LSM - 40 L1.fllS ~~~~~~~~APRIL 1988 Living Standards Measurenent SLt dx Working Paper No. 40 The Effects of Household and Community Characteristics on the Nutrition of Preschool Children Evidence from Rural Cote d'Ivoire John Strauss LSMS Working Papers No. I 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. II Three Essays on a Sri Lanka Household Survey No. 12 The ECIEL Study of Household Income and Consumption in Urban Lafin America: An Analytical History No. 13 Nutrition and Health Status Indicators: Suggestions for Surveys of the Standard of Living in Developing Counfries 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 Markef and Social Accounting: A Framework of Data Presenfation 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 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 (List continues on the inside back cover) The Effects of Household and Community Characteristics on the Nutrition of Preschool Children Evidence from Rural 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 policy makers. 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 40 The Effects of Household and Community Characteristics on the Nutrition of Preschool Children Evidence from Rural C6te dIvoire John Strauss The World Bank Washington, D.C., U.S.A. Copyright (© 1988 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 April 1988 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 dcassroom use is not required, though notification of such use having been made will be appreciated. The most recent World Bank publications are described in the catalog New Publications, a new edition of which is issued in the spring and fall of each year. The complete backlist of publications 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 of each of these 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'I6na, 75116 Paris, France. John Strauss is an associate professor at the Economic Growth Center, Departmnent of Economics, Yale University and 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 Strauss, John, 1951- The effects of household and community character- istics on the nutrition of preschool children. (LSMS working papers, ISSN 0253-4517 ; no. 40) Bibliography: p. 1. Children--Ivory Coast--Nutrition. 2. Ivory Coast-- Rural conditions. 3. Households--Ivory Coast. I. Title. II. Series: LSMS working paper ; no. 40. TX361.C5S86 1988 363.8'209666'8 88-10648 ISBN 0-8213-1043-7 ABSTRACT This paper estimates reduced form equations, derived from an economic model of household production, to analyze the impact of household- and community-level variables on child nutrition in rural C6te d'Ivoire. Of particular concern are the contributions made by parental education, household wealth, and community characteristics - some of which are manipulable by government policy. Their impacts on child height and weight-for-height are estimated using a random effects approach (most households have more than one child in the sample) in order to account for common household-level unobserved variables. Estimates for variables which vary within households are also obtained using household dummy variables (fixed effects) to purge possible correlation between unobserved household characteristics and community variables. The results show that both a mother's and a father's education have positive effects on weight-for-height, and that the mother's education has positive effects (though not precisely estimated) on height. The impact of community characteristics is strong. Of these, local wage rates, the health environment and the quality of health infrastructure seem to matter most. Unobserved household-level factors are shown to be quite important, which suggests the need to account for them in an explicit way. In addition there seem to be strong effects of intrahousehold distribution, particularly for children of household heads and their senior wives. - vi - AKOLEDQHENTS This research has been funded by the Living Standards Measurement Study of the World Bank and by the National Institute of Child Health and Human Development research grant number HD21098-01. The paper has benefited from comments and suggestions provided by Nancy Birdsall, Angus Deaton, Paul Gertler, Jean-Pierre Habicht, Yair Mundlak, T. Paul Schultz, Duncan Thomas and Jacques van der Gaag. The general encouragement given by Dennis de Tray has been invaluable as has the assistance of Kalpana Mehra and Michelle Siegel. - vii - TABLE OF CONTENTS 1. Nutrition Outcomes in Rural C6ted'Ivoire.....................l 2. Past Studies. ......**..e.**.......*o..*o*.**o**o**** *o. .....o5 3. Model ...................................99999.*. 8 4.* Data ..................... .................... .......12 Dependent Variables........................................13 Household Variables ......oo................. ..9 e 99...oo 13 Community Variables....o.................................... 15 Interaction Terms .oo......oe.o..e..e..oo....o..o........o 17 5. Resultso.... *99**o ..o........ . .....e.. .o..o.. ..o..21 Household Fixed Effects Estimates ........*...... 21 Mother Fixed Effects Estimates.... .o..o.............. .. 26 Random Effects Estimates..o.................................... 6. Explaining Interregional Differences in Child Nutrition Outcomes. 9o99999o99o999o999999999o9999o9..34 7. Summary and Conclusionso. .......................... .o...37 References ..................................... .39 1. NUTRITION OUTCOMES IN RURAL COTE D'IVOIRE The Cate d'Ivoire has been generally regarded as one of the few economic success stories in sub-Saharan Africa (World Bank, 1978; World Bank, 1982; Zartman and Delgado, eds., 1984), with a 7.1 percent annual growth rate of GDP from 1965 to 1973 and 3.7 percent from 1973 to 1984 (World Bank, 1986). Even its agriculture sector has grown well, with a sectoral GDP growth rate of about 3.5 percent annually from 1965 to 1984. Yet the Cote d'Ivoire has been widely attacked for lagging in social and health investments as well as for regional inequalities (World Bank, 1978). However, looking at indicators of nutritional status (child height and weight for height) for rural children, it would seem that nutrition problems in rural Cate d'Ivoire are mild by African standards. Child height is often used as an indicator of long run nutrition status and weight-for-height as an indicator of short run status by nutritionists (see for instance the volume edited by Falkner and Tanner, 1986). As a component of health, they thus have direct implications for household welfare. In addition, they may have functional consequences. While the functional consequences of moderately short height or moderately low weight-for-height are debated it is clear that at severely short or low levels serious consequences, such as considerably higher risk of death, can arise (see, for example, Chen, Chowdhury and Huffman, 1980, and the survey by Martorell and Ho, 1984). In Table 1 we see that the percent of rural children under six years who in 1985 were less than 90 percent of median height for U.S. children, a widely used standard of stunting for international comparisons (WHO, 1983), was only 10 percent. This contrasts with over 20 percent in surveys taken in the mid-1970's in seven other African countries (see Kumar, 1986). By comparison only 0.5 percent of 2 year old boys in the U.S. fall below this standard (Waterlow et. al., 1977). Some 4.4 percent of Ivoirian pre-school children fall under 80 percent of the U.S. median weight-for-height (compared to 0.6 percent of two year old girls in the U.S.). This is the same percentage as was found for rural Kenya in a 1977 survey (Kumar). However, there do exist African countries (such as Cameroon, Liberia and Sierra Leone) with smaller percentages (on the order of 1 to 2 percent) of so-called "acutely undernourished" children. Nonetheless, compared to non-African countries, particularly in Asia, the 4.4 percent figure is low (Kumar). Thus it is not obvious from these data that the nutrition of rural Ivoirian children, as measured by their height and weight-for-height, is poor, at least by African standards. TABLE 1: Preschool Child Nutrition Indicators for Rural Cote d'Ivoirel! All Rural East Forest West Forest Savannah Percent of NCHS median height 98.2 99.1 98.0 97.2 Percent of NCHS median weight for height 96.6 97.7 95.5 96.1 Percent less than 90% of NCHS median height 10.5 6.8 10.6 14.9 Percent less than 80% of NCHS median weight for height 4.4 4.1 4.7 3.4 al Using full rural sample of 608 children less than 72 months old. Nevertheless substantial variation does exist within the country. Separating the rural area into regions, for instance, shows that nearly 15 percent of children under six fall under the 90 percent of U.S. median height standard in the northern part of the country (the Savanna). This area is drier and has considerably lower incomes than in the south. In the relatively wealthier eastern rural area, part of which is near the capital city of Abidjan, this percentage falls to just under 7. The purpose of this paper, then, is to examine the extent to which such differences can be explained by such regional factors as local wage rates, health, schooling and water infrastructure; by such household level factors as land availability or wealth; and by parental factors, notably education of each of the parents and parental height. The paper presents estimates of a model for the determination of child nutrition outcomes as measured by height and weight-for-height. The contributions of the above mentioned variables are estimated while accounting for common household-level unobserved variables by using a random effects approach. Fixed household effects estimates for variables which vary within households are also obtained, to purge possible correlation between unobserved household characteristics and community variables. A multi-purpose household level data set is used, which contains detailed individual and household socioeconomic information, and in addition has data on many community characteristics. Few prior studies of child nutrition have had available the wealth of such information, particularly on household socioeconomic and community variables. The results show positive effects for both mother's and father's education on weight-for-height, and positive maternal education effects (though not precisely estimated) on height. The impact of community - 4 - characteristics is strong. Of these local wage rates, the health environment and the quality of health infrastructure seem to matter most. In addition there seem to be strong effects of intrahousehold distribution particularly for children of household heads and their senior wives. The outline of the paper is as follows: Section 2 summarizes prior studies of child anthropometric outcomes focusing on some of the methodological problems which have inhibited a causal interpretation of results. Section 3 outlines an economic model of household behavior to motivate the empirical work. Section 4 discusses the data and variable construction. Section 5 reports the fixed effects and random effects estimates. Section 6 uses the random effects results to conduct the counterfactual experiment of what changes would be predicted in child anthropometric outcomes in the poorer Savanna if households had characteristics of representative East Forest household. An additional counterfactual is examined: what predictions would be made if the representative Savannah household lived in the environment of the East Forest. Finally Section 7 concludes. 2. PAST STUDIES There have been a number of past attempts to model nutrition outcomes (see the surveys by Cochrane, Leslie and O'Hara, 1982, Martorell and HabichT, 1986, Strauss, 1985, and Behrman and Deolalikar, 1986). As pointed out in some of these reviews many studies have confused the concepts of production function and reduced form, estimating a hybrid of the two. Most have also ignored the economic model of the household (e.g. Becker, 1965), which treats households as simultaneously choosing inputs into nutrition such as food consumption, length of breastfeeding or whether inoculations are given. For instance, in a recent paper Martorell, Leslie and Moock (1984) use such household choice variables as the months a child is breastfed and the types of foods eaten to explain height and weight of Nepalese children. Many other examples exist of using such endogenous variables.!1 While those factors undoubtedly belong in a well specified nutrition production function, estimating that function using OLS rather than a simultaneous equations estimator will cause biased, and possibly quite misleading, coefficient estimates. As one example, Martorell, Leslie and Moock find no significant / Wolfe and Behrman (1982) include average caloric intake, use of refrigeration and length of breastfeeding as regressors in a child height equation using Nicaraguan data. Ryan, Bidinger, Rao and Pushpamma (1983) use energy and protein consumption, weight, arm circumference, evidence of morbidity signs and labor market participation of the mother in explaining height of rural Indian children. Battad (1978) also uses a food consumption variable on a rural Philippine sample, and Longhurst (1984) uses child immunization, breastfeeding, medical history and birth order variables on rural Nigerian data. Even Heller and Drake (1979), who treat illness as endogenous, take food consumption, lagged height and weightt lagged illness, age at weaning, birth order and birth interval as exogenous in an analysis of Colombian children. Other studies, mostly by nutritionists, just use crosstabs, not regression analysis. Some of these are summarized in Martorell and Habicht (1986). effect of mother's or father's education on height or weight-for-height in their Nepal sample. However they are holding constant variables, such as the type of food eaten and the number of months breastfed, which these education variables could be expected to affect. A few studies have been more careful to specify what is exogenous and what is not. In a paper closely related in spirit to this one, Rosenzweig and Schultz (1982) estimate reduced form equations for Colombian fertility and mortality outcomes using individual, household and community level variables, as is done here. Pitt and Rosenzweig (1985) use a farm household model to derive reduced form equations for reported illness in Indonesia. Chernichovsky and Coate (1983), Chernichovsky et al. (1983), Blau (1984) and Behrman and Deolalikar (1985) estimate reduced form equations for height and weight. The first three data sets had very few variables to work with, especially on the community side. In the Behrman and Deolalikar analysis which reanalyzed the rural Indian data used by Bidinger et al. (1983), very few strong results emerge for the weight-for-height reduced form equations. Finally Horton (1984, 1986) and Barrera (1987) also estimate reduced form equations for height-and-weight-for-height using data from Bicol, Philippines. Both find stronger effects of parent's education than most previous papers. Horton's 1984 paper is also very closely related to this because she also looks at household fixed effects estimates, in her case of the impact of birth order and its interaction with other variables, on height and weight-for-height. One interesting finding is the positive effect of low birth order on height outcomes, an effect which is reduced for children of more educated mothers. By carefully choosing variables to correspond to those which can more reasonably be viewed as given to the household, the full effects of variables such as parental education or household wealth can be measured. The next section considers a model by which careful variable selection can be made. - 8 - 3. mODEL The reduced form equations for children's anthropometric measurements can be derived from a model of a multi-member household in which production and consumption choices are made. The basic model for one-person farm households is outlined in Singh, Squire and Strauss, 1986. An extension incorporating a health production function and multiple household members is provided in Pitt and Rosenzweig (1985), and that is the approach followed here. Since the model is only used to provide guidance on the choice of exogenous and endogenous variables only a cursory overview will be given. The household behaves as if maximizing a long run utility functionz- having as arguments the nutritional and morbidity status (including height and weight-for height) of each member, each person's leisure time, consumption of a bundle of foods, and consumption of non-food. Health and nutritional status of the household members as well as food consumption enter directly into the utility function because good health is desirable in itself and because foods are consumed partly for reasons other than their nutritive values. Four sets of constraints face these households. Each individual faces a time constraint. There is the usual household budget constraint in which individual market labor supplies are taken to be choice variables. There is a farm production function, and finally there exist nutrition and morbidity production functions. The nutrition production functions, particularly for height and weight, should ideally be represented in terms of 2/ In principle, a dynamic problem with a dynamic nutrition production function would be the appropriate approach. With only cross-section data estimating such a model is not possible. - 9 - changes, or growth, since both height and weight are stocks.31 Growth can be represented as a function of past levels and growth of height (or weight), of current morbidity, food intake, and energy expenditure, as well as of individual and parental characteristics (including time use), some of which are observed in the data and some not. Given a function for initial conditions, i.e. birthweight and length, the technology would be closed. Similarly a production function exists for current morbidity, which may include the current morbidity of other household members and health inputs such as inoculation and consumption of medicines, in addition to inputs in the nutrition production function. Modeling nutrition outcomes in terms of flows rather than stocks results in the derived reduced forms for the stock variables which include lagged exogenous variables as arguments.41 These exogenous variables include five categories: individual level variables such as age and sex; parental variables such as education, age, and height; household level "exogenous" variables such as land area and production assets (or non-labor income)5/; community-level variables having to do with the cost of using (and information concerning) health and schooling facilities; and community prices of consumption items such as foods, non-foods, and leisure. Observations on the 3/ The same issue arises for educational production functions. Hanushek (1979) discusses this issue well. 4/ A second year of data will allow for reduced form and production function equations for growth to be estimated at a later date. 5/ With a time persistent unobservable such as land or management quality in the farm production function, asset accumulation may be correlated with the reduced form equation error term, which will include these unobserved farm heterogeneity effects. This is ignored here. - 10 - lagged values of these exogenous variables are usually not available. Fortunately, many of these variables, such as parents education, will be time invariant. The community variables may change over time as new investments in facilities are made or as prices change. In this paper only young children are considered which should mitigate the absence of lagged community variables. Furthermore, national community rankings in terms of infrastructure typically change only slowly; high infrastructure communities tend to remain so, certainly over short time horizons as exists in these data. The same argument applies, though to a lesser extent, to food prices and wages. A potential complication which needs discussion has to do with the two assumptions implicit in treating the community level variables as exogenous. Health and schooling infrastructural investments may depend on underlying community factors such as income and distance to major towns which in turn may be correlated correlated with unobserved household characteristics. If, for instance, hospitals are located in healthier areas for both demand (incomes are higher) and supply (better infrastructure) reasons, then one would expect a negative correlation between distance to a hospital and nutrition outcomes. This correlation would have resulted from both variables being correlated with the underlying community factors (see Rosenzweig and Wolpin, 1986). A second potential complication is that households may selectively migrate partly in response to different health and schooling infrastructure of communities. This too would cause a correlation between unobserved household characteristics and community variables, as shown by Rosenzweig and Wolpin (1984). It is not clear whether these two sources of correlations between community and unobserved household variables will be - 11 - important for the C6te d'Ivoire sample. We will test below for this possibility. One way to eliminate the impact of unobserved household heterogeneity is to estimate a household fixed effects model. This is one approach taken here. Unfortunately, community-level variables are eliminated by that procedure, although not interaction terms between community variables and variables varying within a household. Provided correlation does not exist between community variables and unobserved household factors, a household random effects model can be estimated to obtain unbiased estimates of the effects of the community variables. This will result in more efficient estimates than OLS provided that there is more than one child under six years in most households.- Given both random and fixed effects parameter estimates, the hypothesis of no correlation between community variables and the underlying error structure can be tested using a Wu-Hausman specification test. 6/ Only 57 children out of 504 in the sample have no siblings under six years old. - 12 - 4. DATA The data were collected by the World Bank's Living Standards Measurement Studies unit in collaboration with the Department of Statistics of the Ministry of Finance in C8te d'Ivoire (see Ainsworth and Munoz, 1986, and Grootaert, 1986, for a detailed description). A national probability sample of 1600 households were visited twice, two weeks apart, out of which 636 households had height and weight measurements taken, in principle, of all household members 7 Of these households, 365 reside in rural villages, and since community data were only collected for rural communities, the analysis is limited to these households. There are 657 children under six years who are residing members of these households, and of those 608 (92.5%) were measured.8/ Of the children measured, 85 children are dropped because no measurements were taken on their mothers, either because the mothers do not 7/ The anthropometric measurements did not begin until seven months into the survey. Communities were in principle randomly distributed across months in data collection, so that the households with measurements taken should be close to a random sub-sample of the entire sample. 8/ The percentage measured is slightly higher for boys than girls (93% vs. 90%) and is reasonably stable across the 0-6 year age group. These are high percentages in a developing country. Since the children are preschoolers school is unlikely to be a factor in their absence. While a potential selectivity issue is thus raised, nothing is done to correct for that in these results. Partly this is because it is not obvious what instruments are available for identifying the selectivity equation parameters other than functional form assumptions. A slightly different issue is that some 91 children under 6 are living elsewhere, usually with relatives, despite their being family members. Sending children to live with relatives, fostering, is not unusual in Africa. As one would expect, the proportion of children living outside the household rises with school age, though this is not a factor for these results on preschool aged children. Counting those children residing elsewhere, 81 percent of all living children under six years were measured. - 13 - live in the household or because they were absent during the two w9/ Finally some 19 children had to be dropped because of interviews. incomplete data, leaving a sample size of 504 for analysis. Dependent Variables Heights and weight-for-height are standardized using the age-sex specific median from the U.S. NCHS standards (see WHO, 1983). Evidence exists from a number of developing countries (see Habicht et al., 1974, and Martorell and Habicht, 1986) that heights for well nourished children in developing countries (outside of east Asia) are close to the NCHS median, while heights of poorly nourished children are not. The coefficients which will thus be most affected by the standardization are those on the child age dummies and on child sex. These will have to be interpreted as the effects relative to the U.S. standard.10/ Household Variables Mother's height is also standardized using NCHS standards. Standardizing adult heights raises more questions than for children because of ethnic differences in the onset and length of the adolescent growth spurt. Standardization is used here to try to control for mothers who are still 9/ These include children who are fostered in. Comparing variable means between the two groups shows a difference in mother's education, with the absent mothers having slightly more. Given the importance of mother's height in the height equation it was considered that the omitted variables bias in excluding mother's height but including the extra 85 children would be greater than any selectivity bias from excluding children of unmeasured mothers. 0/ Regressions using the unstandardized height and weight-for-height were run, including a cubic spline in child age, with the coefficients for variables other than child age and sex essentially unchanged. - 14 - adolescents, between 13 and 17. For those greater than 17 years the same (sex specific) standard is used. The education variables are based on years of completed education. Dummy variables were constructed for each parent, equal to one if the parent has completed four or more years of school. Two dummy variables are used which measure effects of household organization on allocation of resources within the household. These are whether the child's mother is the household head or senior wife, or whether she is the junior wife of the household head. The omitted category comprises children whose parents are not head of household. Age of mother at the child's birth is included to capture the possibility that very young mothers may have less well nourished children. The variable is entered using a linear spline. This piecewise linear specification allows mother's age at birth to have different effects before and after the cut-off point(s), while the two pieces are constrained to be equal at that point. The cut-off point used in the estimation is eighteen 11/ years* The asset variable used in this study is calculated as cultivated land per adult. In addition, we include the percentage of cultivated land planted to tree crops as a crude attempt to proxy for agro-climatic conditions. Coffee, cocoa, coconut and oil palms are major sources of 11/ A quadratic specification resulted in a worse fit, and gave the unlikely result that the turning point for the positive impact of mother's age at birth was over 45 years. - 15 - agricultural income, typically with higher net returns than food crops, but can only be grown in more humid areas found in the southern part of the country. Asset variables are used rather than non-labor income for two reasons. First the non-labor income, which includes both net farm and non- farm enterprise returns (but not subtracting the value of family labor), has major measurement problems.121 Second, essentially all of the children reside in farm households. For such households, farm income net of family labor would be uncorrelated with unobserved household effects provided the household's production and consumption decisions were recursive, or separable (see Singh, Squire and Strauss, 1986, and Pitt and Rosenzweig, 1985, for an application). However if production also depended on consumption decisions, as it would if health or nutrition of adults entered the farm or non-farm production functions, then net returns would be correlated with the child nutrition reduced form disturbance and thus raise a simulataneity issue. As stated, the appropriate concept in a recursive model are the returns netting out household labor, unless family labor supply is assumed to be fixed. Complete household labor data are not available for own account farm or non- farm enterprises making construction of such a variable impossible. Community Variables The community-level variables were obtained independently from the household level information by interviewing a group of village elders, 2/ The value of self-produced food consumption was derived by using purchase prices, an upperbound to the shadow value. In addition only a one week reference period was used, which ignores seasonal variations which can be considerable. The income from non-farm self-employment is also weak, being based on retrospective rather than prospective interviews. - 16 - including the chief, (Ainsworth and Munoz). These variables represent the overall availability of various services (schools, health facilities) within a village and the distance to the nearest facility, if they are not available in the village, thus proxying for prices and information. Data on the service use of each household, although available, is not used because it reflects household choices.131 In addition a variable measuring the "average" village agricultural wage for male labor is used. This should proxy both for overall income levels as well as for wage opportunities in the village. We also include water source dummy variables that show the dominant village source (only one per village) for the rainy season (there are a few, but not many differences during the dry season). The categories included are wells without pumps and natural sources (which includes rain, lakes and rivers). The omitted category includes both private tap connections and village wells with pumps. The major community health problem and health service problem data were derived from two open ended questions which asked about the four major health or health service problems. Answers were subsequently coded into dummy variables based on whether or not a disease was mentioned as being one of the three categories: malaria, dysentery, measles or chicken pox. Not having medical problems was the omitted category. A large proportion of villages gave no answer, which may or may not indicate absence of problems. It is reasonable to presume that those villages reporting such problems are 13/ Food prices, are omitted because of missing data. This does not affect the household fixed effects estimates since they are eliminated by the within household differencing. - 17 - seriously affected.14/ Unfortunately more objective epidemiological information is not available except in the form of hospital reports. Since hospitals can only report cases when someone has sought treatment any probabilities based on such cases will be biased. For the health service problem variable, there are very few villages which did not provide answers, although the answers might be modeled as an outcome of an information process (see footnote 14). Absence of medicines was coded as a separate response, as was having congestion problems (not enough room) at the nearest hospital. A third category, water problems, include not having drinking water, no sanitation services or no well. The residual category is comprised of no problem or having no services. The latter is already accounted for by the distance variables. Interaction Terms We include interactions between mother's education and the community variables, particularly the distance variables, to capture substitutability or complementarity between provision of public services and parental education (see Rosenzweig and Schultz, 1982). Mother's education is used since the mother's time is likely to be intensively used (relative to the father's) in child rearing. In addition to capturing technical relationships between education and use of health, education and water services in the nutrition production function, these interactions will also capture inequality of access to the extent it exists and is related to education levels of the household. 14/ One can think of the reported health problems variables as an outcome of an information process which is a function of the underlying true community health state, health services provided to the community, and the average level of education in the community. - 18 - Sample variable means and standard deviations are reported in Table 2 for all of rural Cote d'Ivoire and broken down by three regions: East Forest, West Forest and Savannah. Table 2 indicates the decline in resources from the East Forest to West Forest to the Northern Savanna. The Savanna in particular has very few educated parents, little land planted to the lucrative tree crops of coffee and cocoa, and a consequent low male agricultural wage rate. Reported disease problems are also much more common there. These differences in resources are reflected in lower standardized heights and weights for height, both for children and their mothers. TABLE 2: Variable Means and Standard Deviations: By Region All Rural East Forest West Forest Savannah Variable Mean (sd) Mean (sd) Mean (sd) Mean (sd) Age (months) 32.0 (19.1) 33.8 (19.4) 29.9 (18,5) 31.7 (19.1) Standardized Height (percent) - All agesal 97.9 (7.1) 99.3 (6.8) 97.7 (7.2) 96.7 (7.1) Age < 6 months - 100.5 (5.8) 6 mos. < Age < 24 mos. 99.3 (6.7) 24 mos. < Age < 48 mos. 96.6 (7.5) 48 mos. < Age < 54 mos. 96.4 (6.9) 54 mos. < Age < 72 mos. 98.4 (6.8) Standardized Weight for Height (percent) - All agesal 96.4 (10,9) 97.6 (10.4) 95.2 (12.6) 96.0 (10.0) Age < 6 months 101.9 (10.9) 6 mos. < Age < 24 mos. 92.8 (12,0) 24 mos. < Age < 48 mos. 97.0 (10.3) 48 mos. < Age < 54 mos. 95.7 (10.4) 54 mos. < Age < 72 mos. 97.7 (9.3) Mother's Standardized Height- All agesa/ 96.6 (3.5) 97.2 (3.4) 96.3 (3.7) 96.1 (3.2) Log of standardized child height - All agesa/ -.024 (.07) -.010 (.07) -.027 (.07) -.036 (.07) Log of standardized child weight/height!' -.043 (.12) -.030 (.11) -.058 (.14) -.046 (.11) Child Age > 6 mos. (S) .91 .92 .91 .89 Age > 24 mos. (%) .67 .71 .63 .66 Age > 48 mos. (%) .31 .35 .27 .31 Age > 54 mos. (%) .19 .23 .14 .18 Male child (%) .52 .52 .55 .50 Mother's education > 4 (%) .13 .20 .17 .01 Father's education > 4 (%) .21 .32 .33 .01 Mother's age at birth (years) 27.9 (8.4) 28.5 (8.7) 27.1 (8.3) 27.9 (8.0) Table 2 (continued). All Rural East Forest West Forest Savannah Variable Mean (sd) Mean (sd) Mean (sd) Mean (sd) Child of senior wife (%) .43 .36 .53 .43 Child of junior wife (%) .21 .23 .20 .21 Land per adult (hectares) 1.14 (1.09) 1.18 (1.00) 1.22 (1.37) 1.04 (0.96) % land in treecrops .45 .59 .68 .13 Community level variables: Male daily agricultural wage (1OOOCFA) .549 (.20) .648 (.21) .524 (.04) .463 (.19) Distance to doctor (l00km) .31 (.24) .27 (.11) .36 (.21) .32 (.33) Distance to nurse (100km) .12 (.10) .05 (.10) .16 (.09) .16 (.08) Distance to primary school (100km) .002 (.01) .001 (.003) .007 (.01) 0.0 (0.0) Traditional healer In village (%) .89 1.0 .58 1.0 I Well without pump (%) .21 .18 .57 0.0 Natural water sources (%) .44 .14 .43 .76 Major community health problems: Malaria (%) .15 .12 0.0 .29 Dysentery (%) .13 .12 0.0 .24 Measles, chickenpox (%) .13 0.0 .23 .18 Major community health service problems: No medicines (%) .07 .10 .10 0.0 Water problems (P .28 .31 .53 .06 Congestion at nearest hospital (P .15 .27 0.0 .12 Sample Size 504 191 128 185 a/ NCHS standards are used. - 21 - 5. RESULTS Household Fixed Effects Estimates Table 3 reports the specifications for the household fixed effects model (Columns 1,2 and 3) and the mother's fixed effects model (Column 4, see paragraph below, Mother Fixed Effects Estimates) There are 211 households represented by the 504 children. Because many of these households are extended, many contain children having different parents (there are 320 mothers). In consequence even when the household fixed effects model (which is equivalent to using household level dummy variables) is used, coefficients for mother level variables can be estimated. Three fixed effect specifications are reported in this table. Columns (1) contain estimates which omit the interactions between community variables and mother's education. Columns (2) adds the two household organization variables: whether the child's mother is the senior wife of the household head (or household head in a female headed unit) or the junior wife. 151 These variables potentially represent effects of intrahousehold distribution, which has not been carefully examined using African data.16/ While household formation might reasonably be treated as endogenous (indeed results reported below are consistent with this interpretation), the fixed effects estimation removes unobserved household level effects, a major source of endogeneity. The third specification, columns (3), add the community infrastructure-mother's education interactions. 15/ Separating being a child of the senior wife of a head with one wife or of a head with multiple wives shows no statistical improvement in fit at the .10 level and so was not incorporated. 6/ An exception is a study of child heights in Gambian rural households showing a positive effect of being a child of the household head (Von Braun, Puetz and Webb, 1987). TABLE 3: Household and Mother Fixed Effects Estimatesal Height/Age b/ Weight/Height b/ Household Effects Mother Effects Household Effects Mother Effects Variable (1) (2) (3) (4) (1) (2) (3) (4) Child age > 6 mos. -.022 -.021 -.020 -.027 -.082 -.080 -.083 -.097 (1.53) (1.52) (1.42) (1.46) (3.49) (3.41) (3.47) (2.83) Child age > 24 mos. -.027 -0.29 -.030 -.029 .051 .050 .052 .075 (2.96) (3.21) (3.22) (1.29) (3.41) (3.28) (3.34) (1.82) Child age > 48 mos. .024 .023 .022 .020 -.034 -.037 -.038 -.009 (2.02) (1.95) (1.92) (0.95) (1.76) (1.89) (1.92) (0.23) Child age > 54 mos. -.002 -.003 -.003 -.009 .047 .047 .048 .052 (0.16) (.025) (.023) (0.49) (2.12) (2.12) (2.12) (1.61) Male child -.003 -.007 -.008 -.002 -.0002 -.001 .0002 .002 (0.45) (.090) (1.02) (0.28) (0.02) (0.10) (0.01) (0.13) Log mother's standardized height .475 .478 .504 -.127 -.125 -.150 (3.03) (3.09) (3.20) (0.49) (0.48) (0.57) Mother's education > 4 years .021 .025 .056 .043 .047 .040 (1.21) (1.43) (1.45) (1.47) (1.59) (0.61) Father's education > 4 years .002 .002 .0007 .059 .064 .066 (0.15) (0.11) (0.04) (2.33) (2.49) (2.54) Mother's age at birth .003 .002 .003 .005 .026 .026 .026 .036 (0.48) (0.36) (0.41) (0.29) (2.26) (2.22) (2.21) (1.16) (Mother's age at birth-18)* -.003 -.003 -.004 -.008 -.027 -.027 -.027 -.030 Dummy = 1 if > 18 years (0.42) (0.41) (0.49) (0.28) (2.24) (2.25) (2.23) (1.16) Child of senior wife .033 .031 .028 .029 (2.21) (2.10) (1.14) (1.15) Child of junior wife -.007 -.006 .030 .031 (0.40) (0.37) (1.10) (1.12) Interactions with mother's education: Distance to doctor (100km) -.220 .209 (0.76) (0.43) Distance to nurse (100km) -.300 -.371 (0.65) (0.47) Distance to primary school (100km) -.588 -3.388 (0.15) (0.53) .58 .60 .60 .75 .54 .55 .55 .71 Standard error .062 .062 .062 .061 .103 .103 .104 .104 N 504 504 504 504 504 504 504 504 Number of Groups 211 211 211 320 211 211 211 320 a/ T-statistics are in parenthesis. b/ Log of standardized measure using NCHS standards. - 23 - The child age coefficients suggest a lag in height growth, relative to U.S. median children, starting at six months, accentuated at around two years, and bottoming out around four years. This is a typical pattern in developing countries,-7/ which may be related to the introduction of solid foods and the termination of breast feeding.18/ For weight for height the relative lag begins at six months, bottoming out at two years. A male sex dummy is not significant. The coefficient is actually negative for both height and weight for height, but negligible in magnitude. Sex bias relative to U.S. standards has not been found in other anthropometric surveys in sub-Saharan Africa (see Svedberg, 1987), nor has it been found there in child mortality rates. The effects of both mother's and father's education are non-linear, the impact being negligible until the fourth year.19/ The education coefficients are stronger for weight for height than for height. Mother's education does have a small positive effect on height, which is consistent with a role for mother's education outside of producing income. Mother's height has an important impact on child height but none (the point estimate being slightly negative) on weight for height. Mother's height encompasses both genetic and human capital investments -- the latter 7/ See, for instance, the review of African evidence by Svedberg (1987) or the work of Barrera (1987) for the Philippines. 18/ By six months 81 percent of children are still being breastfed, which declines to 51 percent by twelve months, 23 by eighteen months, and 4 by two years. No information is available on the timing of introduction of solid foods. 19/ Results including dummies for both 1-3 years and 4 or more years of education, not reported, indicate this. - 24 - including mother's nutritional status during pregnancy. The negative coefficient in the weight for height equation seems inconsistent with an interpretation of mother's height as only proxying for underlying unobserved health endowment. On the other hand a pure genetic effect does not seem to fit well either, since when father's height is added on a considerably reduced sample2'/ the mother's height coefficient retains its magnitude and significance but the effect of father's height is imperceptible (a coefficient of -.11 and t-statistic of .3). While a lower impact of father's as compared to mother's height on infant length is found in the medical literature (see Garn and Rohman, 1966 or Mueller, 1986), height of older children are usually associated with both parents' height. Having a mother who was under 18 years at birth is associated with worse anthropometric outcomes, with a p-value of under .05 for weight for height, but not significant for height. This is holding constant her height relative to age-specific standards as well as her education. The effect seems to be very non-linear, ending at age eighteen.21/ 20/ To include father's height it is necessary to drop an additional 103 children whose fathers were not measured. Not only is sample size thus reduced, but its characteristics, particularly for parents' education is drastically altered. The percent of children having mothers with a fourth grade education or more declines from 13 to 5 while the percent with fathers having 4 years or more education falls from 21 to 13. Given the unimportance of father's height in the reduced sample and the enormous reduction in variation of the education variables, not to mention the greater possibility for contamination of random effects results because of selectivity bias it was decided to drop father's height while retaining the larger sample. 21/ Using mother's age at the time of the survey gives very similar results. Age at birth is reported because it is easier to interpret. - 25 - Being a child of the senior wife of the household head has a positive impact with similar magnitudes on both height and weight-for-height, though only the effect on height is precisely estimated. Being a child of the junior wife has essentially the same impact on weight-for-height as does being a child of the senior wife, suggesting that it is being a child of the household head which is making the difference here (not being a child of the head being the omitted category). For height outcomes this does not seem to be the case however; being a child of the senior wife is associated with a higher height than is being a child of a junior wife. The senior wife's children are likely to have been born earlier than those of the junior wife, when household resources may not be so stretched from any unplanned births. Apparently this disadvantage gets eliminated over time, which is reflected in improved weight- for-height, but not height. This result is very similar to Horton's (1986) result of the effect of birth order221 on height and weight-for-height for children living in the Bicol region of the Philippines. The interactions of distances to health and schooling facilities with maternal education fail to be significant at standard levels; F-statistics are 0.5 for the height regression and 0.3 for weight for height. This partly reflects the unimportance of these distance variables in explaining child anthropometric outcomes (see the random effects section), and partly the low level of education in rural areas, especially in the Savannah. The negative coefficients imply a complementarity of these services with mother's education. 22/ A pure mother birth order variable is available for only a very small subset of the children so it was not possible to use it in the analysis. - 26 - Mother Fixed Effects Estimates The last set of columns provide estimates when mother level fixed effects are used rather than household level effects. The question here is do the mother specific variables used in the household fixed effects specification carry most of the mother-related information. This is addressed by an F-test, possible since the child level variables (the age and sex dummies) are in both the household and mother effect specifications, while all the other variables in the household effects case (including father's education and the household head dummy) take on identical values for a given 23/ mother.- The F-statistics are 1.04 for the height equation and 0.94 for the weight-for-height equations with 104 and 177 degrees of freedom. Neither is significant at the .10 level. By contrast, testing the household fixed effects against OLS estimates with community and household level variables (or alternatively, with community dummy variables and household variables) show the household effects 23/ We can think of two equations where i indexes households, j indexes mothers and k indexes children: (1) Yijk Z Xi + jk + ai Eijk (2) Yijk =Xijko + .ij + ijk The vector of mother specific variables, Zij, and the household dummies, ai, are contained in the column space of the mother dummy variables, the zij's. Equation 1 is therefore nested within the second equation so an F-test will be a likelihood ratio test in this case. - 27 - to add significantly, with p-values under .01 for the height equation and at .05 for the weight/height equation (F-statistics of 1.51 and 1.34 respectively for the non-interaction specification with 195 and 281 degrees of freedom). In consequence it is reasonable to conclude that while unobserved household level effects are clearly important in these reduced form child height and weight/height equations, unobserved mother effects are not. Random Effects Estimates Household level random effects estimates are reported in Table 4. Three sets of estimates appear there. Columns (1) contains estimates which omit community level variables and interactions. While this specification is incorrect given the presence of community effects exogenous to the household, it may serve as a useful benchmark since many previous studies have likewise used only individual and household level variables. Columns (2) and (3) report estimates adding community variables both with and without interactions with mother's education. All regressions are run excluding the child of senior or junior wife variables because Wu-Hausman specification tests indicate correlation between explanatory variables and the household random effects when these household composition variables are included. Chi-square statistics for the specification tests in this case (excluding interactions) are 24.8 (12 degrees of freedom) for height and 10.3 for weight for height. While the latter statistic is not significant at usual levels the former has a p-value of under .025. Omitting the two wife variables, however, results in test statistics of 14.6 and 8.5 (with 10 degrees of freedom) respectively for the height and weight for height regressions. Neither is significant at the - 28 - TABLE 4: Household Random Effects Estimates-' Variable Height/Age-/ Weight/Height-/ (1 (2) (3) (1 (2) (3) Age > 6 mos. -.019 -0.15 -.016 -.098 -.092 -.091 (1.60) (1.30) (1.37) (5.09) (4.78) (4.69) Age > 24 mos. -.026 -0.30 -.030 .050 .048 .046 (3.34) (3.80) (3.73) (3.93) (3.67) (3.55) Age > 48 mos. .003 .006 .006 -.017 -.017 -.017 (0.33) (0.59) (0.59) (1.08) (1.07) (1.02) Age > 54 mos. .018 .016 .016 .029 .029 .028 (1.61) (1.47) (1.46) (1.58) (1.61) (1.54) Male child -.004 -.005 -.005 .004 .0005 .001 (0,73) (0.89) (0.87) (0.40) (0.05) (0.13) Log mother's standardized height .337 .360 .354 .026 .035 .025 (3.58) (3.82) (3.74) (0.17) (0.23) (0.16) Mother's education .014 .014 .007 .027 .032 .059 > 4 years (1.16) (1.14) (0.30) (1.47) (1.65) (1.62) Father's education -.001 -.002 -.002 .016 .018 .019 > 4 years (0.13) (0.17) (0.18) (1.02) (1.14) (1.17) Mothers age at birth .0003 -.0001 .0001 .016 .019 .019 (0.05) (-0.03) (0.04) (1.83) (2.15) (2.13) (Mother's age at birth -18)* .0005 .0008 .0005 -.017 -.020 -.020 Dummy = 1 If > 18 years (0.10) (0.15) (0.08) (1.85) (2.19) (2.17) Land per adult (hectares) -.002 -.001 -.001 .008 .009 .009 (0.58) (0.35) (0.34) (1.57) (1.67) (1.64) % Land in tree crops .018 .012 .011 .002 .007 .003 (1.92) (1.11) (1.09) (0.16) (0.40) (0.17) Community level: Male agricultural wage .063 .062 .100 .089 (1OOOCFA) (2.24) (2.18) (2.20) (1.94) Distance to doctor (100km) -.008 .009 -.018 -.017 (0.47) (0.47) (0.56) (0.56) Distance to nurse (lOOm) -.029 -.033 .039 .070 (0.60) (0.64) (0.49) (0.84) Distance to primary school -.670 -0.344 -.317 .547 (100km) (1.06) (0.49) (0.31) (0.48) - 29 - Table 4 continued Variable Height/Agew-' Weight/Heightt/ (1) (2) (3) (1) (2) (3) Dummy if village has: Traditional healer -.041 -.036 -.025 -.012 (2.32) (2.01) (0.88) (0.42) Well without pump -.047 -.045 -.046 -0.39 main water source (3.43) (3.20) (2.06) (1.74) Natural sources main -.012 -.011 .008 .004 water sources (0.94) (0.89) (0.40) (0.18) Dummy = 1 if Major community health problems: Malaria -.027 -.025 -.064 -.077 (1.72) (1.60) (2.52) (2.61) Dysentery -.036 -.036 -.008 -.009 (2.55) (2.51) (0.35) (0.41) Measles, chicken pox -.011 -.013 -.040 -.403 (0.81) (0.96) (1.83) (1.96) Dummy = 1 if Major community health service problems: Absence of medicines -.053 -.052 -.063 -.061 (2.65) (2.61) (1.93) (1.89) Water or sanitary service -.014 -.013 -.048 -.045 problems (1.01) (0.91) (2.08) (1.95) Congestion problems -0.31 -.031 -.030 -0.33 (2.40) (2.39) (1.44) (1.57) Interactions with mother education dummy: Distance to doctor (100 km) .026 -.044 (0.36) (0.37) Distance to nurse (100 km) .054 -.117 (0.42) (0.56) Distance to primary school (100 km) -.974 -3.665 (0.76) (1.75) Constant .001 .058 .049 -2.84 -.327 -.355 2 (0.01) (0.59) (0.49) (1.86) (2.04) (2.06) R .16 .22 .22 .17 .20 .20 F 7.38 5.14 4.66 7.51 4.47 4.17 Standard error .062 .062 .063 .104 .104 .104 N 504 504 504 504 504 504 a/ T-statistics are reported in parenthesis. b/ Log of standardized measurements using NCHS standards. - 30 - .10 level or better. This suggests that these two variables are the source of mispecification, consistent with viewing household formation as being subject to household choice.24- When community variables are excluded child age and mother's height have strong impacts. The percentage of land in tree crops is positively related to height, as are land, mother's education and age at birth to weight- for-height. None are very precisley estimated however. When community variables are added the variable measuring the percentage of land in tree crops loses significance, its coefficient being reduced by one third, while coefficients for parental education, maternal stature and age at birth, and land per adult all rise in levels and significance. Comparing the weight-for- height and height equations the impacts of parental education and land area cropped are larger for the former, suggesting a larger income effect there.-5/ Maternal education has a greater effect than does father's for both outcomes, though the difference is not statistically significant. This is consistent with the hypothesis that maternal education affects child nutrition in ways different from a pure income effect. Mother's age at birth, if less than 18 years, continues to have a positive effect on weight for height, as was the case for the fixed effects estimates. 24/ Choices include whether to extend the household vertically, by living with parents; or horizontally by being polygynous (see Amyra Grossbard, 1976 for an economic analysis) or living with siblings. 25/ When the land variables are replaced by non-market income (but not netting out the value of family labor) the results are very similar. F-statistics of the joint significance of non-market income and its square are 1.27 for height and 1.91 for weight for height (p-values of .28 and .15 respectively). - 31 - The community variables are jointly significant with p-value less than .01 for height (F-statistic of 2.41 with 13 and 478 degrees of freedom) and at the .15 level for weight-for-height (F equal to 1.47), with many being individually significant at the .05 level. The daily male agricultural wage is significant at the .05 level with an elasticity of .034 for height and .055 for weight-for-height. Consistent with the coefficients for household level variables the income effect is larger for weight-for-height than for height. The coefficients imply that if the daily wage in the poorer Savanna region should rise from its mean of 463 CFA to 648 CFA, the average in the wealthier East Forest, that mean children's height would rise by 1.2 percent and weight for height by 1.7 percent. The implied rise in predicted standardized heights is from 96.6 to 97.8 percent of the U.S. median which would put heights on par with those in the West Forest region. For weight for height it is from 95.5 to 97.2 percent approximately the mean level of weight for height in the East Forest. The effects of distances to health facilities or primary schools are negative as expected except for distance to nurse in the weight-for-height equation, but the magnitudes are small in both equations. This negative impact could be related to the extra time cost of using the facilities when they are situated farther away, the potential loss in information, or unobserved community health infrastructure being proxied by these distances. Jointly the distance coefficients are not significant at any standard level. The height distance elasticities are -.002, -.003 and -.001 for doctor, nurse and primary school distances, while the weight-for-height elasticities are -.006, .005, and -.0006 respectively. - 32 - What seems to have a larger impact than distances judging by the coefficient magnitudes (as well as significance levels) are the quality of service dummy variables. Not having medicines available, having water or sanitary problems, or congestion at the closest hospital have large impacts, particularly on weight-for-height. This would be expected if the frequency or intensity of disease in the community is higher as a consequence, and is supportive of the notion, advanced by Birdsall et al. (1983), that health service quality is critical to the demand for health services, although here we see the impact on an outcome variable rather than an on input. In addition to the health service problems, indicators for malaria, dysentery, or measles and chickenpox being major community health problems are also important. Malaria especially has a large impact, being associated with shorter heights by 2.5 percent and lower weight-for-height by 6.5 percent. Three additional community variables are considered. The absence of a traditional healer is associated with better nutrition outcomes. Relying on wells without pumps (used largely in the West Forest) as the main rainy season water source is associated with smaller heights and lower weight-for-heights than is having private taps or wells with pumps, while use of rainfall, rivers or lakes (concentrated mostly in the northern Savannah) does not seem to have much impact, holding other regional factors constant. It is not necessarily the case that wells without pumps will be a better quality source of water than natural sources, and these result suggest that they may not be. Much depends on well depth, well construction (such as the lining), whether livestock are kept away, as well as on the quality of the area underground water supply. - 33 - As is the case for the fixed effects estimates, the interactions between mother's education and the community variables are not significant at standard levels in either equation (F-statistics of 0.5 and 1.2). The coefficient signs remain negative for weight-for-height but are positive for height. - 34 - 6. EXPLAINING INTERREGIONAL DIFFERENCES IN CHILD NUTRITION OUTCOMES There exist clear interregional patterns in child nutrition outcomes as well as in income, education and infrastructure (Table 2). The random and fixed effects estimates indicate that observable household and community characteristics do indeed explain part of this variation. The implications of improving household and community characteristics as a group are explored by interchanging these characteristics between representative households in the East Forest and Savannah, two regions which represent contrasting levels of living and infrastructure. That is, we take a representative household in the Savannah and predict a child's (of average age and sex for that region) height and weight for height given first that the household now has the characteristics of an average East Forest household, and second that the same household faces community characteristics of the East Forest. Predictions of the log of standardized height and weight for height for a representative household in each region are shown in Table 5. In making the predictions, account was taken of the random effects (see Taub, 1979).261 The second row of Table 5 shows the effect of exchanging 26/ To compute the prediction it is necessary to use the number of observations (measured children) in the representative household. For this an average was taken over each region separately. The prediction is then: ^2 ^ _ ^ N~Nra i Yr = xr A + ^2 2 er a +N a ^ ^~~~~~~~~~~~2 where r indicates region, er is the average residual for a region, ae is the estimated variance of the child specific error term and a is the estimated variance of the random household effect. - 35 - TABLE 5: Predictions of Log Standardize, Measurements: East Forest and Savannah! Height Weight for Height East Forest Savannah East Forest Savannah Predictions Using all own Characteristics -.011 -.035 -.030 -.046 Exchanging Household Characteristics -.018 -.028 -.043 -.033 Exchanging Community Characteristics -.028 -.019 -.032 -.044 a/ From random effects specification without interactions. parental and household characteristics, while keeping constant child (age and sex) and community variables. Row three contains predictions using community characteristics of the other region, but using parental, household and child characteristics of their own region. For height, observed community variables explain much of the difference in predicted standardized height between the Savanna and the East Forest. The predictions imply that a child with household and child characteristics of the average East Forest household but living in conditions of the Savanna would be 97.2 percent of U.S. median height, rather than 98.9 percent. Likewise children in the Savanna exposed to the improved infrastructure of the East Forest would have heights higher by 1.6 percent. Observed household characteristics have some effects on the height predictions, but not as large as for community effects. This reflects the small magnitudes of the coefficients for the height equation. It should be - 36 - pointed out that this is the effect of observed household characteristics. As indicated by the fixed effects results, the effect of unobserved household level characteristics is large. This effect does not show up strongly in these predictions because household effects are being averaged, and there is little correlation between household effects and region (as per the Hausman-Wu tests). The weight for height predictions show a different pattern. Now observed household characteristics have a rather large joint impact, while community characteristics as a group have only a small effect. While individual community characteristics, such as the male agricultural wage or the reporting of malaria as a health problem, have strong effects, some work in opposite directions. - 37 - 7. SUMMARY AND CONCLUSIONS C6te d'Ivoire ranks well among Sub-Saharan African countries in its overall nutrition situation as judged by child anthropometric measurements. This is consistent with aggregate food balance sheet data suggesting that domestic calorie availability exceeds estimated requirements by 15 percent (World Bank, Social Indicators). In rural areas 10.5 percent of children under six years of age are too short for their age by WHO standards (stunting) while 4.4 percent are too light given their stature (wasting). These averages, particularly for stunting, do not look large as compared to other countries in Sub-Saharan Africa, though they are by themselves large enough to be taken seriously. Moreover, regional variation is considerable: for instance, the prevalence of wasting ranges from 7 percent in the rural south eastern region to 15 percent in the Savannah. The results of the multivariate analyses show that, in the long run, policies aimed at reducing the prevalence of wasting will benefit from a general improvement of the living standards in rural areas, as well as from an increase in the education levels of the (future) parents. Health policies that result in a reduction of major disease (especially malaria) as well as in improving the quality of the available health infrastructure will also contribute positively to nutritional outcome. In addition, improvement of the sources of drinking water in rural areas is likely to reduce malnutrition among young children. Many of such policies will also have a beneficial impact on chronic malnutrition, as measured by age and sex adjusted height. An additional interesting result is the expected effect of population policy that results - 38 - in a reduction of pregnancies among young women (under 18 years of age). This will have strong beneficial effects on the nutritional status of children. This study thus presents the first evaluation of nutritional conditions in rural C6te d'Ivoire, and their determinants. It is the first step in a larger research program that includes the analyses of the second year C6te d'Ivoire Living Standards data. Since the sample size for anthropometric data for the second year is more than twice as large as for the first year, we anticipate to be able to estimate the effects of the variables mentioned above with more precision. We expect to find that some of the results that turned out to be not statistically significant will become so. Finally, since part of the first year households were reinterviewed during the second year, the subsequent research will include analyses of changes in height and weight over time, and their determinants. The results presented in the current paper seem to indicate that these scheduled studies will further our understanding of the determinants of nutritional status and thus will contribute to the formulation of policies that aim at reducing malnourishment among young children in an environment where overall food supply is not a binding constraint. - 39 - REFERCES Ainsworth, Martha and Juan Munoz (1986). "The Cote d'Ivoire Living Standards Survey: Design and Implementation," Living Standards Measurement Study Working Paper No. 26, World Bank, Washington, D.C. Barrera, Albino (1987). "Maternal Schooling and Child Health," Ph.D. Dissertation. Department of Economics, Yale University, New Haven, Ct. Battad, Josephine (1978). "Determinants of Nutritional Status of Preschoolers," Philippine Economic Journal, 17:154-167. Becker, Gary (1965). "A Theory of the Allocation of Time," Economic Journal, 75: 493-517. Behrman, Jere and Anil Deolalikar (1986). "Health and Nutrition," in Hollis Chenery and T.N. Srinivasan, eds., Handbook of Development Economics, forthcoming, Amsterdam: North Holland Publishing Co. _____ _ (1985). "How Do Food and Product Prices Affect Nutrient Intakes, Health and Labor Force Behavior for Different Family Members in Rural India?", Processed, Department of Economics, University of Pennsylvania. Birdsall, Nancy, F. Orivel, Martha Ainsworth and P. Chuhan (1983). "Three Studies on Cost Recovery in Social Sector Projects". (Country Policy Discussion Paper 83-8) Population and Human Resources Operations, Country Department, LAl1 World Bank, Washington, D.C. Blau, David (1984). "A Model of Child Nutrition, Fertility, and Women's Time Allocation: The Case of Nicaragua," in T. Paul Schultz and Kenneth Wolpin, eds., Research in Population Economics, Vol. 5, Greenwich, Ct.: JAI Press. Chen, Lincoln, Alauddin Chowdhury and Sandra Huffman (1980). "Anthropometric Assessment of Energy-Protein Malnutrition and Subsequent Risk of Mortality Among Preschool Aged Children," American Journal of Clinical Nutrition, 33: 1836-1845. Chernichovsky, Dov and Douglas Coate (1983). "An Economic Analysis of the Diet, Growth and Health of Young Children in the United States," in David Salkever, Ismael Sirageldin and Alan Sorkin, eds., Research in Human Capital and Development, Vol. 3, Greenwich, Ct.: JAI Press. Chernichovsky, Dov, N.S. Kielmann, Arnfried Kielmann, and William Reinke (1983). "Correlates of Preschool Child Growth in Rural Punjab," in Arnfried Kielmann and Associates, eds., Child and Maternal Health Services in Rural India: The Narangwal Experiment, Baltimore: Johns Hopkins University Press. Falkner, Frank and J.M. Tanner (1986). Human Growth: A Comprehensive Treatise, Vol. 3., 2nd edition, New York: Plenum Press. - 40 - Carn, Stanley and Christabel Rohmann (1966). "The Interaction of Nutrition and Genetics in the Timing of Growth and Development". Pediatrics Clinics of North America, 13:353-379. Grootaert, Christiaan (1986). "Measuring and Analyzing Levels of Living in Developing Countries: An Annotated Questionnaire," Living Standards Measurement Study Working Paper No. 24, World Bank, Washington, D.C. Habicht, Jean-Pierre, R. Martorell, C. Yarbrough, R.M. Malina, and R.E. Klein (1974). "Height and Weight Standards for Pre-School Children: How Relevant are Ethnic Differences in Growth Potential?" Lancet, 1:611-615. Hanushek, Erik (1979). "Conceptual and Empirical Issues in the Estimation of Educational Production Functions," Journal of Human Resources, 14:351-388. Heller, Peter and William Drake (1979). "Malnutrition, Child Morbidity and the Family Decision Process," Journal of Development Economics, 6:203-235. Horton, Susan (1986). "Child Nutrition and Family Size in the Philippines," Journal of Development Economics, 23:161-176. Horton, Susan (1984). "Birth Order and Child Nutritional Status: Evidence From the Philippines," Processed, Department of Economics, University of Toronto, Toronto, Canada. Kumar, Shubh (1987). "The Nutrition Situation and Its Food Policy Links," in John Mellor, Christopher Delgado, and Malcolm Blackie, eds., Accelerating Food Production in Sub-Saharan Africa, Baltimore: Johns Hopkins University Press. Martorell, Reynaldo and Jean-Pierre Habicht (1986). "Growth in Early Childhood in Developing Countries," in J. Tanner and F. Falkner, eds., Human Growth: A Comprehensive Treatise, Vol. 3, 2nd edition, New York: Plenum Press. Martorell, Reynaldo and Teresa Ho (1984). "Malnutrition, Morbidity, and Mortality," in Henry Mosley and Lincoln Chen, eds., Child Survival: Strategies for Research, supplement to Vol. 10, Population and Development Review. Martorell, Reynaldo, Joanne Leslie and Peter Moock (1984). "Characteristics and Determinants of Child Nutritional Status in Nepal," American Journal of Clinical Nutrition, 39:74-86. Mueller, William (1986). "The Genetics of Size and Shape in Children and Adults," in F. Falkner and J.M. Tanner eds. Human Growth: A Comprehensive Treatise, Vol. 3, 2nd edition, New York: Plenum Press. - 41 - Pitt, Mark and Mark Rosenzweig (1985). "Health and Nutrient Consumption Across and Within Farm Households," Review of Economics and Statistics, 67:212-223. Rosenzweig, Mark and T. Paul Schultz (1982). "Determinants of Fertility and Child Mortality in Colombia," Health Policy and Education, 2:305-368. Rosenzweig, Mark and Kenneth Wolpin (1986). "Evaluating the Effects of Optimally Distributed Public Programs: Child Health and Family Planning Interventions," American Economic Review, 76:470-482. (1984). "Migration Selectivity and the Effects of Public Programs," Economic Development Center Discussion Paper No. 84-5, University of Minnesota. Ryan, James, P.D. Bidinger, N. Prahlad Rao and P. Pushpamma (1983). "The Determinants of Individual Diets and Nutritional Status in Six Villages of South India," Processed, International Crops Research Institute for the Semi-Arid Tropics, Hyderabad, India. Singh, Inderjit, Lyn Squire and John Strauss, eds. (1986). Agricultural Household Models: Extensions, Applications and Policy, Baltimore: Johns Hopkins University Press. Strauss, John (1985). "The Impact of Improved Nutrition on Labor Productivity and Human Resource Development: An Economic Perspective," in Per Pinstrup- Andersen, ed., "The Political Economy of Food Consumption and Nutrition Improvements", Washington, D.C.: International Food Policy Research Institute, manuscript. Svedberg, Peter (1987). "Undernutrition in Sub-Saharan Africa: A Critical Assessment of the Evidence," World Institute For Development Economics Research, Helsinki, Finland. Taub, Allan (1979). "Prediction in the Context of the Variance-Components Model." Journal of Econometrics, 1:103-107. von Braun, Joachim, Detlev Poetz and Patrick Webb (1987). "Effects of Rice For Production, Consumption and Nutrition in a West African Setting." Report to International Fund For Agricultural Development. Washington D.C.: International Food Policy Research Institute. Wolfe, Barbara and Jere Behrman (1982). "Determinants of Child Mortality, Health, and Nutrition in a Developing Country," Journal of Development Economics, 11:163-193. World Bank (1986). World Development Report 1986, New York: Oxford University Press. (1982). Accelerated Development in Sub-Saharan Africa. Baltimore: Johns Hopkins University Press. - 42 - (1978). Ivory Coast: The Challenge of Success, Baltimore: Johns Hopkins University Press. World Health Organization (1983). Measuring Change in Nutritional Status, Geneva: WHO. Zartman, William and Christopher Delgado, eds., (1984). The Political Economy of Ivory Coast, New York: Praeger. DISTRIBUTORS OF WORLD BANK PUBLICATIONS ARGENTINA FRANCE KENYA SOUTH AFRICA Cados Hirsch. SRL World Bank Publications Africa Book Service (E.A.) Ltd. For singkd titks: Galeria Guemnes 66, avenue d-Idna P.O. Box 45245 Oxford University Press Southern Africa Florida 165. 4th Floor-Ofc. 453/465 75116 Paris Naimobi P.O. Box 1141 1333 Buenos Aires GERMANY, FEDERAL REPUBLC OF KOREA, REPUBLIC OF Cape Town 8000 AUSTRALIA, PAPUA NEW GUINEA, FII, UNO-Verlag Pan Korea Book Corporation For subscriptior orders: SOLOMION ISLANDS, VANUATU, AND PoppelsdorferAle 55 P.O. Box 101. Kwangwhamun International Subscription Service WESTERNSAMOA D-5300 Bonn I Seoul PO. Box 41095 Info-Linc ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~Craighall Info-Line GREECE KUWAIT Johannesburg 2024 Oversea Document Delivery KEME MEMEB Box 506. GPO 24, Ippodarnou Street P.O. Box 5465 SPAIN Sydney. NSW 200c Athens-I 1635 MALAYSI Mundi-Prensa Libros. S.A. AUSTRIA GUATEMALA University of Malaya Cooperative Bookahiop 28001 Madrid Gerold and Co. Librerias Piedra Santa Limited A-l01 I Wien Centro Culturl Piedra Santa P.O. Box I 127. Jalan Pantai Baru SRI LANKA AND THE MALDtVES Graben 31 11 calle 6-50 zona I Kuala Lumpur Lake House Bookshop BAtRAIN Guatemala City MEXICO P.O. Box 244 MEMB R Information Services G MEXICO 100, Sir Chittampalam A. Gardiner Mawatha PMEMBox 2750ain cvie HONG KONG, MACAU INFOTEC Colombo 2 MPna. Town 375 7 Asita 2000 Ltd. Apartado Posta 22-860 Manatna Town 317 6 Fl., 146 Prince Edward Road, WC. Apa Pobre SWEDEN BANGLADESH Kowloon 14060 Talpan. Mexico D.F. For single titles: Micro Industries Development Assistance Society Hong Kong MOROCCO ACE Priest Kungl. Hobokandel (MIDAS) 1IUNGARY Societe d'Erudes Maketing Mamcaine R-e103s27 Stockholm G.P.O. Box SW Kultura 2 Rue Moliere. Bd. d'Anfa Dhaka P.O. Box 139 CasabFnca Forsrbscription orders: BELGIUM 1389 Budapest 62 THE NETHERLANDS Wennergren-Williams AB Publications des Nations Unies INDIA InOr Publikaties Box 30004 Av. du Roi 202 Allied Publishers Private Ltd. Noorderwal 38 5-104 25 Stockholm 1060Brussels 751 Mount Road 7241 BL Lochem SWITZERLAND BRAZIL Madras-600 002 NEW ZEALAND Librairie Payot Publicacoes TeGomc Inte. ciona Ltda. ISiJ.N. Heredia Marg Hills Library and Infonnation Sevice 6 Rue Grenus Rua Peixoto Gomide, 209 Bollard Estate Private Bug Case postal 381 01409 Sao Paulo, SP Bombay-400 038 New Market CH 1211 Geneva 1I CANADA 13114 Asaf Ali Road Auckland TANZANIA Le Diffuseur New Delhi-110002 NIGERIA Oxford University Press C.P. 85.1501 Ampere Street Unvertty Pte Limted PO. Box 5299 Boucherville, Quebec 17 ChinaranJan Avenue niTeCrswns BuildingJeritdo Dares Salaam J4B 5E6 Calcua-70 072 Private Mai Bag 5095 THAILAND COLOMBIA Jayadeva Hostel Building Ibadan Central Depanment Store Enlace Ltda. 5th Main Road Gandhrangar NORWAY 306 Silom Road Carrera 6 No. 51-21 Batgalot 0- 009 Tanum-Kart Johan, A.S. Bangkok Bogota D.E. 3-5-1129 Ktchiguda Cross Road P.O. Box 1177 Sentrum TRINIDAD & TOBAGO, ANTIGUA, Apanado Acme 4430 Hyderabad-500 027 Oslo I BARBUDA, BARBADOS, DOMINICA, Cali, Valle Prartham Flats, 2nd Floor OMAN GRENADA, GUYANA, JAMAICA, COSTA RICA Near Thakome Baug. Navrangptua MEMRB Infornation Services MONTSERRAT, ST. IIITS AND NEVIS, ST. LibreriaTtejos Ahmedabad-380 009 P.O. Box 1613, Seeb Airpot LUCIA, ST. VINCENT & GRENADINES Calle 11-13 Patiala House Muscat Systematcs Studies Unit Av. Fernandez Guelt 16-A Ashok Marg PAKISTAN Curepe San Jose Lucknow-226 001 Mirza Book Agency Trinidad, West Indies COTE D'IVOIRE INDONESIA 65. Shshrah-e-Quaid-e-Azam Entre d'Edition et de Diffusion Africaines Pt. Indira Limited P.O. Box No. 729KE A (CEDA) it. Saes Ratulangi 37 Lahore 3 Haset Kitapevi S. 04 B.P s41 Pusat ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~469, Istiklal Caddesi 04 B.ij 541 Jakarta PutPERU Beyogu-Istnbul Abidjan 04 Plateau RO. Box 181 Editorial Desarrollo SA CYPRUS IRELAND Apaftado 3824 UGANDA MEMRB Infornnation Services TDC Publishers Lima Uganda Bookshop PO. Box 2098 12 North Frederick Stee tE PHILIPINES KamPalBx Nicosia Dublin I National Book Store DENMARK ISRAEL 701 Royal Avenue UNITED ARAB EMIRATES SamfundsLitteratur The lenirsaem Post Metro Manila MEMBRGulf Co. Rosenoerns Alle II The Jetusat Post Building OANP.O. B 6097 DK- 1970 Frederiksberg C. P.O. Box St ORPAN Saji DOMIINICAN REPUBLIC Romema Jerusalem 91000 Palac Kultury i Nauki UNITED KINGDOM Editora Tatler, C. por A. ITALY 0-9o01 WARSZAWA Microinfo Ltd. Restauracion ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~PO. Box 3 Restaupd cion Licosa Comesissionaria Sansoni SPA PORTUGAL Alton, Hampshire GU 34 2PG Apdo. posal2190 Via Lamarmora45 Liveria Portugal England Santo Domiingo Casella Postale 552 Rua Do Casno 70-74 EGYPT, ARAB REPUBLIC OF 50121 Florence 1200 Lisbon VENEZUELA Al Ahram JAPAN SLbrerna del Este Al Golan Street ~~~~~~~~JrN ~ ec SAUDI ARABIA, QATAR Aptd. 60.337 Al Gaiaa Sur Eastem BoDk Service Jarir Book Store Caracas 1060-A Cairo 37-3, Hongo 3-Chome. Bunkyo-ku 113 P.O. Box 3196 FINLAND Tokyo Riyadh 11471 YUGOSLAVIA Akateeminen Kit~~~~~~~~~~~~~~~akaupps JOEDAN ~~~~~~~~~~~~~Jugosloveaska Krkjiga Akateemoinen 1 i2 akauppa JORDAN f M SINGAPORE, TAIWAN, BURMA, BRUNEI YU-1 1000 Belgrade Ttg Republike SF000tP. Box 3143dn etrfr akfngRsac Informiation Publicationis eS-0Ink P.O. Box 3143 Private. Ltd. ZIMBABWE Helsinki 10 Jabal 02-06 Ist Fl., Pei-Fu Industrl Textbook Sales Pvt. Ltd. Amttn Bldg.. 24 New Industial Road Box 3799 Singapore Harare LSMS Working Papers (continued) 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 Cote d'Ivoire No. 36 Labor Market Activity in Cofe 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 Un The World Bank rr| 'n Headquarters European Office Tokyo Office 'AiI 1818 H Street, N.W. 66, avenue d'Iena Kokusai Building Washington, D.C. 20433, U.S.A. 75116 Paris, France 1-1 Marunouchi 3-chome Telephone: (202) 477-1234 Telephone: (1) 47.23.54.21 Chiyoda-ku, Tokyo 100, Japan Telex: WUI 64145 WORLDBANK Telex: 842-620628 Telephone: (03) 214-5001 RCA 248423 WORLDBK Telex: 781-26838 Cable Address: INTBAFRAD WASHINGTONDC