An Analysis of Childhood Malnutrition in Rural India: Role of Gender, Income and Other Household Characteristics
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
World Development Vol. 27, No. 7, pp. 1151±1171, 1999 Ó 1999 Elsevier Science Ltd All rights reserved. Printed in Great Britain www.elsevier.com/locate/worlddev 0305-750X/99/$ ± see front matter PII: S0305-750X(99)00048-0 An Analysis of Childhood Malnutrition in Rural India: Role of Gender, Income and Other Household Characteristics SARMISTHA PAL * University of Wales, Cardi, UK Summary. Ð There are controversies regarding the role of individual and household characteristics in childhood nutritional status measured by anthropometric indicators. Using a nutrition index based on weight-for-age of children in rural India, the paper re-examines this issue. Ordered probit estimates of nutritional status suggest female literacy improves the nutritional status of boys at the cost of girls while higher per capita current income improves that of both boys and girls, though the impact is higher for boys; however, eect of income is not robust when we use instruments of longer-run income. But more income and literacy give more ways to discriminate between boys and girls. Ó 1999 Elsevier Science Ltd. All rights reserved. 1. INTRODUCTION robust in equations predicting child height; its signi®cance depends critically on the choice of Even after half a century of independence, instruments for income. In this context, we childhood malnourishment and infant mortal- shall use the WIDER dataset from rural West ity are of grave concern in India. ``Though Bengal to re-examine the role of income vis-a- quieter than famine, it (persistent under-nutri- vis other relevant individual and household tion) kills many more people slowly in the long characteristics on the nutritional status index run than famines do'' (DreÁze and Sen, 1989). based on weight-for-age among male and Using an ordered probit model, the paper female children. analyzes the nature of nutritional status derived Existing studies use various anthropometric from childhood weight-for-age among male indicators such as weight, height, height-for- and female children in the late 1980s in rural age (e.g., see Thomas, Strauss and Henriques, India. 1990; Thomas and Strauss, 1992) to measure There are controversies regarding the role of child health in dierent countries which often dierent individual and household characteris- give rise to varying results depending on the tics in child health. The World Bank view is choice of particular anthropometric indicator. that ``malnutrition is largely a re¯ection of While weight or height measures short-term poverty: people do not have enough income for thinness or wasting, height-for-age and weight- food.'' Behrman and Wolfe (1984), however, for age measure child growth relative to its argue that the World Bank tends to overem- potential and, therefore, re¯ects the extent of phasize the role of income, ignoring the signif- icance of other household characteristics, especially female literacy. Using the empirical * I am grateful to Amartya Sen and Sunil Sengupta for data from Nicaragua, they suggest that the allowing me to use the data and to two anonymous elasticity of women's schooling is much higher referees of this journal for very perceptive comments. I than that for income or household size. would like to thank Haris Gazdar for his help to access Thomas, Strauss and Henriques (1990) suggest the data and also Jean Dreze, Gale Johnson and Jocelyn that in most regions of Brazil, improvements in Kynch for their helpful suggestions on earlier drafts of household income increase the probability of the paper. Any remaining errors are my own. Final children's surviving. Income however, is not revision accepted: 22 December 1998. 1151
1152 WORLD DEVELOPMENT long-term deprivation and acute nutritional land. By 1967±68 however the incidence of ru- crisis (Kynch and Maguire, 1998). Though ral poverty was above-average in West Bengal height-for-age is considered to be a better and the situation did not improve perceptibly in indicator of stunting among children, weight- the 1980s under the Left Front regime which for-age prescribed by the World Health came to power in 1977. For example, though Organization is most commonly used for child infant mortality rate (IMR) in rural West welfare work in India. The latter has, therefore, Bengal has declined over 1981±90, the state's the added advantage of being directly related to own rate of decline in the 1980s was not much immediate policy intervention in India. Fol- faster than the Indian average; in fact, it was lowing this Indian tradition, the WIDER da- surpassed or equalled by Bihar, Uttar Pradesh, taset at our disposal has constructed childhood Gujarat, Punjab, Kerala and Tamil Nadu nutritional status in terms of a child's weight- (Sengupta and Gazdar, 1997). for-age (see further discussion in Section 3 a). Certain clari®cations are, however, in order. With the cross-section (single-period) dataset at (a) Socio-economic characteristics of the study our disposal, we shall in this paper consider the villages number of children and their birth order to be predetermined and focus on the determination The analysis of childhood malnutrition in of childhood nutritional status or health1 in this paper is based on the information from six terms of a number of household and individual villages in West Bengal for 1987±89.2 Much of characteristics, including income, female liter- the social and economic data of the WIDER acy, family size and birth order. Second, in view survey were based on complete enumeration of of the controversial results on child health ob- all households; likewise, all relevant individuals tained from various anthoropometric indica- were enumerated for individual information tors like weight, height or height-for-age, we such as educational attainment, wage earnings would ideally like to compare and contrast and nutritional status. In all, the survey covered between these alternative anthropometric indi- 749 households and 3972 individuals. The pa- cators. The WIDER dataset however contains per makes particular use of the nutrition survey information only on childhood nutritional sta- data collected from 436 children below the age tus based on weight-for-age commonly used in of ®ve (based on complete enumeration of this India and not on any other alternative indica- category of children) in all six study villages. tors. Accordingly, our analysis is based on These six villages taken together capture a childhood weight-for-age only. Finally, there good deal of the diversity present in rural West are pronounced gender biases in child health in Bengal. The study villages are drawn from many South Asian countries including India dierent agroclimatic regions of West Bengal. (e.g., see Dasgupta, 1987; Kishor, 1993). While While four villages are drawn from southern most existing studies use a gender dummy in Bengal, other two are located in North Bengal. the regression for all children taken together to Generally being located in dierent districts, examine the incidence of gender bias in child they display interesting regional variation even health (based on some anthropometric indica- within the state (Table 1).3 tor), we determine separate male and female Bhagabandasan situated in the Medinipore health functions to examine their dierence district of southern Bengal is the most pros- with respect to dierent individual and house- perous of the study villages while Simtuni is the hold characteristics. poorest in terms of average per capita income, modal wage rate or incidence of poverty. There is also a demographic variation among the 2. SAMPLE CHARACTERISTICS sample households across the study villages. Family size as well as the proportion of female Our study focuses on six villages (hereafter members is the highest in Kalmandasguri in ``study villages'') in India drawn from the North Bengal. Simtuni is a tribal village where eastern state of West Bengal. In the post-inde- 86% of the population belongs to the scheduled pendence period, West Bengal started its caste category while Kalmandasguri is the only economic development in a relatively good village with a signi®cant Muslim population.4 position among the Indian states measured by Literacy rates are generally higher in the south its high rate of urbanization, strong industrial Bengal villages as compared to those in the infrastructure and very high productivity of North Bengal, with the exception of Simtuni.5
CHILDHOOD MALNUTRITION IN RURAL INDIA 1153 a Table 1. Selected socioeconomic characteristics of the study villages Variables Study village Kuchly Sahajpur Bhagabandasan Simtuni Kalman-dasguri Magurmari District Birbhum Birbhum Medinipur Purulia Kochbehar Jalpaiguri Household no. 142 227 134 75 89 49 FSize 6.9 6.7 5.48 6.55 7.04 6.04 Female (%) 55.6 49.7 38.9 51.3 57.5 57.8 Landless 65 131 39 2 42 ÿ SC (%) 38.4 37.5 15.5 1.3 33.4 2.6 ST (%) 11.7 22.3 11.8 86 8.3 1.13 Muslim (%) ÿ ÿ ÿ ÿ 40.8 ÿ Literacy [1] 0.38 (0.29) 0.40 (0.30) 0.66 (0.55) 0.10 (0.01) 0.52 (0.39) 0.35 (0.23) Land reform 64 56 49 13 13 Pn®nc 0.21 0.53 0.35 0.32 0.31 0.65 PCINC [2] 1647 1545 2213 1160 1212 1441 Modal wage 3.42 3.3 3.6 2.57 2.66 ÿ Poverty 40.4 52.3 16.5 62.5 72.7 56.6 a Fsize: family size; Female: Average proportion of female members; Landless: number of landless households; Land reform: number of household who have gained from the land redistribution programme. Pn®nc: proportion of in- come earned from nonfarm activities while PCINC is mean income per head measured in rupees; Modal wage: Kilogram of rice per day in 1988; Poverty: % of households below poverty line. Female literacy in parentheses. There are also much ¯uctuations in the per capita income quartiles across the study villages. Three quartiles were (903, 1111, 1455), (670, 1170, 1541), (1280, 1694, 2355), (714, 804, 951), (905, 1167, 1406) and (683, 920, 1226) respectively in Kuchly, Sahajapur, Bhagabandasan, Simtuni, Magurmari and Kalmandasguri. All villages except Magurmari (close to some There is an intervillage variation in the centers of traditional industry such as biri provision of public services too (Table 2). The making6) are predominantly agricultural. percentage of villages not covered by all- Average land holding per household varies weather roads (pucca roads) is as high as 49% widely among the study villages; on average, in West Bengal. Among the study villages, land holding is higher in the south than in the Kuchly and Kalmandasguri are not connected north. Simtuni has almost no landlessness, but by a ``pucca road.'' Though there are primary extremely poor soil and water conditions. In schools in all the study villages, access to high all other villages, most landless households schools is dicult in some villages like predominantly belong to the lowest caste Kalmandasguri, Simtuni and Kuchly. The categories. scheme of household electri®cation for the Besides land, nonland allied activities have rural poor had little impact in these villages. recently emerged as the alternative source of Only Sahajapur is enjoying the bene®ts of it rural employment in West Bengal. In this while Kalmandasguri and Simtuni are not elec- respect, too, an intervillage variation is tri®ed. Even in the villages with formal electri- pronounced; whereas in Kuchly only 21% of ®cation, the rate of utilization is often as low as total income is earned from nonfarm activities, 7.9%. It appears that there is a somewhat in- the proportion is as high as 65% in Magurmari. creasing health awareness in these villages. Table 2. Distance of the village from the nearest facility (in Km s) Village Railway station Pucca road Headlth centre Primary school High school Market centre Bhagabandasan 6 0 6 0 1 0 Magurmari 4.5 0 2.5 0 2.5 2.5 Kalmandasguri 9.5 3 3 0 33 3 Simtuni 68 0 3 0 30 2 Kuchly 18 3 8 0 18 18 Sahajapur 8 0 3 0 8 8
1154 WORLD DEVELOPMENT Even in Sahajapur and Magurmari, where there to the highest expenditure quartile. The dier- are no tube-wells for potable water, some poor ences in mean per capita food expenditure households were found collecting water from across per capita expenditure quartiles are sta- the tube-wells of the neighboring villages rather tistically signi®cant7 for all the study villages. than taking drinking water from open and The contrast between average per capita food stagnant tanks. There is no health center in any expenditure between the lowest and the highest of these villages, however, and villagers have to quartiles is particularly striking (the dierence travel a few kilometers to see a doctor; the is also signi®cant at 1% level). Some intervillage situation is worst in Kuchly and Bhagabanda- variation is also observed: the absolute dier- san. Even when there is a health center, getting ence between the lowest and the highest quar- expected health care is not easy. There is always tile is the greatest for the most prosperous a scarcity of hospital beds or of quali®ed doc- village Bhagabandasan (and the dierence is tors or nurses in these rural areas. statistically signi®cant too). Finally, we compare household demographic (b) Household expenditure and well-being and economic characteristics of those in the lowest and in the highest per capita expenditure In the context of nutrition and poverty quartiles (see Table 4). In all the study villages analysis, household expenditure on basic needs (except Magurmari) the households in the of lifeÐincluding those on food, clothing, lowest expenditure quartile are generally land- housing, education and medical careÐis often less or land-poor, have a larger family size and used as a measure of household welfare a strikingly lower female literacy rate, and (Glewwe, 1991; Alderman, 1993). Such expen- predominantly belong to the scheduled caste diture re¯ects household command over re- (SC), scheduled tribe (ST) or Muslim families. sources and, therefore, to a large extent, the There is a signi®cant correlation between caste health status of household members. The and size of land holding in the villages (except analysis in this section is based on the expen- Magurmari): households belonging to the lower diture dataset in ®ve villages except Simtuni for caste categories usually possess less land. This is which the data were not available. also re¯ected in the signi®cant correlation8 be- Food is the major component of expenditure tween income and caste classi®cation.9 It also in all the study villages; the average proportion implies that there is a one to one correspon- of total expenditure spent on food varies be- dence between low caste households and lower tween 80% in Kalmandasguri and about 68% in per capita income in the study village (see ap- Simtuni. Since food is one of the crucial inputs pendix for the de®nition of the caste variable). for the production of human health, we exam- ine the distribution of average per capita food expenditure across per capita expenditure 3. ANALYSIS OF CHILD HEALTH quartiles (Table 3). We also test the statistical signi®cance of the dierence between any two Persistence of endemic hunger and malnu- quartile means for a given village and also the trition is a complex socioeconomic phenome- dierence of the means between two villages for non, giving rise to controversial empirical a particular quartile with a view to assess the evidence and necessitating careful analysis. nutritional status of the sample children. Before we examine the empirical evidence The average food expenditure per capita in- from the WIDER villages, we consider the creases signi®cantly as we move from the lowest analytical arguments to determine the nature a Table 3. Average annual per capita food expenditure (in Rupees) across per capita expenditure quartiles Village First quartile Second quartile Third quartile Fourth quartile Kuchly 636 (105.45) 772 (95.27) 1281 (133.72) 1498 (555.22) Sahajpur 635 (127.07) 720 (89.85) 814 (104.90) 1394 (365.25) Bhagabandasan 605 (287.11) 756 (136.87) 898 (156.40) 1566 (380.25) Kalmandasguri 682 (97.42) 760 (103.57) 842 (88.07) 1389 (241.65) Magurmari 547 (156.05) 660 (81.50) 764 (55.64) 1346 (190.11) a Expenditure data for Simtuni are not available. The numbers in the parentheses denote standard deviation.
CHILDHOOD MALNUTRITION IN RURAL INDIA 1155 a Table 4. Socioeconomic characteristics of households in the lowest and the highest per capita expenditure quartiles Kuchly Sahajpur Bhagabandasan Magurmari Kalmandasguri Q1 Q4 Q1 Q4 Q1 Q4 Q1 Q4 Q1 Q4 Land 0.4 4.08 0.45 4.17 0.68 3.21 0.34 0.51 0.45 1.83 Fsize 5.79 5.4 5.52 3.75 5.18 4.82 5.68 4.96 5.57 3.91 Flitrate 0.17 0.58 0.14 0.61 0.2 0.69 0.18 0.54 0.21 0.47 Hindu 34.3 74.3 7.4 72.2 39.4 90.9 40 32.4 ÿ 27.3 SC 48.6 20 66.7 14.8 39.4 3 60 66.1 40.9 40.9 ST 17.1 5.7 25.9 13 21.2 6.1 ÿ 1.4 13.6 9.1 Muslim ÿ ÿ ÿ ÿ ÿ ÿ ÿ ÿ 45.5 22.7 a Q1 is the lowest quartile while Q4 the highest. Land is measured in acres while other religious entries are given as the percentage of the total. Fsize is the family size and Flitrate, the female literacy rate. and characteristics of child quality using an- For the purpose of child health classi®cation, thropometric measures like weight-for-age. WIDER survey constructed an index of un- dernourishment as follows. It used the standard (a) Use of anthropometric indicators in de®ning ``weight curves from birth to ®ve years of age'' child health commonly used in most health centres in India. Curve I refers to as bounding the weight for One can identify two strands of the existing average well-fed healthy children (at a given empirical literature on child quality: one relat- age) and is related to the international anthro- ing to the availability of dierent nutrients in- pometric standards known as the Harvard cluding total calory (Bouis and Haddad, 1992; Standard. For dierent levels of malnutrition, Alderman, 1986) and the other relating to the there are accepted norms giving rise to weight determination of child health including survival curves II, III and IV (for further details see Sen rate, mortality, height, weight, height-for-age and Sengupta, 1983; Kynch and Maguire, etc. (Alderman, 1993). Our analysis makes use 1998). For example, for children aged 12 of the latter approach involving anthropomet- months, the points on these weight curves I, II, ric measure of a child. III and IV correspond to 7.8, 7, 6 and 5 kg There has been some debate whether small respectively; in terms of percentages, points on size (weight, height, height-for-age, or weight- weight curves II, III and IV for a 12 months old for-age) is an indicator of poor child quality or child correspond respectively to 90%, 77% and health. Nutritionists, however, accept the 64% of the accepted normal weight (7.8 kg) for damage associated with smallness as due to the this age. Similarly, for children aged 24 months, process of becoming small, i.e., growth falter- points on weight curves II, III and IV corres- ing, rather than to smallness per se. Faltering pond to 86%, 75% and 63% of the accepted generally occurs between six months and two normal weight 10 kg. (corresponding to I) for years of age. the age. Using these weight curves I, II, III and IV, WIDER survey records the nutritional Though a stunted child may have some catch up status of a child below age ®ve as follows: growth, for the most part a child whose growth has faltered in the ®rst two years of life will be on a dier- NUTST 0 if well-nourished ent growth trajectory during rest of his/her life (Alder- man, 1993). if weights at or above line I 1 if slightly undernourished Three types of anthropometric index have if weights between lines I and II commonly been used: 2 if moderately undernourished (i) weight-for-height which measures the short-term thinness of the body. if weights between lines II and III (ii) Body Mass Index (BMI) which measures 3 if severely undernourished the adult risks of morbidity. if weights between lines III and IV (iii) Weight-for-age or height-for-age which measures the child growth relative to poten- 4 if disastrously undernourished tial. if weights fall below line IV
1156 WORLD DEVELOPMENT Any analysis based on these measures will have among caste Hindu children. The chi-square an added advantage as compared to other an- statistic (27.02) between caste and nutritional thropometric measures in that they directly status is signi®cant at the 1% level of signi®- relate to immediate policy interventions in In- cance for all the villages taken together. dia. Health workers and paramedical sta in It also follows from Table 5 that irrespective India are instructed on the basis of the nutri- of the caste classi®cation there is a gender bias tional status as de®ned above: slightly under- in child nutrition in the study villages: pro- nourished children require nutrition education portion of female children well-nourished is less of the mother and supplementary feeding at than that for male children for every caste. home; moderately undernourished children re- Similarly, compared to male children, the pro- quire supplementary feeding at the health cen- portion of a female child of being malnourished ters; severely malnourished children require to is much higher, especially for severe malnour- follow doctor's advice; and disastrously mal- ishment. A chi-square likelihood ratio statistic nourished children need to be hospitalized for between sex and nutritional status (17.33) sug- treatment. It is worth noting here that the gests that there is a signi®cant association be- WIDER dataset contains information on the tween these two variables. nutritional status index of the children, but not the actual weight-for-age measure for these (b) A cross-section analysis of child health sample children; that is why we need to use the ordered variable NUTST instead of the con- Economic rationale for the analysis of child tinuous weight-for-age measure to model child health is usually derived from household deci- health. sions regarding the allocation of resources There are 436 observations in our sample, of which had originated with Becker (1965) and which only 15.6% are well-nourished. About Becker and Lewis (1965). In the standard 15% fall under the category of severely or di- model, a household maximizes its utility from sastrously undernourished while 69% are the quantity and quality of the children and also slightly (36%) and moderately (33%) under- the consumption of other commodities subject nourished. Distribution of nutritional status to a budget constraint which in turn determines NUTST across caste (Hindu, SC, ST and the optimal values of consumption and also Muslim) and sex (male and female) is shown in quantity and quality of children. But for the Table 5. About 50% of scheduled caste and short-run analysis of child health (since we 58% of scheduled tribe children are moderately, consider cross-section variation among the severely and/or disastrously malnourished in sample children for a given year), we assume the study villages while the proportion is 39% quantity of children and their birth order to be predetermined and thus ignore the dynamics of fertility and consumption choices and their Table 5. Distribution of nutritional status across caste and implications for child health. In particular, we gender a assume that at a given time, a representative Caste NUTST household maximizes a quasi-concave utility (assuming household's preferences are inter- 0 1 2 3 4 temporally separable) as a function of average consumption c of commodities by household Hindu 18 42 26 12 2 members and child health index w (based on, Male 24 48 13 12 3 say, child's weight-for-age as in our case) sub- Female 11 37 39 11 2 SC 21 29 37 10 3 ject to the current period budget constraint Male 27 26 38 8 1 (which depends on household income and Female 13 32 39 12 4 wealth and also prices of consumption and ST 7 35 36 20 2 child health goods) and the child's survival Male 8 50 27 15 ÿ function (which depends on the duration of Female 6 20 44 26 4 breast feeding, calorie and protein intake, child Muslim 10 41 41 4 4 health care practices and also the individual Male 13 50 31 ÿ 6 incidence and severity of diseases). Along with Female 7 31 54 8 ÿ determining the optimum value of average a Each entry refers to the percentage of each group in consumption c , this constrained maximization the respective NUTST category. The sum total of all the exercise determines the household demand for i- numbers in a row is 100. th child's health wi (in implicit form) as follows:
CHILDHOOD MALNUTRITION IN RURAL INDIA 1157 wi g XI ; Xh ; XP ; vi ; 1 tional intakes has not always been signi®cant where XI the individual characteristics (e.g., (Deaton, 1989), but Morduch and Stern (1997) gender, physiology, birth order) of the child, Xh ®nd that there is gender bias in childhood is the set of household characteristics (e.g., height-for-age in Bangladesh. family size, parental health, parental care, in- come) and XP the public environment (e.g., (ii) Household characteristics medical and health care practices and facilities available) into which the child is born. As- Income13 is one of the most signi®cant vari- suming, that all the right hand side variables ables in the child's health function. To a large are exogenous, Eqn. (1) can be considered as a extent, it determines the amount of dierent reduced form equation which forms the basis of inputs (e.g., food, clothing, residence, sanita- much discussion of the socioeconomic litera- tion, medical care etc.) into child health pro- ture on child anthropometry (e.g., see Heller duction function (Behrman and Wolfe, and Drake, 1979; Behrman and Wolfe, 1982; 1982, 1984; Thomas, Strauss and Henriques, Thomas, Strauss and Henriques, 1990; Thomas 1990, Strauss and Henriques, 1991). Given the and Strauss, 1992). Care must be taken how- close correlation between caste and income ever to interpret the estimated coecients if all distribution in the study villages (see Section 2), the explanatory variables in reality are not ex- caste may also have a similar eect on the ogenous.10 earning capacity of a household in the study We also extend Eqn. (1) to include the pos- villages. Family size re¯ects the number of units sible discrimination against female children in among which household resources need to be survival and nutrition observed in many south allocated according to the weights of each unit. Asian countries including India (Behrman, Family size may have an ambiguous role in 1988a, b; Dasgupta, 1987). It has been argued nutritional status depending on the relative that household's allocation of resources be- strength of size economies in consumption as tween male and female children is overwhelm- against the diminishing returns to scale in nu- ingly determined by the household's expected tritional status. Some empirical studies suggest future gain from male and female children (e.g., that the correlation between family size and see Rosenzweig and Schultz, 1982; Sen and child health is weak (Lanjouw and Ravallion, Sengupta, 1983). This gender discrimination 1995). A particularly distinctive role has been among children is largely household-speci®c attributed to literacy (male/female/overall) in though it may depend on the market opportu- that it determines the technology of a child's nities as well as the social set-up of the region. health function. The available evidence sug- This necessitates us to modify the child health gests that female literacy does have a signi®cant function (1) to derive gender-speci®c health impact on child's health while male literacy fucntion wis of the i-th child of a given sex s, does not (Behrman and Wolfe, 1984; Thomas, s m, f as follows: Strauss and Henriques, 1991; Murthi et al., 1995). wis / XIs ; Xhs ; XPs ; vis ; 2 where vis is individual-speci®c random term of (iii) Public environment the i-th child belonging to s-th sex. Heller and Drake (1979) and also Thomas (i) Individual characteristics and Strauss (1992) suggest the signi®cance of public environment on a child's nutritional For a child of a given sex, age is an important status. Among other things, this includes the determinant of the physiological characteristics provision of public utility services such as which convert consumption into nutrition11 sewerage, drinking water, medical facilities and and nutrition into higher productivity and, market opportunities. To a certain extent, therefore, higher earning potential. Sex availability of these public services aects (Rosenzweig and Schultz, 1982; Morduch and household resources as well as child's health. Stern, 1997) and birth order (Dasgupta, 1987; For example, the greater the distance of a Behrman, 1988a, b) may also re¯ect the weights primary health center from the village, the a household attaches to dierent children in the greater the cost of getting some medical care allocation of its resources.12 The empirical ev- during illness, thus worsening the health situ- idence of bias against female children in nutri- ation.
1158 WORLD DEVELOPMENT (iv) Random disturbance term ished children with those more seriously mal- nourished, especially in a population where vis refers to any individual, household or only a minority of the children are adequately community speci®c unobservable characteris- nourished (see discussion in Section 3). A bet- tics that may aect health of a child of a given ter way to model NUTST is, therefore, to use sex. For example, among other things, health an ordered probit model where one can dis- status of a child may be aected by the random tinguish the children according to each level of illness of the child or breakdown of parental their nutritional status as coded in NUTST. health, a sudden outburst of an infection or Ordered probit model diers from a univar- pollution in the local community. iate probit one in that the dependent variable is no longer a dummy variables, but an ordered variable taking values 0, 1, 2, 3, 4 according to 4. MODELING NUTRITIONAL STATUS the level of nourishment of the children. As in a univariate probit model, the model is built In this section we econometrically analyze around a latent regression variable. Suppose the factors determining childhood nutritional w0 0 is x b vis ; 4 or health status of boys and girls below age ®ve in the study villages. Given the ordered nature where w0 isis unobserved, b the set of regression of the nutritional status index NUTST provid- parameters and v the random disturbance term ed by the WIDER data, we shall use an ordered following a normal distribution with zero mean probit model to analyze the nutritional status and constant variance r2 . What we observe is of male and female children. as follows: wis 0 if w0 is 6 0 if wellnourished (a) An ordered probit model 1 if 0 6 w0is 6 l1 Suppose we observe the nutritional status if slightly undernourished NUTST of male and female children in the 2 if l1 6 w0 is 6 l2 study villages. This can be modeled at least in two ways. if moderately undernourished We could model the probability of ®nding a 3 if l2 6 w0 is 6 l3 child that is (slightly, moderately or severely) if severely undernourished malnourished. To this end, we generate a variable MALNUT which takes a value 1 if any 4 if l3 6 w0 is child is malnourished (i.e., if NUTST P 1) and if disastrously undernourished zero otherwise. Now suppose for each child i of a given gender s m, f, there is an underlying Here l0 s are the unknown threshold parameters response variable wis de®ned by the relation- to be estimated along with the regression pa- ship: rameters bs. Given this classi®cation, we can derive the probabilities of malnutrition of dif- wis x0 a uis ; 3 ferent degrees as follows: where a is the vector of parameters and u the Probw 0 U ÿx0 b random disturbance term which follows a normal distribution with zero mean and con- Probw 1 U l1 ÿ x0 b ÿ U ÿx0 b stant variance. In practice wis is unobservable; Probw 2 U l2 ÿ xb ÿ U l1 ÿ x0 b 5 what we observe is the variable MALNUT 0 0 Probw 3 U l3 ÿ x b ÿ U l2 ÿ x b de®ned as follows: Probw 4 1 ÿ U l3 ÿ x0 b MALNUT 1 if wis > 0; where U is the cumulative normal distribution 0 otherwise: function such that the sum total of above Using MALNUT as the dependent variable, probabilities is equal to one. Also note that one can use a univariate probit model to esti- here we drop the subscripts `is' for notational mate the probability if the i-th child being simplicity. We maximize the log-likelihood malnourished or not. It means however that we function to obtain the estimates of b0 s and l0 s. are not using all the information contained in Following our speci®cation of Eqns. (1) and NUTST: we are combining slightly malnour- (2) in Section 3, three sets of explanatory
CHILDHOOD MALNUTRITION IN RURAL INDIA 1159 variables are included: Individual characteristics PCINC by PCLAND (e.g., LANDFLIT, of the child, namely, sex (FEMALE), age LANDC2, LANDC3, LANDC4), and PCEXP (AGE), birth order (ORDER); and Household respectively (e.g., EXPFLIT, EXPC2, EXPC3, characteristics namely, family size (FSIZE), EXPC4; also see note to Table 6). family literacy rate (LITRATE), presence of an Finally, in order to pick up any nonlinearity adult, literate female (FLIT)14 and caste (SC, present in the data, we include some additional ST, Muslim respectively for scheduled caste, squared terms for the continuous variables: the scheduled tribe and Muslim households). So far square of age (SQAGE), square of family size as the income measure is concerned, one needs (SQFSIZE), square of income (SQPCINC, to be careful in the choice of right instrument. SQPCEXP, SQPCLAND) and square of liter- To the extent that household smooths con- acy rate (SQLITRT). The means and standard sumption, permanent income or household ex- deviations of the explanatory variables are penditure may be regarded as a better measure given in Table 6. of long-run resource availability than current Since the dependent variable is an ordered income which tends to have a larger transitory variable, estimated parameters do not re¯ect component. Cross-section nature of the WID- the marginal eects. The derivation of the ER data at our disposal does not, however, marginal eects in the ordered probability allow us to construct a measure of permanent models is quite complex; we calculate the eects income. Consequently, we have considered of change in covariates on the cell probabilities three instruments of income, namely, (i) per as follow: capita current income (PCINC),15 (ii) per cap- ita expenditure (PCEXP, for all villages except @Probcell j / ljÿ1 ÿ b0 xk ÿ / lj ÿ b0 xk Simtuni since the expenditure data for Simtuni @xk are not available) and (iii) per capita land- b holding (PCLAND) to instrument current in- come. Instruments (ii) and (iii) could be where /(.) is the normal density function, lj the considered as longer-term income measure in threshold parameter and xk the k-th explana- the analysis of child health. (c) Locational fac- tory variable. tors, namely, ®ve village dummies for six vil- lages are also considered (however, in case (b) Parameter estimates PCEXP is used as an instrument of income, we use four village dummies for the ®ve villages for First we consider the ordered probit esti- which data are available). Inclusion of these mates of b and l for all sample children as village dummies account for the village-level shown in Table 7.17 Columns (1), (2) and (3) of variation not only in the provision of public the table gives the estimates for three instru- services16 but also in prices and in market op- ments of income, namely, PCINC, PCEXP (for portunities across the study villages. all villages except Simtuni) and PCLAND, It follows from our discussion in Sections 2 other variables remaining the same. A com- and 3 however, that there has been some cor- parison of the likelihood ratio statistic suggests relation between/among variables like income/ that the goodness of ®t of the regression expenditure, literacy and caste. This in turn equation is the highest when we consider means that these variables are not randomly PCINC variable instead of PCEXP or distributed in our sample, which in turn may PCLAND. Moreover, estimates presented in introduce some bias in the estimates. For ex- columns (2) and (3) generally have lower t-ra- ample, wealthier households generally belong tios. AGE is highly signi®cant among the sam- to upper caste Hindu families or literate female ple children such that older children have members are more likely to belong to wealthier, greater likelihood of being malnourished; upper caste households. In order to control for however, SQAGE is signi®cantly negative sug- this possible bias in our estimation, we include gesting that probability of being malnourished a set of interaction terms between: family in- increases less than proportionately with age. come and female literacy (INCFLIT); female This result holds good in all three speci®ca- literacy and caste (FLC2, FLC3, FLC4); and tions. Second, PCINC and SQPCINC are both income and caste variables (INC2, INC3, highly signi®cant: children from households INC4). In addition, for the instruments of in- with higher per capita income are much less come, namely, PCLAND and PCEXP, we likely to be malnourished. But, the coecients generate similar interaction terms by replacing of PCLAND and PCEXP though negative are
1160 WORLD DEVELOPMENT a Table 6. Mean and standard deviation of explanatory variables Variables Male Female All AGE [1] 29.65 (1.717) 3.142 (1.598) 3.054 (1.658) SQAGE [1] 11.7248 (10.7547) 12.4122 (10.4508) 12.0732 (10.5951) ORDER 0.02 (0.15) 0.04 (0.19) 0.03 (0.18) SIZE 6.55 (3.28) 6.55 (3.05) 6.55 (3.16) SQSIZE 53.58 (62.76) 52.14 (52.15) 52.85 (57.56) PCINC [2] 1.389 (0.9358) 1.346 (0.8871) 1.367 (0.9107) SQPCINE [2] 2.8009 (5.4092) 2.5950 (5.0069) 2.6965 (5.2042) PCEXP [2] 1.2180 (0.5198) 1.2091 (0.5876) 1.2135 (0.5545) SQPCEXP [2] 1.75 (1.9296) 1.8055 (2.2980) 1.7793 (2.1217) LITRATE 0.40 (0.39) 0.43 (0.50) 0.42 (0.39) SQLITRT 0.31 (0.38) 0.34 (0.39) 0.33 (0.39) FLIT 0.38 (0.49) 0.43 (0.49) 0.41 (0.49) SC 0.33 (0.47) 0.38 (0.49) 0.36 (0.48) ST 0.24 (0.43) 0.24 (0.43) 0.24 (0.43) MUSLIM 0.07 (0.26) 0.06 (0.24) 0.07 (0.25) FLC2 0.0884 (0.2845) 0.1222 (0.3282) 0.1055 (0.3076) FLC3 0.0047 (0.0682) 0.0136 (0.1160) 0.0092 (0.0955) FLC4 0.0186 (0.1354) 0.0181 (0.1336) 0.0183 (0.1344) INCFLIT 713.0361 (1187.7772) 732.3180 (1153.8353) 722.8098 (1169.3880) EXPFLIT 561.0961 (828.1882) 615.7612 (855.7972) 588.8048 (841.7728) LANDFLIT 0.1401 (0.2807) 0.1456 (0.2938) 0.1428 (0.2871) INC2 364.1399 (572.6027) 439.3844 (622.0920) 402.2799 (598.6991) INC3 266.0134 (520.2095) 247.6766 (461.0789) 256.7189 (490.6470) INC4 73.5633 (269.6706) 57.6186 (241.3429) 65.4812 (255.5340) EXPC2 326.7020 (484.2186) 395.5128 (552.3618) 361.5809 (520.4217) EXPC3 183.2426 (430.9795 159.3441 (379.9271) 171.1289 (405.6133) EXPC4 73.4927 (265.5296) 57.4559 (241.0020) 65.3640 (253.2290) LANDC2 0.0244 (0.0823) 0.0385 (0.1096) 0.0316 (0.0973) LANDC3 0.0434 (0.1837) 0.0400 (0.1306) 0.0417 (0.1588) LANDC4 0.0034 (0.0277) 0.0057 (0.0369) 0.0046 (0.0327) Kuchly 0.15 (0.36) 0.18 (0.39) 0.17 (0.37) Sahajapur 0.34 (0.48) 0.33 (0.47) 0.34 (0.47) Bhagabandasan 0.15 (0.36) 0.10 (0.29) 0.12 (0.33) Simtuni 0.09 (0.28) 0.09 (0.29) 0.09 (0.29) Kalmandasguri 0.14 (0.35) 0.19 (0.39) 0.17 (0.37) No. of observations 215 221 436 a AGE: age of the child in months; FEMALE: 1 if the child is female and zero if male; ORDER: 1 if the child concerned is the eldest in the birth order, 2 if second eldest and 3 for the third eldest in the family and so on; SIZE ± number of members in the household the child belong to; PCINC: annual per capita income of the household the child belongs to; FLIT: If some female member of the household is literate; INCFLIT: Interaction between PCINC and FLIT; LITRATE: literacy rate of the members in the household; SC: 1 if the child belongs to scheduled caste household and zero otherwise; ST: 1 if the child belongs to scheduled tribe household and zero otherwise; MUSLIM: 1 if the child belongs to Muslim household and zero otherwise [Reference group is upper caste Hindu]. SQAGE, SQSIZE, SQPCINC, SQLITRT: square of age, family size, PCINC, LITRATE respectively. FLC2, FLC3, FLC4: interaction among FLIT and SC, ST and MUSLIM respectively; INC2, INC3, INC4: interaction among PCINC and SC, ST and MUSLIM respectively; EXPC2, EXPC3, EXPC4: interaction among PCEXP and SC, ST and MUSLIM respectively; LANDC2, LANDC3, LANDC4: interaction among PCLAND and SC, ST and MUSLIM respectively; INCFLIT, EXPFLIT, LANDFLIT: interaction among FLIT and PCINC, PCEXP and PCLAND respectively. Age has been scaled down by a factor of 10; PCINC, PCEXP have been scaled down by a factor of 1,000. This scaling was necessary to ensure the convergence of the log-likelihood function of the ordered probit model. both insigni®cant (while the coecient of termining childhood health status among the SQPCEXP is signi®cant). In other words, while sample children. These estimates may, there- current per capita income is highly signi®cant, fore, cast doubts about the role of permanent longer-term instrumemt of income like land or income instruments on child health derived expenditure per capita is not signi®cant in de- from weight-for-age. Finally, the gender dummy
CHILDHOOD MALNUTRITION IN RURAL INDIA 1161 a Table 7. Ordered probit estimates of childhood nutritional status for all Variables Coecient (T-ratio) Coecient (T-ratio) Coecient (T-ratio) Intercept 1.44 (2.659) 0.84 (1.190) 0.44 (0.928) AGE 0.35 (2.724) 0.32 (2.350) 0.33 (2.580) SQAGE ÿ0.04 (2.058) ÿ0.04 (1.747) ÿ0.04 (1.926) FEMALE 0.35 (3.225) 0.34 (2.895) 0.36 (3.211) ORDER ÿ0.02 (0.157) 0.04 (0.410) 0.03 (0.248) SIZE 0.09 (1.224) 0.07 (0.928) 0.09 (1.257) SQSIZE ÿ0.008 (1.887) ÿ0.006 (1.469) ÿ0.007 (1.644) PCINC ÿ1.28 (3.938) ÿ ÿ SQPCINC 0.17 (3.702) ÿ ÿ PCEXP ÿ ÿ0.80 (1.324) ÿ SQPCEXP ÿ 0.25 (2.351) ÿ PCLAND ÿ ÿ ÿ1.03 (0.902) SQPCLAND ÿ ÿ ÿ0.045 (0.082) LITRATE 0.18 (0.285) 0.22 (0.339) 0.11 (0.170) SQLITRT 0.17 (0.267) 0.003 (0.004) 0.14 (0.204) FLIT ÿ0.55 (1.317) 0.14 (0.221) ÿ0.59 (1.560) SC ÿ0.95 (2.321) ÿ0.78 (1.356) ÿ0.46 (1.974) ST ÿ0.47 (0.991) ÿ0.49 (0.650) ÿ0.31 (1.098) MUSLIM ÿ1.12 (0.906) ÿ0.72 (0.466) ÿ0.53 (1.131) INCFLIT 0.08 (0.349) ÿ ÿ EXPFLIT ÿ ÿ0.52 (1.095) ÿ LANDFLIT ÿ ÿ 0.45 (0.443) FLC2 0.34 (1.046) 0.27 (0.832) 0.48 (1.343) FLC3 1.09 (1.656) 1.12 (1.634) 1.23 (2.014) FLC4 1.20 (2.077) 1.09 (1.856) 1.39 (2.389) INC2 0.48 (1.637) ÿ ÿ INC3 0.32 (0.913) ÿ ÿ INC4 0.71 (0.631) ÿ ÿ EXPC2 ÿ 0.40 (0.825) ÿ EXPC3 ÿ 0.33 (0.517) ÿ EXPC4 ÿ 0.36 (0.249) ÿ LANDC2 ÿ ÿ ÿ0.002 (0.002) LANDC3 ÿ ÿ 2.13 (1.896) LANDC4 ÿ ÿ ÿ0.49 (0.190) Kuchly 0.23 (0.840) 0.11 (0.400) 0.06 (0.223) Sahajapur 0.28 (1.089) 0.20 (0.766) 0.09 (0.311) Bhagabandasan 0.38 (1.246) 0.08 (0.267) ÿ0.05 (0.149) Simtuni 0.33 (0.999) ÿ ÿ0.11 (0.228) Kalmandasguri 0.29 (1.129) 0.20 (0.764) 0.06 (0.223) l1 1.18 (13.924) 1.15 (13.528) 1.16 (13.966) l2 2.27 (20.723) 2.22 (20.023) 2.22 (21.960) l3 3.32 (17.971) 3.19 (16.228) 3.26 (17.671) L ÿ557.6878 ÿ515.2573 ÿ564.921 L0 ÿ599.023 ÿ542.1376 ÿ599.023 LR 82.6703 53.7606 68.2040 Observations 436 397 436 a FLC2, FLC3, FLC4: interaction among FLIT and SC, ST and MUSLIM respectively; INC2, INC3, INC4: interaction between PCINC and SC, ST and MUSLIM respectively; EXPC2, EXPC3, EXPC4: interaction among PCEXP and SC, ST and MUSLIM respectively. LANDC2, LANDC3, LANDC4: interaction among PCLAND and SC, ST and MUSLIM respectively; INCFLIT, EXPFLIT, LANDFLIT: interaction among FLIT and PCINC, PCEXP and PCLAND respectively. The dependent variable of the ordered probit model is NUTST. LR 2(LÿL0 ) is the likelihood ratio statistic with a chi-square distribution where L is the log-likelihood function, L0 the restricted log-likelihood function. Signi®cant at the 10% level. Signi®cant at the 1% level.
1162 WORLD DEVELOPMENT FEMALE is highly signi®cant and positive so Table 8. Ordered probit estimates of childhood nutritional that other things remaining identical female status children are more likely to be malnourished Variables Female coecient Male coecient than comparable male children. (T-ratio) (T-ratio) In view of the latter result, we next examine the nature of separate nutritional status func- Intercept 0.81 (0.766) 2.06 (2.445) tions for male and female children in the study AGE 0.77 (3.337) 0.11 (0.501) villages vis-a-vis dierent individual and SQAGE ÿ0.11 (2.943) ÿ0.001 (0.037) household characteristics as speci®ed in ORDER 0.02 (2.118) 0.058 (0.300) Eqn. (2). Hence, we re-estimate the nutritional SIZE 0.13 (0.903) 0.06 (0.614) SQSIZE ÿ0.01 (1.394) ÿ0.005 (1.072) status function by excluding the gender dummy PCINC ÿ1.17 (1.372) ÿ1.48 (3.067) (see Table 8). The likelihood ratio statistic re- SQPCINC 0.23 (3.173) 0.15 (2.312) ¯ecting the goodness of ®t of the model is sig- LITRATE 0.37 (0.362) 0.21 (0.230) ni®cant in each case.18 Parameter estimates for SQLITRT ÿ0.42 (0.408) 0.23 (0.241) alternative speci®cation with respect to longer- FLIT 0.71 (0.674) ÿ1.35 (2.205) term instruments of income, namely, PCEXP SC ÿ0.42 (0.462) ÿ1.34 (2.197) and PCLAND are shown in the Appendix ST ÿ0.18 (0.142) ÿ1.39 (1.379) MUSLIM ÿ1.56 (0.693) ÿ0.93 (0.616) (Table 14). INCFLIT ÿ0.42 (0.569) ÿ0.41 (1.084) Results obtained establishes the contrast in FLC2 0.32 (0.581) 0.13 (0.280) the nature of nutritional status functions be- FLC3 0.89 (0.856) 1.46 (0.000) tween male and female children. In particular, FLC4 0.73 (0.653) 1.7 (1.287) age, birth order and family income are found to INC2 0.11 (0.152) 0.85 (1.586) be signi®cant for girls; signi®cant variables for INC3 0.17 (0.169) 1.06 (1.259) boys' health are income and female literacy.19 INC4 1.2 (0.611) 0.42 (0.300) This at once implies that the nutritional status Kuchly 0.07 (0.199) 0.33 (0.660) Sahajapur 0.48 (1.426) 0.06 (0.130) estimates obtained from the combined sample Bhagabandasan 0.42 (0.964) 0.47 (0.883) (with a gender dummy) fail to capture the Simtuni 0.39 (0.799) 0.44 (0.751) complex dierences in the nature of the nutri- Kalmandasguri ÿ0.09 (0.244) 0.61 (1.301) tional status functions between male and fe- male children, thus providing a further l1 1.19 (8.309) 1.28 (10.452) justi®cation for determining child health func- l2 2.47 (14.123) 2.22 (13.391) tions separately for male and female children. l3 3.56 (12.796) 3.32 (9.539) The likelihood of a girl of being malnour- L ÿ272.8278 ÿ266.8102 ished increases with age (though the rate of L0 ÿ299.9716 ÿ290.3857 LR 54.2876 47.1511 increase decreases as age increases). The prob- Observations 221 215 ability of malnourishment is signi®cantly higher if the girl has a higher birth order, possibly a The dependent variable of the ordered probit model is suggesting a discrimination against female NUTST. LR 2(L ÿ L0 ) is the likelihood ratio statistic children (this issue will be further examined with a chi-square distribution where L is the log-likeli- later in this section) while birth order is not hood function, L0 the restricted log-likelihood function. signi®cant for boys. But, per capita household FLC2, FLC3, FLC4: interaction between FLIT and SC, income and female literacy are highly signi®- ST and MUSLIM respectively; INC2, INC3, INC4: in- teraction among PCINC and SC, ST and MUSLIM cant determinants of boys' health. Per capita respectively; INCFLIT: interaction among FLIT and family income is also a highly signi®cant factor PCINC. in improving the health of baby girls: nutri- Signi®cant at the 10% level. tional status of a boy or girl signi®cantly im- Signi®cant at the 1% level. proves as per capita family income increases.20 Female literacy rate however, exerts opposite eects on boys and girls: presence of a literate childhood nutritional status: the variable SIZE adult female member in the family improves the is insigni®cant in all speci®cations. This per- nutritional status of a baby boy, ceteris paribus, haps suggests that the positive eect of size while it lowers that of a girl (even after con- economies of scale in consumption just com- trolling for the interaction between income and pensates the negative eect of diminishing re- female literacy).21 turns to scale in nutritional status so that the There is, however, no evidence that family total eect is insigni®cant. In addition, the ag- size in the WIDER sample signi®cantly aects gregate family literacy rate does not have a
CHILDHOOD MALNUTRITION IN RURAL INDIA 1163 signi®cant eect on the nutritional status of female literacy is, however, dierent for male male or female children. Finally, we consider and female children: it improves the probability the eect of caste variables on child health. of being well-nourished by about 35% for male After controlling for the interactions between children while it lowers that for female children per capita income, caste and female literacy by about 11%. Increase in per capita income by rate, none of the caste variables is signi®cant in one unit would raise the probability of being the determination of female health. Boys from well-nourished for both male and female chil- scheduled caste households however are less dren though the extent is still higher for male likely to be malnourished (since the dummy for children (38% as against only 17% for female SC is negative and signi®cant).22 children). Finally, given that the dependent variable of our regression NUTST is an ordered variable, (c) Predicted probability of nutritional status we calculate the marginal eects of a unit change in a number of explanatory variables In this section, we use the male and female for both male and female children which, to ordered probit estimates to calculate the pre- some extent, would re¯ect the eect of a unit dicted probability of childhood nutritional change in any explanatory variable on the status of dierent degrees (see Table 10). These probability of the child of being well nourished predicted probability estimates further charac- (NUTST 0), slightly undernourished terize the extent of the dierence in male-female (NUST 1), moderately undernourished health status in the study villages. To this end, (NUST 2), severely undernourished we only consider the estimates corresponding (NUST 3) and disastrously undernourished to per capita current income which plays a (NUTST 4). These estimates are shown in signi®cant role (Table 8) unlike the estimates Table 9. These marginal eects ®gures further using longer term instruments of income. Sub- strengthen the inferences obtained from the stituting these ordered probit estimates from parameter estimates (see Table 8). In particu- Table 8 for male and female into Eqn. (5), we lar, we focus on the marginal eects of age, per calculate the predicted probability of a child of capita income and female literacy which are a given sex aged 30 months from a scheduled signi®cant in determining male/female nutri- tribe household with (shown in the parenthesis) tional status. An increase in age by one unit and without literate female of being slightly, would lower the probability of being well- moderately, severely and disastrously malnour- nourished and slightly malnourished for both ished, values of continuous regression variables male and female children, though the extent is being maintained at their gender-speci®c vil- higher for female children. Marginal eect of lage-level averages unless otherwise stated. a Table 9. Marginal eects of selected explanatory variables Variable NUTST 0 1 2 3 4 F M F M F M F M F M AGE ÿ0.11 ÿ0.03 ÿ0.19 ÿ0.01 0.13 0.02 0.14 0.02 0.03 0.002 SQAGE 0.02 ÿ0.0003 0.03 ÿ0.0002 ÿ0.02 2Eÿ04 ÿ0.02 0.0002 ÿ0.004 0 ORDER ÿ0.003 ÿ0.015 ÿ0.005 ÿ0.007 0.003 0.012 0.004 0.009 0.001 0.001 SIZE ÿ0.02 ÿ0.016 ÿ0.03 ÿ0.008 0.02 0.012 0.03 0.009 0.005 0.001 SQSIZE 0.002 0.001 0.003 0.0007 ÿ0.002 ÿ0.001 ÿ0.002 ÿ0.0008 ÿ0.0005 ÿ0.0001 PCINC 0.17 0.38 0.28 0.18 ÿ0.19 ÿ0.29 ÿ0.22 0.23 ÿ0.05 ÿ0.03 SQINC ÿ0.03 ÿ0.04 ÿ0.06 ÿ0.02 0.04 0.03 0.04 0.02 0.01 0.004 FLIT ÿ0.11 0.35 ÿ0.17 0.17 0.12 ÿ0.27 0.13 ÿ0.21 0.03 ÿ0.03 a Each entry refers to the marginal eect of a unit change in the respective explanatory variable listed in column 1 for male (M) and female (F) children on the probability of being well-nourished (NUTST 0), slightly undernourished (NUST 1), moderately undernourished (NUST 2), severely undernourished (NUST 3) and disastrously un- dernourished (NUTST 4).
1164 WORLD DEVELOPMENT a Table 10. Predicted probability of nutritional status in the study villages Village Sex NUTST 0 NUTST 1 NUTST 2 NUTST 3 NUTST 4 Kuchly Female 0.09 (0.006) 0.34 (0.01) 0.44 (0.20) 0.12 (0.41) 0.01 (0.37) Male 0.37 (0.39) 0.46 (0.45) 0.15 (0.14) 0.02 (0.02) 0.0014 (0.0012) Sahajapur Female 0.04 (0.0001) 0.24 (0.005) 0.48 (0.09) 0.09 (0.33) 0.04 (0.57) Male 0.45 (0.53) 0.42 (0.39) 0.11 (0.08) 0.01 (0.0092) 0.007 (0.0004) Bhagabandasan Female 0.05 (0.0001) 0.28 (0.006) 0.47 (0.10) 0.17 (0.33) 0.03 (0.56) Male 0.54 (0.62) 0.38 (0.33) 0.07 (0.05) 0.008 (0.005) 0.003 (0.0001) Simtuni Female 0.02 (0.0001) 0.18 (0.007) 0.47 (0.12) 0.27 (0.35) 0.06 (0.53) Male 0.09 (0.08) 0.39 (0.36) 0.35 (0.36) 0.15 (0.17) 0.0002 (0.03) Kalaman dasguri Female 0.08 (0.0003) 0.34 (0.013) 0.44 (0.16) 0.13 (0.39) 0.02 (0.44) Male 0.36 (0.44) 0.46 (0.44) 0.15 (0.11) 0.03 (0.01) 0.0002 (0.0008) a These probabilities are calculated for children aged 30 months belonging to a scheduled tribe household without a literate female, values of other variables being kept at their respective village-level averages. Numbers in the pa- rentheses denote the probability for households with a literate female. First, we consider the case when the house- male and female health status if Simtuni had hold concerned does not have any literate the same per capita current income as the female. In this case, the likelihood of being richest village Bhagabandasan (see Table 11). well-nourished (NUTST 0) and of slightly In this case, the probability of being well- malnourished (NUTST 1) is much higher for nourished increases signi®cantly for both male boys while the likelihood of being moderately and female children, though the margin is still or seriously malnourished (NUTST P 2) is higher for boys. For example, other things re- much higher for girls. There is also a pro- maining unchanged, the probability of being nounced intervillage variation in child health: well-nourished (NUTST 0) increases by (i) probability of male and female children be- about 35% for male and only 2% for female ing well-nourished is less in the tribal domi- children in Simtuni; however, the probability of nated village Simtuni; (ii) the likelihood of slight malnourishment (NUST 1) increases by female children being well-nourished is rela- 4% for male and 8% for female children. The tively higher in the Muslim dominated North probability of disastrous malnourishment Bengal village Kalmandasguri and (iii) the dif- (NUTST 4) decreases by 15% for male and ference in male and female health status is rel- 3% for female children.24 atively less pronounced in the tribal dominated village Simtuni or Muslim dominated village (d) Signi®cance of birth order on female Kalmandasguri compared to the villages dom- nutritional status inated by upper caste Hindus. Second, we consider the nature of child In the Indian society with a pronounced health in the presence of literate female in the preference for male children (Kishor, 1993), household. In this case, the probability of boys signi®cance of birth order for girls cannot be being well-nourished increases while that of ignored. In particular, Dasgupta (1987) has girls decreases in all the study villages. The emphasized the disadvantage of being a second probability of severe malnourishment (or higher) order daughter. Our regression re- (NUST > 2) also increases for female children. sults from the WIDER villages in West Bengal In other words, the eect of female literacy on (as presented in Table 8) too suggest that birth childhood health status diers between male order matters for girls though not for boys. We and female children: it improves the health of shall, in this section, further examine the eect the boys at the cost of girls in the study villages. of birth order on the nutritional status of baby Finally, given the relative importance of in- girls only. In particular, following Dasgupta's come in the existing literature, we examine the argument, we shall examine if second or higher eect of increasing per capita current income23 order girls in the sample are signi®cantly at a on child health when FLIT 0, using the esti- disadvantage with respect to their health status. mates shown in Table 8 as before. To this end, The simplest way to model this is to include a we focus on the poorest of the study villages dummy for whether the female child has an Simtuni and examine what would happen to older sister or not (OLDSISTER). Alterna-
You can also read