Determinants of child mortality, health, and nutrition in a developing country

Journal of Development

Economics

11 (1982) 163-193. North-Holland

Publishing

DETERMINANTS OF CHILD MORTALITY, HEALTH, NUTRITION IN A DEVELOPING COUNTRY*

Company

AND

Barbara L. WOLFE University OJ Wisconsin, Madison, WI 53706, USA

Jere R. BEHRMAN University of Pennsylvania, Philadelphia, PA 19104. USA London School of Economics and Political Science, London W.C2A 2AE. UK

Received May 1981, ha!

version received December 1951

Child health and nutrition status are important determinants of adult productivities and earnings; and, thus, are an important channel of intergenerational socioeconomic mobility. Building upon economic models of household behavior, we investigate the determinants of child mortality and health and nutrition status in regions defined by the degree of urbanization in the developing country of Nicaragua, and offer some policy suggestions. Our estunates suggest that: determinants differ signilicantly across regions, income is not an important factor, there is an inverse relation with number of sib!.ngs, and there are positive associations with calorie intake. schooling (except in the relatively low-ina me rural areas), the availability of refrigeration, and the quality of sewage systems.

1. Iiatruduction

What determines child mortality, health and nutrition status in developing countrier is a critical question, for a whole host of reasons. First, the levels of *This paper is one of a series resulting from a survey and research project to investigate

the

social, economic and demographic roles of women in the developing country of Nicaragua. Initial funding for 1977-1978 was provided by the Population and Development Policy Research Program sponsored jointly by the Ford and Rockefeller Foundations. Basic funding has been provided by the Agency for the International Development Negotiated Contract AID/otr-C1571 for 1977-1981 and thr: National Science Foundation Grant SES-8025356 for 1930-1982. Behrman was a Guggenheim Foundation Fellow for the 1979-1980 academic year and a Compton Foundation Fellow for part of the 1980-1981 academic year. The project is being jointly conducted by the Universities of Pennsylvania and Wisconsin, the Centro de Investigaclones So&es Nicaraguenese (CISNIC), and the Banco Central de Nicaragua. Humberto Belli, Direction ajf CISNIC, and Antonio Ybarra, former head of the Division of Social Studies and Tnfrastructure, Banco Central de Nicaragua, were co-principal investigators with Behrman and Wolfe for the early stages of the project. Belli supervised the collection of all survey data. The authors would like tu thank, but not implicate, the funding agencies, a referee for this journal, their principal co-investigators, and associates in the project, especially David Blau, Kathleen Cairns and Nancy Williamson. Wolfe and Behrman equally share responsibility for this paper.

0304-3878/82/CKWO4OW/$02.7.‘i 0

1982 North-Holland

indexes of current &~ltJ mortality, health. and nutrition are important socictcconomic welfare, and these levels are often low in developing countries [World Bank (1980}]. Second, the health and nutrition status of children significantly conditions the subsequent intelligence, health and nutrition status of adults; ;,I1 of these have a direct impact on adult productivities, earnings and the quality c,f life.’ Third, child mortality, health, and nutrition status may be relateal to the level of such investments in children as impact on adult and thereby have an indirect long-run education, productivit,les ar;d earningsd2 Fourth, these may be important intervening variables through which mtcrgcnerational socioeconomic mobility is limited. Parental education or income. for example, can affect a child’s health and nutrition directly ot indirectly (say. through the number of siblings), and this child’s adult options nr,dy bc directly k2r indirectly affected. P,:bverty may then be transmitted through such a channel. Fifth, in the ilaited States. family income is relatively unimportant among those farll!lj characteristics that affect child health status, but parental education is impo:iant [Chernichovsky and Coate I 19793. Edwards and Grossman (1978)]; do similar results hold for developing ,ountrics in which incomes are much lower’! Or are generalized income incrc:scs very powerful in improving health and nutrition, as the World flank (1980, p. 591, Ward and Sanders (1980), and others claim? The answer may substantially assist policymakers in choosing policy interventions with :hc highest payoffs. Sixth. child mortality, which is higher in Latin Amel: :a than in the more developed countries, because health interventions are crirrentlq’ less effective in reducing it thaIi in reducing adult mortality i “derrick 1197Li)].has a feedback on fertility and overall child quality.3 Better Lnt~l~dgc of the determinants of child mortality may shed some light on thus process. Seventh, there is evidence for the United States that pre- and

in Wolfe and Behrman ti982a;. we consider the detemlinants ol’ adult health and nutrition \PIIIUSm the same develooing country for which the present study is undertaken. Unforiunalely, halueker. our data set da .es not permit the exploration of a direct link between the child and .uduh health of an indivlduai. Set Anthony (19791, Rlau (1977 2nd 19801, Selowsky (1976), and SzEn%sky and Tallor (1373), and .he references therein, regarding evidence of the link between ,hdd and adult he3Rth and nutriL:m status. See Behrman and Woll’e (1979, 1980a, b, 1981a, b. d, I hr. lkhtrnm~ W&e and Blau (1980), Bet.rman Wolfe and Tunali (1980), Blau (1977 and 1980). .rnd \f’oife. Behrman and BIail (1982), regarding the impact of adult health and nutrition status .*r! ~&it productl\itics. earnmgs, labor force vamcipation and fertility. -I-dward, and Grrssman (13Xj tind a significant association betw. m health and intellectual Jsrciopmt-nt for children !;I the United State:. We consider the qvestment in children’s ~xfer~at~m m a devel,:otng L, Yntry elsewhere fBehrman and Wolfe (198Ar, 1982)J In Behrman dnd Wolfe (l%Ob, l%Ra, S, h). Behrman, Wolfe and Bfau (1980), Behrman, Wolfe and Tunali li9UOj and Wolfe and Behrman (1982b) we investigate the impact of schooling on adult s:;..&;tr-=ities and earn:qg~ ‘Lndsr tht- assumptron th-lt krtiiity is partiali> mder control of the parents, for example, f:::i !LCK~;“~‘%rnh-.*nc4tl he kjuered. See Rehrman a ad Wolfe (1079). a~!d Schultz (1976).

B.L. Wnl/e and J.R. Behrman,

Child mortnlity, heuith and mttritiotl 111CI 1)~

I h5

peri-natal care has an impact on child health, and through c!lild health, on education, adult health, and adult productivity and earnings [Davis rind Schoen (1978)]. In order to evaluate the returns to improved pre- and perinatal care in Latin America, some quantification of this link is useful. Eighth, many studies in the human capital tradition for developed countries, and some on developing countries [e.g., Blau (1977)] suggest that there is a quality-quantity trade-off for children. If this trade-off exists, improved child health may be associated with higher quality, and hence lowered fertility and lessened population pressure. For all of these reasons, knowledge of the determinants of child health, mortality and nutrition status in developing countries is very important. But the current state of such knowledge is quite poor. In this paper, therefore, we present the results of our efforts to investigate this topic to the extent possible, using data we collected in a cross-sectional multipurpose survey of women of childbearing age in Nicaragua.4 Tn section 2 we present our (I priori rationale for the specification of multivariate relatior 3 to determine child health and nutrition status. In section 3 we introduce our data set and define the relevant variables. We present and discuss our multivariate estimates in section 4, and conclude in section 5.

2. Model specification We build upon economic models of household behavior developed in the past two decades, placing a ‘Chicago-Columbia schcol’ emphasis on the role of human capital, household production and the allocation of time, and a ‘Pennsylvania school’ emphasis on intergenerational considerations, hiological factors, imperfect knowledge and the endogeneity of tastes.5

‘This survey was collected as part of an extensive study on the rule of women itI &:veloptng countries. Among the resulting studies are all of the items in the reference lis: whiclh are authored by Blau and by both of the present authors (in some cases. with additional collaborators). 5Warren Sanderson uses these labels in his recent review of the Easterlin. Pollak, and Wachtcr which provides a number of earlier r+rences to (1980) paper on fertility determinants. Pennsylvania studies. The standard references to the seminal Chicago-Columbia school studies can be found in Becker and Tomes (1976). Of course the distinctions are not perfectly sharp. The Pennsylvania school incorporates the iChicago-Columbia insights regarding human capital. time allocation and household production, at least on a general level. Michael and Willis (1976). who generally would be classified in the Chicago-Columbia school, incorporate biological factors within a demographic ‘renewal model’ of contraceptive use. Becker and Tomes (1976) of the Chicago-Columbia school, discuss intrzfamilial allocations and intergenerational decisions using a model that is similar in many respects to one developed by Behrman, Pollak and Taubman (1982) of the Pennsylvania school. Nevertheless, the distinction is useful because of certain ongoing differences concerning the endogeneity of preferences, the .importance of biological factors. the usefulness of ‘full’ or ‘social’ income measures and shadow prices. etc.

:fh

B.L.

Wolfe and J.R.

Bchrmaw.

Child

worfalit?;, hcwlth und nutrition

in (I DC

WC posit a household (parental) utility function (U)” which depends upOn :ion (Z), number of children (C), exrscted average full commodity cons1 ineom,e of the chjldreti when they become adults (E), average child mortality (iv), health (;Y) and nutrition status (IV), all of which are conditional on ~ORPI.S concerning commodity consumption (Z*), and number of children (C*):

U = I:(Z, C, E, M, H, N, Z”, C*).

(1)

We need not disc-lss the standard inclusion of commodity consumption and number of child;b:n in this function. We follow the widespread ‘practice of including the aver-age characteristics of children, with the exception that. we also consider sex..ial differences, in our empirical work below.’ We include expcted average earnings of children, because we posit that parents are concerned about their children’s adult prospects as well as their current welfare.” The former may be either a pure concern about intergenerational family welfare, or a concern about potential transfers from children to parents in the parents’ old age, which are much more important in developi.ng than in developed countries due to the relative inadequacies of capital markee T and pension systems. We could reduce the number of argumenis in the utility function with no impact on our empirical analysis by assuming that parents are concerned with average child characteristics only insofar as they .affect expected average child earnings as adults, as some other studies do [Becker s.nd Tomes (I976), Behrman, Pollak and Taubman I I %82)], but we think it useful to emphasize that current average child welfare may be a relevant fac:or even if we are not able to identify whether or not in fact it enters separately from expected average earnings in the utility function.Y Even if we were to assume that average child health and nutrition status entered into household utility only through its impact on average expected adult earnings, we VNould think it useful to include child mortality separately as a reminder that there must be some disutility costs to such mortality, or infanticide would be much more widely practiced than it apparently IS. We include variors norms to emphasize tha: preferences are conditional ‘We follow the well-established

practice of ignoring the dificult question of how such a preferences of family members, In one dimension this procedure is tnore satisfactory for our empirical work than for many similar applications. Since wc focus on tbe slatus of children! under 5 years of age we can ignore more safely than can many others questlons of how children’s preferences enter into the household utility function. -For consideration of allocation witbin families, see Becker and Tonnes (1976), Behrman. PollaL 2nd Taubman (19821, and Behrman and Wolfe (1982). ‘We @ore bequestsand m-kind transfers. &:t the first two references cited in footnote 7 for a dlicussion of such aiternatr~&s. Also see Behrman and Wolfe (1981g) for an argument that marriagemarketconsiderationsshould also be incorporated into such analysis. we a& could inaqorate average child quality instead of expected average child earnings and measures J cwre.t average child welfare, but we find the more specific representation of &lfd characteristics more satisfactory. hmclion is defined given differer,tial

upon them. Such norms may vary across communities or ova- tinlc. WitIlin the context of our particular empirical exploration, the specific implications of these norms are twofold: First, certain activities, such as schooling, may change these norms. Therefore, realistically, it may be difficult to identify whether such activities change efficiency in household production or change tastes.” Second, it is widely hypothesized that reference norms differ among communities in developing countries depending on thei degree of urbanization and modernization. Therefore it may be important to subdivide a national sample into subsamples by such criteria. The household faces a number of constraints:

(1) The traditional budget constraint in which total monetary expenditures on goods and services (X) cannot exceed total earnings plus other inflows. (2) TIT,: traditional time-budget constraints in which total time spent in pa’d labor force activities plus household production (including brcastfeeding), ztc., cannot exceed total time available for each individual. Hclwever, in developing countries the distinction between market and no+market activities is often fuzzier than in more developed countries. For example, a number of household activities, particularly child care, can be combined with market participation in the informal sector [Behrman and Wolfe (1980b. 1981a, h), Behrman, Wolfe and Tunali (1980)]. market prices (W), including those for goods and (3) The traditional services, child investments., labor dependent on skills, etc. production function in which commodities (Z) ;1rc (4) A household produced by inputs of market-purchased goods and services cX) and time in efficiency Lnits (If). The latter may depend upon skills and upon the extent to which market activities and household production are carried on simultaneously, as is fairly frequent in developing countries. Given sexual division of labor with women specializing in home production, the work conditions and skills of women may be more important than those of men [e.g., Birdsall (1979), Leibowitz (I 977)J. (5) .4r: average child’s expected earnings function (E) which depends on various human capital investments, including those in average child health (H) and nutrition (N), and on average genetic endowments (Gj. (6) A biological childklirth function that depends on parental commodity consumption (Z) and public environment considerations rzlating to sewage and water supplies and population densities (I’) (as well ;is on genetics and earlier human capita.1 investments in the parents). The function emphasizes parental health, fecundity status and parental “For indirect evidence that schooling may represent taste effects and not only efkiency effects in other studies with this data set, see Behrman and Wdk (19Rla-c. e) an4 Wolfe and Behmxa (1982b. c).

of caition, contraceptive choice, and chokces regarding practices such as breastfeeding (B). (7) Functions for biological child mortality (M), health (H) and nutrition status (N) that depend upon average child genetic endowments (G), commodity consumption (Z), the public environment (Pf and such as breastfeeding (B). In the pro&ction of parental practices commodities that enter into these functions, parentai educalion may be particularly important under the plausible assumption that more educated parents tend to be more knowledgeable than these who are less educated about preventive and curative health anil nutrition measures, more capable of following medical and nutritional advice or instructions, less fatalistic about illness and therefore more prone to seek medical help for a sick child, and more child-oriented due to the larger role mothcr:j play in intrafamilial decisions [e.g., see Behm (1979), C’aldweli (1979) and Cochrane (1979)]. 181 f-.xpectcd average child earnings functions (E) that depend on investments in children (including their health and nutrition status) and their average genetic endowments. I% Parental skill functions which depend on their genetic endowments and various investments in them, including schooling (S). (IW Other adding-up constraints, such as the number of surviving children, which equals natality minus child mortality. practices

such as the frequency

If parents wrceive all of these constraints, they can maximize household utilily with respect to all of them and determine optimal levels of child morAity (.44). health (If) and nutrition status (N) with other simultaneous tariables (e.g.. 2, C, E). If parents only partially perceive the constraints due to what Easterlin, Pollak and Wachter (1980) term ‘unperceived jointness’, they can maximize utility with respect to the constraints that they perceive, but in the process unknowingly determine outcomes for the other variables in the system. In either case, child mortality, health and nutrition status depend on al1 of the exogenous variables and parameters and forms of the constraints. Howeve:, if there is ‘unperceiked jointness’ for these variables as ..sould seem to be more likely for the more traditional areas of developing countries, the incomplete pature of the maximization would lead to different results than with complete maximization. This is another reason that for empirical work it is preferable to subdivide samples into more and less traditional groups. One could specify exact functional forms for the utility function and all the canstraints and derive explicit solutions for the optimal levels of child mortality, health and nutrition status with either complete knowledge or any particular degree: l->funperceived jointness.’ 1 We do not here follow such a ’ ! Under

the

aswmphw that the conditions

exist

k i a maximization.

procedure for three reasons: (1) The choices of functional forms are somewhat arbitrary. (2) The analytical expressions which would result would be sufficiently complex that the signs of most partial derivatives would be indeterminate without very specific information not only about the general functional forms but also about the magnitudes of all of the parameters. (3) We do not have the data nor other required resources 1.0 estimat.e the complete system. Instead, we proceed from this overall general framework to the estimation of simple reduced-form solutions fof the child mortality, health and nutritional status variables of primary interest for this study. Our simple (generally linear) relations can be interpreted as local approximations to the more complex expressions which would result from maximization of the complete system (or some subset thereof) which explicit functional forms.” In our specification, estimation and interpretation of these relatio‘ls. we benefit from the above outline of the overall theoretical system in a number of respects: (1) The theoretical discussion suggests what variables should be included, if possible, in the estimated relations. (2) It also suggests that such relations may differ between more traditional versus more modern communities because of different norms, market prices and public environments. and degrees of unperceived jointness. Therefore subdivision of our mttional sample is desired. (3) It further implies that care must be taken in interpreting certain effects since some variables may ahe;’ the outcomes in a number of ways. For example, more schooling for the parents may alter through changing tastes or through changing efficiency in outcomes household production. However, parental schooling also may be serving partly as a proxy for genetic eK !owments. (4) Finally, this discussion suggests that in reduced-form estimates of the type t5at WC are undertaking, disaggregation to particular fairly specific variables may be important because norms, market prices, and public environments may differ across the sample. For instance, because of such differences, it may not suffice to include a proxy for aggregate commodity consumption alone. In addition it may be desirable to include specific representations of nutrient intakes, use of medical services, etc. 3. Data We conducted a stratified random survey of socioeconomic characteristics of women of childbearing ages (15-45) in the central metropolis, other urban. ‘lAfter we ‘finished this study we became aware of the very interesting hasenzweig and Schultz (1980) study which assumes specific functions for the parental utility and child heillth functions (i.e., Cobb-Douglas), eliminates by assumption most of the other complications noted in this section, assumes complete knowledge, anti derives explicit demand functions for child health inputs in terms of the original structural parameters and the exogenous variables. In their empirical applications to U.S. data, however, they conclude that the necessary assumption:c are too restrictive to be realistic, so they emphasize the results from more general bit approxirnatc demand relations which do not permit identification of all of the structural parameters.

170

B.L.

Worfe and 2.R. Behrman,

Child mortals’iy, health anti nutrition

in u DC

and rural areas of Nicaragua in 1977-1978. For 1,871 cases we have child

mortality data. For 1,281 women we have health and nutrition status information on a randomly-selected child under 5 years of age. We now discussour definitions for the variables which enter into the determination of average child mortality, health and nutrition status oa the basis of the discussion in the previous section. We first consider the dependent variables and then the right-hand side variables. In all cases we dist.inguish among the three regions noted above (i.e., central metropolis, o.ther urban, and rural) due to probable differential (but unobservable) norms, prices, public endowments, and degrees of unperceived jointness. Table .I gives the means and standard distributions for the relevant. variables for the three regions and far the overa~i sample. 3. I. Dependent tuaiahles Child mortality.

Our reported child mortality rates are 1.7% for the central metropolis, 6.3”/, for the other urban areas, and 8.50/dfor rural areas.l” The inverse association with urbanization is not surprising given the positive association between urbanization and incl me, schooling, health care facilities, good water and sanitation. But the levels Le-:’ low relative to other estimates, Probably they are underestimates becau.. r!:y are based on recall data. We expect in particular that deaths of very . - _‘I infants may be underreported, so we control for child’s age in our es:.~ 2s below. Perhaps there is an inverse association between reporting su~il deaths and the respondent’s education. If so, this would cause a bias towards zero in an estimate of the impact of wornenS education on the probability of child mortality. Also there may be cultural differences in regard to underreporting which are associated wi:h the degree of modernization or urbanizatioc. If so, this is another reason for subdividing our sample by the degree of urbanization. Standard weight.

To

construct this variable we took the weight of the child (which we measured), subtracted the mean weight for a child of identical age and sex according to widely used international standards based on health and well-nourished United States children L-National Center for Health Statistics (1976)], and di,&ed the difference by the sex and age-specific standard deviation from the same international standardsI This measure “We always give statistics for our three Tegions in d-creasing order of urbani&ion, unless otherwise noted. To keep the presentation concise, however, we do not always repeat the n:gional !identifications. ‘%stead of standardizng for age-sex group, it would be possible as an alternative to include sonic function of age and XX as an additional right-hand side variable in the relationship. We do not estimate such an alternative because we believe that age and sex affect weight and height more generally tka i n additive function could capture. The considerations are analogous on the question of akrnative controls for agr: and len;rh of exposure for fern&y variables to those th;tt we discuss in Behrman and Wolfe (198&t).

B.L.

WoJ/r and J.R.

Means and standard

Behrman,

Child mortalit_v, health wd

nwritiorz

it1 II DC

Table 1 deviations of regional distributions of variables used ii1 analysis mortality, hea!th and nutrition status.=

171

of child --

Central metropolis

Other urban --

Rural ----_

Variables

Mean

S.D.

Mean

S.D.

Mean

S.D.

Commodity consumption and income (100’s of cordobas per fortnight)b Other income Woman’s predicted income

7.1 2.2

7.0 1.8

6.8 2.0

8.5 1.7

3.5 0.9

9.2 0.6

5.3 27.8

3.2 6.5

5.2 28.2

3.8 6.6

1.3 29.0

1.9 6.8

Other factors in household production Woman’s schooling (years Age (years) . Participation in informal sectore Male companion present’ Schooling (years) Household size Births/year in live years

0.38 0.88 6.1 7.3 0.39

0.49 0.33 3.9 3.6 0.28

0.79 0.85 5.7 7.2 0.43

0.41 0.36 4.6 3.2 0.23

0.89 0.86 I.2 7.2 0.50

0.32 0.35 2.0 3.1 0.32

Nutrition-specific commodities Average caloric intake .Refrigeration’ Length of breastfeeding (months)

0.62 0.28 5.3

0.15 0.45 5.7

0.75 0.28 5.0

0.17 0.45 6.5

0.49 0.02 10.1

0.16 0. I 5

Public en’rironment Sewers’ Population density (people/km”) Parasitesc

0.75 72 0.41

0.43 41 0.49

0.30 172 0.57

0.46 32’1 0.50

0.03 59 0.54

0.18 7! 0.50

Other variables and controls Male’ Never migrated” Child’s age (months)

0.48 0.50 26

0.50 0.50 17

0.48 0.55 35

0.50 0.50 LB

0.46 0.34 40

0.50 0.4s 15

Dependent child mortality, health, and nutrition variables 0.13 0.063 0.24 0.085 Mortality’ 0.017 -0.21 -0.23 1.31 -0.88 1.50 Standardized weight 1.61 1.50 -0.68 -0.90 - 1.87 Standardized height 0.09 0.91 0.96 0.95 0.08 Standardized biceps circumference -._______--“S.D. = standard deviation. bin 1977-1978, 7 cordobas equaled one United States dollar. ‘Dichotomous variables with value of one indicated state: otherwise 2 value of zero.

7.4

0.27 1.16 1.55 0.07

(and the other two below) is not subject to the recall error that may contaminate the child mortality rates, although there probably is measurement error (which causes no bias if it is random). The ‘use of the international norms gives a reference point based on a population in which

172

B.L. Wdf~ ad J. R. Bchrmmt, Chilci mortnlitv,

hdth

~utrl ttuirition

irt N DC

malnutrition is not present. We use these norms because they control partially I5 for genetic ag e- sex patt:ms of child growth and because they facilitate comparison with other studies. The standard interpretation of the standardized weight measure is that it refers primarily to relatively current or short-run health and nutrition status, and not only to differences in genetic pools across popwlations [Blau (1980), Edwards and Grossman (1977 and 1978), Habicht et al. (1974), Heller and Drake (1979), Zerfas et al. (1975)]. For our three regions the mean values of the standardized weight measures are -0.21, -0.23, and -0.88 (with standard deviations of 1.50, 1.31 and 1.16). Children in all three regions on the average tend to be below the U.S. standard, and more so in the rural than in the urban arcas (although the dispersion is positively associated with the degree of urbanization). Thus short-run malnourishment and poor health are widespread in general, but more extensive in the rural areas. However there may be an interaction with age occurring, which underlies part of the higher average malnourishment in the (on the average, older) rural sample. Srattdtirdizrd height.

This variable is defined in a fashion parallel to that for the stanaardized weight, and similar comments apply. The standard interpretation of the standardized height measure is that it refers to relatively long-run or permanent health and nutrition status [Edwards and Grossman t 1977 and 1978). Heller and Drake (1979)]. It assumes that potential height is the same across ethnic groups [Habicht et al. (1974)]. For our three regions the mean values of the standardized height . measures are -0.68, -0.90 and - 1.87 (with standard deviations of 1.50, 1.61 and 1.55). Thus long-run malnourishmen: :‘k,prevalent in all three regions and is Inversely associated with ur’banization (although again there may be an interaction with age), but in this case without any increase in dispersion in the more urban areas. The standardized height measure of long-run nutritional and health status is significantly correlated with the short-run standardized weight measures (0.7, 0.6, 0.7). However, these correlations indicate that there may be important differences between short- and long-run health and nutrition status since the variance in one measure is consistent with half or less of the variance in the other. biceps circumference. This variable is standardized relat’ive to the mean international value for a child of the identical age [Seoane and Latham (19?1)]. We use such a standardization to control for age since we

Standardized

“We qualify our statelnent about genetz controls with the adrerb ‘partially’ because we do not believe that the distributions of weighrs and heights for the base sample for these norms (i.e., chi?dren in Yellow ‘Springs,Ohio) are independent of environmental factors. They probably are, hInever, free of sultnourichmeot. See Berg (1973) and Heller and Drake (1979) and Zerfas et al. I 1’1751and references therein for further di:;cussion of such indexes.

rj.L. Wolfe and J. R. Behrmar?. Child mortality, hrulth and nutrition irr N DC

172

do not have age-specific standard deviations of IhI. international norms. The standard interpretation of the biceps measure is that it pertains to muscular development associated with relatively long-run or permanent health and nutrition status. For our three regions the mean values of the standardized biceps circumference are 0.96, 0.95, and 0.91 (with standard deviations of 0.09, 0.05, and 0.07). In this case, as with the standardized weight measure, the regional averages are positively associated with urbanization (once again, with a possible age interaction), but the standard deviaiions are inversely related. The correlation:; of the standardized biceps measure with the standardized height measure (0.7, 0.7, 0.6) are higher than are those with the standardized weight measure (0.5, 0.4, 0.4). This is not surprising since both the standardized height and biceps measures supposedly represent long-run health and nutrition status, while the star.dardized weight measure supposedly represents short-run health and nutrit(on status. But note that the standardized biceps measure is even less correlated with the standardized weight measure than is the standardized height measure. Moreover, the correlations between the two long-run measures imply that one is consistent with less than half the variation in the other. Therefore, either measurement error is important or the different measures are referring to different dimensions of long-run nutritional status. We expect that measuremeot error is a factor, probably more so for the bicep measure than for the height measure because of the difficulty in assuring uniformity regarding exactly where on the arm the circumference measurement is made. If any such measurement error in our variables is random. once again. it does not bias our coefficient estimates. 3.2. Right-hand side variables Because of limitations of data we have to approximate some of the variables that the discussion of the previous section indicates are relavant. We do not have even proxies in other cases. We now discuss the variables which we use. Commodity consumption (2) and income. In the model of section 2, both commodity consumption and income are endogenous variables. In our data set we have observations on income, :Jut not on overall commodity consumption. We treat income other than women’s earnings (‘other income’) as an exogenous variable representing the generalized purchasing power that largely accounts for the market goods and services (X) that enter into the household production function. The largest component of this other income is earnings from male companions [Behrman, Wolfe and Blau (1980). Wolfe.

171

EL.

Wolfe crtd S. R Behrman,

Child nlortalilv,

heulrh and nutri[ion

in a DC’

Behrman and Blau (1982)]. The assumption of exogeneity in this case is not too troublesome since prime-age male participation rates are very high, hours worked are relatively fixed for males in comparison to females, and wages are determined by past human capital investments. The other two components of other income are income from assets and transfers. An important part of the latter in many cases is child support. The assumption of exogeneity also does not seem troublesome for these two categories, although some transfers could be in response to perceived needs in the form of child illness or malnutrition. The distributions across regions for other income have means that are about twice as high for the two urban areas as for the rurai areas, while they have standard deviations that are inversely associated with the degree of urbanization. If prices are constant across regions, the rural sample has Power mean income and greater intraregional income inequality than do the urban ones.16 We do not use actual women’s earnings in our income measure because women’s labor force participation is far from universal and many women participate primarily or exclusively in non-paid household production. lr!stead we use the WOIIMFZ’S predicted earnings, as estimated from earnings functions with a control for selectivity in labor force participation [Behrman and Wolfe (198Ob, 1981a, h), Behrman, Wolfe and Tunali (1980)]. This gives us a much better measure than would actual earnings of the relative contribution of women in our sample to household commodity production, whether it be entirely in direct household production or partly through goods and services purchased by earnings from women’s paid labor market activities. This estimate also lessens the possibility of confounding earnings viith breastfeeding in the multivariate context; actual earnings may be inversely associated with breastfeeding, since the latter normally is not practiced while working at high-earning jobs.” We include predicted women’s earnings separately from other income because of the difference in definition. Across regions, predicted women’s earnings tend to te much higher in the urban than in the rural areas, both because of the regional distribution of wornen’s !luman capital and because of geographically segmented labor markets [Bchrman and Wnlie (:980b)]. In contrast to the distributions for other incon&e,however, the dispelsion in the distributions of **o.men’s?redict;;d earnings is much lower in the rural than in the urban areas.

“‘in Bebrman.Wolfeand Bku (1980) and Wolfe,Bebrmanand Blau (16821, we examine the rofe of demographic and human capital variables ia the determination of the rlpgional and
B.L. Wo(fk and J. R. Behrman. Child mortality.

healtk and rwritio,:

irz a DC

175

Other factors in household production. We include other factors related to efficiency in household production and the time available on a per child base. We suggest above that more educated women may be more efficient in household production. Therefore we include the woma~‘s schooling as a separate variable to see if there are any added efficiency effects beyond those captured in the predicted earnings variable. We remind the reader, however, that this variable may be representing in part differential norms or genetic endowments. Across the regions, the big disparity in the distributions of women’s schooling is between the two urban areas on one hand and the rural on the other, with means of 5.3, 5.2 and I.3 years. Women may become more efficient in household production with general experience and maturity. Therefore we also include the \t*otnun’s age as a factor. The distributions of age do not vary much across regions, although there is a slight inverse association between urbanization and mean ages (27.8, 28.2 and 29.0 years). There is a fair amount of variance, however, within each region. A third factor which is important regarding the women’s contribution to household production is the sector in which she works if shz participates in the paid labor force. As we note above, women working in the inforrntrl sector generally can combine their paid-labor participation with at least the child-care component of household production.18 Across regions, the proportions of working women in the informai sector are inversely associated with urbanization (0.38, 0.79, 0.89). If a male companion is present, he too may affect household norm;, and child genetic stocks and may aid in household production, although probably much less than do the women respondents. Across regions there is not much difference in the proportion of households with a companion present (0.88, 0.85, 0.86). But there are substantial differences for mean years of companion’s schooling between the urban and rural areas (6.1, 5.7 and 1.2). In regard to the impact on average child health, finally, the number of persons over which household commodities have to be spread would seem to be relevant.” Mean household size does not vary significantly with urbanzation across regions (7.3, 7.2, 7.2). However, more important than the total number of household members is the number of small children. because of the time-intensive demands of pregnancy and infancy. Across, regions the ‘sIn Behrman and Wolfe (19801). 198la, h) and in Behrrnan. Wolfe and Tunali (1980) we present statistical evidence that the presence of small children _md the absence of home child care selects working women away from formal and domestic employment and in to the mformal sector. In this analysis we also differentiated between the formal sector anr domestic employmxt, but it was never statistically significant, even at the 207; level. 19At least as long as such commodities are not entirely public goods within the family. which would seem quite unlikely. However our other work on household nutrition demand suggests considera.Se economies of scale with respect to household size [Behrman and Wutfe (198 tel. Wolfe and Behrman (1982c)J.

176

B.L. Wo@eand J. R. Behrman. Child morrahv, health amt nutrithl

rn a IX

average number of liar births per year in the past quinquennium (0.39, 0.43, 0.50), as well *as the variance. is itiversely associated with urbanization. Of course both the LGZs&G&inumber and the live births per year may be determined simultaneously with the dependent variables of interest (particularly with mortality if there are replacement births). Therefore, we have estimated our relationships both with and without these variables. Excluding them does not Seem to alter significantly the coefficient estimates of the other right-hand side variables, so for the sake of economy we present below only the relationships in which these variables are included. But tb: coeflicient estimates of these variables sC1 need to be interpreted with care because of possible sirilultaneity bflses. Sutrition-sgecijic form of commodiiy consumption. As we argue in section 2 above. the use of reduced forms, with no observations on some important variables (particularly related to prices and norms), shows that different households with the same income make different consumption choices even under the maintained hypothesis of identical household utility functions across families. For this reason we include three specific nutrition-related variables. (For a smaller subsample in table 7 we also include a specific represeniation of prenatal medical care.) Our ;irst specific nutrition-related form of consumption is the average standardized cufutic intake of the family. We constructed this variable from summing the caloric content of the food which the household consumed in the previous week and normalizing by internationai standards.“’ Among our thra regions those in the smaller urban areas tend to be best off by this measure and those in the I-Ural areas worst, with residents of the central mcrropolis in between (0.62, 0.75, 0.49). Our second specific nutrition-related form of consumption is really a form of capital equipment in household production - reftigeration. Earlier work suggests that the presence of a refrigerator improves nutrient intakes and corresponds to higher given income, education and a number of other characteristi-3 >Behrrzan and Wolfe (1981e), Wolfe and Behrman (1982c)]. The proportion of househgfds with refrigeration once again renects a dichotomy between the two !,rban and the rural areas (0.28, 0.28, 0.02). Our third specific nutrition-related form of consu*nn_lon is the length of keastfeeding in months; (averaged for all of a woman’s children). The urbanrural dichotobny is strong once again (5.3, 5.0, lOa1!. This may reflect the greater preeafe?ce of traditional norms regarding breastfeeding and related

‘“We-?lwconst:rlL%xl similzpmeasures of protein and vitamin A and iron intakes, since these al,io are in relativelylimited supply. However, within the development literature, calaric defic+esr:~~ are consider -d the best single measure of nutrient deficiencies. Multirollinearity prcciudc-. the inclusiotn of other measures simultane,usly. For further discussion see the two i&rence3 in the previous note.

Tatle 2 Regression estimates of average length of breastfeeding (in months) for combined .-___ Right-hand side variable< Estimates (~-statistics)” ___-

sample.

_

Commodity consumption and income Other income Woman’s predicted earnings

-0.oc2 0.35

(0.1)

- 0.004 0.28

(0.1)

(2.2)

Other factors in household Woman’s schooling Woman’s age Informal sectorb Household size Live births/year

-0.49 0.13 0.45 0.09 - 2.6

(6.2) (4.1) (1.0)

-- 0.40 0.17 -0.30

(3.6)

-I

(3.X) (4.1 I (0.h) (0.5) il.‘)

- 4.3 -0.70

(3.X) (1.4)

-2.3 - 0.61

11.51 (I.‘)

- I.4 - 0.0 I

(3.2) (0.0)

- 0.2 --0.32

(0.3) (O-7)

- 0.16 - 0.15

(0.8) (2.5)

6.2

(3.5)

Nutrition-specific Caloric intake Refrigerationb

(1.6)

production

(1.6)

0.83

5

commodities

Public environment Sewersb Parasitesb Other variables and controls Trimesters of medical care Expected average schooling

7.7

Constant R’

(5.9)

0.17

F

0. I 5

19.5

0.2

1281

Sample size -____I__

745 ____.__

___

-.

“To the right of the pint estimates in parentheses are the absdute values of z-statiks. For a two-tailed test at standard signific;nce levels of 5”,, (lb?‘,). a value equal to or greater than 2.0 (1.6) is cignificant. ‘See note c in table 1.

practices, or the lesser availability of substitutes in the rural as compared to the urban areas. Because of some puzzling results concerning the coefficient estimates for this vartable. we estimated some relations for the length of breastfeeding as a function of a number of other variables in our system. The results in table 2 indicate no significant association with other income. although there is a significantly positive association with the woman’s predicted earnings if the expected average education of her children is not have a significantly negative included. Births per year. not surprisingly, coefkient estimate, since the currently youngest child must be weaned with the coming of the next child. The other significantly non-zero coefficient estimates suggest that longer breastfeeding is a lower socioeconomic class phenomenon, perhaps due to traditional norms or to the opportunity cost of a woman’s time: these include a significantly positive estimate for her LIP:.

1711

8. L. Wolfe und J. R Beltman.

;~nd significantly tihe standardized children.

Child ntortalif~.

heal!lt and rzrrtrition in u DC

negative ones for her I:ducation, for having sewers, and for caloric intake and the expected average education of her

Ptrhlic enuironmerat (P). We posited in section 2 that the biological functions of child mortality, health and nutrition status depend directly on certain elements in the envir0nmen.t outside the household. One of the reasons that H’Csubdivide our sample into three regions is that such environments vary so much with the degree of urbanization. We also include three proxies for irnporlant within-region differences in these environments. The first of these is whether or not the home is integrated into a sew~~ge s rsfenl *’ The proportion of households so connected is strongly associated a fth tt;c degree of urbanization (0.75. 0.30, 0.03). The .second is the population density. Although the availability of public services altd of integrated markets probably increases with population densities up to a pointVt2 eventually crowding and congestion are offsetting. F’or the central metropolis we use data on the population densities within neighborhood sectors. Because of the sparse distribution of population in many of these sectors, particularly after the 1972 earthquake devasted the comi;lcrcial center of the city, on the average population densities are less in the central metropolis than in other urban areas, although greater than in rural areas [Wolfe. Behrman, Belli, Cairns and Williamson (1980)]. The third variable relates in part to the quality of the water supply, although it also reflects the habits of the household and particularly of the woman respondent. As such it pertains to dimensions of household poducdion as b,ell as eo the extra-household environment. This variable is whether or not the woman has had parasites. The proportion of households in which the woman h.as had parasites is lowest in thle central metropolis, and sfightly higher in the other urban areas than in rural areas (0.41, 0.57, 0.541? 3 Other variables and ccncrols. We also are able to include several other ~Gables of interest. One of’ these is child’s sex. The proportions that are rrlltle are slightly below half ir. all three regions (0.48, 0.48, 0.46). This may represent genetic differenws between the sexes, with males presumably ’ 'Xntonovsky ( 19791, Dyson (1978) nnd Puffer s.nd Sxrano (1973)report inverse associations h~tawn the quality of housing (particularly regarding w.lter and sanitatic%) and infant and child mkrtahi! m developing countries. “In W:)lfe and Behrnnan [19Y2c) we tind some evidetxc consistent with this pat::ern for food mrkets in the central metr~apoiis. 23This variable is based on r1xa1I self diagnos s, and undoubtedly is subject to considerable .mtzuremcnt error. We do nor. find evidence trf it being significant at the 5% ltvel in the z~matcs pre.serrted below, perhaps due to such error. If it is drllpped from thr: etitimated rrr..~irrnc. thz other coef&ient estimates are not effected ~ tbstantially, .jo its inclusion apperently t, =:--eras’CIIUSChiasef in the other estimates.

CL..

Wo!fe and J.R.

Behrmon.

Child wxwtulit~.

ltcal~h md rlutritirm

:II u DC

179

tending to be weaker than females. Son preference, however, based r3n higher expected earnings or status for males, may of&et those diffet,ences and result in intrafamilial allocations that favor males over females.24 A second variable is whether or not the respondent h2.s ever migrated. Migration may reflect the self-selection of individuals with greater ability and motivation for socioeconomic success. It also may result in a wider exposure and a change in reference norms. The lowest proportion of nonmigrants is in the rural areas, with a slightly higher value in the other urban areas than in the centre! metropolis (0.50, 0.55, 0.34).‘5 A third variable is child’s age. As we note above, we include this variable in the mortality relationships to control for expected systematic underreporting of deaths of very young infants. Based on the literature on breastfeeding and infant health, we expect lower mortality rates for l--6 months after birth than before or after this period. We include it in the weight and height relations to see if there are systematic associations with age in divergences below standards. For our sample there seems to be some inverse association between children’s ages and urbanization, although the differences are not statistically significant. For a smaller subsample for which we have data, we also include three additional variables. The first is reported low birth weight, as indicated by 8’?;, of our total population. This variable relates to a combination of earlier genetic and environmental factors. The second of these is the number of trimesters of medical care during pregnancy, which averages 1.8 for our overall sample. This is possibly an important earlier form of specific commodity consumption. (For our probability of mortality estimates we also are able to include a related variable for whether or not there was medical attention at birth). The third is the expected average years ufschnnlir~g of the respondent’s children, for which we have a mean of 11.1 years among respondents. This can be seen. as an alternative form of huma?. capital accumulation to health and nutrition investments, which are generally in the complements or substitutes for ikz6 As such, it may be endogenous model. However, it also may reflect norms and expected household incomes. Unobserved vuriubies. We remind the reader that there are several variabies which are included in the model above on which we have no observations or very poor proxies. One important example is market prices, and a seconcl is 241n Behrman and Wolfe (1981e), Behrman, Wolfe and Blau (1980). and khrman, Wolfe and Tunah (19X0), we find that expected average earnings are higher for males than for females, even though the marginal returns to schooling are, ii’different, greater for the latter. 25We explore the micro-determinants of migration in terms of person-specific earnings options, marriage options, and other personal characterit rics in Behrman and Wolfe (1981d). Z61f the expected earnings function is log licear, and human capital investments affect parental utility only through these expected earnings, inrrafamilial relative allocations of investments in schooling are proportional to those in health a!ld Lutrition.

I XI)

B.J..

Wolfe and S.R

Bchrtnatt,

Child mortality.

health attd tnttritJon

in N DC

norms. The absence of these is among the factors that led us to subdivide our sample into regions by degree of urbanization, A third important example is grnetic endowments and a myriad of other possible family effects. Of course, the absence of these variables does not cause biases in our estimated coefkients if the unobserved variabaes are uncorrelated with our observed variables, In some cases, however, t-his seems a very strong with assumption. In particular, schooling quite possibly is correlated unobserved abilities, motivations and norms.*’ 4. Estimates Table 3 contains probit estimates for the probability of reported child mortality. Tables 4, 5 and 6, respectively, give ordinary least squares estimates with the dependent variables representing standardized weight, height and biceps size. In all four of these tables, esrimat:s are presented for rhe combined sample and for each of the thrte regions. In general the estimated relationships are significant at standard levels. Slope coefficients often differ among regions, and the constant differs as well as for the bicep measure. Therefore the subdivision of the snmple is important, although we cannot identify whether this is so because of differential prices. norms, or degrees of unperceived jointness across regions. Table 7 gives alternative estimates for the standardized weight, ‘height and bictlrx dependent variabks for the subsample (combined across regions) for which the three additiorlal varia’tles discussed at the end of section 3 are available. We organize our discussion of these estimates with reference to the estimated effects of the right-hand side variables in the same order as in the previous section. Commodity consumption (z) and income. We iind very little evidence of a stibstantial impact of either orher income or the woman’s predicted earnings. The only significant coefficient estimates are positive ones in the weight and height relationships for other income in the rural areas at the 10% level, which is ‘also significaint in the reiationship with biceps circumference, as is the coefficient estimate in th:: combined sample of the woman’s predicted earnings in table ?. That th,: rural areas are relatively poor suggests that ihere may be ;.n Engel curve phenomenon with more response at low-income levels. Even the magnitudes of these significant estimates, however., are not vc’ry large. Therefore we conclude that our results are not consistent with th~rc being an important generalized income effect (whether nominal or more ‘fr:ll’ a~ in the predicted earnings variable) on child health and nutrition in thar CSI nates of the returns to schooling in the United States mri:y be biased to thr faiiure to control for ability .nd motivation, see Ekhrman. Hrube; Wales i19XQ~.En Behrman and Wolfe (19) If) we find evidence of a smaller upward ~-EKZL :I-~GF the same s;tmple tRst we use in this study. “I-or

~dence

up.>rds due f 4~~b~ll-n md

Probit estimates of probability

Variables Commodity consumption and income (100’s of cordobas per fortnight) Other income Woman’s predicted income Other factors in household production Woman’s schooling (years) Age (years) Participation in informal sectorb Male companion presentb Schooling (years) Household size Births/year in five years

Table 3 of child mortality Combined sample”

for combined sample and regions. Central metropolis”

Other urban

Rural’

- 0.02 (0.8) - 0.09 (0.7)

- 0.0 1 (0.4) 0.04 (0.6)

0.4 (0.3) -0.01 10.0)

-0.09 (3.0) . -- 0.06 ( 1.O) o.ocJ(0.3) 0.00 (0.0)

-0.10 (2.6) 0.00 (0.3)

- 0.01 10.5) 0 02 iO.3)

- 0.08 (0.6) -0.12 (0.7) _ 0.01 (0.4) -0.02 (1.1) 0.58 (2.8)

-- 0. I2 (0.4) 0.08 (0.2) 0.06 (I 2) 0.01 (0.2) 0.79 (2.2)

-0.28 (1.3) 0.07 (0.3) tJ.00 (0.0) -0.05 (1.9) 0.04 (0. I)

I46 (1.2) 0.21 (1.4)

1.5 (1.5) 0.10 (0.3)

0.37 r3.8) 0.23 (1.11

Nutrition-specific commodities Average caloric intake Refrigeraiiorrb Length of breastfeeding (months)

-O.Ol(I.ti)

-0.03(1.1)

-- 0.00 (0. I )

-0.05 (I.61

Public environment Sewers” Population density (people/km’) Parasitesb

-0.55 (4.0) 0.00 (0.9) 0.16 (1.4)

-0.32 (1.1) 0.00 (1.1; - 0.08 (0.3)

-0.48 (2.2) 0.00 (0.6) 0.30 ( 1.9)

-- 0.00 (0.6) --0.3x I I .01

Other variables and controis Maleb Never mifxatedb Child’s age (months) Low birth weightb Trimesters of medical care Medical attention at birthb Rural’

t-I.02(0.2) 0.15 (I.31 0.1 I (2.7) 0.61 (3.9) -0x5 (1.0) -0.09 (&7) 0.10 (0.5)

--0.24 0.35 0.26 0.42 0.11 - 0.30

(O.!)) (I.>) (2.6) (1.1) (1.0) (1.O)

I) 1-t (0.9) 0.05 (0.3) 0.07 ( 1.4) 0.76 (3.5) -O.OP (1.2) - 0.03 (0.2)

- 0.24 -0.48 0.15 9.85 0.01 0.18

Constant

-2.0

-4.2

ix.!)

- 0.70 ( I .O)

-0.77 (0.5)

2* log likelihood Sample size of which number died ____~_.__

88.4 1871

(4.1)

32.8 88X

61.1 830

0.03 (OX)!

(0.6) 10.X) (1.2) (1.6) (0.0) (0.3)

20.4 153

“see note a in table 2. bSee note c in table 1.

this developing country, a conclusion which is similar to that attained for the United States [Chernichovsky and Coate (1979), Edwards and Grossman (197711 and for Colombia [Heller and Drake (1979)], as well as for nutrition intakes for this country [Behrman and Wolfe (1981e), Wolfe and Behrman ( 1982c)j. Therefore the emphasis of the World Bank (l980), Ward and

B.L. Wolfe and J.R. khrman,

182

Child mortality, health and nutrition in a DC ‘Table 4

Regression estimates of standardized

Right-hand side variables Commodity consumption and income (loo’s of cordobas per fortnight) Other income Woman’s predicted mcome Other factors in household production Woman? schm!Ing (years) Age (years) Participation in

child weight for combined sample and regions.

Combjned sample”

Central metropolis”

0.00 (0.0) --i).CU (0.8)

-0.01 0.07

(0.8) (1.3)

0.02 (2.1) 0.02 io. 1)

0.04 (2.4)

0.05 (1.6) 0.00 (0.1)

0.04 -0.01

(1.6) (0.6)

0.09 (2.1) o.otr (0.4)

0.3x -0.18 0.01 - 0.02

(2.‘) (I.11 (0.7) ( 1.O)

-0.57

(2.2)

0.11 (0.4)

(2.4) (0.9)

1.2 (2.1) -1.1 (1.6)

(0.5)

-0.00 (0.1)

(1.6)

0.60 (1.2)

-0.OGO4 (2.0) 0.11 (0.9)

- 0.00 (0.0) 0.02 (0.1)

- 0.06 -0.02 -0.01

(0.6) (0.2) (3.3)

-0.30 (2.1) 0.01 (0.1) -0.01 (1.4)

-0.52

(1.0)

- ! .4 (2.2)

0.00 (0.0)

0.43 0.06 0.03 s-003

- 0.34 (2.3j

--0.63 (2.3)

0.54 (2.2) 0.26 (2.6)

-0.34 (0.7) 0.38 (2.4)

-0.01 (1.8)

-0.02 (2.1)

0.19 (2.1)

0.15 (1.0)

-0.00 (1.6) o.rt-! (1.4)

0.00 (1.1) o.r2 (0.9) 0.03 (0.2) -0.02 (0.I) -0.02 (5.0)

Ruralb

-0.07 (1.0) - o.u3 (0.4) - O.Ul (6.4) -0.04 (0.4)

Constant

-0.38 (1.3)

0.09 (0.2)

Household size Births/year in five Ye=s

Rural”

0.00 (tI.3) o.co ‘0.1)

0.27(3.0) 0.02 (0.2) 0.01 (1.2) - 0.03 (2.4)

informal sectorb Male companion presenth Mioohng (years)

Other urban’

(2.9) (0.3) (1.4) (1.6)

0.08 0.24 - 0.02 -0.04

(0.3) (1.1) (0.5) (1.5)

Nutrition-specific commodities

Average caloric intake Refrigerationb Length of breastfeeding (months) Public environment Sewers” Population density IpeopJe&m*) Parasite9 Other variables and controls Makb Kever migrateb

Child’s age (months)

RL F

G.14 ‘IO.2

Sample size

-

1281 ---__-

0.13 3.8 517

0.9 1 0.13 -0.01 0.23

0.15

0.12

4.2

1.7

499

265

%e note a

in table 2. ‘Sre note c in table I.

Sanders (1980) and others r3n imprrwing health and nutrition generalized income increases seems questionable.

status through

O&r Jackws in Izousehold production. We find evidence caf a fairly widespread impact on child he&h Iof :he woman’s educatton and participation in the in$x-mai sector (although not of her age),. At the stsndard S’Y,level of

8. L. Wolfe md

J. R. Behrman. Child mortality, health artd mtrition in (I DC

Regression estimates of standardized

183

Table 5 child height for combined sample and regions.

-

Right-hand

side variables

Commodity consumption and income (100%of cordobas per fortnight) Other income Woman’s predicted home Other factors in household product non Woman’s schooling (years) Age (years) Pit rticipation in informal se: torb Male companion presentb Schooling (years) Household size Births/year in five years Nutrition-specific commodities Average caloric intake Refrigerationb Length of breastfeeding (months)

Combined sample’

Central me2ropolis”

0.01 (1.0) -0.02 (0.5)

0.05 (2.7) 0.01 (0.7) 0.02 0.14 0.03 -0.02 -0.72

(0.2) (1.1) (2.3) (1.9) (4.3)

0.71 (It) 0.15 (1.4) -0.2

(2.4)

Other urbana

Rural’

0.00 (0.3) -0.01 (1.3) -O.OJ (0.8) 0.06 (1.0)

0.X 0.01

(1.9) 0.06 ( 1.9) (1.3) -0.01(0.7)

0.02 (0.2) 0.41 -0.07 (0.3) 0.08 0.05 (2.2) 0.04 -0.04 (2.0) 0.01 -0.87 (3.;) --0.97

(1.9) (0.4) (2.0) (0.4) (3.1)

0.03 (3.4) 0.07 (0.3)

0.10 (1.9) 002 (1.3) 0.18 0.41 - 0.04 -0.05 -0.26

(0.5) (1.5) (0.8) (1.7) (0.8)

-C..L 0.32

(0.9) (2.2)

0.95 (2.1) 0.03 (0.1)

2.6 (3.6) - 0.72 (0.9)

-0.2

(2.2: -0.02 (1.6)

-0.01 (0.6)

Public environment Sewers” Population density (people/km*) Parasitesb

0.11 (1.1)

0.18 (1.2) -0.04 (0.2)

0.61 (1.0)

-0.00 (1.4) 0.14 (1.7)

o.c!03 (1.7) -0.00 (1.2) 0.15 (1.‘) 0.05 (0.4)

0.00 (0.1) 0.25 (1.4)

Other variables and controls Maleb Never migrztedb Child’s age (months) Ruralb

0.03 0.08 -0.03 -0.13

(0.3) (0.9) (10.1) (1.0)

-0.00 (0.3) 0.12 (0.9) 0.12 (0.9) 0.09 (0.6) -0.03 (7.6) -0.02 (4.7)

-0.10 (0.6) 0.12 (0.6) -0.02 (2.9)

Constant

-(I.%!

(2.6)

-0.31

-3.2

R2 F

0.22 18 1

Sample size

1281

0.21 6.8 517

(0.6) - 1.3 (2.1) 0. I 8 5.3 499

(4.2)

0.22 3. -, 6 265

‘See note a in table 2. bSee note c in table 1.

significance, the woman’s education is inversety associated with the probability of child mortality in other urban areas and in the combined sample, positively associated with weight and biceps measures in the rural area and with height in the combined sample. At the 10% level it is positively associated with height in all three regions and with the biceps measure in the two urban areas. As we note above, these results may reflect the efliciency of the woman in household production, tastes, or genetics. We suspect that the

B. L. WoI/p and J. R. Behmatt. Child ttrorra1il.r.health and nrrlritioti itt a DC

1%

Table 6 Regression estimates ol’ standardized biceps circumference for combined sample and regions.

Right-hand side variables Commodity consumption and income (loo’s of cordobas per fortnight) Other income Wonan’: predicted income Other Ealors in household productton Wotwn’\ schooling (years) AWL!tyears) P;:rticipatron in informal Lt4torh Male companion presentb Sclmoling (years1

Central metropolis”

Combined sample”

0.00 Q.00

Other urban’

Rural”

(0.9)

0.00 (0.8)

0.00

( 1.O)

0.00 (0.6)

0.00 (1.1)

O.aOl (1.7) -0.01 (1.5)

0.002 (1.1) 0.00 (0.2)

0.003 ( 1.9) 0.00 (0.6)

0.01 (2.X) -0.00 (0.7)

O.O!)! (2.6) o.f)o (0.0)

(0.2)

0.01 0.01 0.00 -0.001 0.03

(1.5) (C ‘) (G.3) ,i.ti) (3.21

0.02 -0.00 0.00 -0.002 -0.05

(2.1 I (0.21 (0.8) 11.8) (2.9)

0.01 (1.3) -0.01 (0.5) 0.00 (0.2) -0.00 (0.6) -0.03 (2.3)

-0.01 (0.5) 0.03 (2.1) -0.00 (0.9)

0.00 0.01

10.2) (2.4)

-0.04 (1.3) 0.01 (1.2)

0.02 (0.4) 0.02 (2.2)

0.06 (1.9) -0.03 (0.7)

-0.001

(2.6)

-0.001 (1.8)

-0.001

(1.7)

-0.001 (0.9)

0.01

(2.6)

0.01 (1.1)

0.01

(i.0)

0.01 (0.5)

~-0.00 0.00

(2.4) (0.l I

0.00 (1.5)

0.01 (2.5) -o.orJ (0.9, 0.004 (2.6) -0.01 (2.0)

f rrnrtant R’ i-

x.5

Hou~:hold tizt ISit-lhs. year in live years

~ulrltrim-spcilic wmmnditiea Average caloric intake Rcfrtgrattonb

0.00

-0.00

(0.5)

(0.1)

Length of breastfeeding tmonthr

1

Puhhc environment Sewers” Populatwn

density

tpople km21 I’.. %l,:tL4h (&her earrabIes and controls \l aleb Yww; migratedh Child’s age (months)

Rural’

Enmpre 47s %c

now

IS1 a in

-scr nutc’ c

In

l&k

-0.00

- 0.003 (2.6) (0.4)

-0.00 (0.1) 0.01 (0.7)

-0.01 (1.3) -0.00 (0.2) 0.002 (0.8)

0.01 (2.0) -0.00 (0.1) 0.00 (1.2)

0.00 (0.4) -0.02 (1.7) 0.001 (3.0)

0.92 (53.7)

0.96 (30.1)

0.90 (31.1)

0.85 (23.4)

0.12

0.09

0.13

0.14

0.m 10.1)

2.5 5:7

3.5 499

2.0 265

2

table I.

schooling is not representing only efficiency. Dimensions of household efficiency that are associated with market efftciency would seem to be at least partly represented by her predicted earnings. Therefore we think that her schooling is representing in part, at !ras%,taste and genetic effects, with more ambi_guous welfare interpretations than would be the case if only clficiency were involved. In any case, our estimates lead to a conclusion +niiar to that of Edwards and Grossman (!s’77) for the United States, ~oman’s

B.L.

Wolfe and J.R. Behrman.

Child mortality,

heclth and nutrition

in u DC

185

regarding the relatively greater importance of a woman’s education in comparison to income in determining her children’s health. The results contrast with the more ambiguous impact of female production on child health and nutrition in Colombia found by Heher and Drake (1979). The participation of working women in the i@ormaf sector has significant positive coefficient estimates in the weight relationships for both urban areas and the combined sample, in the height relationship for other urban areas (at the 100/o level), and in the biceps relationship for the central metropolis (and in the combined sample of table 7). As we note above, we interpret the?e estimates to reflect the fact that women who ‘work in the informal sector often are able to combine child care with work and thus to devote more total time to household production than working women in the formal and domestic sectors. We find much less evidence of an impact of a male companion than of the woman’s characteristics. The only significant coefftcient estimates are positive ones f:3r the presence of a male companion in the biceps relationship for rural areas and for the male companion’s schooling for the height relationship in both urban areas and in the combined sample. That the coefficient estimates for male’s schooling are significant only for height may be due to genetic endowments, which wo,uld stern to be mere important for permanent than for transitory health and nutrition status. However, the same results might hold if the male’s education were a better measure of permanent income than other income or if norms were affected by his schooling. The more frequent significance of the estimates for women’s than for men’s schooling, together with the predominance of the former in household production, is consistent with an efficiency interpretation for women’s sLhooling rather than a genetic one. However, it also is consistent with an interpretation that schooling affects tastes, and that wornen’s tastes predominate in child-health-felated decisions. Finally, we find evidence of a widespread inverse impact of household size, and especially of other recent births, on child health and nutrition status. Household size has Ggnificantly negative coefficient estimates for the combined sample for weight, for height in the central metropolis (and at the loo//, level in the combined sample), and for biceps circumference in the smaller combined sample of table 7. The number of live births per year in the past quinquennium significantly increases the probability of child mortality in the central metropolis and in the combined sanple (as well as in the rural areas at the 10% level) and reduces weight, heighr and biceps circumference in both urban areas and in the combined sample. Subject to the above caveat about simultaneity (particularly regarding replacement births in relation to mortaiity), these results suggest that greater numbers of children reduce average child health and nutrition levels both transitorily and

186

B.L. Wove rrnd J.R. Behrmun, Child mortality,

health and nutrition

in a DC

Table 7 Regression estimates oli standardized

child weight, height, and biceps circumference combined sample.

for smaller

--

Right-hand asidevariables -

Standardized weight”

-

Standardized height”

Standardized biceps circumferenxa

Commodity consumption and income ( IWS al cordobas per fortnight) - 0.00 (0.2) 0.02 (0.4)

- 0.00 (0.4) 0.00 (0.0)

Other factors in household production Woman’s schooling (years) Ape (years) Participation in Informal sectorb Male companion presentb Schooling (years1 Household size Births/year in five years

0.02 (0.9) -0.01 (1.1) -3.29 (2.4) 0.08 (0.5) 0.00 (0.3) - 0.04 (2.2, -0.55 (2.7)

0.04 -O.Oi 0.15 0.20 0.02 -0.01 -0.93

(1.6) (1.1) (1.2) (1.2) (1.1) (0.8) (4.3)

0.00 -0.00 0.02 0.00 -0.00 -0.002 -0.04

( I.O) (0.6) I Y..3) ((,.4) iris) G1.1) (i.9)

Sutrition-sp&fic commodities Average caloric intake Refrigerationb Length of breastfeedlng (months)

0.29 (0.8) 0.35 (2.8) -002 (2.1)

0.03 (0.1) 0.25 (1.8) -0.02 (2.3)

-0.02 0.02 -0.002

[!.2) (2 1) (3 3)

Zub!ic cnv~ronment Ewe& Population density (people:km’) Parasitesb

0.13 (1.2) -0.00 (0.8) 0.09 (0.9)

0.03 (0.2) - 0.00 (0.4) 0.13 (1.1)

0.01 -0.00 -0.00

(2.0) (1.1) (0.2)

Other variables and controls Vale” Xescr mligratedb ChilQ’s z.gc (months) 1.o~ birth weightb Tnmestr:rs of medica! care Expected average schooling R~ra:~

- 0.05 -0.05 - 0.02 -0.53 - 0.00 0.02 0.13

(0.5) (0.5) (4.9) (2.9) (0.0) (1.6) (0.5)

0.04 (0.4) 0.05 (0.4) - 0.03 (7.3) -0.53 (2.7) - 0.04 (0.8) Q.02 (1.6)

0.01 (1.6) -Q01 (1.0) 0.13004(1.9) -0.04 (3.2) 0.00 (0.1) 0.00 (0.1)

Constant -i

0.25 (0.6)

- 0.01 (0.0)

0.97 (38.0)

R-

0.13

0.16

0.12

F

4.7

6.4

4.5

Other incsome

Woman’s predicted income

Sample sii:e

745

745

0.00 0.01

(0.3) (1.9)

745

“%z no1.e a in table 2. ‘See note c in table I.

pertnancntly by causing a given level of household commodities to be spread more thinly. 3hrritkwspecifi form of commodity cnnsumpdon. We find fairly general cvidew of the impact of nutrition-specific forms of commodity consumption on child health and nutrition status, which contrasts with the limited

B.L. Wulfe and J. R. Eehrmun. Child mnrtalir~~, hculrh and nu~ririort itt 11DC

187

relevance of the generalized income (expetditure) variables.28 As we discuss above, this pattern suggests that the composition of given aggregate levels of household commodity consumption varies across households be:;ause of some combination of different prices, norms and/or utility functions. The a\ -trage household standardized caloric inrak~ has significantly positive coeCicient estimates for weight and height for other urban areas, other rural areas, and for the com’bined sample (and at the lOn/, level for biceps circumference in rural areas). Apparently in both the short and the long run, ‘you are what you eat’. The lack of significance in the central metropolis, however, is somewhat surprising. In this case, however, the quality of nutrient inputs may be captured better by refrigeration, which hias significant positive coefficient estimates for weight and height in the central metropolis (as .:,:I1 as for weight and biceps circumference in the combined sample and for biceps circumference in other urban areas). The average length oj breastfeeding has a siguificantly negative coefficient estimate at the loo/, level for the probability of child mortality in rural areas. However, the other significant coefficient estimates for this variable are somewhat puzzling. They are negative for weight in the central metropolis (and at the 10% level in the combined sample), for height in the central metropolis and in the combined sample (and at the lOoi, level in ether urblan areas), for biceps circumference in the combined sample (and at the 10% level in both urban regions), and for all three estimates for the smaller combined sample of table 7. is that prolonged breastfeeding leads to One direct interpretation nutritional deprivation in older infants, despite the advantages of initial breastfeeding in transferring immunities, providing a well-balanced diet and avoiding problems due to contaminated ?Kater. In such a case, one might expect that the effect of breastfeeding would be quadratic - but adding the square of breastfeeding to the relationships does not eliminate the estimated negative linear effect. Alternatively, prolonged breastfeeding may be correlated with a low level of household income and overall innuts into the child health and Trutrition functions because it limits women’s working hours and reduces the probability of their participation in the high-earnings formal sector, particularly in the central metropolis. However, it is difficult to differentiate this possibility from the likelihood that prolonged breastfeeding simply serves

“‘These variables are not simply proxies for income. Nutriricr intake, for example. has a very small income elasticity [Behrman and Wolfe (1981ej, Wolfe and Behrman (1582~jj LX; t!-.ough the expenditure elasticity for Lad is relatively high. Apparently with higher income. the composition and degree. of processing of food change considerably, but not the nutrients themselves.

as a proxy for low socioeconomic class conditions estimates in table 2 above might suggest.

and

norms,

as the

encironment (P). Access to sewers significantly lowers the probability of child mortality in o;rher urban areas and in the combined sample, and increases weight and biceps circumference in the combined sample (and weight at the 10% level in other urbtin areas). Populatio~t density is significantly negatively associated with biceps circumference in other urban arcas and in the combined sample (and at the loo/, level with weight in the combined sample,but also positively with height in the central metropolis). Parusites have no significant associations at the 5% level.“’ These estimates arc mixed, but suggest that the extra-family environment, particularly regarding sewers, may be important in determining child heaEth and nu!rition. Public

0: k? cariables ancC controls. We find that males have significantly less weight relative to the age-sex specific international standards in rural areas.

This suggests either that males are favored less or are inherently more vulnerable to transitory health and nutrition problems in these rural areas than are males in the U.S. reference population. We also find a significantly positive association with biceps circumference for other urban areas and the overall sample, but we expect that this merely reflects sexual genetic differences that are not controlled for in the normalization.. We obtain no significantly non-zero coefficient estimates for never migrated at the SS/, level. At the 10% level this variable has a ne:gative coefficient estimate in the relationship for biceps circumference in rural areas. This result quite weakly suggests that either women who have lived all of their lives in the same rural areas nave narrower perspectives (and less knowledge or more traditional tastes) than urban women, to the possible detriment of their children’s health and nutrition st.atus. A chMs age has a signilicantl;, positive estimated impact on the probability c-f ch.ild mortality in the central metropolis and in the combined sample. As we note above, we interpret this to reflect the systematic measurement error dicussed above In the reporting of deaths of young infants3@ Cl&l’s age also hab signif&.ntly negative coefficient estimates for a?eight and height in all of thos regions (except for rural areas for the former) and in the combined samples. These estimates are quite robust. They suggest that children in our sample fall progressively further below international :‘.\I the lo”;, level there is a positive one with child mortality in height in the combined sample (ihe latter of which is surprising). See ‘“Alternatively, this result ma:’ reflect a pattern in which mortality 1 5 month range, so that the impact of age is non-lkear. Due to datex xe have not explore4 pssible non-linear effects with quadratic

other urban areas and with footnote 23 above. is relatively low during the measurement error in birth term:s in age.

standards for transitory and permanent health and nutritinll status as they age. Finally, child’s age has significantly positive coefficient estimates for biceps circumference for rural areas and for the combined sample. Apparently this reflects better muscular development, retrris paribus, in the rural than in the urban environments (which carries over to the overall sample because of the somewhat older ages of the rural children). Low birth weight significantly increases the probability of child mortality ir. other urban areas and in the combined sample (and at the 1036 level in rural areas). It also has significantly negative effects on our other three mcasurcs of health and nutrition status in the smaller combined sample of table 7. Apparently it has considerable correlation over time with infant and child health status, although we cannot identify whether the cause of the initial low weight was genetic or environmental. In contrast we find no evidence of significant imp:uct of prenatal or partnm medical care (trimesters of medical cnre or tnedicuI c’crrcat hi,-th). Finally, we obtain positive coefficient estimares w?ich are significantly zero only at the 10% level for auerage expected education of children in the smaller combined satnple for height and weight. The association between various human capita1 investments is not obviously very strong, particularly in light of the fact that simultaneity probably causes an upward bias, if anything. 5. Concluding remarks As we discussed in the introduction, knowledge of the determinants of chilti health and nutrition status is important to understand and to alter both current and future socioeconomic welfare in the developing countries, On the basis of the theoretical framework sketched in section 2. we have obtained empirical estimates of the determinants of four indexes of child mortality, health and nutrition status in a developing country. A number of important conclusions come out of this analysis. First, it is important to distinguish among regions identified by the degree of urbanization because the estimates differ significantly among them, particularly between the urban and the rural areas. For example, men’s education has significant effects only in the former and other income only in the !atter. Our theoretical model suggests that such difference<: may originate in our inability to control for different relative prices, different norms I-or preferences and different degrees of unperceived jointness across regions. Second, our estimates uncover important determinants of current health and nutritional status (i.e., weight) and of long-run or permanent health and nutrition status (height and biceps circumference). Some of thl: C:.terminants are fairly similar (e.g., nutrition-specific commodities and cornpetition from siblings), but others differ. For example, the additional time for child car-c

1Yn

B.1.. b’olfc and J.R. Behrman. Child mortality. heahh and nutrition in a DC

which women who work in the informal sector have tends to improve current, but not permanent, chiBd health in urban areas. On the other hand,

the male’s schooling is associated significantly with permanent, but not current. health - perhaps it represents basic genetic endowments (if he is the father) or permanent income. Third, income or generalized pur’chasing power is not a major determinant of chiid mortality, health and nutrition status. Only for the relatively lowincome rural areas, in fact, does it even have significant coefficients at standard levels, and in that case not of particularly large magnitudes. Increasing income levels in the general process of economic development will not quickly imy?:ove chiid health and nutrition status for this sample, in sharp contrast to World Bank (198Q)and Ward and Sanders (1980) ciaims. F-ourth, parental schooling, particularly that of the mother, does have a wldespread positive association with child hea!: h and nutrition status. Changing formal education may have important positive effects on child health and nutrition, as Edwards and Grossman (1977) suggest is the case for the United States, but in contrast to the Heller #and Drake (1979) results for Colombia. But to the extent that these &ets are: through changing tastes, as we argue may be an important channel, the private w&fare effects are much more ambiguous than usually is assumed. 31 But, whatever the mechanism for such schooling the investment period is quite long. Adult- education programs directed towards health and nutrition practices may be more efficient, although we do not have data to test this possibility. Flflh, again with a qualification about simuhaneity, family size and the number of young siblings in particular are inversely associated with child health and nutrition. ‘Thus .there does appear to be a quantity-quality trade* oiT with possible implications for fertility. Sixth, although generalized income does not have much of an effect, some specific commodities do. In particular the average household caloric intake and the presence of refrigeration are quite important. On the theoretical level the reiative importance of specific versus general purchases suggests that there are within-region vatiations in relative prices and norms, etc., so that specific nutrient (as opposed to food) purchases are not tied to general

income in a simple Engel-curve rAanner. On a practical level, this pattern suggests that there mai:4be a high payoff, in terms of current child health and future adult productivities,

to specihc pragrams

which improve

the nutrition

of small children by subsidizing the necessary inputs [see Popkin et al. f 19801, for a similar pattern for the U.S.]. However, our estimates do not support the importance of ai] plausible specific interventions. For example, we find no evidence of a substantial effect for formal medical care. “111a parallelviewfor ferililty, Easterlin. Pollak acd Wachter (1980) discuss the ambiguous prl-slo wiTarc: interpretation if tastes hre Aanged, and S_ presem some empirical evidence c~~-.+~stcnlweth. wch a possibility in Be&man and Wolfe (19@ ,, (981c, f).

B.L.. Wo&

and J.R. Behrman.

Child mort:Aty,

health and nutrition

in LI DC

191

Seventh, our results suggest that there may be a payoff in terms of better child healrh and nutrition to some public sector investments, particularly in sewer systems. Better water systems also may have payoffs, but we have not been able to explore this possibility very satisfactorily. Eighth, we do not find support, and if anything find counterevidence, for the frequent hypoth.eses that longer breastfeeding has a payoff in terms of better child health and nutrition and that male children are favored in intrafamilial allocation. These insights, if supported by other studies, should provide a better basis for prediction and policy analysis regarding child health and nutrition status in developing countries - and thus for improving current and long-run welfare. References Anthony, R., 1979, Health, population and income in rural Nepal, Unpublisllcd Pd.D. dissertation (University of Pennsyb ania, Philadelphia, PA). Antonovsky, A., 1979, Implications 0.’ socio-economic differentials in mortality for the health system, Paper presented at WHO Meeting on Socioeconomic Determinants and Consequences of Mortality, June 13-25 (Mexico City). Becker, G.S. and N. Tomes, 1976, Child endowments and the quantity and quality of children, Journal of Political Economy 84, S143-5162. Behm, H., 1979, Socioeconomic determinants of mortality in Latin America, Paper presented at WHO Meeting or Sociaeconomic Determinants and Consequences of Mortality, June 19-X (Mexico City). Behrman, J.R. and N. Birdsall, 1981. The quality and quanti’y of schooling: Is the standard focus on years of schooling alone seriously misleading? Mimeo. (World Bank, DEDPH. Washington, DC) Behrman, J.R. and 13.L. Wolfe, 1979, The impact of health and nutrition on the number of surviving children in a developing metropolis. Mimeo. (University of Pennsylvanin. Philadelphia, PA). Behrman, JR. and B.L. Wolfe, 1980a, Fertility determinants in a developing country: of human capital. Nonresponsive, incomplete fenility, and a broader representation Discussion pape: no. 610-80, Institute for Research on Poverty (Llniversity of Wisconsin. Madison, WI) revised 1981. Behrman, J.R. and B.L. Wolfe, 1980b, Sectoral and geographic labor market segmentation and earnings functions in a developing country, Mimeo. (University of Pennsylvania. Philadelphia. PA) revised 1981. Behrman, J.R. and B.L. Wolfe, 1981a, Earnings and labor force participation iu a developing country: Are there sexual differentials? Mimeo. (University of Pennsylvania, Philadelphia. PA). Behrman. J.R. and B.L Wolfe, 1981b, Fertility determinants in a #developing country: Urbanrural differentials, non-linear eNec,s, interactions, an extended representation of human capital, and response selectivity, Mimeo. (University of Pennsylvania, Philadephia, PA). B,ehrman, J.R. and B.L. Wolfe, 1981c, Knowledge and use of modern contraceptives in a developing country: Are the effects of womer’s schooling often misunderstood? Mimeo (University of Pennsylvania, Philadelphia, PA). Behrman, J.R. and B.L. Wolfe, 1981d, Micro determinants of female migration in a developing country: Are labor market or marriage market considerations more important? Mimeo. (University of Pennsylvania, Philadelphia, PA). Behrman, J.R. and B.L. Wolfe, : 981e, More evidence on nutrition demand: Still income seems overrated and women’s schcboling underemphasized, Mimeo. (University of Pennsylvania. Philadelphia, PA).

I92

B.L. U’olfe and J. R. &hrtn:~Vt. Child nlc~rtrrlir~. hrahh and nulrition

in a DC

B&rrman. JR, and B.L. Wolfe, 198!f, The returns to schooling in a developing Counlry: Are the[e biases oue to omitted family background controls? Mimeo. (University of Pennsylvania. Philadelphia, PA). B&man, JR. and B.L. Wolfe, 1?8lg, Who marries whom? - And how it affects the returns to cchooling, Mimeo. (University,of Pennsylvania, Philadelphia, PA). Behrman, J.R. and B,.L. Wolfe, 1981h, Women’s labor force participation and earnings detamirants in a developing country, in: Proceedings of International Economic Association sixth World Congress, 1980 (Mexim City) forthcoming. Behmran. J.R. and B.L. Wolfe, 198Z!,Parental prelirences, multiple outcomes, marriage market t:onsiderations, and intrafamilial allocntions of: hum;m capital investments (University of Pennsylvania. Fhiladelphia, PA) in proms. R&man. J.R.. R.A. Pollak and P. Taubman, 151%2,Parental preferences and provision for :lrogeny. Journal of Political Economy, forthcoming. fs:hrman, J.R., B.L. Wolfe and D. Blau. 2980, Human capital and income distribution in a fcteloping country. Mimeo. (University of Penn:;ylvania, Philadelphia. PA) revised 1981. Ikhrman. J.R.. B.L. Wolfe and I. Tunali, 1980.. Determinants of women’s earnings in a ,Jevelofling country: A double selectivity, extended human capital a approach, Discussion :upe: r o. 596-g (Ilniversity of Wisconsin, Madison, WI) revised 1981. tkhrman. J.R.. 2. Hrubec. P. Taubman and T.J. Wales, 1980, Socioeconomic success: A study of the etfects of genetic endowments, family environment and schooling (Norlh-Holland, &nsterdamt Bert I R.I... 1973. Health status indexes, (Proceedings of a conference conduct,ed by Health S~~vtces Research. Tucson, Arizona, Oct., 1972 (Hospital Research and Educational Trust. C hreago, IL). fiir&& N., 1979. Siblings and schooling in urban Colombia, Unpublished Ph.D. dissertation (Yale University. New Haven, CT). Ilau. D.. 1977. On the relation between child malnutrition and economic growth in less tieveloped count&, Mimeo. (University of Wisconsin, Madison, WI). Blati. I).. 1980. Nulntion, fertility md labor supplg in developkg countries: An economic analy%a. Unpublished Ph D. dissertation (University of Wisconsin, Madison, WI). c’,ddwell. J.C., 1979. Education as a factor in mortality decline: An examination of Nigerian data. Pa?er prcseotcd at WHO Meeting on Socioeconomic Determinants and Consequences cif S%ortahty. June 19-25 (Mexico City). E hemichovsky. D. and D. Gate, 1979. An economic analysis of diet, growth and health of young children in he United States, NBER working paper no. 416 (Cambridge, MA). !‘r=!rrane. S.H.. 197% Fertility and &ucation: Wh.at do we really know? (Johns Hopkins Press, Baltimore. MD). LEasis. K. and C. Schoen, 1978, Health and the walr on poverty (Brookings, Washington, DC). f&son. T.. 1979. Levels, trends, diiferentiatc :.,ld tames of chihi mortslity - A survey World Health Statistics Reports 30 282-311. e-~erhn. R A., R-A. PoUak and M Wachter, 19110.Towards a more general model of fertility determination: Endogeuous pr ..fecen.ses a Id national fe&lity, in: R.A. Easterlin, ed.. Population and economic Chaq2 itI l~s-.klreloped countries (NBER and University of Chicago Press. Chicago, iL) 81-140. i Ad+. I..?; and %I. Grr&sman. 1977. The relationship krween childrens’ health and rnfclfectual development. NBER working paper no. 213 (New Yotkj. 4 dn:~rd%. L N. and !vI. Grcssnaan, 1978, Children’:3 heal&h and the family, NBER working paper nar 256 [Vew York). Hab~bt. J.P, R.E. Klein. R.M. Malena, R. Malt~arell awd C. Y&rough, 1974, Height, weight +randardl for pre-school children: How relevant are ethnic differences in growth potential’? kancet t.611-615. tf&r. P.S. and W.D. Drake. 1979, Malnutrition, child mortality. and the family decision v%L%~. Jo~lr~a~ ot Development Economics 6, 203-235. 8c.‘~ar’~!?/_ ~1~ 1977. br:?i!y b&&rcJuI!d and economic success: A review of the evidence, in: P. y~~%WXI cd ‘ Kinoms:ric. iktcrminants of socioecono,* ‘c success within and between FIW~~ ~~~,~t~~-~4~~~a~d. Amicerd;~m) I? 34.

B.L.

Wolfe nnd J. R. Bekwun.

Child ttrortu!iiy.

iwnl~h trrui rttrtritiotr itr r~ I)(

193

Derrick, T.W., 1979, Infant and child mortality differentials in Brazil. Mimco. tGeorgetou_tr University, Washington, DC). Michael, R.T. and R.J. Willis, 1976, Contraception and fertility: Household production under uncertainty, in: N.E. Terleckyj, cd., Household production and consumption (NBER. New York) 27-93. National Center for Health Statistics, 1376. Growth charts (U.S. D.H.E.W., P.H.S., Health Resources Administration, Rockviile, MD) HRA76-1120.25,3. Popkin, B.M., J. Akin, P. Haines, M. MacDonald and D. Spicer. 1980, Nutrition program options for maternal and child health, Institute for Research on Poverty special report no. 28 (Madison, WI). Puffer, R.R. and C.V. Serrano. 1973, Patterns of mortality in childhood (Pan American Health Organization, Washington, DC). Rosenzweig, M.R. and T.P. Schultz, :980. Birthweight. the production of child health, and input demand, Yale University Economic Growth Center discussion paper no. 352 (New Haven. CT). Schultz, T.P.. 1976. Interrelations between mortality and fertility. in: R. Ridker, ed.. Population and development (The Johns Hopkins University Press for Resources for IIIC Future. Baltimore, MD) 239-289. Selowsky, M., 1976. A note on preschool-age investment in human capital in developing countries, Economic IIevelopment and Cuitural Change 24, 707-720. Selowsky, M. and L. Titylor. 1973, The economics of malnourished children: An example of disinvestment in human capital. Economic Development and Cultural Change 22 no. I, :2 .?i. Seoane, N. and M.C. La.tham, 1971, National anthropometry in the identification of _malnutrition in childliood. Journal of Tropical Pediatric and Environmental C!nld Health, Sept., 98-105. Ward. J.O. and J.H. Sar,ders, 1980, Nutritional determinants and migration in the Brazilian northeast: A case study of rural and ur3an Ceara, Ecormmic Development and Cultural Change 29, 141-163. Wolfe, B.L. and J.R. Behrman, 1982a. De.erminants of adult women’s health status in a developing country (University of Pennsylvania, Philadelphia, PA) in process. Wolfe, B.L. and J.R. Behman, 1982b, Educational productivity in a developing country. in: A. A. Summers, ed.. New directions for testing and measurement: Mexurement of educational productivity (Jossey-Blss, San Francisco, CA) forthcoming. Wolfe, B.L. and J.R. Behrman, 1982c, Is income overrated in determining adequate nutrition? Economic Development and Cultural Change, forthcoming. Wolfe, B.L., J.R. Behrman and D. Blau, 1982, The impact of demographic changes Lm the distribution of income in a deveioping country. Journal of Developmer,t Economics. forthcoming. Wolfe, B.L., J.R. Behrman, H. Belli, K. Cairns and N. Williamson. 1980, Demoeclmomic characteristics of wotnen and different degreos of urbanization in a developing country (Institute for Research on Poverty DP 655-81, University of Wisconsin, Madisxt. WI). World Bank, 1980, World development report, 1980 (Worid Bank. Washington. DC). Zerfas. A.J., I.J. Shon, D.B. Jelliffe and E.F.P. Jelliffe. 1975. Selected indicators of null-rlioil:r) .