schooling depresses fertility, with the effect of income more ambiguous (cf. Cochrane 1979, T.P. Schultz 1974, 1981). Studies of the economic determinants of African fertility have been limited by the lack of household data sets with adequate demographic and economic information. This paper examines the likely impact of the spread of schooling and rising incomes on fertility in Cote d'Ivoire with data from the 1985 C6te d'Ivoire Living Standards Survey. Per capita income in C8te d'Ivoire was $660 in 1985, making it a relatively prosperous country by African standards (World Bank 1987). GDP grew by an average 6.8 percent per year between 1965-80 but declined at a rate of 1.7 percent per year from 1980-85. The population of 10 million is growing at 3.8 percent per year, the combined effect of a high rate of natural increase (3.1 percent annually) and immigration from neighboring countries. Roughly 30 percent of the population is foreign born and about 40 percent of the total population resides in urban areas. Schooling has spread rapidly since independence, but lags behind other Subsaharan countries with similar incomes. The Ivorian gross primary enrollment ratio for females is only 63 percent, -2- compared to 94 percent for Kenya, 97 percent for Cameroon and 127 percent for Zimbabwe.' Despite rapid economic growth, high urbanization and the spread of schooling, the average woman in C6te d'Ivoire has 6-7 children by the end of her childbearing years. The 1980-81 Ivorian Fertility Survey found that only 3.8 percent of currently married, fecund women were using any method of family planning and only 0.6 percent were using an effective method such as the pill or IUD (R.C.I. 1984). The government has been pronatalist since independence and there are virtually no family planning services available except from private sources in the largest city, Abidjan. This paper is organized into two parts. The first part provides descriptive analysis of the separate relationships between fertility and three main economic correlates: female schooling, household income and area of residence. The second part uses multivariate techniques to estimate a reduced form equation of the determinants of fertility. The results show that female schooling reduces fertility and household income raises it, when both variables are included in a regression with controls for the woman's age and area of residence. Relationships among subgroups of the population and the sensitivity of results to the choice of income variable and specification of schooling are also examined. An annex to the paper compares the CILSS fertility results with the results of other demographic surveys of C6te d'Ivoire. -3- PART I: SOCIOECONOMIC CORRELATES OF FERTILITY The information presented here is based on interviews of 1468 women by the C8te d'Ivoire Living Standards Survey (CILSS) between February 1985 and January 1986. CILSS interviews 1600 households annually, spread out over 12 months. It obtains a complete list of live births from one woman age 15 or older randomly selected In each household. It also collects detailed consumption and income information at the household level and socioeconomic data on all household members. The survey methodology is documented in Ainsworth and Mufnoz (1986) and Grootaert (1986). Of the 1599 households surveyed the first year of the survey, 1488 had women age 15 or older. Twenty women have been dropped from the analysis because of inconsistent fertility records. The average woman in the sample is 34 years of age with a mean of 3.9 children.2 Seventy-two percent of the women are currently married, 14 percent have never been married and 14 percent are divorced, widowed or separated. Table 1 presents the distribution of women in each 5-year age cohort by marital status. By the early twenties, about three-quarters of the women have been married and by the age of 25, about 95 percent have been married at least once. Fertility and area of residence The measure of fertility studied here is "children ever born", which is the number of live births a woman has had during her lifetime. Miscarriages and stillbirths are excluded, while children born alive who - 4 - TAMLE 1: Distribution of women by marital status (proportions) AGE GROUP MARITAL -----------------------------------________________ STATUS 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50+ TOTAL CURRENTLY .361 .671 .853 .917 .868 .920 .829 .634 .715 MARRIED DIV/WID/ .100 .085 .094 .059 .104 .080 .162 .366 .142 SEP NEVER .538 .244 .052 .024 .028 .000 .009 .000 .143 MARRIED TOTAL 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 (n) (249) (234) (191) (169) (144) (113) (111) (257) (1468) Note: Divorced, widowed, and separated includes women who have ever cohabited but are not currently living with a man. Never married includes women who have never married or cohabited. subsequently died are included. The women in the sample live in three areas: Abidjan (the largest city, 20 percent), other urban areas (22 percent) and rural areas (58 percent). Figure 1 shows the relation between children ever born and the area of residence. For most age groups, fertilty is higher in rural areas than in urban areas and, among urban areas, is lowest in Abidjan.3 This relation may be observed because urban areas provide more employment opportunities for women, raising the opportunity cost of children, or because of underlying differences in female schooling and income across areas. FIGURE 1: Children ever born by age and location Children ever born 10 8.1 8- 7. 0 6 ~~~~~~~~~~~~ ~~6.1 6.1 6.0 6.2 15-19 20-24 25-29 30-34 35.4 5.4 .4 4 .7 4. 8 4 3. 8 -4.0 3. 22.9 2.1 2 -1. 61 0. 80 0 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50+ Current age Rural E Other urban Ll Abidjan (all women) -6- Fertility and schooling One of the reasons for fertility differences among regions is that women In each region have different amounts of schooling. About half of the women in Abidjan have some schooling, while only 10 percent of rural women have gone to school. In other urban areas, the figure is 38 percent. Most notably, 31 percent of women in Abidjan have some secondary schooling, compared with only 18 percent in other urban areas and 1 percent in rural areas. Mother's schooling can affect fertility by altering both the demand and supply of children. Women with more schooling can command a higher wage, so the opportunity cost of an educated mother's time in childrearing is greater than for uneducated mothers. Schooling may also affect women's preferences for children, inducing them to have fewer births and invest more in each child. If educated mothers are better at maintaining their children's health, then education can reduce fertility by raising child survival and increasing childspacing intervals. Schooling can directly affect lifetime fertility by delaying age at marriage and the onset of childbearing. For all of these reasons, we expect more education to be associated with lower fertility. On the other hand, a more educated mother may be healthier, potentially raising her fecundity and the number of live births if other things are equal. According to Figure 2, women with secondary schooling in C8te d'Ivoire have substantially lower fertility than those with no schooling. In the 25-29 cohort, for example, women with secondary schooling have had on average almost two children fewer than those with no instruction. The FIGURE 2: Children ever born by age and schooling Children ever born 7 6.5 6 -6.1 6.0 5 4.8 4l74*7 4 -0 3-2 7 33. 8 3.1 2 1. 2.0 i$1.9 1 0.706 0.8 0 15-19 20-24 25-29 30-34 35-39 40-44 45-49 5Q+ Current age No schooling m Primary Lii Secondary or higher (all women) - 8 - relationship for primary schooling is less clear; not until age 30 does primary schooling seem to have a negative impact on fertility. Since explicit limitation of family size after marriage is not widespread in C8te d'Ivoire,* the major impact of schooling is probably on raising the age at marriage. The mean age at marriage for ever-married women in the sample is 17.6 years. Figure 3 shows clearly that both primary and secondary schooling tend to delay marriage for women under 25. Among 15-19 year olds, for example, each additional level of schooling raises the probability of being single by about 20 percent. In Figure 4, secondary schooling raises the mean age at first cohabitation for women 20-24 by two years and for older women by three or more years, compared to women with no schooling. The effect of primary schooling on mean age of marriage, by contrast, is relatively small. Comparison of mean age at marriage for the three age cohorts between 25 and 39 (the cohorts for which almost everyone has been married) seems to show no temporal trend in age at marriage when schooling is controlled for. Fertility and income Holding other factors constant, households with higher nonlabor income can afford to have more children and may want more in the same way that they may want to consume more of all goods. The same may be true for households with higher labor income if the household members earning it are not involved in fertility decisions or childrearing. For women and those involved in childrearing, however, higher labor income implies a higher opportunity cost of time and thus a higher shadow price of children, reducing demand for them. In the aggregate, high income households also FIGURE 3: Proportion of women never married by age and level of schooling Proportion never married 1 .84 0.8 0.6 .1 .51 0.4 -.38 .28 0.2 .13 14 0~~~~~~~~~~~~~~~~~~~~~~~~~0 15-19 20-24 25-29 30-34 35-39 Current age _ No schooling E Primary L Secondary and higher FIGURE 4: Mean age at first cohabitation by current age and schooling Age at first cohabitation 25 20.83 21.24 20.80 20 - ~~18.17 18.23 16s1416.74 17.30 17.09 17.08 15( - CD 1 0 5 0__ 20-24 25-29 30-34 35-39 Current age No schooling Primary IISecondary and higher (ever-married or cohabited women) - 11 - tend to have the most schooling, making It difficult to disentangle the potentially opposing effects of income and schooling on fertility. Unless it is possible to control for income, the impact of schooling on fertility may be understated if schooling is positively correlated with income. Fortunately, the CILSS data set includes detailed income and consumption data at the household level. Figure 5 shows the relationship between children ever born and income for the subsample of women with no schooling. The proxy for income used here is total household consumption expenditure (including the value of home production) divided by the number of adults in the household. The values for per adult expenditure were ordered from lowest to highest and divided into terciles (thirds) -- lowest, middle or highest. Figure 5 shows that for unschooled women 25-50 higher income is associated with higher fertility. Unfortunately, cell sizes were too small to confirm this relation for women with primary and secondary schooling. FIGURE 5: Children ever born by age and per adult expenditure tercile Children ever born 8 7.4 6.6 6. 6 6.2 6. 2 59.6.2 6. 1 6 596. 5.8 4.5.0 5.2 4 5-19 20-24725-29 30-34 3 4 4 ~~~~~~~~~~~~3. 7r 3 4~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~N 2.2 2- 0 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50+ Current age Lowest tercile ME Middle tercile Highest tercile (Women with no schooling) - 13 - PART II: MULTIVARIATE ANALYSIS OF FERTILITY In this part a reduced form equation of the determinants of fertility is estimated, allowing the effects of age, schooling, income and residence to be isolated. The first section describes the theoretical model of the demand for children that motivates the reduced form equation, the second section discusses the data, variables and estimation techniques and the third section presents estimation results. A model of the determinants of fertility The choice of variables in the reduced form equation is motivated by a theoretical model of the demand for children in the tradition of Becker (1965, 1981). The model adopts the perspective of an individual woman,s who maximizes a long-run utility function over children (C), market goods (X) and her own leisure (L): U = U ( C , X , L ), U' > 0, U" < 0 (1) The utility function is maximized subject to a household production function for children and to time and budget constraints. The production of children is described by a linearly homogeneous production function with mother's time in childrearing (T.) and purchased child goods (Xe) as inputs: C = I ( T6, Xe) 1' > O0 T" < 0 (2) Since the model takes the perspective of individual women, married or not, the husband's time input does not enter this production function. This is a common assumption in modelling fertility in the U.S. and is probably more realistic in Subsaharan Africa. - 14 - The woman's time constraint allocates total time (a) among childrearing (T.), market production (Tm) and leisure: a = T_ + Tm + L (3) The full-income budget constraint sets the total value of the woman's time plus nonlabor income equal to consumption "expenditure": wR + V = i,C + p.X + wL (4) where w = the woman's market wage; V = nonlabor income; it = the shadow price of children; and p. = the price of other market goods. If we assume that the time allocation of other household members (such as the husband) is exogenously given and that their leisure does not enter the woman's utility, then their income can be considered exogenous and included in the woman's nonlabor income. The shadow price of children is the sum of the value of the marginal inputs in their production: = wt= + pS=xc (5) where t. is the marginal time input in child production (6T./6C), p.. is the price of purchased child inputs and x. is the marginal input of purchased goods (6x.-/6C). Maximizing equation (1) subject to the constraints (2) - (4) yields equations expressing the demand for children, market goods and leisure as a function of exogenous prices and nonlabor income. The comparative statics for the demand for children can be signed with some assumptions. If children and market goods are substitutes in consumption and the substitution effect exceeds the negative income effect of a price increase, an increase in the price of market goods will raise the demand for children. If we assume that children are normal goods, then an increase in nonlabor income will also raise the demand for children.' The effect of an - 15 - increase in the woman's wage on the demand for children can be divided into two components: the effect of an increase in the shadow price of children (which is unambiguously negative) and the effect of an increase in the price of leisure (which is negative if children and leisure are complements and ambiguous if they are substitutes). Empirical studies have generally found the net effect of the woman's wage on fertility to be negative and that is what we expect here. The demand for children and posited signs are summarized as: + _ _ + D. = D= ( p., p.-, w, V ) (6) The number of children observed is the result of the interaction between the demand for and supply of children. The supply of children is biologically determined by the age of the woman (A) and a variable ()) that measures a woman-specific component of fecundity (Rosenzweig and Schultz, 1985). The supply of children increases with age but at a decreasing rate, declining absolutely as the woman reaches the biological end of childbearing. + + S. = S. ( A , p ) (7) The reduced form equation for the determinants of fertility includes both demand and supply-side variables, since there is insufficient information to separately identify supply and demand. Note that this long-run model assumes that women make a "once and for all" decision about the number of children to have, based on their perceived lifetime wage, income and exogenous prices. Yet fertility decisions are clearly dynamic. Preferences change over the life cycle and expectations about the future may - 16 - not be realized. With this cross-sectional data set, estimation of a dynamic model of fertility is not possible. The reduced form The estimated reduced form equation regresses children ever born, the endogenous choice variable, on a set of five exogenous variables: age, age squared, years of schooling, urban residence and one of three household income variables. The equation is estimated for the entire sample, for urban and rural women separately and for three age cohorts. Aqe and age scquared are included to control for the biological supply of children. Since the reduced form is estimated for women of all ages, many of whom are still of childbearing age, these variables control for exposure to the risk of preganacy. Completed years of schooling is used as a proxy for female wages, which were not available for most women. As a proxy for wages, an increase In schooling should lower the demand for children. Maternal schooling may have independent effects on the demand for children other than as a proxy for wages, however. It may improve maternal health, raising the supply of children, or by improving child health (lowering child mortality) it may Increase childspacing intervals and reduce the supply of children. If the mother's demand for children is really a demand for surviving children, the lower child mortality associated with mother's schooling may lead her to have fewer pregnancies. Schooling may also affect women's preferences, inducing them to demand fewer children of higher "quality".' Several - 17 - different specifications of schooling are used, Including linear and quadratic forms and dummy variables for individual years of schooling. A dummy variable for urban residence is included to reflect greater wage-earning opportunities for women in urban areas.> It also measures greater availability of services of all types -- market services, schools, health facilities and other economic infrastructure. Urban residence should be associated with a higher shadow price of children and thus lower demand. Better health services In urban areas would lower the supply of children through reduced child mortality and longer birth intervals, induced by the extended period of breastfeeding and postpartum amenorrhea if the child survives. on the other hand, better urban health services could equally raise fecundity and the supply of children by reducing the prevalence of sexually transmitted disease and infertility problems.9' The theoretically correct income variable to use is "nonlabor income". It should exclude the woman's own earnings, which are endogenous through her labor supply. The earnings of other household members can be included, since their labor supply is considered exogenous to fertility decisions here. Unfortunately, except for the 5.5 percent of the women in the sample who had wage income, it was impossible to attribute income to individuals and to purge household income of the woman's earnings. Three different household income variables are used in the empirical estimation. The principal income variable is a proxy for household permanent income that includes annual consumption expenditure, the value of home production consumed and an imputed value of services from durables.t° - 18 - Consumption is used as a proxy for permanent income because it tends to fluctuate less over the life cycle than current income. The variable used is the natural log of "permanent income" as defined above, in CFA francs, per adult household member age 15 or older. The two other household income variables are: (a) current income, the sum of income from wages, home agricultural production, home businesses, the value of services from durable goods, receipt of transfers, imputed rent for owners of housing in urban areas and other income; and (b) nonlabor income, which includes the value of services from durable goods, receipt of transfers, income from rents on property, dividends, imputed rent for owners of housing in urban areas and all other household income not tied to labor supply. Social security and pension income are not included in household nonlabor income. Both income variables have been divided by the number of adults in the household and are expressed as natural logarithms."1 All three of the income variables have shortcomings. The permanent and current income variables suffer from endogeneity, since the woman's consumption and earnings could not be netted out from the rest of the household. Further, although dividing through by the number of adults makes the permanent income variable less dependent on the left hand side variable, children ever born, the presence of children in the household will nevertheless drive up the level of consumption per adult. Ten percent of all households and 15 percent of rural households had no nonlabor income, and the value of owner-occupied housing could be assessed only for urban households. The nonlabor income variable is also sensitive to the assumptions used to value the services from housing and durable goods. - 19 - Landholdings are not included as a regressor because ownership is not well defined in much of land-abundant rural Cote d'Ivoire. The land cultivated is not an acceptable alternative because it is endogenous. Respondents were generally unable to cite with accuracy the area owned or cultivated and in most of the rural communities there is no land market, making valuation of land impossible. Exogenous health service variables, such as the distance to the nearest maternity ward, maternal and child health clinic and hospital, were tentatively included in the regressions but subsequently dropped due to problems in interpreting their coefficients. In rural areas, services of all types tend to be clustered at the same administrative level; the distance to a health facility is also the distance to all economic and administrative infrastructure. Access to services on the date of Interview is unlikely to be a good proxy for lifetime access to health services, which is clearly more relevant to fertility decisions. Finally, access to services may be endogenous if people selctively migrate to areas with better service availability. Alternatively, the government may place health services in areas with the worst health and the highest fertility. For all of these reasons, it is difficult to interpret coefficients on distances as the effect of access to a service on fertility.1: The means and standard deviations of the dependent and independent variables are presented in Table 2, for the sample of 1444 women for whom fertility records were consistent and consumption variables had been computed. t3 - 20 - TABLE 2: Sample means and standard deviations All women Urban Rural VARIABLE Mean SD Mean SD Mean SD Children 3.91 3.30 3.14 3.05 4.46 3.37 ever born Age 34.31 15.07 30.38 12.93 37.07 15.85 Age squared 1403.8 1266.7 1090.2 1019.0 1624.9 1373.6 Years of 1.69 3.43 3.40 4.47 0.48 1.59 schooling Schooling 14.64 39.22 31.48 55.53 2.77 10.53 squared Dummy, 1-6 yr 0.136 0.343 0.194 0.396 0.094 0.292 Dummy, 1-2 yr 0.029 0.168 0.027 0.162 0.031 0.173 Dummy, 3-6 yr 0.107 0.309 0.168 0.373 0.064 0.244 Dummy, 7+ yr 0.107 0.309 0.243 0.429 0.011 0.103 Dummy, 1 year 0.015 0.123 0.017 0.128 0.014 0.118 Dummy, 2 years 0.014 0.117 0.010 0.100 0.017 0.128 Dummy, 3 years 0.003 0.059 0.005 0.071 0.002 0.049 Dummy, 4 years 0.010 0.101 0.013 0.115 0.008 0.091 Dummy, 5 years 0.022 0.147 0.040 0.197 0.009 0.097 Dummy, 6 years 0.071 0.256 0.109 0.312 0.044 0.204 Urban dummy 0.41 0.49 1.00 0.00 0.00 0.00 Ln permanent 12.59 0.82 13.09 0.72 12.24 0.68 income/adult Ln current 12.29 1.43 12.81 1.33 11.92 1.39 income/adult Ln nonlabor 8.31 3.32 10.10 2.52 7.04 3.23 income/adult N 1444 597 847 (continued) - 21 - TABLE 2 (cont.) Age 15-24 Age 25-34 Age 35 + VARIABLE Mean SD Mean SD Mean SD Children 1.13 1.28 3.90 2.27 6.06 3.29 ever born Age 19.53 2.73 28.94 2.73 48.74 11.38 Age squared 388.78 106.70 844.84 159.00 2505.3 1246.4 Years of 3.06 3.88 2.26 4.20 0.31 1.56 schooling Urban dummy 0.53 0.50 0.45 0.50 0.30 0.46 Ln permanent 12.65 0.77 12.77 0.88 12.44 0.79 income/adult Ln current 12.44 1.00 12.47 1.52 12.04 1.62 income/adult Ln nonlabor 8.76 2.99 8.42 3.17 7.89 3.59 income/adult N 473 355 616 Estimation techniques The reduced form equation is estimated using three econometric models: ordinary least squares; maximum likelihood Tobit; and a Poisson count model. Ordinary Least Souares. Least squares estimates have been widely used in fertility analysis (see the studies cited in Cochrane 1979, T.P. Schultz 1974, 1981, and T.W. Schultz, 1974). OLS expresses the dependent, continuous random variable (y) as a linear function of exogenous variables (x) plus an error term (a), where it is assumed that a is independent, - 22 - identically distributed and uncorrelated with the regressors. Under these assumptions, the least squares slope estimates are best linear unbiased. Children ever born is not a continuous variable, however, and It is censored at zero. 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). When both y and the regressors are normally distributed, OLS coefficients are biased downward in proportion to the proportion of nonzero observations (Amemiya 1984, Greene 1981, Maddala 1983). Thus, the smaller the degree of consoring, the closer are OLS and Tobit coefficients. All of the OLS specifications exhibited hetero- skedasticity and the standard errors have been re-estimated using the White heteroskedastic-consistent covariance matrix. Maximum Likelihood Tobit. Let y* be the true demand for children (which can be positive or negative) and y the observed number of children ever born, which can take a value of zero or positive integers. The Tobit model takes into account the fact that negative values of y* are not observed, but it still assumes that y is a continuous variable in the nonnegative range. The Tobit model is: Y*= X' + a y = y* if y* 2 0 = 0 otherwise a N (0, a2) E (a) = 0 ; E (x's) = 0 E (ea t ) = 02I , i j =0 , i not equal j E (y*) = x'1 - 23 - Tobit coefficients are estimated using maximum likelihood methods."4 Although Tobit takes the censoring of the dependent variable into considera- tion, results are sensitive to the assumptions of homoskedasticity, normality of the errors and that the dependent variable, y, is measured without error. Violation of any of these assumptions produces inconsistent estimates (Amemiya 1984, Maddala 1983, Stapleton and Young 1984). Poisson count model. The Poisson count model assumes that the dependent variable is generated from a Poisson process and that it takes on values that are nonnegative integers. Thus, both the censored and integer aspects of children are taken into account. The Poisson model has the additional advantage that it models some heteroskedasticity, since the variance of the dependent variable is a function of x'3. The model is: exp (-xt ) xit Pr (y) =------------- yt i where xt = exp (x'W) = E (yt) = variance (yt). The Poisson model is estimated using maximum likelihood methods.1~ These ML estimates are consistent even if the data are normally distributed, provided that the mean is correctly specified (Cameron and Trivedi 1986). One problem with the Poisson model is that it restricts the condi- tional mean of the dependent variable to equal the conditional variance. "Overdispersion" of the data results in spuriously small standard errors for the estimated coefficients (Cameron and Trivedi 1986, Portney and Mullahy 1986). The standard errors reported for Poisson regressions are calculated - 24 - using a method that is robust to departures from the assumption that the variance equals the mean, as described in Portney and Mullahy (1986).±& Estimation results 1. Determinants of fertility Estimates of the reduced form equation for all women and various subsamples are presented in Tables 3 and 4. The first figure in each cell is the coefficient (6) for the variable in the regression. To facilitate comparison across models, the slope coefficients on the expected value functions of the Tobit and Poisson models have been calculated at the sample means and are presented in brackets; it is these figures that should be compared with the OLS coefficients.:' Reduced form estimates for the entire sample using the permanent income variable and completed years of schooling as regressors are in the first three columns of Table 3. All of the coefficients are highly significant and of the expected signs. At the mean, an additional year of schooling reduces the number of children ever born by about 0.14. The coefficients on permanent income per adult range from 0.32 to 0.38 and correspond to income per adult elasticities of +0.082 to +0.098. The estimates for urban and rural subsamples reveal that the schooling coefficient is significant only for urban women (ranging from -0.09 to -0.12), while the income coefficient is significant only for rural women. The permanent income per adult elasticity for rural women ranges from +0.12 and +0.13 at the mean. That is, a ten percent increase in income per adult at the mean is associated with an increase In the number of TABLE 3: Estimates by location Dependent variable: Children ever born ALL WOHEN UREAN WOMEN RURAL WOMEN EXPLANATMRY __ _ ... _ .. VARIABLES OLS Tobit Poisson OLS Tobit Poisson OLS Tobit Poisson Age 0.4296** 0.5414** 0.1379** 0.4891** 0.6686** 0.2009** 0.4106** 0.4106** 0.1163** (.0218) (.0240) (.0077) (.0308) (.0362) (.0301) (.0285) (.0280) (.0084) (0.48421 (0.4286] (0.54201 10.48151 10.37801 [0.48621 Age2 -0.0038** -0.0050** -0.0013** -0.0046** -0.0066** -0.0020** -0.0036** -0.0036** -0.0010** (.0003) (.0003) (.0001) (.0004) (.0004) (.0004) (.0003) (.0003) (.0001) (-0.00451 [-.00401 1-0.00531 [-0.00481 [-0.0033] [-.0042] Years of -0.1113** -0.1562** -0.0443** -0.0990** -0.1428** -0.0366** -0.0633- -0.0633 -0.0238 schooling (.0158) (.0356) (.0071) (.0168) (.0355) (.0141) (.0391) (.0781) (.0169) [-0.13971 1-0.13771 (-0.11581 1-0.08771 1-0.05831 1-0.09961 Urban -0.4467** -0.6259** -0.1338** dum_y (.1606) (.1881) (.0425) [-0.55981 [-0.41581 LA Ln perm. 0.3215** 0.4287** 0.1091** -0.0842 0.0391 0.0158 0.5827** 0.5827** 0.1399** inc/adult (.1015) (.1095) (.0260) (.1286) (.1622) (.0745) (.1444) (.1242) (.0315) (0.38341 (0.33911 10.03171 (0.03791 (0.53651 10.5848] Constant -9.0986 -12.9165 -3.015 -5.2096 -10.5349 -3.132 -12.0359 -12.0359 -2.956 (1.233) (1.414) (.3415) (1.588) (2.164) (.9922) (1.810) (1.626) (.4230) Ra .44 .53 .37 Sigma 2.7961 2.4879 2.6903 (.0509) (.0725) (.0541) LogL -3119.8 -3181.2 -1143.5 -1144.5 -1969.3 -2005.4 N 1444 1444 1444 597 597 597 847 847 847 Notes: 1. The first figure in every cell is the coefficient of the model. Standard errors are In parentheses. Figures In brackets are slope coefficients of the expected value functions, calculated at the mean (see footnote 13). 2. OLS standard errors are corrected for heteroskedasticity using the White heteroskedastic-consistent covariance matrix; Polsson standard errors are corrected to account for overdispersion (see text). 3. ** significant at .01; * significant at .05; - significant at .10. - 26 - children by about one and a quarter percent. The weak results for schooling of rural women are probably due to the very low levels of schooling and small variation in the subsample. Alternatively, schooling may not be a good proxy for wages in rural areas where there are few wage-earning opportunities for women. An F-test on the OLS regressions found significant structural differences at the .01 level between the coefficients for urban and rural subsamples: F(5,1434) = 4.75. Likelihood ratio tests on Tobit and Poisson estimates came to the same conclusion (LRT, XZ(5) = 96 and 79, respectively). The results in Table 3 show that schooling has a negative effect on current fertility but not necessarily on completed fertility. Women at different points in the life cycle are included in the regressions; younger women may be marrying later but having just as many children as their mothers by reducing childspacing intervals. Estimates by age cohort are presented in Table 4.*e Among the oldest cohort (35+ years), schooling has had no discernable effect on fertility but the effect of income is positive and significant. The absence of a schooling effect is not surprising, given the very limited amount of schooling in this cohort (mean of 0.3 years). The Tobit and OLS coefficients are virtually identical, as only 5 percent of the women have never had a live birth. Expected value coefficients are similar across models. Among the youngest cohort (15-24), only the schooling and urban dummy coefficients are significant. The negative relation between schooling and fertility Is probably due to the effect of schooling on delayed marriage. TABLE 4: Estimates by age qroup Dependent variable: Children ever born AGE 15-24 AGE 25-34 AGE 35 + EXPLANATORY _ _ _ _ VARIABLES OLS Tobit Poisson OLS Tobit Poisson OLS Tobit Poisson Age -0.0226 1.0834* 1.1160** 0.4340 0.4340 0.1960 0.1950* 0.1950* 0.0335* (.2690) (.4875) (.3030) (.8734) (.9422) (.2184) (.0899) (.0854) (.0161) t0.65261 10.9709] (0.41941 (0.74731 (0.18831 (0.20481 Age& 0.0066 -0.0177 -0.0220** -0.0038 -0.0038 -0.0024 -0.0018* -0.0018* -0.0003* (.0071) (.0123) (.0074) (.0151) (.0161) (.0037) (.0008) (.0008) (.0002) 1-0.01071 [-0.01911 1-0.00371 1-0.00921 1-0.00171 (-.00181 Years of -0.0685** -0.1242** -0.0666** -0.1565** -0.1565** -0.0466** -0.0700 -0.0700 -0.0121 schooling (.0130) (.0245) (.0140) (.0268) (.0384) (.0088) (.0770) (.1064) (.0140) [-0.0748] (-0.05791 [-0.15121 (-0.17771 (-0.0676] (-0.07401 Urban -0.3359** -0.5312** -0.2952** -0.2990 -0.2990 -0.0744 -0.6354- -0.6354* -0.1056- dummy (.1177) (.1920) (.1037) (.2766) (.2828) (.0686) (.3340) (.3099) (.0560) [-0.32011 1-0.25681 (-0.28901 1-0.28371 (-0.61351 1-0.64551 Ln perm. 0.0719 0.1003 0.0712 0.3245* 0.3245- 0.0858* 0.4253* 0.4253* 0.0703* inc/adult (.0742) (.1251) (.0676) (.1678) (.1695) (.0434) (.2020) (.1807) (.0334) (0.06041 (0.06191 (0.3136] 10.32711 10.41071 10.42971 Constant -1.5207 -14.3497 -13.92 -9.1276 -9.1276 -3.263 -4.0964 -4.0965 0.0905 (2.651) (5.062) (3.082) (12.79) (13.08) (3.259) (3.531) (3.326) (.6024) Re .33 .14 .02 Sigma 1.5485 2.1214 3.2758 (.0728) (.0645) (.0958) LogL -639.4 -587.4 -766.0 -760.4 -1590.3 -1695.4 N _ _ 473 473 473 355 355 355 616 616 616 Notes: 1. The first figure in every cell is the coefficient of the model. Standard errors are in parentheses. Figures In brackets are slope coefficients of the expected value functions, calculated at the mean (see footnote 13). 2. OLS standard errors are corrected for heteroskedasticity using the White heteroskedastic-consistent covariance matrix. Poisson standard errors are corrected to account for overdispersion (see text). 3. ** significant at .01; * significant at .05; - significant at .10. Age and age squared are jointly significant at .01 for all models and all subsamples. - 28 - Many of these women are still in school or not yet married and thus have not begun childbearing. They may also be delaying marriage or childbearing because of greater income-earning opportunities in urban areas. The OLS and Tobit coefficients differ greatly due to the high degree of censoring (42 percent of this cohort have had no children). The expected value coefficients show greater spread across models for this cohort, the Poisson estimates being the smallest absolute value. Most of the women in the middle cohort (25-34) are married and have borne children -- almost 4, on average. They have been married for enough time to to have compensated for any delay in marriage by having children at closer intervals. The schooling coefficient for the middle cohort is negative, highly significant and 2-3 times the size of the coefficient for younger women in absolute value. This finding is consistent with the hypothesis that a fertility decline is underway among educated women. The permanent income coefficient for women 25-34 is also positive and significant and corresponds to a larger income elasticity (+0.08) than for the oldest cohort. The urban dummy variable is not significant as it is for the youngest and oldest groups. Poisson coefficients on schooling and income are greater in absolute value than Tobit and OLS, but the differences are not great. Table 5 summarizes the point elasticities of fertility with respect to schooling and income for all subsamples in Tables 3 and 4. It is difficult to assess these elasticities, as potentially comparable studies include different sets of regressors, different measures of dependent and independent variables, and are often based on aggregated rather than - 29 - Individual data. The female schooling elasticity of approximately -0.06 falls at the low end (in absolute value) of schooling elasticities in other studies of developing countries surveyed by T.P. Schultz (1974). Raising mean schooling from 1.7 to 3 years, holding all other variables constant, would lower mean fertility from 3.91 to about 3.73 children. Note that schooling elasticities for urban women and the youngest cohort (15-24) are two to three times as large as for the entire sample, however. The Income elasticity of about +0.09 implies that a 10 percent increase in per adult permanent income would raise mean fertility from 3.91 to 3.95 children. TABLE 5: Point elasticities Permanent income Schooling SUBSAMPLE OLS Tobit Poisson OLS Tobit Poisson All women .082** .098** .087** -.048** -.060** -.059** Urban -.107** -.125** -.095** Rural .131** .120** .131** -.007W Age 15-24 -.185** -.202** -.156** Age 25-34 .083* .080 .084* -.091** -.088** -.103** Age 35+ .070* .068* .071* Note: Elasticities calculated at the mean using OLS coefficients and Tobit and Poisson expected value function coefficients in brackets in previous tables. Elasticities are not reported for insignificant coefficients. ** indicates significant at .01; * indicates significant at .05; indicates significant at .10. - 30 - 2. Choice of income variable The sensitivity of results to the choice of income variable is presented in Table 6. (Results for permanent income are repeated from Table 3 for ease of comparison.) The coefficients on current income are positive, but about a third the size of the coefficients on permanent Income. Only the Tobit current income coefficient attains reasonable significance. The fact that current income has a positive, if weaker, effect compared to permanent income is reassurance that the coefficient on permanent income is not simply reflecting the positive correlation between per adult consumption and the number of children in the household. The stronger effect of the permanent income measure indicates that fertility is more sensitive to permanent than current income, as theory would suggest. None of the coefficients on nonlabor income are significant. The schooling coefficient for the specification using permanent income is greater in absolute value than the schooling coefficients for the other specifications. A more relevant question for many fertility data sets with no income variables at all is the effect on the schooling coefficient when income cannot be controlled for. Income and female schooling tend to be correlated but, as has been shown, have opposing effects on fertility in C8te d'Ivoire. An inability to control for income might therefore weaken the coefficient on schooling. Table 7 compares OLS regressions with and without permanent income. (Again, permanent income results from Table 3 are repeated for ease of comparison.) Excluding income from the regression for all women reduces the coefficient on schooling by 0.02 in absolute value, or by about 18 percent. Separate regressions for urban and rural women reveal that the TABLE 6: Sensitivity of results to income specification Dependent variable: Children ever born Permanent Income Current Income Nonlabor Income EXPLANATORY VARIABLES OLS Tobit Poisson OLS Tobit Poisson OLS Tobit Poisson Age 0.4296** 0.5414** 0.1379** 0.4362** 0.5497** 0.1393** 0.4375** 0.5512** 0.1395** (.0218) (.0240) (.0077) (.0217) (.0240) (.0240) (.0217) (.0241) (.0078) [0.4842] [0.42861 [0.4974] [0.44551 [0.4988] 10.44991 Age* -0.0038** -0.0050** -0.0013** -0.0039** -0.0051** -0.0013** -0.0039** -0.0051** -0.0013** (.0003) (.0003) (.0001) (.0003) (.0003) (.0002) (.0003) (.0003) (.0001) [-0.0045] [-0.00401 [-0.00461 [-0.00421 [-0.00461 [-0.00421 Years of -0.1113** -0.1562** -0.0443** -0.1001** -0.1385** -0.0397 j -0.0966** -0.1342** -0.0383** schooling (.0158) (.0356) (.0071) (.0155) (.0352) (.0203) (.0152) (.0352) (.0070) [-0.13971 [-0.13771 (-0.1253] 1-0.12701 (-0.12141 [-0.1245] Urban -0.4467** -0.6259** -0.1338** -0.2882* -0.3999* -0.0762* -0.3080* -0.4196* -0.0799* dumy (.1606) (.1881) (.0425) (.1473) (.1783) (.0504) (.1518) (.1919) (.0405) [-0.55981 [-0.41581 [-0.3619] [-0.24371 [-0.37971 [-0.25771 Ln income 0.3215** 0.4287** 0.1091** 0.0965- 0.1088* 0.0295 0.0305 0.0336 0.0076 per adult (.1015) (.1095) (.0260) (.0566) (.0519) (.0187) , (.0243) (.0246) (.0058) [0.3834] (0.33911 (0.09851 [0.0944] I [0.03041 (0.0245] Constant -9.0986 -12.9165 -3.015 -6.4398 -9.1244 -2.056 -5.5227 -8.0849 -1.756 (1.233) (1.414) (.3415) (.7365) (.8258) (.6460) (.3952) (.5606) (.1565) Rs .44 .44 .44 Sigma 2.7961 2.8054 2.8077 (.0509) (.0511) (.0512) LogL -3119.8 -3181.2 -3125.2 -3191.9 -3126.2 -3195.1 N 1444 1444 1444 1444 1444 1444 1444 1444 1444 Notes: 1. The first figure in every cell is the coefficient of the model. Standard errors are in parentheses. Figures In brackets are slope coefficients of the expected value functions, calculated at the mean (see footnote 13). 2. OLS standard errors are corrected for heteroskedasticity using the White heteroskedastic-consistent covariance matrix. Poisson standard errors are corrected to account for overdispersion (see text). 3. ** significant at .01; * significant at .05; - significant at .10. - 32 - exclusion of income has virtually no effect on the schooling coefficient for urban women but that the schooling coefficient for rural women loses significance altogether when income Is excluded. Failure to control for income among rural women would have led to the erroneous conclusion that schooling has no effect on fertility. The impact of omitting income from the regression is thus most severe for the group of women with the lowest levels of schooling. TABLE 7: Specifications with and without income Dependent variable: Children ever born ALL WOMEN URBAN WOMEN RURAL WOMEN EXPLANATORY -------------------- -------------------- -------------------- VARIABLES (1) (2) (3) (4) (5) (6) Age 0.4296** 0.4374** 0.4891** 0.4856** 0.4106** 0.4176** (.0218) (.0217) (.0308) (.0299) (.0285) (.0288) Age2 -0.0038** -0.0039** -0.0046** -0.0046** -0.0036** -0.0037** (.0003) (.0003) (.0004) (.0004) (.0003) (.0004) Years of -0.1113** -0.0918** -0.0990** -0.1044** -0.0633- -0.0353 schooling (.0158) (.0146) (.0168) (.0150) (.0391) (.0389) Urban -0.4467** -0.2272** dummy (.1606) (.1423) Ln perm. 0.3215** -0.0842 0.5827** inc/adult (.1015) (.1286) (.1444) Constant -9.0986 -5.3133 -5.2096 -6.2300 -12.0359 -5.0297 (1.233) (.3571) (1.588) (.4589) (1.810) (.4905) Ra .4404 .4365 .5283 .5288 .3630 .3500 N 1444 1444 597 597 847 847 Notes: 1. Standard errors are in parentheses. They are corrected for heteroskedasticity using the White heterosdastic-consistent covariance matrix. 2. ** significant at .01; * significant at .05; significant at .10. - 33 - 3. Specification of schooling The regressions so far have used completed years of schooling as an independent regressor. This specification restricts the effect of schooling to be linear. It has been suggested in the literature that the effect of schooling is nonlinear: the first few years of primary schooling actually raise fertility through a positive effect on maternal and child health (Cochrane 1979, World Bank 1984). Table 8 presents OLS regression results for several alternative specifications of schooling. According to the first regression, with a quadratic specification of schooling, fertility reaches a "maximum" at 0.4 years of completed schooling and declines afterward. This is confirmed by the second, third and fourth regressions, in which dummy variables for different combinations of primary years have been used, with no schooling as the left out category. In the second regression, the effect of any primary schooling is negative and significant at .10; in the third regression, the effect of the first two years Is negative and significant at .05. In the fourth regression, dummy variables for individual years of primary schooling have been entered. The only two primary years that are significant are the first and sixth years, both of them with negative coefficients. The rather large and significant negative coefficient on the dummy for one year may be partly attributable to underreporting of higher levels of completed schooling. For all of regressions (2)-(4), the negative effect of secondary schooling on fertility is far greater than the effect of primary schooling. The relation between schooling and fertility is thus negative in all ranges of primary schooling and we find no evidence of a positive effect of the first years of primary school when income and urban residence are controlled for.1' - 34 - TABLE 8: Various schooling specifications, all women Dependent variable: Children ever born EXPLANATORY VARIABLES (1) (2) (3) (4) Age 0.4380** 0.4332** 0.4334** 0.4336** (.0221) (.0220) (.0220) (.0221) Age2 _0.0039** -0.0039** -0.0039** -0.0039** (.0003) (.0003) (.0003) (.0003) Urban -0.4749** -0.4688** -0.4763** -0.4784** dummy (.1610) (.1611) (.1620) (.1621) Ln perm. 0.3315** 0.3030** 0.3026** 0.3039** inc/adult (.1015) (.1012) (.1012) (.1010) Dummy for -0.6352* 1 year (.2484) Years of 0.00891 Dummy for -0.2370 schooling (.0343)1 2 years (.3128) ** Schooling -0.0110 Dummy for 0.3472 squared (.0024)] 3 years (.5566) Dummy for -0.2505 Dummy for -0.1013 1-6 years (.1366) 4 years (.3581) Dummy for -0.4461*1 Dummy for 0.1536 1-2 years (.2069) 5 years (.2965) Dummy for -0.1940 Dummy for -0.3418' 3-6 years (.1547) 6 years (.1790) Dummy for -1.0866* -1.0813* Dummy for -1.08)7*e 7 years + (.1839) J (.1843) j 7 years+ (.1842) J Constant -9.4371 -8.9793 -8.9764 -8.9957 (1.241) (1.232) (1.231) (1.230) RZ .4424 .4381 .4378 .4368 N 1444 1444 1444 1444 Notes: 1. Standard errors are In parentheses. They are corrected for heteroskedasticity using the White heteroskedastic- consistent covariance matrix. 2. ** significant at .01; * significant at .05; significant at .10. Brackets indicate Joint significance. - 35 - CONCLUSIONS In a cross-section of Ivorian women of all ages, female schooling is associated with lower fertility while household income raises it. Raising women's schooling from a mean of 1.7 to 3.0 years (an increase of 76 percent) would lower mean children from 3.9 to 3.7. With such low levels of schooling in the sample, it is risky to predict what impact universal female primary schooling -- a quadrupling of the mean -- would have on fertility. Neverthe- less, the effect of additional schooling on fertility was found to be negative for all levels of schooling, even for the early years of primary school. The fact that schooling coefficients for the youngest women are negative and highly significant and remain so into the 25-34 year age group is consistent with the hypothesis that a fertility decline may be underway among women with schooling. There is additional evidence from Part I that secondary schooling leads to delayed marriage, thereby reducing lifetime fertility. Experimentation with different income measures found that the consumption-based proxy for permanent income has a stronger relation with fertility than does current income and that nonlabor income has no effect. When all income variables are excluded from the fertility regression, the coefficient on schooling is reduced and, in the case of rural women, is rendered insignificant. Failure to control for income in fertility analysis can, therefore, result in underestimates of the impact of schooling. Finally, experimentation with three econometric models revealed that, except for subsamples with substantial censoring (such as the youngest cohort), least squares yields estimates that are similar in magnitude and significance to the expected value coefficients of the more sophisticated Tobit and Poisson models. - 36 - FOOTNOTES 1. The gross primary enrollment ratio for females is the total number of females attending primary school as a percentage of all females of primary school age. Since older children are often enrolled in lower classes, this ratio may exceed 100 percent. 2. Summary statistics for the sample are not nationally representative because only one woman was interviewed per household. Women from larger households thus had a smaller probability of being included. The results must be weighted to get nationally representative figures. The nonrepresentativeness of the sample does not invalidate conclusions on relationships between fertility and socioeconomic variables, however. 3. Data underlying Figures 1-5 are presented in Annex A. 4. The widespread desire for additional children in C6te d'Ivoire is confirmed by the results of the 1980-81 Ivoirian Fertility Survey: among 4,224 currently married, fecund women, 90.4 percent wanted another child and only 4.3 percent said they wanted no more children (R.C.I. 1984, Volume I, p. 106). Among women age 45-49, only 12.9 percent wanted no more children; among those with 8 children already, only 13.5 percent wanted no more (Volume I, Table 6.1). The mean number of additional children desired by women who gave a numerical response was 4.8 children (Volume I, Table 6.4). Desired family size among all currently married women who gave a numeric response was 8-9 children; 26 percent of married women gave non-numeric responses such as "up to God" (Volume I, p. 112). 5. Treatment of the woman as the decision-making unit is more than a convenient assumption in the African context. Recent work by Oppong and Bleek (1982) and Etienne (1979) on matrilineal societies in Ghana and C6te d'Ivoire, respectively, underline the self-sufficiency of women and the "marginality of men". Children automatically belong to the mother's lineage and extramarital fertility is not discouraged in the urban area of southern Ghana studied by Oppong and Bleek. Etienne finds that Baul6 women behave as "autonomous social agents" and may even adopt the children of relatives as their own dependents, not to be shared with their husbands. There are other practical and theoretical reasons for considering the woman as a decision-making unit, however. I wish to estimate a reduced form model of fertility for both ever married and never married women; were estimates confined to married women only, the effect of schooling on delayed childbearing through delayed marriage would not be reflected in the results, nor would non-marital fertility. Further, although husbands clearly play a central role in childbearing decisions, to the extent that marital decisions and the choice of a spouse directly reflect childbearing preferences, marital status and husband's characteristics are jointly endogenous with fertility and are Inappropriate as independent variables in the reduced form regressions. - 37 - 6. Becker (1981), p. 102, points out that if there is an interaction between the quality and quantity of children, the effective price of children might increase with income and the sign on income could be negative. 7. The tradeoff between the quantity and quality of children is not studied here. 8. The urban dummy is treated as exogenous, although the household conceivably could have moved to an urban area in order to take advantage of work opportunities and greater availability of services. This would introduce a self-selection problem into the estimation, where women who prefer working in the market and having fewer children move to urban areas and those who prefer home production and more children stay in rural areas. 9. Lee, Pol and Bongsuiru (1986) found that the fertility of rural-urban migrants in Cameroon is essentially no different from that of rural stayers; they attribute this to the lower incidence of infertility (presumably due to better health services) and higher marital stability in urban areas, both of which have a positive effect on the supply of births. 10. The household income and consumption variables were computed by Kozel. Details are in her 1987 dissertation. 11. All households had positive permanent income; 9 had zero or negative current income and 144 had no nonlabor income. Per adult income values less than or equal to zero were arbitrarily assigned a value of 1 CFA franc (about half a cent) before conversion into logarithms. 12. The distance to health and schooling services was available for rural households only. The OLS results for these variables, when added to the variables of the reduced form for 773 rural women, are as follows: 0.0479 DPRIM + 0.0082 DSEC + 0.0100 DHOSP + 0.0036 DPMI - 0.0197**DMAT (.0577) (.0073) (.0067) (.0026) (.0080) where DPRIM is the distance to the nearest primary school, DSEC the distance to the nearest secondary school, DHOSP the distance to the nearest hospital, DPMI the distance to the nearest maternal and child health clinic, and DMAT the distance to the nearest maternity ward, all in kilometers. Standard errors corrected for heteroskedasticity are in parentheses. 13. Income and consumption data for a few households were inadvertently left off the original data tape. They have since been added, but the variables have not yet been computed. These omissions were random. 14. The Tobit log likelihood function is: y- X't r X'8 Log L = d-ln I(1/a)W( ------- )] + (1 - d) ln {1 - #( --- )I - 38 - where d = 1 when y* > 0 and d 0 when y* < 0, t and * are the normal density and distribution functions, respectively, and the t subscript denoting the observation has been left off of d, y, and x for convenience. 15. The Poisson log likelihood function, first (g) and second (h) derivatives are: Log L = - E exp (x'A) + E y-(x'B) - Z log(y!) g = z x-(y - exp (x'8)) h = - E exp (x'B)-xx' where the t subscript denoting the observation has been left off. The first order conditions are set equal to zero and solved iteratively using the Newton-Raphson method. The likelihood function is globally concave, so convergence to a global maximum is relatively rapid. 16. Portney and Mullahy (1986, p. 37) obtain consistent estimates of the covariance of estimated B using: IMB-1 I E (d5lt;/dB )(d lu/dB)@ I(B)-I t where -62l/6366B is denoted as I(A), lt is the contribution of the "t"th observation to the log-likelihood function and the expression is evaluated at maximum likelihood estimates of A. The unconditional variance/mean ratio from children ever born is 2.79 to 1 in the sample. This non- equivalence does not necessarily imply that the Poisson model is inappropriate, however: even when the conditional mean and variance are equal, the unconditional variance may exceed the mean (Gourieroux, Monfort and Trognon 1984, footnote 3). 17. The coefficients of the Tobit model represent dy*/6x. The coefficients for E(y) are obtained as follows (Maddala 1983, Rosenzweig and Schultz 1985): E (y) = Pr (y > O)*E (yly > 0) + Pr (y = O) E (yly = 0) = * (A'x + a */#) + (1 - C)- 0 = A'x + u-, 'x B'x dE(y)/6x± = [ + --- - --- 6E(y)/6xi = * where t and 9 are the normal density and distribution functions of B'x/a, evaluated at the mean, and the i subscript denotes an explanatory variable. The coefficients for the expected value locus of the Poisson model are calculated as follows: - 39 - E (y) = exp (i'x) dE(y)/6x± = (3 - exp (B'x) where exp (B'x) is calculated at the mean. 18. Tests for structural differences in regressions for age subsamples were significant at the .05 level for OLS coefficients (F(12,1426) = 2.07) and .01 for Tobit and Poisson coefficients (LRT, X2c±e - 248.2 and 276, respectively). 19. A comparable study of the determinants of fertility in Kenya, which also controlled for income, found that primary schooling had no effect on fertility, while all schooling above primary level had a significant negative effect (Anker and Knowles 1982). That study, based on a 1974 household survey, was confined to currently married women 15-49, while the C8te d'Ivoire results reported here are for all women 15 and older, regardless of marital status. Further, the proportion of Kenyan women with any schooling in 1974 (45 percent) was almost double the proportion for C8te d'Ivoire in 1985 (24 percent). - 40 - ANNEX A: Data underlying figures 1 - 5 TABLE Al: Mean children ever born by woman's age, location and schooling AGE GROUP 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50+ TOTAL Location ABIDJAN 0.31 1.17 2.93 4.04 5.21 5.38 8.05 4.83 3.08 (55) (64) (41) (46) (34) (13) (19) (23) (295) OTHER 0.54 1.58 3.20 4.70 5.37 6.08 7.00 5.44 3.20 URBAN (83) (59) (41) (33) (27) (25) (13) (39) (320) RURAL 0.77 2.06 3.84 4.44 6.13 6.29 6.04 6.23 4.46 (111) (111) (109) (90) (83) (75) (79) (195) (853) ---- Rural areas ---- East 0.62 1.97 3.51 4.38 6.33 6.36 7.15 7.05 4.68 Forest (34) (34) (45) (21) (33) (14) (27) (61) (269) West 0.70 1.95 4.48 4.44 6.39 5.80 5.83 5.89 4.39 Forest (37) (42) (33) (32) (23) (30) (29) (70) (296) Savanna 0.95 2.29 3.65 4.49 5.67 6.74 5.00 5.83 4.33 (40) (35) (31) (37) (27) (31) (23) (64) (288) Schooling NONE 0.73 1.91 3.67 4.75 5.92 6.10 6.50 5.99 4.53 (130) (132) (135) (119) (126) (107) (108) (256) (1113) PRIMARY 0.57 1.96 3.83 3.89 4.71 - - - 2.23 (70) (53) (35) (28) ( 7) ( 2) ( 2) ( 1) (198) SECONDARY 0.24 0.84 1.90 3.05 4.73 - - .. 1.63 AND HIGHER (49) (49) (21) (22) (11) ( 4) ( 1) ( 0) (157) TOTAL 0.59 1.70 3.51 4.38 5.77 6.14 6.50 5.98 3.91 (249) (234) (191) (169) (144) (113) (111) (257) (1468) Note: The number of observations is in parentheses. Means are not computed for cells with fewer than 5 observations. - 41 - TABLE A2: Mean children ever born by woman's age and expenditure per adult tercile AGE GROUP ----------------------------------------------------- TERCILE 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50+ TOTAL All women LOWEST 0.64 1.92 3.34 4.27 5.11 5.88 5.82 6.16 4.08 (81) (61) (53) (44) (36) (43) (45) (120) (483) MIDDLE 0.58 1.72 3.60 4.75 6.04 6.00 6.68 5.74 3.88 (83) (86) (57) (57) (51) (37) (31) (81) (483) HIGHE$T 0.55 1.52 3.55 4.05 5.89 6.64 7.28 6.12 3.78 (78) (86) (80) (66) (56) (33) (32) (52) (483) Women with no schooling LOWEST 0.71 1.94 3.41 4.36 5.15 5.88 5.82 6.16 4.36 (59) (47) (46) (42) (34) (43) (45) (120) (436) MIDDLE 0.76 1.75 3.67 4.83 6.16 5.97 6.62 5.75 4.48 (37) (52) (43) (41) (49) (36) (29) (80) (367) HIGHEST 0.74 2.16 3.93 4.97 6.19 6.61 7.39 6.12 4.85 (31) (32) (45) (34) (42) (28) (31) (52) (295) TOTAL 0.59 1.69 3.50 4.35 5.75 6.13 6.50 6.02 3.91 (All women)(242) (233) (190) (167) (143) (113) (108) (253) (1449) Note: The number of observations is in parentheses. The sample for expenditure terciles is 1449 women. Cell sizes were too small to yield meaningful results for subsamples with primary and secondary schooling. - 42 - TABLE A3: Proportion of women never married, by woman's age, location and schooling AGE GROUP ----------------------------------------------------- 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50+ TOTAL Location ABIDJAN .709 .406 .122 .044 .088 .000 .000 .000 .254 (55) (64) (41) (46) (34) (13) (19) (23) (295) OTHER .639 .378 .024 .030 .000 .000 .000 .000 .216 URBAN (83) (59) (41) (33) (27) (25) (13) (39) (320) RURAL .378 .153 .037 .011 .012 .000 .013 .000 .077 (111) (111) (109) (90) (83) (75) (79) (195) (853) ---- Rural areas ---- East .471 .177 .067 .000 .000 .000 .000 .000 .093 Forest (34) (34) (45) (21) (33) (14) (27) (61) (269) West .324 .214 .000 .000 .000 .000 .035 .000 .074 Forest (37) (42) (33) (32) (23) (30) (29) (70) (296) Savanna .350 .057 .032 .027 .037 .000 .000 .000 .066 (40) (35) (31) (37) (27) (31) (23) (64) (288) Schooling NONE .385 .129 .037 .008 .024 .000 .009 .000 .069 (130) (132) (135) (119) (126) (107) (108) (256) (1113) PRIMARY .614 .283 .057 .071 .000 - - - .313 (70) (53) (35) (28) ( 7) ( 2) ( 2) ( 1) (198) SECONDARY .837 .510 .143 .046 .091 - - .. .452 AND HIGHER (49) (49) (21) (22) (11) ( 4) ( 1) ( 0) (154) TOTAL .538 .244 .052 .024 .028 .000 .009 .000 .143 (249) (234) (191) (169) (144) (113) (111) (257) (1468) Note: The number of observations is in parentheses. Proportions are not computed for cells with fewer than 5 observations. - 43 - TABLE A4: Mean age at first cohabitation, ever married or cohabited women, by woman's age, location and schooling AGE GROUP 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50+ TOTAL Location ABIDJAN 15.87 17.37 18.36 19.34 18.77 19.15 17.79 18.36 18.27 (15) (38) (36) (44) (31) (13) (19) (22) (218) OTHER 15.67 17.47 18.23 17.41 17.64 18.88 17.54 19.13 17.79 URBAN (30) (45) (39) (32) (25) (25) (13) (39) (248) RURAL 15.63 15.77 17.00 17.18 16.68 18.03 17.84 18.57 17.28 (67) (94) (105) (88) (79) (75) (76) (188) (772) ---- Rural areas ---- East 16.29 16.00 17.26 16.33 16.91 19.50 18.33 18.87 17.57 Forest (17) (28) (42) (21) (32) (14) (27) (61) (242) West 15.12 15.18 16.39 17.25 15.77 17.43 16.71 17.28 16.54 Forest (25) (33) (33) (32) (22) (30) (28) (65) (268) Savanna 15.68 16.15 17.30 17.63 17.20 17.94 18.71 19.63 17.77 (25) (33) (30) (35) (25) (31) (21) (62) (262) ------------------------------------------------------------------__------ Schooling NONE 15.85 16.14 17.30 17.09 17.08 18.47 17.85 18.63 17.51 (78) (115) (129) (117) (118) (107) (105) (248) (1017) PRIMARY 15.31 16.74 16.67 18.23 16.71 - - - 16.76 (26) (38) (33) (26) ( 7) ( 2) ( 2) ( 1) (135) SECONDARY 15.13 18.17 20.83 21.24 20.80 - - .. 19.36 AND HIGHER (8) (24) (18) (21) (10) ( 4) ( 1) ( 0) (86) TOTAL 15.67 16.54 17.54 17.80 17.34 18.35 17.80 18.64 17.56 (112) (177) (180) (164) (135) (113) (108) (249) (1238) Note: The number of observations is in parentheses. Means are not computed for cells with fewer than 5 observations. - 44 - ANNEX B: Comparison of CILSS fertility results and other demographic surveys This annex compares the fertility results of the 1985 CILSS with selected results of other demographic surveys in Cote d'Ivoire. These include: (a) The 1975 Census, which enumerated 6,709,600 persons, of which there were 1,548,131 women age 15-50. The census did not collect fertility data directly, but serves as a guide to the age structure of the population. (b) The Enquete D6mographique & Passages R6p6t6s (EPR, or multiple-round demographic survey), conducted in 1978-79 to supplement the census with information on population movements and demographic rates. A detailed fertility history was obtained during the first visit, and fertility between visits was recorded during the second and third visits, 3-6 months apart. The survey covered a nationally representative sample of towns and villages. (c) The Ivorian Fertility Survey (IFS), conducted in 1980-81 as part of the World Fertility Survey. The IFS contacted 3770 households and administered individual fertility questionnaires to a nationally representative sample of 5764 women age 15-50. This is the most recent and most thorough survey of fertility, nuptiality, family size preferences and knowledge and use of contraception. The 1985 CILSS interviewed a nationally representative sample of 1599 households and asked the fertility module of one woman 15 or older per - 45 - household, resulting In a sample of 1468 women 15 and older. The selection of only one woman per household means that women from large households had a lower probability of being selected than women from small households. CILSS results must be adjusted to arrive at a nationally representative fertility estimate for comparisons with the other surveys; it is not apparent in which way the selection procedure would affect the results. The principal source of information for the 1975 Census and the EPR is Ahonzo et al (1984) and for the IFS is the 2-volume final report entitled "Enqufte ivoirienne sur la f6condit6" (R.C.I. 1984). Age distribution of the samples Table B1 compares the age distribution of the women age 15-50 or 15-49 in the various samples. The distribution of women who responded to the CILSS fertility module and of all women from the household roster are both presented. The IFS sample is all women responding to the individual fertility questionnaire. The women who responded to the CILSS fertility module are slightly older than women in the Census and EPR, while women in the IFS sample are younger than the Census and EPR samples. The CILSS fertility sample is therefore notably older than the IFS sample. This remains true for the sample of women from the CILSS household roster as well, although the major differences arise from "bunching" at the ends of the CILSS distribution. - 46 - Table B1: Distribution of samples by age group, all women 15-50 AGE GROUP SURVEY 15-19 20-24 25-29 30-34 35-39 40-44 45-50 TOTAL CILSS (f) 20.02 18.81 15.35 13.59 11.58 9.08 11.58 100.0 1985 (249) (234) (191) (169) (144) (113) (144) (1244) CILSS (h) 24.23 18.46 15.09 11.29 10.26 8.95 11.23 100.0 1985 (777) (608) (484) (362) (329) (287) (360) (3207) IFS (f) 22.91 21.77 16.69 12.96 10.18 8.55 6.92 100.0 1980-81 (1321) (1255) (962) (747) (587) (493) (399) (5764) EPR 21.40 19.17 16.96 13.77 11.50 9.29 7.91* 100.0 1978-79 Census 21.04 19.27 19.06 14.10 11.61 8.46 6.46* 100.0 1975 Notes: * indicates for ages 45-49, not 45-50. CILSS (f) is the sample of women asked the fertility module. CILSS (h) is the sample of all women household members 15-50 on the household roster. IFS (f) is the sample of women asked the individual questionnaire. Cell sizes are in parentheses. Sources: RCI 1984, Volume II, Table 0.1.1A, p.1; Ahonzo et al 1984, Annexl, p. 298, and Annex 8, p. 305. Marital status There are great differences in the proportion of women never married across samples (Table B2). The definition of never married for the IFS and in the CILSS fertility module was essentially the same: never married or cohabited. The definitions used for EPR and the census are unknown, but probably did not take cohabitation into consideration. - 47 - Table B2: Proportion of women never married AGE GROUP ---------------------------------------------- TOTAL SURVEY 15-19 20-24 25-29 30-34 35-39 40-44 45-49 15-49 CILSS (f) .538 .244 .052 .024 .028 .000 .009 .173 1985 IFS (f) .440 .104 .045 .019 .007 .002 .003 .134 1980-81 EPR .479 .178 .084 .049 .035 .023 .014 .165 1978-79 Census .505 .187 .100 .070 .050 .056 .053 .242 1975 - - Note: For IFS and CILSS, never married includes women who have never been in a formal or Informal union, Including marriage and cohabitation. Sources: RCI 1984, Volume 1, Table 4.3, p. 70; Ahonzo et al. 1984, Table 5.5, p. 219, and Annex 7, p. 304. CILSS finds much larger proportions of women never married in the two youngest cohorts than does IFS. There is no particular reason why we should expect the two surveys to yield the same results, however, given that marriage will be further delayed as schooling spreads and the surveys are roughly 5 years apart. Table B3 shows that women in the CILSS sample, particularly those in the younger cohorts, have substantially more schooling than those in the IFS, which could account for the differences in marital status. Note in particular that about 20 percent of the 15-24 year olds in the CILSS sample have some secondary schooling, compared to only 14 percent of 15-19 year olds and 10 percent of 20-24 year olds in the IFS sample. - 48 - Table B3: Distribution of samples by schooling (proportion with given schooling level within each age group) AGE GROUP -------TA--------_-_-__---- TOTL SURVEY 15-19 20-24 25-29 30-34 35-39 40-44 45-50 15-50 CILSS 1985 (f) None .522 .564 .707 .704 .875 .947 .979 .715 Primary .281 .227 .183 .166 .049 .018 .014 .158 Secondary+ .197 .209 .110 .130 .076 .035 .007 .126 IFS 1980-81 (f) None .624 .707 .782 .896 .966 .974 .972 .792 Primary .235 .194 .154 .062 .022 .022 .018 .135 Secondary+ .142 .098 .064 .043 .012 .004 .010 .072 Notes: Primary includes any amount of primary schooling; secondary + includes any amount of secondary or higher education. Source: IFS, Volume II, Table 0.1.1A, p. 1. Children ever born Table B4 compares mean children ever born (CEB) by age group across surveys. The first CILSS entry is the mean CEB for the respondents to the fertility module. The second CILSS entry is mean CEB for the entire sample of women on the household roster, averaging together actual CEB for women who responded to the fertility module and imputed CEB for women on the household roster who did not respond to the fertility module. Values were imputed based on the woman's age, schooling, area of residence and the logarithm of per adult household expenditure, using coefficients from the ordinary least squares regression reported in Table 3, column 1 of the text. The CILSS results for the women who answered the fertility module show lower CEB for all but two age groups, compared to IFS. Since the age - 49 - structure of the CILSS women is older, however, mean CEB for the CILSS sample exceed that for IFS: 3.52 vs. 3.35, respectively. Within age groups, CILSS, IFS and EPR results are quite similar. Using the imputed CEB for all women 15-50 in the CILSS survey, mean CEB drops from 3.52 to 3.25 and now is lower than the IFS estimate. The results by age group don't change much (with the exception of 15-19 year olds), but younger women receive greater weight, lowering the total. Table B4: Mean children ever born AGE (AZOUP TOTAL, ----------------------------------------------- AGE SURVEY 15-19 20-24 25-29 30-34 35-39 40-44 45-50 15-50 CILSS (f) 0.59 1.70 3.51 4.38 5.77 6.14 6.20 3.52 1985 CILSS (h) 0.45 1.78 3.30 4.30 5.45 5.96 6.31 3.25 1985 IFS 0.51 1.91 3.34 4.74 5.87 6.73 6.84 3.35 1980-81 EPR 0.6 1.9 3.4 4.7 5.7 6.0 6.1* ** 1978-79 Notes: * indicates ages 45-49. ** indicates total for ages 15-50 not available. CILSS (f) is mean CEB directly from the sample. CILSS (h) is imputed CEB for all women listed on the household roster. The latter is the average of actual CEB for those women who completed a fertility module and imputed CEB for other women, given their age, schooling, residence and income, based on OLS regression coefficients from Table 3 of the text. Sources: RCI 1984, Volume I, Table 5.1, p. 86; Ahonzo et al 1984, Table 4.1, p. 116. - 50 - SOURCES Ahonzo, Etienne, Bernard Barr4re and Pierre Kopylov (1984). "Population de la C6te d'Ivoire: Analyse des donn6s d6mographiques disponibles". Abidjan: R6publique de C6te d'Ivoire, Minist6re de l'6conomie et des finances, Direction de la Statistique. Ainsworth, Martha and Juan Mufioz (1986). "The C6te d'Ivoire Living Standards Survey: Design and Implementation". World Bank, Living Standards Measurement Study Working Paper No. 26. Amemiya, Takeshi (1984). "Tobit Models: A Survey". Journal of Econometrics 24, 3-61. Anker, Richard and James C. Knowles (1982). Fertility Determinants in Developing Countries: A Case Study of Kenya. Li6ge: Ordina Editions. Becker, Gary S. (1965). "A Theory of the Allocation of Time". Economic Journal 75, 493-517. -------------- (1981). A Treatise on the FamilY. Cambridge, Mass.: Harvard University Press. Ben-Porath, Yoram (1974). "Economic Analysis of Fertility in Israel" in T. W. Schultz, ed. Economics of the Family. Chicago: University of Chicago Press. 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. Cochrane, Susan Hill (1979). Fertility and Education: What do we Really Know? World Bank Staff Occasional Papers No. 26. Baltimore: Johns Hopkins University Press. Etienne, Mona (1979). "The Case for Social Maternity: Adoption of Children by Urban Bau16 Women". Dialectical AnthroPoloqy 4 (3). Faruqee, Rashid, K.C. Zachariah, Ken Smith, Nwanganga Shields and Frederick Golladay (1980). Kenya: Population and Development Washington, D.C.: The World Bank. Gourieroux, C., A. Monfort and A. Trognon (1984). "Pseudo Maximum Likelihood Methods: Applications to Poisson Models". Econometrica 52(3): 701-720. Greene, William H. (1981). "On the Asymptotic Bias of the ordinary Least Squares Estimator of the Tobit Model". Econometrica 49 (2). Grootaert, Christiaan (1986). "Measuring and Analyzing Levels of Living in Developing Countries: An Annotated Questionnaire". World Bank, Living Standards Measurement Study Working Paper No. 24. - 51 - Kozel, Valerie (1987). "Income, Consumption and Savings Behavior In the C6te d'Ivoire". PhD Thesis submitted to the Department of Urban Studies and Planning, Massachusetts Institute of Technology. Lee, Bun Song, Louis Pol and Lamlenn Bongsuiru (1986). "The Influence of Rural-Urban Migration on the Fertility of Migrants in Developing Countries: Analysis of Cameroon Data". Draft Final Report. U.S.A.I.D. Bureau of Science and Technology Contract AID/OTR-5412-C-00-1205-00. Maddala, G.S. (1983). Limited Dependent and Qualitative Variables in Econometrics. Cambridge: Cambridge University Press. Mullahy, John (1986). "Specification and Testing of Some Modified Count Data Models". Journal of Econometrics 33, 341-365. Nelson, Forrest D. (1981). "A Test for Misspecification in the Censored Normal Model". Econometrica 49 (5), 1317-1329. Newman, John L. (1987). "Labor Market Activity in C8te d'Ivoire and Peru". World Bank, Living Standards Measurement Study Working Paper no. 36. Oppong, Christine and Wolf Bleek (1982). "Economic Models and Having Children: Some Evidence from Kwahu, Ghana". Africa 52 (4). Page, Hilary and Ron Lesthaeghe, eds. (1981). Childspacinc In Tropical Africa: Traditions and Change. New York: Academic Press. Portney, Paul R. and John Mullahy (1986). "Urban Air Quality and Acute Respiratory Illness". Journal of Urban Economics 20, 21-38. R&publique de C8te d'Ivoire, Minist&re de l'Economie et des Finances, Direction de la Statistique (1984). "Enqufte Ivoirienne sur la F6condit6, 1980-81". Rapport principal (2 vols), Abidjan. Rosenzweig, Mark R. and T. Paul Schultz (1985). "The Demand for and Supply of Births: Fertility and its Life Cycle Consequences". American Economic Review 75 (5). Schultz, T. Paul (1981). Economics of Population. Reading, Mass.: Addison-Wesley Publishing Company. ---------------- (1974). "Fertility Determinants: A Theory, Evidence, and an Application to Policy Evaluation". Santa Monica, California: Rand Corporation. R-1016-RF/AID. Schultz, T.W., ed (1974). Economics of the Family. Chicago: University of Chicago Press. Stapleton, David C. and Douglas Young (1984). "Censored Normal Regression with Measurement Error on the Dependent Variable". Econometrica 52 (3): 737-760. - 52 - Strauss, John (1986). "The Effect of Household and Community Characteristics on the Nutrition of Children: Evidence from Rural C8te d'Ivoire". World Bank, Living Standards Measurement Study Working Paper no. 40. Westlake, Andrew (1987). "Ivory Coast Fertility Surveys: A Comparison of Selected Results from the WFS and LSMS Surveys". International Statistical Institute Research Centre. Processed. World Bank (1984). World DeveloPment Report 1984. New York: oxford University Press. World Bank (1986). Population Growth and Policies in Sub-Saharan Africa. Washington, D.C. World Bank (1987). World Development Report 1987. New York: Oxford University Press. Distributors of World Bank Publications ARGENTINA FLEAND KUWAIT SPAIN Carmo Hiz.I ESRL AkteinsKjakappa MEbR3 Mundi-Presa ibn, SA Calria Gu P.O. Box 128 P.O. B SKS Castdlo37 Flaida 165, 4ih Roor-Of 453/465 SF-0101 38011 Madrid 1333 BuasosAis HadilO 10 MALAYSIA Tlnivemity ofMalsyaCooperatave SRI LANKXA AND THE MAIDIVBS AUSTRALIA, PAFUANEWGUINEA, FRANCE Booksap, LImiStd Lake Hogefbooshop FIULSOLOMONISLAND,S W1d Banck Publicatit P.o. Box 1127, jalan P." Ba P.Box 244 VANUATU, AND WESTERN SAMOA 66, a-star d'IEoa Koaa LIsmpi 10D, Sir Chiftnpalam A. Gardina DA.Book&Jourmab 7SI16Pawis Mawatha 11-13 Statn Stleet MEXICO Colombo 2 Mitdium 3132 GERMANY, FEDERAL REPUBLIC OF INFOlSEC Vicloria UNO-Vag Apartado Posta 2DA60 SWEDEN Poppladarfer AlleSS 14060`1lal pMedcofDF. Fariagto AUSTRIA D-53W0BomI FritsFackbokAloret Geoid and Co. MOROCCO R-npSogatan 12, By 16356 Craben 31 GREECE Sodeted'EtdaM tingMaroceine S-10327 Stlockhol A-1011 Wle IKEME 12e Moztrt,3d.d'Asfa 24, ippodaouSttPa FLaPabs Ciabmca oor BAHRAIN Atih-11635 Weneqrs-WliamaAB Bahrain Rerch and Conyultanq NETHERLANDS B 3X4 Assoiates Lid. GUATEMALA InOr-PubWioH bv. 5-104 25 Stodobn POf Box 22103 UbLeriasFPiedSanuta PO Boxl4 Manar Town 317 Centro Cultural Pledra Snut 7240 BA Lodrin SWITZERLAND 11 mbe 653zma I Fniqlr ktfla BANGLADESH Guatemala City NEW ZEALAND Labralte Paynt Miros Industies Drv neLnt Hilo Ubraryandi nd alonServo 6, rue,rG u Asiniae Soety (MIDAS) HONGIKONG,MACAO PivaBog Case Fpl 381 Houe 54 Road 7A Asia 20D Ltd. New Market CH 1211 Cava 11 DLan,ondi E/Area 6 FL 146 Prnce Edward Rod, W. Auddad Dhaka 1209 KXoloon Foro4kia aiomr Hong Keg NIGERIA Ubraiae Payot BELGIaUM Univerity Press limited ServicedwAboon-ntns Publi-iiad-NakloeUnies HUNGARY Thr CrownaBuldiagJaid Case ptl 3312 Av. du Ri202 KYltih PrivateMallBag95 CHlIOD2 Laiuanne 1060 Bruslas PFO Box 139 padm 13E° Budapest 62 TANZANIA BRAZIL NORWAY O