| Including the poor |
|Part I. Concepts and measurements|
|5. New research on poverty and malnutrition: What are the implications for policy?|
Understanding the relation between income and nutrition requires an understanding of how nutrients produce nutritional status or other measures of health. The standard model of individual utility maximization, in which home-produced goods (including nutritional status) contribute to utility, provides the basis for most empirical studies of nutrient consumption or nutritional outcome. These models are now well known,4 and only a few features need to be discussed here. Of particular importance is that food enters the utility function both directly as a consumer good and indirectly as an input in the production of health. Consequently, the demand for food is both the demand for a productive input and the demand for certain taste and other characteristics that provide utility. It is generally not possible to distinguish between these two roles.5
Estimates of income elasticities of demand for calories and proteins vary widely in the literature (Bouts and Haddad 1992; and Alderman 1986). The differences may reflect differences in approaches to estimations and choice of variables as much as differences in time and place that lead to behavioral differences for each sample population.
Recent studies that reach different estimated income elasticities within a single population when they use different econometric approaches can be particularly instructive. Bouis and Haddad (1992), for example, indicate the potential for bias in estimation due to measurement error. As accurate data on income are difficult to collect, total household expenditure is often used as a proxy for permanent or long-run income. This causes a bias in measuring calorie elasticity, because errors in reported food acquisition and in total expenditures are correlated.6 This bias, first pointed out thirty years ago by Liviatan (1961), is generally ignored when the total budget comprises many commodities. Bouis and Haddad, however, reported that the use of an instrumental variable for expenditures reduced the average income elasticity for their sample from the Philippines from 0.41 to 0.25. Similarly, Alderman (1989), in a panel study of five districts in Pakistan, found that using total expenditures in period t - 1 as an instrument for expenditures for period t reduced calorie elasticities by an average of one-third. Behrman and Deolalikar (1990), however, did not find substantially different estimates when using instrumental variables.
In a related vein, Behrman and Wolfe (1984) argue that income elasticities are often biased upward because of missing variable biases; higher income is often associated with higher energy consumption not only because one causes the other, but because both of them are associated with better schooling for women. When such information is available, it should of course be used, but often such biases can be controlled for by using community or household fixed effects estimation. When Bouis and Haddad (1992) and Alderman (1987b) estimated calorie demand with household panel data, however, they found that such fixed effects estimations had very little effect on the income response. Although fixed effects estimations do not eliminate errors in variables,7 these results support the view that estimates of income elasticities are not seriously biased by the inability to include all household characteristics. But the absence of an omitted variable bias does not imply that the excluded variable is unimportant in its own right. For example, disease incidence may well influence household energy consumption at a given household income, even if estimates of the effect of income (or expenditure) on energy consumption are not biased when the effect of disease incidence is excluded.
These two categories of potential bias in estimates of calorie-income elasticities—correlated measurement errors and, to a smaller extent, missing variables—have some empirical importance and potential policy relevance. But a potentially more important source of bias is indicated by the marked difference between elasticity estimates using quantities imputed from food expenditures and those based on data provided by individuals on the food they consumed in the previous twenty-four hours (Bouts and Haddad 1992; and Behrman and Deolalikar 1987). In the Bouis and Haddad study, conducted in a poor rural region of the Philippines, elasticities drop from 0.26 to 0.05 when twenty-four-hour food recall data are substituted for data derived from monthly expenditures using the same panel technique. In the Behrman and Deolalikar study, conducted in six very poor villages of rural south-central India, elasticities fall from a range of 0.77 to 1.18 to a range of 0.17 to 0.37 with the same substitution of data. Bouis and Haddad's lower estimates imply that transfer programs or income growth have a very small effect on nutrition. Behrman and Deolalikar's results imply a greater scope for such income-mediated nutrition programs, although the parameters are estimated imprecisely and the authors emphasize that the elasticities are considerably smaller than earlier estimates.
Two explanations have been offered for the appreciably smaller nutrient elasticities estimated using twenty-four-hour food recall data compared with those derived from expenditure data. Bouis and Haddad suggest that there is a data problem. Data on food expenditure, even when data on quantities of food are also available, often include purchases for guests or laborers. These purchases for nonfamily consumption correlate with income and, hence, lead to an upward bias. Although one round of twenty-four-hour recall data can be shown to be inaccurate relative to repeated recall information, there is less evidence that such inaccuracy varies systematically with income.8 Lipton (1983c), however, points out that if twenty-four-hour food recall data compress the extremes (as studies that he cites from both India and the United States suggest), they may well bias elasticities downward.
Behrman and Deolalikar attribute the much smaller elasticities derived from twenty-four-hour food recall data to a behavioral source: with rising income, households increase expenditure on food at a much faster rate than they increase the quantity purchased. This means that an increase in expenditure on, say, cereals would lead to an overestimate of the increase in calories from cereals— because the higher expenditure reflects in part the improved quality of cereals purchased (and hence fewer kilograms per unit of spending). This effect was seen in Prais and Houthakker's (1955) decomposition of expenditure elasticities into quality and quantity elasticities. Alderman (1986) presents evidence that more than half of total food expenditure elasticities may reflect such quality effects. Behrman and Deolalikar's results, however, are based on disaggregated commodity groupings and imply significant within-group quality effects—in the above example, as income increased significant shifts occurred from, say, coarser to finer (and costlier) rice, and not only from coarse grains to rice and wheat, as income increased. This finding is in contrast to most other results in the literature (Deaton 1987; Case 1987; and Alderman 1989). Moreover, many of the within group quality effects in the South Indian communities studied pertain to grains from home production, for which quality effects are not expected to be as pronounced as for goods obtained from the market; if production decisions and consumption decisions are separable—if, as seems likely, transactions costs are not large in most grain markets in rural India—the variety of grain grown should reflect ecological considerations and not consumer preferences. These concerns cast some doubt on the interpretation offered by Behrman and Deolalikar.
The two interpretations are not mutually exclusive, however, and the potential policy implications come as much from the magnitude of the income response observed as from the interpretation. Do these recent results, then, actually suggest that earlier expectations for income responses—that is, that reducing poverty will swiftly and substantially improve energy intake—are inappropriate? Behrman and Deolalikar's warning about deriving nutrient responses using expenditures, although useful, does not invalidate the many previous studies based on quantities, not expenditures (Alderman 1986; Pinstrup-Andersen and Caicedo 1978; and Sahn 1988). The data problems reported by Bouis and Haddad counsel caution, but there are too few other studies that have explored how robust Bouis and Haddad's results are to alternative estimates and variables to assess how general the results are.
The surprising aspect of Behrman and Deolalikar's results is not the average value of the elasticity—even the lowest estimate of 0.17 is not especially low, measured at the mean income of a population—but the finding that there is little difference between the average elasticity and the elasticity for the poorest members of a poor community. This finding differs sufficiently from that in a number of studies that have found marked nonlinearities in the income response to raise questions about any generalizations from Behrman and Deolalikar's results.
Strauss and Thomas (1990), for example, find appreciable curvature in the calorie-expenditure curve using both parametric and nonparametric techniques. They find that the curve is virtually flat for the top seven income deciles, but that the expenditure elasticity for households in the lowest income-per-person decile is 0.26—eight times higher than that in the top decile.9 Likewise, Ravallion (1990) finds appreciable curvature in an Engel curve estimated for Indonesia. Moreover, Ravallion adds the important new observation that, in his sample, the observed calorie intakes are densely clustered around normal requirements, so that the income elasticity of the probability of meeting those requirements is far higher than the income elasticity of demand for calories. 10 Similarly, Pinstrup-Andersen and Caicedo (1978) find nutrient elasticities for the poorest quintile in Cali, Colombia, to be three times those of the wealthiest quintile, and Alderman (1986) finds that calorie elasticities from eleven data sets average 0.48 at incomes that correspond to the average income of populations consuming 1,750 to 2,000 calories per capita per day.
To be sure, many of these earlier studies do not use instrumented variables to control for errors in variables or for endogeneity of income due to labor supply choices. Nevertheless, there is enough variability in the literature regarding such instruments, and enough doubt about twenty-four-hour recall data, so that it would be rash to exclude these earlier studies in forming our expectations about the response of the poor to extra income. Behrman and Deolalikar (1987) and Bouis and Haddad (1992), among others, indicate important methodological improvements, but their results are not necessarily representative of a wide range of socioeconomic conditions.
Engel curve estimation is not a new field of econometrics, yet there are enough uncertainties remaining and new techniques available to allow studies of the range of income responses over populations and over income groups within a population to continue yielding important results.11 There is still work to be done in, for example, estimating the response of those at the lowest tail of the income distribution to changes in income. They are likely to benefit most from programs to improve nutrition—or other health interventions—yet their response to changes in income is the most difficult to estimate. They are likely to be underrepresented in surveys; even when present, they will have few assets or transfer incomes that provide information for constructing instruments for income. Furthermore, many techniques for constructing such instruments reduce the variability of the income variable, and hence the probability of measuring significant curvature in Engel estimates even when it exists.
In circumstances such as famines, marginal increases in incomes may have an important effect on the quantity of food consumed. In a similar manner, a particularly harsh "hungry season" or an unanticipated employment shock may increase a household's calorie-income elasticity. The behavior of households in famines or following shocks may differ from that of households in chronic poverty. There are still few sound empirical studies on the short-term response to famine conditions or employment shocks.
Short-term responses may differ from the long-term response typically measured with cross-sectional data, in part because credit constraints and asymmetries in savings and dissavings possibilities are likely to be more pronounced among low-income households. Consequently, such households may be less able to maintain their level of calorie consumption in the face of transitory shocks than when there is a permanent change in the level of income, as indicated by average returns to physical and human capital or similar instrumental variables.
Short-term and long-term changes in food consumption in response to changes in income may also differ because current demand reflects past investments. Calorie consumption reflects in part the size of the individuals in the household. Measurement issues aside, change in physical size should make the long-run response to a change in income greater than that in the short run. But as household assets, including education, increase, the labor intensity of work usually decreases. Over the long term, or over a cross-section, increased income will be associated with occupations (and perhaps work days) that tend to reduce the demand for energy, thus making the calorie-income curve somewhat flatter. In the short term, however, a reduction in, say, real wages or crop yields is unlikely to affect the energy intensity of work. In such situations—with work input largely given—an income transfer or food subsidy might well have a larger effect on individual and household food intake than a similar increment to income stemming from, say, a move from manual to managerial work.
A broader issue remains, however: Do the policy implications of the more recent studies actually differ much from policies widely advocated? Behrman, Deolalikar, and Wolfe (1988), for example, challenge the view expressed in World Development Report 1980 that malnutrition is largely a reflection of poverty: people do not have in come for food. Given the slow income growth that is likely for the poorest people in the foreseeable future, large numbers will remain malnourished for decades to come....The most efficient long-term policies are those that raise the income of the poor. (World Bank 1980: 59)
Although it is difficult to assess efficiency with the data available, Behrman and his colleagues are clearly correct in saying that unrealistic expectations would be raised if income elasticities for nutrients are assumed to be closer to one than to zero. Overambitious predictions often lead to disillusionment, which could discourage policymakers from pursuing income-mediated nutrition programs.
Nevertheless, the realization that calorie-income elasticities are moderate, on average, does not in itself invalidate the proposition that income-mediated nutrition policies are an important means of reducing malnutrition. Indeed, the view cited above from World Development Report 1980 is based largely on conclusions reached by Reutlinger and Selowsky (1976), who assumed an income elasticity for calories of 0.15 at a level of intake equal to calorie requirements reported by the World Health Organization at the time12—well within the range of "low" estimates now proposed by the findings of Behrman and Deolalikar (1987). This assumption was used to explore the distribution of caloric inadequacy over income distributions and the likely change in undernutrition with income growth. Because of the income disparities found in most populations— as well as in most data sets from which estimates are derived—a much larger income elasticity would not be consistent with known distributions of food intakes. Upper-income groups often have incomes five to ten times higher than those of the poorest groups in the population, and the ratio of the two groups' calorie intakes cannot be more than two or three to one.13 It is important to note, however, that the increase in energy intake that occurs as income rises might be compressed into the lower end of the range of incomes.
Reutlinger and Selowsky demonstrated that normal income growth neutrally distributed would have only a moderate effect on energy intake, and hence on malnutrition, in the short run. They therefore argued that a policy goal of reducing malnutrition would require targeted nonmarginal income transfers and other nutritional programs. Behrman and Deolalikar's recent results, then, do not negate these conclusions; they merely keep expectations in perspective.