Nutritional status is measured anthropometri-cally,1 using three indicators: height for age (stunting), weight for height (wasting), and weight for age (underweight). Height for age is an indicator of long-term malnutrition because it
measures nutrition cumulatively over time.
Weight for height is sensitive to short-term fluc-tuations in nutritional state. Weight for age is a composite measure of weight for height and height for age; it is criticized due to its inability to distinguish children who are truly thin from those who weigh less because they are shorter.
However, it has been shown to be more closely related to mortality risk in Bangladesh than other indicators (Chen and others 1980).2
Child nutritional status is determined by com-paring the anthropometric indicator for each child in the survey population to the expected measurement for a child of the same age/height and sex from a healthy reference population.3 This comparison can be done in a variety of ways, the most effective of which is to normalize the anthropometric measure by calculating z scores:
where xiindicates the anthropometric measure for individual i and m and s are the respective median and standard deviation of the anthropo-metric measure for the appropriate age or height y and gender g in the reference popula-tion.4
The basic model of child nutritional status is due to UNICEF (1990), in which nutrition de-pends proximately on dietary intake and health status. Thus, high rates of child malnutrition are likely to be closely linked to the underlying household food insecurity and the exposure to infectious disease such as diarrhea, due to poor sanitary environment and inadequate medical treatment. A third underlying factor is the quan-tity and quality of child care, which determines
z x
i y g
i y g
y g ( , )
( , ) ( , )
= − µ σ
how effectively income is converted into nutri-tion and the share allocated to children (and, im-portantly, which children receive the most), as well as the healthiness of the home and com-munity environments. Note that these three underlying factors may be complements or sub-stitutes in the production of child health.
The meta-analysis conducted as a back-ground paper for this study (Charmarbagwala and others 2004) provides a summary of the lit-erature modeling nutritional status, as indicated by height for age z score. In multivariate analysis, nutritional status can be modeled as a function of child-specific factors, d, household-specific factors, h, and community factors, e:
z = n(d, h, e).
Child-specific factors include demographic variables such as child’s age, gender, birth order and birth interval, and behavioral factors such as breastfeeding and immunization. The relation-ship between age and nutritional status depends on whether the anthropometric indicator mea-sures short-term or long-term malnutrition. For
short-term measures, age should exert a non-linear concave impact on nutritional status: nu-trition deteriorates in the period in which children are most susceptible to disease—from the onset of weaning (which should be at age 6 months) until age 24 months—and improves thereafter. For long-term measures the relation-ship between age and malnutrition is also likely to be concave, though because stunting is often permanent, the function is unlikely to fall after 24 months of age. Gender-based discrimination in South Asia means that the coefficient on a fe-male dummy may be negative; girls who have older sisters or who are competing with the first-born male are particularly likely to be discrimi-nated against (Croll 2001).5 Surprisingly, the meta-analysis found that boy children were sig-nificantly more likely to be malnourished than girls (Charmarbagwala and others 2004). How-ever, this finding may be due to mortality selec-tivity, which was not allowed for in previous studies (see below).
More generally, children of high birth order (born latest) and children born closely together Malnutrition Prevalence in Bangladesh
F I G U R E D . 1
0 0.1 0.2 0.3 0.4 0.5 0.6
Stunting
Percent
Severe stunting
Underweight Severe underweight
Wasting Severe
wasting 1997 2000
Source: Calculated from DHS data.
C H I L D M A L N U T R I T I O N D U R I N G T H E 1 9 9 0 s
are more likely to have lower nutritional states, due to socioeconomic factors (resource compe-tition) and biological factors (e.g., physically de-pleted mothers may give birth to low birth-weight babies and may be unable to breastfeed).
Mozumder et al. (2000) find that short sub-sequent birth interval impacts negatively on weight for age, but no statistically significant im-pact of preceding interval, which they attribute to the likelihood that “the new infant holds an advantageous position with the mother, com-pared with any older siblings, because of breast-feeding” (p. 295). One issue concerning use of household composition and demographic var-iables is that they may be endogenous to de-cisions on child health, leading to biased regression estimates. As Horton (1986) argues, parents are likely to make joint decisions on child quantity and child quality (demonstrated, for example, by nutritional status). Regression estimation should account for this source of bias, e.g., through use of instruments or by dropping endogenous demographic variables to assess robustness of the other coefficient estimates.
Child immunization is an important variable to control for—vaccination reduces chances of contracting debilitating disease. A mother’s de-cision to immunize her child depends partly on availability of health services and income, but also her preference regarding modern medical care, which is unobservable.6Due to unobserv-able heterogeneity, behavioral variunobserv-ables such as immunization may be correlated with the error term in the regression equation, leading to biased regression estimates (Rosenzweig and Schultz 1983; Thomas and others 1991; Alder-man and Garcia 1994). Consistent estimation requires use of instrumental variables (the “pro-duction function” approach).
In addition to the above, determinants of long-term malnutrition (height for age) should control for genetic health endowment and birth weight, which can be an important determinant of ill-health and height throughout childhood and onwards. Commonly used indicators of health endowment are (log of) mother’s height and, due to seasonal variation in food availabil-ity, month of birth may be a good proxy for birth
weight; DHS round 2000 also provides data on mother’s recollection of low birth weight.
Household-specific factors include parental or family resources such as income, household size and composition, parental education, mother’s age and mobility, and sex of the house-hold head; these are proxies for food security and the quality and quantity of care provided to children.
Income—broadly defined to include imputed own production—is a key determinant of house-hold nutritional intake. In the health production function literature, income is considered jointly determined with nutrition and health, leading to biased estimates of the coefficient on the in-come variable in standard regression analysis (see Annex C). The extent to which current in-come and child health are endogenously deter-mined can be questioned for extreme cases such as long-term malnutrition (stunting) and mor-tality (see Charmarbagwala and others 2004).
However, estimation techniques usually instru-ment for income, a suitable instruinstru-mental vari-able being wealth.7 The meta-analysis found income/wealth to be strongly correlated with child nutrition, with the clear majority of studies finding a significantly positive effect (see Char-marbagwala and others 2004).
Food availability is highly seasonally depen-dent in rural areas of Bangladesh (see Annex J).
Models of short-term nutritional status should therefore include the month that measurement was taken (interacted with a rural sector dummy). The wet season in July-October, which occurs before the main rice harvest, aman, is a critical time of year for child health, due to greater prevalence of water-borne disease (diar-rhea, malaria) and food scarcity (Muhuri 1996;
Annex I). DHS data were collected between November and March, limiting the possibilities to explore seasonal variation in short-term mal-nutrition to this 5-month period. DHS 2000 round does, however, provide data on the household’s (self-reported) food availability dur-ing the year.
Household size is likely to be positively cor-related with child nutrition, reflecting availability of replacement caretakers, including older chil-dren and grandparents (people who may
other-wise be considered economic “dependents”).
The meta-analysis found evidence for a sig-nificantly positive effect of household size on nutrition, and a significantly negative effect of presence of young children in the household (Charmarbagwala et al. 2004). Estimation should account for potential endogeneity of household composition and demographic variables, as ex-plicated above.
Maternal education usually has a positive ef-fect on child nutrition, as indicated by the meta-analysis, which found education to be positively correlated with child nutritional status, par-ticularly of mothers. The relationship between maternal education and child well-being reflects a number of factors: greater educational attain-ment means greater income earning oppor-tunities; schooling may impart knowledge of modern caring techniques directly; literacy means better ability to assimilate new informa-tion from newspapers; exposure to new envi-ronments due to schooling makes women more receptive to modern medical treatment; educa-tion improves self-confidence and therefore de-cision-making ability in the family; and schooling provides the opportunity to form social net-works (Alderman et al. 2003). Some of the ef-fects of education can therefore be broken down by controlling for income, father’s education, mother’s literacy, and knowledge of health and family planning.8
Other indicators of women’s power in the household likely to determine child nutritional status include women’s agency, mobility, and age and whether the household head is female.
The share of economic resources devoted to children is often greater in households where women have greater say in decision making, though in the case of female-headed households (FHHs) the positive impact on child health and nutrition may be counter-balanced by the greater likelihood of both monetary and time poverty.
Community factors include location (urban/
rural, division of residence) and environmental resources including access to clean drinking water, adequate sanitation, electricity and health services. Despite the apparent logic that greater food availability in the countryside would favor
rural child nutritional status over urban, urban location was found to exert a positive impact on child nutritional status in the meta-analysis, pos-sibly due to better access to health facilities in urban areas and other factors relating to better communications and physical infrastructure (Charmarbagwala and others 2004).
Variables such as water and sanitation are key complements to food availability in determining child nutrition, because diseases such as diar-rhea diminish the body’s nutrient intake; the meta-analysis found that water and sanitation were positively correlated with nutrition (ibid.).
These are often termed “community variables”
because access is often determined by location and because there are likely to be positive spill-overs from one person’s consumption of clean water and sanitation to another person (e.g., by reducing exposure to contagious diseases).9 However, it may be so that those households with direct access to a facility, such as electricity, derive greater benefit than do other members of the community.
Mortality selectivity, a final estimation issue, concerns the lower censoring of malnutrition data due to child death. Malnourishment is as-sociated with increased mortality risk (Gomez et al. 1956; Briend and others 1986), the risk of death from malnutrition being greatest between ages 6 and 36 months in one Matlab study (Fauveau et al. 1990). Since anthropometric data cannot be collected on children who are dead and these children are more likely to have been malnourished in life, the sample of live children is unlikely to be random (Lee and others 1997;
Charmarbagwala and others 2004). In conse-quence, the nutrition model should be esti-mated conditional on survival probability; this is possible because DHS collect complete fertility and mortality histories of eligible mothers, en-abling survival probability to be estimated. Lee and others (1997) do not find evidence for non-random selectivity in survival with respect to an-thropometric measures.10