| Including the poor |
|Part I. Concepts and measurements|
|5. New research on poverty and malnutrition: What are the implications for policy?|
Reutlinger and Selowsky, like many other economists concerned with nutrition, focused primarily on hunger or underconsumption of nutrients relative to requirements.14 Despite the emphasis on nutrient intake in much of the economic analysis of nutrition, however, levels of food consumption are not necessarily precise indicators of nutritional status. As mentioned above, food has two roles in the household: as a consumer good and—the focus of this chapter—as an input in the production of health. The chapter next considers what is an appropriate measure of the outcome of this process of production.
Measurement of nutritional outcome
Clearly, one important measure is the production of survival—that is, the avoidance of childhood mortality. Another is the avoidance of disease. Because disease incidence and severity are difficult to measure with a cross-sectional survey, studies often take the growth or stature of individuals as a direct measure of their health. These studies express the outcome variable in terms of such anthropometric measurements as height and weight, alone or in combination. These indicators can be shown to be strongly correlated with the probability of mortality, but generally in a nonlinear manner (Chen, Chowdhury, and Huffman 1980; and Smedman and others 1987). Although severe malnutrition is linked with higher probabilities of childhood mortality, moderate levels of growth retardation are not strongly associated with mortality.
There has been a fair amount of debate, however, on whether stunting or smallness is actually an indicator of poor health. For example, Seckler (1982) argues that small stature may increase long-term survival of the family. But there is some evidence that size is positively associated with work productivity.15 And women's size may also indicate fitness in the biological sense of survival of offspring (Martorell and Gonzalez-Cossio 1987; and Thomas, Strauss, and Henriques 1990). In other contexts, however, such as in studies of the relation between malnutrition and learning, it is often difficult to trace lasting harm due to smallness—as opposed to the factors that created smallness.
Therefore, nutritionists increasingly see the damage that has been associated with smallness as due to the process of becoming small (growth faltering)— which is strongly associated with other indicators of poor health—rather than to smallness per se, except at extremes (Beaton 1989; and Payne and Lipton 1991). This faltering generally occurs between six months and two years of age. The timing of growth faltering has particular importance for monitoring the growth of individuals as well as for nutritional surveillance of communities and the evaluation of nutritional interventions. It is also important for studies of the determinants of malnutrition. Although a stunted child may have some catch-up growth, for the most part a child whose growth has faltered in the first two years of life will be on a different growth trajectory during the rest of his or her life. Consequently, over much of any sample of children observed in a cross-sectional survey, some nutritional outcome variables indicate more about past conditions than about current ones. Information about independent variables in the study that are not time-invariant is often not obtainable for analysis. This, then, counsels caution when analysis is conducted in environments in which incomes, prices, or infrastructure are rapidly changing.16
The research discussed above has been concerned primarily with nutrition as a direct indicator of health or as a correlate of other indicators of individual and family welfare. In other contexts, however, researchers are interested in nutrition primarily as an input into a further process—the production of health. These two perspectives do differ, and failure to distinguish between them occasionally causes confusion.
Income and the production of health and nutritional well-being
The effect of a policy intervention—say, an income transfer—on health or nutritional status depends not only on the demand for nutrients or other inputs but also on the magnitude of the response of household health and children's birth weight, for example, to such inputs in the health production function. Ideally, that pathway could be traced using simultaneously estimated production functions and input demands. Rosenzweig and Schultz (1983), for example, use Cobb-Douglas and translog production functions for birth weight in the United States. Guilkey and others (1989) use a similar model to examine the effect of prenatal care on birth outcomes in the Philippines, and Alderman and Garcia (forthcoming) present a model for Pakistan.
These approaches are data-intensive and often prone to identification problems. An alternative approach is to use a reduced-form model. As is well known, such models do not reveal the full structure of the causal pathways—in this case, the marginal effects of various inputs—but they do indicate the potential response to policy interventions.
Often the magnitude, and even the sign, of the coefficients of various inputs into health change when the variables are considered as choice variables—that is, as endogenous. The study by Behrman and Wolfe (1987) of the effects of instrumented energy and protein intake, sanitation, and use of medical care on child health in Nicaragua illustrates the sensitivity of results to alternative specifications. Their base model shows that income has a significant influence on the demand for these three inputs and that energy and protein intake and sanitation have significant marginal effects on health. Income has a much smaller effect on demand for inputs when community effects and the mother's childhood endowments are controlled for. Moreover, there are no variables that significantly influence child health in Behrman and Wolfe's complete model; hence, their findings are primarily negative.17
The study by Pitt and Rosenzweig (1985) of the demand for and the impact of inputs into health in Indonesia also indicates the sensitivity of models to the choice of variables presumed exogenous. They estimate the probability of individual illness as a function of predicted household nutrient intake. Using coefficients in the nutrient demand equations that they report, one can calculate the effect of a change in the wage of the household head on the probability of illness. The total effect calculated in this manner indicates that an increase in wages increases the intake of some nutrients and, subsequently, reduces the probability of illness. The effect is seven times larger when calculated using the two-stage tobit than when calculated with an ordinary tobit model. The study also indicates the sensitivity of the choice of instruments for income. The estimated impact of farm profits on nutrient demand, and therefore on health, is consistently less than the impact of wages. This is interpreted as indicating greater impact at lower income, although an instrumented variable for profits was noted to increase the coefficients of profits slightly.
Thomas, Strauss, and Henriques (1991) compared alternative specifications of income and instrumented income in their study of the impact of the mother's education on child height in northeastern Brazil. They find a significant income effect only in the rural areas. The effect remains when income is instrumented but drops when community covariates are included. Moreover, in the fixed effects model, it no longer differs significantly from zero. The authors interpret this result as an indication that the income effect works mainly through choice of residence.
The studies cited above provide new perspectives on the long-run impact of income on health and nutrition by indicating alternative interpretations of the correlations. Particularly noteworthy is their challenge to prior expectations. But a number of studies that make similar efforts to consider the potential endogeneity of income or errors in variables also confirm the existence of a significant income response. That these more conventional results are also noteworthy may therefore indicate a general shift of expectations.
One such study is Thomas, Strauss, and Henriques (1990). This study indicates that in most regions of Brazil, improvements in household income (indicated by various measures) increase the probability of children's surviving. The doubling of income, measured either by instrumented expenditure or unearned income, has an effect of the same magnitude as that of maternal literacy. Income is not robust in equations predicting child height, however; its significance depends critically on the choice of instruments.
Sahn's results (1990) from CÃ´te d'Ivoire show that instrumented income had a significant impact on children's standardized height for age, although education did not. Strauss (1990) explored the same CÃ´te d'Ivoire data set, using both fixed-effects models (with different children in the same household) and random effects. Assets and wages, both proxies for earnings,18 are positively associated with weight for height in his study. Unlike the cross-sectional approach used by Sahn, however, Strauss's fixed-effects model indicates that the father's education has a strong effect on weight for height. The coefficients of education in Strauss's community effects model are not significant, however.
Pitt, Rosenzweig, and Hassan (1990) look at how the distribution of food is influenced by the gender and the health endowment of family members in Bangladesh.19 Weight for height is shown to increase with calorie consumption, with the coefficient increasing fourfold when this consumption is instrumented. Moreover, the income elasticity for calorie consumption, when the weight-for-height endowment is controlled for, doubles to 0.12 when endowments are instrumented. The elasticity could well be greater at lower levels of income per person; the study did not investigate whether the coefficients of income, nutrients, or endowments changed as the level of the variables increased.
There are also several studies by nutritionists that—although their statistical techniques differ from the econometrically innovative models discussed above— help bracket expectations about the production of nutritional status. In particular, these studies find that the link between household or even individual food intake and stature or growth is modest (Kennedy 1989; Calloway, Murphy, and Beaton 1988; and Gershoff and others 1988). But income may have a larger effect than that mediated by food consumption (Kennedy 1989). This relatively larger role of income is consistent with the view that other inputs into health also increase with income. These reviews present relatively little evidence, however, on which household choice variables, other than food, are influenced by income and what their subsequent effect on stature is.
The demand for inputs in the production of nutrition is also influenced by the price of those inputs. There is, of course, a large body of literature on the price responsiveness of demand for various food items. The literature on the price responsiveness of demand for nutrients is more limited. The distinction between these subsets of the demand literature is that substitution between food commodities often moderates the effect of a price change on the aggregate demand for nutrients. Pitt (1983) indicates, for example, that both own-price response and cross-price responses in Bangladesh are appreciable for various food items. Consequently, the net effect of a single price rise is small and often positive for some nutrients. But this is not logically necessary. Alderman (1986) reviews a number of cases in which the net effect of a price rise on nutrient consumption is negative and substantial. Moreover, price effects, even income-compensated price effects, are often larger for low-income households.
Most studies of the effects of price on nutrient consumption look at the effect of a single price movement in isolation. But commodity prices often move in parallel. The study of the cross-responsiveness of demand and supply in Cali, Colombia (Pinstrup-Andersen, de Londono, and Hoover 1976), is one of the few models of nutrient intake in interrelated markets. In this study the indirect effects of a change in price for either rice or maize added to the direct effect on calorie consumption rather than moderating it. Although the study used supply shifts due to changes in technology as a point of departure, the approach may also be valid—depending on the nature of international trade—for price shifts due to subsidies, taxes, or movements in exchange rates.
Pitt and Rosenzweig's (1985) reduced-form model of the probability of adult illness also indicates some appreciable net price effects on nutrient consumption, although no consistent patterns emerged. Few other models of health or nutritional status have food prices in the reduced form.
Not only is nutrition an important input into health, it is in turn influenced by aspects of health—the incidence of disease and the use of health care. The price of health care therefore has the potential to affect nutritional status through its effect on the demand for such services. A number of studies indicate substantial price responsiveness for curative health care, particularly among the poor (Gertler, Locay, and Sanderson 1987; and Gertler and van der Gaag 1990). In such cases, however, cross-price responses may also mitigate own-price responses; shifts away from public health care providers may sometimes be shifts to private providers (Alderman and Gertler 1989).
A household's production of health reflects not only the levels of inputs but also the production technology chosen. Many studies have investigated this by exploring the role of parents' education. There is no issue of endogeneity in such studies, but interpretation must separate the role of education in raising incomes from its role—at both individual and household levels—in determining the amounts of inputs used in the health production function (nutrients, health care), and in determining the health production technology through which such inputs influence indicators of health (Thomas, Strauss, and Henriques 1991).
Human capital also influences nutrition through the physical stature of parents. Since studies generally find the effect of the mother's height to be greater than that of the father's height (Thomas, Strauss, and Henriques 1990; Sahn 1990; and Alderman 1990), this effect is not purely genetic.20 Both parents' education and mothers' size are generally less in low-income populations, which accounts for some of the observed correlation between malnutrition and poverty. It also suggests that the long-run effects of interventions and of income growth on health could exceed their short-run effects.
Behrman and Wolfe (1987) and Wolfe and Behrman (1987) have studied such intergenerational determinants of nutrition, arguing that a mother's education may be a proxy for other aspects of her endowment. As mentioned above, few of the variables in these studies turn out to be statistically significant. Consequently, it is not clear which aspects of the maternal endowment influence the child's health or nutrition. The influence may work through maternal stature, but which factors influence education and stature remains a question for research. In particular, if there are factors that influence one generation's health through the previous one's endowment, one would expect that the current generation's endowment would also be affected by current inputs. In a related vein, Thomas, Strauss, and Henriques (1990) find that the inclusion of maternal height, which may be considered a proxy for maternal endowment, does not reduce the additional effect of education on child survival. The coefficient of education was, however, appreciably reduced in equations explaining children's height, though it retained statistical significance.
Also assumed to affect the production of health is the availability of health infrastructure. To a degree, variables indicating that availability are analogous to price variables. For example, the distance to a clinic affects a family's resource allocation in a manner analogous to prices (Gertler, Locay, and Sanderson 1987). Because these "prices" often are higher for low-income households, they have a distributional impact similar to that of price discrimination.
Moreover, variables for the quality of health services, as well as measures of a community's production of health—such as the prevalence of disease—are often found to influence nutritional status (Strauss 1990). Similarly, CastaÃ±eda (1985) found that the most important variable in explaining the reduction of infant mortality in Chile between 1975 and 1982 was the increase in urban coverage of sewerage and potable water. Differences in such coverage were found to be more important than the positive impact of nutrition programs aimed at mothers, which in turn appeared to have more impact than child-oriented programs.
In most countries there is a positive association between the average income of a region or neighborhood and the availability of services there. As indicated in a number of the studies cited above, the failure to include such infrastructure in a model can bias upward the apparent effect of individual income on nutrition and health—entirely if higher-income individuals tend to (choose to) live in, or move to, places with better infrastructure—although the model may nevertheless accurately depict the long-term effect of reduced poverty on health and nutrition.
Also important is the question of the potential interactions of infrastructure variables and income. If the use of health care services is an input into a production function, community factors and household variables can be modeled through the input demand equations. But this is a less satisfactory approach for sanitation and other public health goods, where the reduction in the prevalence of infection vectors has clear externalities beyond a household's demand for services. Burger and Esrey (1989) found, for example, that improved sanitation has a greater impact on poor households than it does on comparatively well-off households. Similar interactions between the health effects of water quality and quantity, and between those of education and public health, are also found; these interactions too may depend on income levels. Burger and Esrey's review does not, however, discuss a household's demand for and utilization of such services from the perspective of individual or family constraints and preferences.
Thomas, Strauss, and Henriques (1991) show that maternal education is a complement, in producing health, to some goods (sewerage) and a substitute for others (health care).21 They also note a strong interaction between education and access to information: the availability of radio and television enhances the effect of education on nutritional status. Indeed, the positive effects on children's growth of literacy, and even of schooling, appear in their study to depend almost entirely on the interaction of education with communications media. The study can be interpreted to imply that the value that schooling provides stems less from the information acquired than from the skills acquired for processing information later in life. Increased income may enhance the contribution of education to family nutrition and health, for example, because information is costly to acquire.