Poverty and Illness in Low-Income Rural Areas

by Andrew D. Foster
Poverty and Illness in Low-Income Rural Areas
Andrew D. Foster
The American Economic Review
Start Page: 
End Page: 
Select license: 
Select License

One of the oldest and most studied ques- tions in development is whether the resource and environmental conditions faced by households in low-income rural areas significantly affect productivity.' The reason for this interest is clear: if productivity is significantly constrained by these factors, then the returns to investment in the form of human capital are likely to be high, and under certain conditions there may be scope for efficiency-improving policies and programs. Given this motivation it may at first seem surprising that most of the existing literature focusing on the health-productiv- ity relationship concentrates on the effects of nutrition. While nutrition is an important contributor to good health, other compo- nents of health, especially illness, may also influence worker productivity. Moreover, in contrast to the case of nutrition, over which households have direct control, illness is affected by exposure to pathogens, something over which individual households may have little control. This public aspect of illness makes an examination of the effects of illness on productivity particularly impor- tant from a policy perspective.

This distinction between the private nature of nutrition and the public nature of illness may at least in part be evident in Tables 1 and 2. These tables present the distribution of illness and calories, respec- tively, by wealth quartiles in three Asian countries. What is most striking about these tables is the fact that despite clear evidence that calories are importantly associated with wealth in at least two of the countries (Bangladesh and Pakistan) there is little sys- tematic relationship between illness and


of Economics, University of Pennsyl- vania, Philadelphia, PA 19104-6297.

'J. Behrman (1993) provides a review of much of the relevant recent literature with a particular focus on nutrition.

wealth in any of the countries. For example, in Bangladesh, the top quartile (in terms of wealth) of the population of men and women consumes 6 percent and 8 percent more calories, respectively, than the lowest quar- tile. By contrast, Bangladeshi males in the upper quartile experience 19-percent more days ill over the same period than do those in the lower quartile; the figures for the women are identical.

There are several possible reasons for this pattern. As suggested earlier, one possi- bility is that households have little control over their exposure to illness. If this is the case, then one might well conclude that the distributional effects of illness are quite lim- ited. A second plausible explanation is that illness is measured or reported differently for better-off households. For example, a better-off individual who is generally healthy may be more readily able to identify when he or she is ill than a poor individual with low caloric intake. Moreover, to the extent that illness is measured as an alternative to work (as is the case for the Pakistani data set), a wealthier individual may be more willing to skip work on a day that he or she is ill. If this latter point is true, then the distributional consequences of illness could be quite important despite what is evident in Table 1.'

In this paper I address these and other related issues using two of the longitudinal data sets from rural Asia from which these tables were derived. In the first section of

'A. Hill and M. Mamdani (1989) provide a good summary of the problems associated with measuring illness using survey data. J. Strauss et al. (1993) note that objective measures such as ability to walk a certain distance provide a useful alternative to measures of reported illness for assessing ill health. Unfortunately, existing data sets with this type of objective data do not have a sufficiently detailed longitudinal component to carry out an accurate assessment of productivity effects.


Country 1 2 3 4

Bangladesh Males


Philippines Males


Pakistan Males


Note: Standard deviations are reported in parentheses.


Country 1 2 3 4

Bangladesh Males


Philippines Males


Pakistan Adults

Note: Standard deviations are reported in parentheses.

the paper, I provide a theoretical frame- work that is used in the subsequent estima- tion. In Section 11, the two data sets are briefly described and the results are presented. The principal conclusion that emerges from this paper is that illness has important distributional consequences de- spite the apparent similarity of reported illness across economic strata.

I. Theoretical Framework

The theoretical framework for this analy- sis is provided by a multiperiod agricultural household model in which parents care about the consumption and health of their children and income is influenced by health.3 I assume that households maximize expected discounted utility:

where the second summation is over house- hold members, u(., .) is a single-period util- ity function, cis and zis are calories and illness, respectively, for individual i over the interval from s to s + 1. In each interval there is a budget constraint

where ps is the price of consumption goods, as is net dissavings, .rr, is period-specific agricultural profits, and ws(zs) is potential labor income when adult illness in the household is zs and the time endowment has been normalized to 1.Assets A, in each period are updated according to the follow- ing equation

where rs is the interest rate. It is assumed that health (body mass, hi,) and illness (zit) are produced according to the following function:

for J ={h, z} where E~~~ is an independently and identically distributed shock that may be correlated across individuals in the same village (as a consequence of epidemiological

3~ee,for example, Foster and Mark R. Rosenmeig (1994) for a more detailed example of the use of dynamic models for studying health in developing countries.

factors). Finally, to capture the idea that the severity of illness is lower for better-off households (for reasons of reporting or overall levels of health) it is assumed that reported illness Z; = cr(At)zit where crr(At)> 0.

11. Data and Results

The two data sets that are used in the subsequent analysis are from Bangladesh and the Philippines. The Bangladesh data are collected for the period 1977-1978 from 135 households from five villages in close proximity to one another. The data from the Philippines were collected in four rounds at four-month intervals in 1984-1985 as part of an International Food Policy Re- search Institute study by H. Bouis and L. Haddad (1990) from 448 households.

The first question I address is whether the effect of illness on body size differs for poor and better-off households in the sense of a body-size production function. Since this expression nets out calorie allocations and work as well as previous body size, the illness effect will only vary to the extent that the severity of reported illness (in terms of its effect on body size) differs between the two wealth groups. There are two reasons one might expect the severity to differ. First, as discussed above, it may be the case that better-off households overreport illness rel- ative to poor households. If this is the case, then one might expect the effect of illness on the better-off households to be somewhat less. Alternatively, it may be the case that poor nutritional status importantly in- fluences the severity of the disease, an ef- fect that could operate in either direction: individuals with poor nutritional status may be less able to fight off infection, or alterna- tively, they may simply have fewer excess reserves to deplete during times of illness.

The results are presented in Table 3. The estimates of the body-size production func- tion are reasonable. Individuals tend to lose weight when they are sick or when they are working a great deal, and they tend to gain weight when they eat more, although the estimates for the calorie effect are quite imprecise. Of particular interest are the co-




First Third All and and

house-second fourth

holds quartiles quartiles Variable (N = 2,273) (N = 1,333) (N = 940)

D~oi~~ s

-0.023 ~-o.206

(1.773) (2.590) (1.337) Lagged BMI 0.887 0.785 0.948

(31.149) (12.712) (32.644)

Log Calories 0.081 0.171 0.0307 (xlO-s) (1.21 1) (1.636) (0.809)

worked -0.058 -0.555 0.008

(0.718) (2.617) (0.099)

constant 2.318 4.263 1.118

(3.925) (3.377) (1.884)

F test: P 40.001 P < 0.001 P < 0.001

Notes: BMI = body mass index, defined as ~eight/(hei~ht)~. variables are

Endogenous instrumented using round X village-wealth interactions. Lagged BMI refers to a four-month interval. Numbers in parentheses are asymptotic t values.

efficients on illness. It is evident that an extended period of illness results in a sub- stantial reduction in body mass for the poorer households. Each additional day of illness yields a -0.206 reduction in the body-mass index. While this figure seems overly high, it should be noted that the fact that the coefficient on lagged body size is considerably less than 1 indicates that there is substantial catch-up growth. Therefore, the effect of an extended period of illness will be diminished: an individual with a body mass that is 1 kg/m2 lower will ceteris paribus grow 0.24 kg/m2 more than other- wise over the subsequent period. The coef- ficient for the better-off households is con- siderably smaller and indeed is not different from zero at traditional levels of significance.

The fact that body size is differentially affected by illness for poorer households does not necessarily imply that illness has important implications for household resources: body size may have little impact on productivity or insurance may operate to mitigate any adverse effects of low body size or illness on income. While in principle it is possible to measure these effects directly, it is instructive to take an indirect approach by examining the changes in a consumption variable that is likely to be closely associ- ated with resource availability to the house- hold. The basic idea is to make use of an equation relating consumption at different points in time. To avoid complications aris- ing from the fact that consumption may fluctuate in response to varying work activ- ity, I focus this component of the analysis on calorie consumption of young children. I estimate an equation of the following form:

+ 4fzfs+ 'is

where zmsand zfsare the illnesses of adult males and females, respectively, and test for

= 4,, = 0. As noted, the idea is to ask whether the consumption of a child tends to decline over a period in which the child's parents are ill and therefore receiving lower income. No effect of illness would be evi- dent if illness did not affect income or if households were perfectly insured against illness shocks. For this component of the analysis we make use of the Bangladeshi data for which the illness information is relatively complete.

The results are presented in Table 4. The effects of the illness of the father are quite striking: for poorer households, each addi- tional day of paternal illness in a given two-month period results in a 5.4-percent drop in calorie consumption for the child. Consistent with the notion that the better-off households have better access to smoothing mechanisms, the corresponding coefficient for these households is not significantly dif- ferent from zero, although the positive sign is surprising. It should be noted that this differential effect between the two strata may also reflect overreporting of illness on the part of better-off households. The ef- fects of mother's illness are not significant in either group, a result that is consistent with the fact that women in this conserva-





First Third
All and and
house- second fourth
holds quartiles quartiles
Variable (N=502) (N= 241) (N= 261)
Days ill (previous      
Log standardized      
Days father ill      
in interval      
Days mother ill      
in interval      

F test: P = 0.082 P = 0.016 P = 0.201

Notes: Endogenous variables are instrumented using initial state variables and round Xvillage-wealth inter- actions. The specification includes round dummies. Parental illness refers to two-month interval, while calories refer to current day and own illness during the previous week. Asymptotic t-ratios are reported in parentheses.

tive area of rural Bangladesh play a much less important role than do men as a source of income, because of restrictions on their mobility.

111. Conclusion

In this paper I have examined the distri- butional implications of illness through a stratified analysis of three data sets from rural Asia. While this analysis is far from comprehensive, a coherent and sensible pic- ture seems to emerge suggesting that illness has important effects for poor households. First, despite the similarity of reported lev- els of illness for poor and better-off house- holds, reported illness has a greater impact on body size for these households. Second, illness appears to reduce resource availabil- ity in the poor households in the sense that the calorie consumption of children declined more when their father was reported ill over the corresponding interval. It should be emphasized here that no attempt has

been made to distinguish between the hy- pothesis that reported illness reflects more severe conditions for the poor and the hy- pothesis that that illness is equally severe but the poor are less able to accommodate these shocks. The point is that regardless of which of these two hypotheses more accu- rately reflects reality, the implication of Ta- bles 3 and 4, coupled with the fact that reported illness levels are similar for the two populations, is that the poor are more affected by illness.


Behrman, J. "The Economic Rationale for Investing in Nutrition in Developing Countries." World Deuelopment , November 1993, 21(11), pp. 1749-71.

Bouis, H. and Haddad, L. Agriculture commer- cialization, nutrition and the rural poor: A study of Philippine farm households. Boulder, CO: Riener, 1990.

Foster, Andrew D. and Rosenmeig, Mark R. "A Test for Moral Hazard in the Labor Mar- ket: Effort, Health, and Calorie Consumption." Review of Economics and Statistics, 1994 (forthcoming).

Hill, A. and Mamdani, M. Operational guide- lines for measuring health through house- hold sumeys. London: Centre for Popula- tion Studies (School of Hygiene and Tropical Medicine, University of London), 1989.

Strauss, J.; Gertler, P.; Rahman, 0.and Fox, K. "Gender and Life-Cycle Differentials in the Patterns and Determinants of Adult Health." Journal of Human Resources, Fall 1993, 28(4), pp. 791-837.

  • Recommend Us