American Economic Association

American Economic Association
Health, Nutrition, and Economic Development

Author(s): John Strauss and Duncan Thomas
Source: Journal of Economic Literature, Vol. 36, No. 2 (Jun., 1998), pp. 766-817
Published by: American Economic Association
Stable URL: http://www.jstor.org/stable/2565122 .
Accessed: 03/12/2014 11:17
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to Journal
of Economic Literature.
http://www.jstor.org
This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

All use subject to JSTOR Terms and Conditions
Journal of EconomnicLiterature
Vol. XXXVI (June 1998) pp. 766-817
Health, Nutrition, and Economic

Development
JOHN STRAUSS
and
DUNCAN THOMAS'
1. Introduction opment. This paper reviews some of the

evidence.
Over the past 20 years, investment in There are many reasons why the
human resources has taken center stage relationship between health and labor
in the study of developing economies. A market outcomes in developing econo-
voluminous set of wage function esti- mies should be of special interest.
mates provides the basis for calculating First, there is a long tradition of
market returns to education for virtu- theoretical models of nutrition-based
ally every country in the world. Studies efficiency wages in the development
have also looked at the effects of literature. Harvey Leibenstein (1957)
schooling on nonmarket outcomes. hypothesized that, relative to poorly
Prominent among those outcomes is the nourished workers, those who consume
health of children and adults. Since more calories are more productive, and
health, like schooling, is a form of hu- that at very low levels of intake, better
man capital, one might expect it to also nutrition is associated with increasingly
be related to labor market success. That higher productivity. Such nonconcavi-
link has received much less attention in ties, which lie at the heart of efficiency
the empirical literature, although in wage models, have powerful implica-
recent years there have been substantial tions for the level and composition of
advances in our understanding of the employment. Employers have an incen-
complex interrelationships between tive to raise wages above the minimum
health, nutrition, and economic devel- supply price of labor, and to exclude
I Strauss: Michiganl State University; Thomas:
those in poorest health from the labor
RAND and UCLA. Acknowledgments: Financial market because they are too costly to
support from NICHD Grant PO1 HD28372, NIA hire. While many variants of the model
Grant PO1 AG08291 and the College of Business, have appeared in the theoretical litera-
Michigan State University is gratefully acknowl-
edged. The comments of John Pencavel, four ture, the models have been subjected to
anonymous referees, Harold Alderman, Partha little direct empirical scrutiny. We de-
Dasgupta, Robert Fogel, Elizabeth Frankenberg, scribe the empirical conditions that
Mark Gersovitz, Kathleen McGarry,Eileen Miech,
David Neumark and James P. Smith have been need to be satisfied if these models are
very helpful. true, and assess the case for the exis-
766

Strauss and Thomas: Health, Nutrition, and Economic Development 767
tence of health-based efficiency wages. problems, some of which stem from

We conclude the case is weak, at least early childhood, and the functional con-
in terms of evidence from modern de- sequences of ill health are likely to be
veloping economies. felt throughout the life cycle (and not
If, however, we were to focus only on primarily at older ages as in advanced
the specific nonlinearities that are im- societies). In addition, the structure of
plied by efficiency wage models, we employment in lower income economies
would miss a very rich set of insights is such that work often relies more
that emerge from studying health and heavily on strength and endurance and,
labor outcomes. There are good reasons therefore, on good health.
to expect improved' health to result in Knowledge of the nature and extent
improved functionality and produc- of links between health and labor mar-
tivity, which has important implications ket outcomes is also important for pol-
for an array of behaviors and invest- icy. The health sector accounts for a
ments. And yet this most basic question sizeable fraction of the public purse in
has received serious attention only re- most countries. If public investment in
cently. health infrastructure and interventions
This is all the more surprising in a yields benefits in terms of higher pro-
developing country context, since the ductivity and economic growth, then
marginal productivity of health is likely those benefits belong in evaluations of
to be higher in those settings relative to health programs.
higher-income industrialized economies. The possibility that the income-
Not only are levels of health far lower generating capacity of the poorest is
in developing countries, but the inci- enhanced more by some health sector
dence and nature of diseases tend to be investments relative to others raises
different. Notably, there is a much issues revolving around the distri-
higher prevalence of malnutrition and butional effects of policies. The litera-
infectious diseases (many of which are ture indicates that some health prob-
preventable) which often interact, caus- lems (such as malnutrition) can become
ing even more serious damage (World debilitating at extremes, and the
Bank 1993; Dean Jamison et al. 1993; biomedical evidence highlights the po-
Per Pinstrup-Andersen et al. 1993).2 tential importance of thresholds, below
These differences are important be- which poor health can have dire conse-
cause they result in an age distribution quences for functionality. This suggests
of ill health that is tilted toward infants that in a developing country context,
and preschool children, as manifest by the labor market consequences of poor
these groups comprising a much higher health are likely to be more serious for
fraction of total deaths than in higher- the poor, who are more likely to suffer
income settings (World Bank 1993). An from severe health problems and to be
important implication for our purposes working in jobs for which strength (and
is that adults in poorer economies are therefore good health) has a payoff.
more likely to be afflicted with health Empirical studies of the links be-
tween health and labor outcomes in the
2 For example, Alan Lopez (1993) estimates that context of developing countries have
45 percent of all deaths in developing economies provided more than documentation on
in 1985 can be attributed to infectious and para- the nature and extent of the relation-
sitic diseases such as diarrhea and malaria; these
diseases account for about 4.5 percent of deaths in ships. They have also proven to be fer-
industrial market economies. tile ground for addressing a broad set of

768 Journal of Economic Literature, Vol. XXXVI (June 1998)
questions that economists have been be invested in health. This would gener-
grappling with for many years. There is, ate potentially important feedbacks be-
for example, evidence that credit mar- tween health and income. Disentan-
ket constraints are binding for low- gling the direction of causality will
income households in several settings, clearly be key if we are to interpret be-
particularly in very poor, rural areas. haviors. We discuss these issues in de-
These constraints may result in health tail in Section 3.
investments being below efficient lev- Although there are many commonali-
els. Several recent studies have ex- ties between health and education, Sec-
ploited this insight to shed new light on tion 4 highlights two aspects of health
the operation of such constraints. Oth- that distinguish it from most other hu-
ers have used observations on health to man capital measures. First, health is
compare worker effort under different fundamentally multidimensional, and it
labor contracts and to draw inferences is important to differentiate among
about incentives underlying alternative these dimensions. Second, measure-
contractual forms. Interactions between ment of health is difficult, and in many
health and labor have also provided op- cases, measurement error is likely to be
portunities to peek inside the house- correlated with outcomes of interest
hold and examine the allocation of like income. We draw out the implica-
health-related inputs to different house- tions for interpreting empirical evi-
hold members, in light of differential dence on the links between health and
activity levels and returns to health labor market outcomes.
among members. Insights from this lit- Section 5 reviews the current state of
erature are highlighted below. the empirical evidence regarding the
Section 2 presents correlations be- relationship between health and labor
tween health and labor outcomes. We outcomes, taking into account both ex-
sh-ow that, in the United States, men perimental and nonexperimental studies.
who are taller and heavier (given In addition to setting out the facts re-
height) earn higher wages. We also garding a causal impact of health on la-
show that the magnitudes of the corre- bor outcomes, we discuss studies that have
lations in the United States are dwarfed addressed broader questions regarding
by the magnitudes in low-income set- the functioning of the economy, and we
tings like Brazil. This is followed by a review the evidence on efficiency wages.
discussion of both economic theory and We conclude by drawing some lessons
econometric issues involved in drawing and discussing gaps in the literature.
causal interpretations from the evi-
dence in Section 2. In contrast with 2. Correlations between Health and
schooling, health varies over the life Labor Outcomnes
course and is the outcome of behavioral
choices both during childhood and in Drawing on historical series on the
later life. If better health is associated stature of adults in the United States
with improved functionality, and there- and Europe, Robert Fogel (1992, 1994)
fore productivity gains, then individu- makes a compelling case for linking ag-
als, families and even society will invest gregate movements in adult height to
more income in health than would be long-run changes in standards of living,
implied by its value in purely utility including income, mortality, and possi-
terms. As productivity increases, so will bly morbidity. Having demonstrated,
income, and that additional income may for example, that the average American

A _ _ _ _ _ _C _ _ _ _ _ _
177 UNITED STATES 164 170 COTE D'IVOIRE

176- , --. 156 - 160
175-
- w~~~~
1625 ~~ 168- 159
w
78174-
1
-
09 ~~~~~~161 1 -
067 1 9 -15
173-
160---
-
172 - - 159 ~~~~~~~~~166-157
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ~~~~~~~~~156
1641910X '1920 1930 M.Mes
1940 1950 ~~~~163 -
15691920 1930 1940 1950 1960 1970 1
B -__
_ _ _ _
_15__ _ _ _ _
BRAZIL 1. AdultStatur VIET

NAMC153
168 BRAIL157 13- -- - 152
162
167 - --- 156 161- 151
-2~~~166
166 - - ~~~~~--- ~ ~ ~7 ~~160 ~~~150
-
~~~~~155~ 49
165 -154~ 5 158 ~-148~
Males 157 - 147
164 Feals153 156 ' 146
I I I I I
~~~~~
~~~1551 I 145
1910 1920 1930 1940 1950 1920 1930 1940 1950 1960 1970
Birth year
Figure 1. Adult Statureby BirthCohort
(or European) today is a good deal men (on the left) and women (on the
taller than his counterpart of two centu- right) are directly comparable within
ries ago, Fogel argues that stature is a each panel, but not across panels.
useful index of the well-being of a popu- Among these people, American males
lation, supplementing data on wages, in- are the tallest. The average American
comes, and economic activity. (See the
survey by Richard Steckel 1995.)
Following this argument, Figure 1 amination Survey (NHANES) 11 (1976-80); we re-
strict attention to only the native born (who are
presents evidence on the relationship taller than immigrants). There are 9,323 men and
between stature of adults and date of women in the sample. The Estudo Nacional da
birth for the United States and three Despesa Familiar (ENDEF) was conducted in
Brazil in 1974-75, and we have 23,107 adult re-
contemporary low-income populations: spondents in the sample. The Viet Nam Living
Brazil (in Panel B), the Cote d'lvoire Standards Survey (VLSS), which was collected be-
(Panel C) and Viet Nam (Panel D). tween October 1992 and October 1993, covers
4,800 households randomly drawn from the entire
Each panel covers a span of between 45 country; the sample included in this figure con-
and 55 years.3 Note that the scales for tains 10,235 adult respondents. The C6te d'Ivoire
Living Standards Study (CILSS) (1986) contains
3 All the figures are based on samples of prime- information on 3,744 native born adults. In the
age adults (ages 23 through 60) and present CILSS, most adult birth dates, or ages, are re-
LOWESS estimates (Cleveland 1979) of the rela- ported in complete years, and there is a good deal
tionship between height (measured by an anthro- of stacking on even digits. These age errors, along
pometrist) and exact birth date (or birth year in with the relatively small sample size, give us less
the C8te d'lvoire). The United States data are confidence in the estimated shapes for this survey,
drawn from the National Health and Nutrition Ex- relative to the other three.

male born in 1950 is nearly 177 cm tall; highlights a key advantage of using
his Ivorian counterpart is about 170 cm, height data as an indicator of social and
a Brazilian male is slightly shorter and a economic development over long peri-
Vietnamese male born that year is al- ods of time in low-income populations
most 15 cm shorter than the American. where data are scarce. Between 1925
It would be naive to attribute all of and 1955, heights of adult men and
these differences to differences in the women increased very rapidly (by be-
standards of living of these people when tween 1.6 and 1.8 cm each decade). But
they were children, since height also re- for birth cohorts since the early 1950s,
flects genotype influences. However, adult stature has remained unchanged.
the differences and the changes over The 25-year period 1950-75, and espe-
time within each country do provide a cially from 1965 to 1975, was obviously
wealth of information about health and a period of tremendous upheaval and
development over the long term. dislocation in Viet Nam. Charles
First, it is immediately clear from the Hirschman, Samuel Preston, and Vu
figure that there have been substantial Manh Loi (1995) estimate that approxi-
increases in attained height by maturity mately 1 million (North and South) Vi-
in all four populations during this cen- etnamese were killed during the intense
tury. Average growth rates lie between war period 1965-75. Panel D of Figure
.75 cm and 1.5 cm per decade. For ex- 1 indicates that during the entire 25
ample, an Ivorian man born in 1950 years of upheaval, the health of children
would be, on average, about 169 cm suffered. The figure also suggests that
tall, which is almost 3 cm taller than his those born toward the end of the period
father would have been had he been suffered the most (since their parents
born in 1930. were taller than earlier birth cohorts,
Second, underlying these average and child height is positively correlated
growth rates, there is considerable het- with parental height).4
erogeneity over time. In the United While the secular increases in heights
States, growth was very rapid for men are sizable in all four countries, the
and women until around the 1935 birth gaps in height between them are even
cohort; growth has substantially tapered larger. However, comparison of means
off among later cohorts. A roughly simi- hides the very large dispersion within
lar pattern emerges for Brazil, with each country. In order to isolate the
growth in heights of men and women cross-section variation, we focus on the
tracking each other closely and tapering 1950 birth cohort in each country.
off since World War II. The Cote d' Among men, the interquartile range of
Ivoire provides a stark contrast. There
has been no tapering off of growth 4 Our interpretation of the evidence will be
among post-War cohorts, and changes wrong if taller men were more likely to die during
in female height have not tracked the the war. There are three reasons why we do not
think this form of mortality selection fully explains
male profile very closely. The stature of the evidence. First, the same pattern is found for
adult males increased substantially and Vietnamese women, who were much less likely to
significantly faster than that of females be soldiers. Second, mortality selection would
have to be increasing over time (to explain the fact
until arounid 1945, when the gender gap that the effect on eight becomes cumulatively
reached its peak; since then, growth larger during the period). Third, wars typically
rates in stature have been roughly equal ravage those at the bottom of the socioeconomic
distribution, and we will present evidence below
for men and women. that this is a fair characterizationof the Viet Nam
The Vietnamese case is striking, and War, particularlyin the North.

TABLE 1
ANNUAL RATES OF GROWTH OF ADULT STATURE: VIETNAMESE MALES
Birth Cohort: 1925-55 1956-70

Birth Place: North South North South
10t1Percentile 0.251* 0.164 -0.086 -0.041
(0.03) (0.03) (0.05) (0.05)
Mean 0.189* 0.150 -0.008 -0.060
(0.02) (0.02) (0.03) (0.04)
90t11Percentile 0.134 0.129 0.039* -0.075
(0.03) (0.03) (0.05) (0.05)
Notes: Coefficients from piecewise-linear regressions of height (in cm) on exact birth date (measured in years) for
least squares (Mean) and quantile regressions (at 10th and 90th percentiles). Standard errors in parentheses.
Quantile regression standard errors calculated using bootstrap. * Denotes significant difference between North and
South at 5 percent level.
height is 9.2 cm in the United States, 50 percent faster than among the short-
slightly higher in Brazil and lower in est in the South, and nearly twice as
the Cote d'Ivoire and Viet Nam. The fast as among those in the top decile of
gap in heights of women is slightly height. This evidence suggests there
smaller both across countries and was improvement in the standard of liv-
within countries; for example, the inter- ing of all Vietnamese and a decline in
quartile range of female height of inequality in terms of height (and, per-
women born in 1950 in the U.S. is 6.9 haps, economic conditions). In the
cm. This within-country heterogeneity post-1955 era, however, growth in at-
does not solely reflect genetic variation, tained height stopped for the average
but is related to socioeconomic condi- Vietnamese male in both the North and
tions (John Strauss and Duncan Thomas South. But in the North, growth ap-
1995b). pears to have declined among the short-
The variation in changes in adult est while increasing slightly among
stature within a country yields insights those at the top of the height distri-
into how the benefits of growth have bution, resulting in an increase in
been distributed within the population. height inequality during the period.
Viet Namnserves as an example. Table 1 This evidence suggests that not only did
reports annual changes in stature growth falter, but in the North the
among men born in the North and worst off were hit hardest by the dis-
South of the country; the periods before ruption from the war period.
and after 1955 are also distinguished. Turning to data at the household
Among the 1925-55 cohort, adult stat- level, there is a good deal of direct evi-
ure increased 1.9 cm each decade in the dence indicating that income and health
poorer North, which is about 25 percent are correlated. Not only does the de-
faster than in the South. Moreover, the mand for health services typically rise
shortest (and the poorest) benefited with income, but there is abundant evi-
most during this period, especially in dence that the poorest are typically in
the North, where average height in- the worst health. (See Jere Behrman
creased by 2.5 cm each decade-over and Anil Deolalikar 1988, and Strauss

and Thomas 1995a for discussions.) For Height, which is but one indicator of
example, child height tends to rise with investments in human capital made dur-
income in many low-income societies, ing childhood, is also related to educa-
and well-nourished children are about tional attainment. Panel B of Figure 2
the same height throughout the world presents the correlation between adult
(except in South and East Asia where stature and years of schooling for two
well-nourished children are shorter but cohorts in the United States (Bi) and
height is rising rapidly across genera- Brazil (B2). Taller men tend to be bet-
tions). ter educated in both countries, although
The central question for this study is the correlation is substantially larger in
whether health and productivity are Brazil. Specifically, among 25 to 34 year
correlated at the individual level. Panel olds, a 10-cm gap in height is associ-
Al of Figure 2 presents nonparametric ated with an additional year of school-
estimates of the bivariate relationship ing in the United States and 1.5 years in
between height and (log) hourly wages Brazil; because levels of schooling are
of men in the United States and urban very different, this translates into an 8
Brazil.5 There is a powerful association percent increase in the U.S. and a 25
between height and wages in Brazil. percent increase in Brazil.
Taller mien earn more: a 1 percent in- Panel B of Figure 2 raises a legiti-
crease in height is associated with an al- mate question: Is height simply proxy-
nmost8 percent increase in wages. While ing for education in the wage function
this dwarfs the magnitude of the corre- in Panel Al? The answer is provided in
lation in the United States, even taller Panel A2, in which relationships be-
Ameiricani-men earn higher wages.6 This tween height and wages are presented
ranking probably reflects differences in separately for two groups of Brazilian
the extent of poor nutrition in the two men: those with no education and those
coun-tries as well as differences in the who have some schooling. While a good
natu're of work that is commonplace in part of the observed positive association
eaclh society, since manual labor-and between height and wages can indeed
thus reliance on physical strength-is be attributed to the role of education,
far more imuportant in Brazil.7 even for those with no education the
correlation persists and is large in mag-
5 Data on U.S. meni are drawn from the 1992 nitude. For them, a 1 percent increase
wave of National Longitudinal Survey of Youth for in height is associated with a 4 percent
Panel A of Figures 2 and 3. The figures are based
on wages in 1992 of 1,955 white males aged 27 to increase in wages. After controlling for
35 years. Data from NHANES II are used in Panel education, the elasticity remains large
B of Figure 2. (The NHANES data do not record and, it turns out, is largest (4.6) for men
wages but they do report height and education of
all cohorts.) The Brazilian data are from ENDEF. who completed secondary schooling.8
6 In aIn OLS ln(wage) regression, the coefficient
oni ln(heiglht) is 7.7 (stancdard error is 0.2) in Bra-
zil; the coefficienit is 1.0 (standard error is 0.4) in
tlhe Uniited States. increase over the life course. Empirically, it is the
7 Th1eIe aIe other reasons that could
eyplain the latter that dominates. Controlling for age (by in-
differenice. The U.S. sample is restricted to rela- cluding a dummy for each birth year), the correla-
tivelv Young meni (aged 26 to 35) whereas the Bra- tion between height and wages is slightly larger.
ziliain samiiple inicludes meni ages 25 thr-ough 60, In fact, restricting the sample to men aged 30 to
and so colhort differences may be confounding the 35, the elasticity rises from 7.7 to 8.3 in Brazil.
correlationi between lheight and wages in Brazil. 8 A similar pattern is evident in the NLSY. Con-
Th-ie direction is niot obvious. On one hand, as trolling for educational achievement, the effect of
demlonistrated above, adult stature has risen over log height on log wages of white males is reduced
time in Brazil, on- the otlher hand, wages tend to by about one-third (to 0.62 with a standard error

Al ______________Bi ________________
United States - - ------- 114 United States
1.25 - --2.25 Ag 53
A2 B2
z.75 -1.75~ 10 -
1).C SEu ,,2 4 Age 55-64
.5 - t71.5 ' a
.25 - Brazil -1.25
160 170 180 190 160 170 180 190
A2 _______________B2
I- Brazil 8 -Brazil
2
Some education W, E 1.5
1.25~~~
Age 25-34
05-
No educationAg5-6
160 1710 180 190 150 160 1710 180

Height CMS H-eiglit CNIS
Figuire 2. Wages, Education, anidHeight of Males in Braziland the United States
Heiglit may be a proxy not just for dex (BMI)9 varies over the life course
the quantity of schooling but also for its and thus may capture both longer- and
quality, another dimension of human shorter-run dimensions of nutritional
capital investments. While this concern status and health. Clearly, BMI is re-
does not apply to those who have no lated to energy intake, net of output; it
schooling, there may be other human has also been shown to be related to
capital investments, such as parental maximum oxygen uptake during physi-
time allocation, which are correlated cal work (VO2 max), which is, in turn,
with height and underlie the positive related to aerobic capacity and endur-
correlation between height and wages. ance, independent of energy intake
It is important, therefore, when inter- (G.B. Spurr 1983, 1988; Reynaldo Mar-
preting these correlations, to recognize torell and Guillermo Arroyave 1988).
that they may reflect both rewards to Whether this is an important pathway
human capital investments early in life through which health may influence
and strength or robustness as an adult. productivity is not obvious, since mnany
In contrast with height, body mass in- jobs do not require sustained physical
effort. Moreover, energy can be stored
of 0.34) and the correlation between height and
wages is largest for those white men who have 9 Weight (in kilograms) divided by h-teight(in
some college education. meters) squared.

774 [ournal of Economic Literature, Vol. XXXVI (June 1998)
in the body and expended when A

needed, which implies that BMI may be 1.25 - - - 2.25 ,
affected by contemporaneous move- N
ments in incomes or prices, so that the :4 1- United States 2 i
dynamics linking them, BMI and pro-

.75- 1.75
ductivity, may be very complicated. X *=
Furthermore, current BMI partly re- = .5 1.5 a

flects previous health and human capi-
tal investments, and so a correlation be- .25- Brazil 1.25
tween BMI and productivity may be
capturing the influence of those prior 18 20 22 24 26 28 30 32 34
investments. BMI kg/,,i2
With these caveats in mind, Figure B
3A presents the relationship between Brazil -- - 1.5 a.0
wages and BMI for United States and - a
*M .75 - Some education , '
Brazilian males. In the United States, 1.25a
the function is an inverted U with opti-
mal BMI (in terms of maximizing - 2 S 1
wages) being around 24. (This translates a 25 - a
into a weight of 177 lbs for a man who
is 6 ft. tall.) The magnitude of the dif-
a
No educationa
~~~~~~~~~~~~~.75
0
ferences in wages across the BMI distri-
I
bution in the United States is dwarfed l
18
l
20 22
l
24
l
26
l l l
28 30 32 34
l
by the magnitude in Brazil, where the BMI kg/rn,2

shape is also quite different. Among
men whose BMI is less than 27, wages Figutre3. Wages and BMI in Braziland the United States
rise dramatically as BMI increases, par-
ticularly for those above 22. Wages are among men with no education, elevated
essentially unrelated to BMI for the 13 BMI is associated with greater physical
percent of men whose BMI is above 27. strength, which is of value for manual
It is instructive to examine the wage- labor, but that strength is of less value
BMI curve within education groups among the better educated (who are
(Fig. 3B). In contrast with height, only more likely to have sedentary occupa-
a small part of the correlation with BMI tions).
appears to be capturing the influence of Figures 2 and 3 include only working
education, further suggesting that nutri- males. But labor force participation
tional status and wages are related over choices may also be related to health.
and above other human capital invest- The relationship between the fraction
ments. The wage-BMI curve flattens at of urban men in Brazil who are not
low BMI only for the poorest (those working and their nutritional status is
with no education), suggesting that reported in Figure 4. It is clear that
poor nutrition (or poor health) takes its shorter men not only earn less, they are
heaviest toll on the most vulnerable. also less likely to be working. Over 10
The positive correlation between wages percent of men who are 154 cm tall
and BMI persists for those with no edu- were not working at the date of the sur-
cation even at high levels of BMI, but is vey, but among those who are about 167
zero when BMI exceeds 25 among the cm tall, the fraction is only 5 percent.
better educated. It is plausible that For men taller than 167 cm, there is a

10-o Height
ured by the fraction of calories from
protein sources). The preponderance of
9- evidence suggests that health and labor
outcomes are, indeed, correlated among
0o -\ men in Brazil; similar conclusions
67- emerge for women (Thomas and Strauss
1997).
There are, however, very good rea-
5-
sons to be cautious about giving these
4 correlations a causal interpretation.
I l l l l l I
155 160 165 170 175 180 185 Thus, before discussing what has been
learned about the impact of health on
10 - BMI Height cms
productivity in the empirical literature,
9- we outline a simple model in the next
section to guide our assessment of that
7t - literature and discuss issues that arise
in interpreting the evidence.
6-
3. EconomnicFramework
5-
It is useful to begin with a simple
4-
. I I l l l l I household production model in order to
18 20 22 24 26 28 30 highlight the main issues in the litera-
BMI kg/m2 ture (Gary Becker 1965); several exten-
sions are discussed below. Assume a
Figure 4. Height, BMI, and Percent Not Working single-person household that maximizes
(Males in Urban Brazil) utility over one period. Let worker pro-
ductivity be perfectly observed by firms
and assume the labor market is com-
slight decrease in the probability of not
petitive.
working as height increases, but the
Our central focus is on the link be-
slope is not significantly different from
tween labor outcomes and health. It is
zero.10 A similar pattern emerges for
important to distinguish health out-
BMI: the probability that a man is not
working decreases with BMI until
comes, such as height, body mass, dis-
ease incidence or physical functioning,
around 24, at which point it is essen-
from health inputs that might include
tially flat.
nutrient intakes, exercise, smoking, and
Other dimensions of nutritional
utilization of preventive or curative
status are also correlated with labor
health care.11 Following the seminal
outcomes in the Brazilian data. Wages
work of Michael Grossman (1972), let
rise as calorie intakes increase (until
there be a generic health production
around 2400 calories per day) and as
function for an individual:
protein intakes increase. In fact, wages
rise as diet quality increases (as meas- 11 The distinction is not always clean-cut since
some outputs may also be intermediate inputs into
10 Looking only at men below 167 cm in height, other outputs, such as disease incidence for body
the slope of the participation-height function is mass and vice-versa. We abstract from these com-
steepest among men with no schooling; a centime- plications here. Note also that some health outputs
ter increase in height is associated with a .5 per- may be jointly produced, while others may be pro-
cent decline in the probability of not working. duced solely.

H = H(N,L; A,B',D,t,eh) (1) puts (such as body mass) may affect

wNhereH represents an array of measured wages through better physical or mental
hiealthl outcomes. They depend on a vec- health and through strength and endur-
tor of health inputs, N, and labor supply, ance. Local community infrastructure, I,
L, both of which are under the control of such as electrification or road density,
the individual. We assume that actual may be related to labor demand or to
health is increasing in inputs, except for various work characteristics that are val-
labor sulpply, which consumes energy ued separately from wages. Wages will
and may in other ways be taxing on also be influenced by unobserved fac-
hiealth. The technology underlying the tors, oc, such as ability or school, quality.
hiealth production function is likely to Random fluctuations in wages-and
over the life course and, possibly, measurement error-are captured in
vary
12
with other socio-demographic charac- eW'
teristics, A, such as gender; in addition Discussions of how to interpret esti-

to family7background, B', such as paren- mates from wage or earnings function
tal health, and environmental factors, in- like (2) have long been central in the
clhiding local public health infrastructure labor economics literature (see, for ex-
aLld treatment practices, as well as the ample, Zvi Griliches 1977, and Orley
disease environment, D. Ashenfelter and Alan Krueger 1994).
Fromii an econometric point of view, The most common and well-known ar-
there are two sources of unobservables gument is that estimated effects of edu-
in the health production function. First, cation are biased because of omitted
the inhlerent hiealthiness of the individ- variables, such as unobserved individual
uti, ~t, is generally not observed in so- ability, which are correlated both with
cioeconomlic data, but may be known, at years of schooling and with wages. Pre-
least in part, to the individual by adult- cisely the same problem arises with
hlood. Second, there is a part which is measures of health, even time-invariant
unknown to both the individual and the ones such as height. There may be (we
econioimietrician, ei,, which includes mea- would argue that there are likely to be)
si-remeent error. For now, we will be ag- omitted factors in the wage regression
nostic regarding the sources of unob- that are correlated, both with wages and
seriables, and return to them in detail with health factors. For example, school
1)be1ow. quality is typically unobserved in survey
Assumie that an individual's real wage data. If higher quality education is asso-
is equal to his marginal product, which ciated with both higher wages and bet-
is costlessly observed by the employer. ter allocative decisions by the individual
A person's (log) real wage, w, varies during adulthood (or by the parents
w7ithblhealth outputs, H, as well as other during childhood) with regard to health
traits including schooling human capi- inputs and behaviors, then estimated
tal ?S, aind family background, B (which effects of health, as well as schooling,
now includes, for example, parents'
12 An alternative model assumes wages are paid
scloolil-lg):
by the task, again costlessly observed, with piece-
it' = tv(H; A,S,B,I,a,e,). (2) rate wages being competitively determined. Wages
per unit time can be thought of as a piece-rate
ApCartf'roml-the inclusion of health, this is wage, wP, multiplied by a function, y (H, S, B, I, oc)
a standard earnings function that is the that relates an hour of clock time to efficiency
bread and btitter of labor economics. As units. This efficiency units (production) function
depends on the same factors as the Mincerian
discussed below in Section 4, health out- wage function.

Strauss and Th-omas:Health, Nutrition, and Economic Development 777
will be upward biased. This point was production function (1). Suppose that
made in the discussion of Figure 2 the individual works for a wage, w, and
above. that asset or nonlabor income is V. The
budget constraint is
3.1 Feedbacks between Health and
Incomne pcC* + pnN = wL + V (4)
where, for expositional simplicity, con-
For the case of health, there are two sumption, C, has been decomposed into
additional substantive issues that are two elements, a vector of health inputs,
likely to be of considerable import: di- N, with prices p, and nonhealth con-
rect contemporaneous feedback, or si- sumption, C* with prices Pc In practice,
multaneity, discussed in this subsection, the division between these groups is not
and measurement of health, which is always clear, and not all health inputs
discussed in the next section. Feed- may be valued directly in the utilitv
backs between current health and in- function.
come will arise, for example, if better Letting X be the marginal utility of
health results in higher productivity, income, and assuming interior solu-
more hours at work, or a higher prob- tions, we see several points emerge
ability of working, and if, in turn, higher from the first-order condition with
incomes are invested in health by respect to usage of the jth health in-
spending more on health-augmenting put:13
inputs. The worker may either buy
more inputs (purchase more calories, au_aH ( aFwaH]
aH j= pn-L aHaNj (5)
for example) or buy higher-quality in-
puts (such as better health care); there
First, if health inputs raise wages
is evidence suggesting investments are
made on both the quality and quantity through improving health outcomes,
then the shadow price of health-aug-
margin, especially among the very poor.
menting inputs declines, inducing
Investments in training provide a direct
greater use of those inputs. Second, the
analogy in the education literature.
degree of decline of the health input
Closing the household production
shadow prices is likely to be greater for
model by allowing for behavioral
those in worse health. Biomedical evi-
choices helps clarify the role that unob-
dence suggests that, in some instances,
served heterogeneity plays and suggests
links between health inputs and outputs
some empirical methods for dealing
are nonlinear. For example, it has been
with it. Assume that an individual's (or
a household's ) welfare depends on la- argued that higher calorie intakes have
no beneficial effect above some thresh-
bor supply, L, and consumption of pur-
old but are associated with improved
chased goods, C, (some of which, such
health below the threshold. If calories
as foods, may be health inputs). Utility
affect productivity in a similar way,
may be conditioned on health outputs,
then the shadow price of calories will
H, observable characteristics such as
decline among those people whose in-
schooling, S, and family background
takes are below the threshold, enhanc-
measures, B, as well as unobserved
ing their incentive to consume addi-
characteristics including tastes, t.
tional calories. This has important
U = U(C,L,H;S, A,B,E,) (3)
13 An input,
Allocations are confined by budget and NM, such as calorie intake mav also
have a direct effect on utility; this effect is ignored
time constraints in addition to the health in (5).

distributional consequences if, as is puts, such as health care in the public

generally thought, the poor are most sector, are heavily subsidized and can
likely to have low intakes. be purchased at either zero cost or for
The second point is apparent from very low (monetary) prices. Thus, finan-
the reduced form demand function for cial outlays may be a small part of the
health inputj: full cost of health care; the dominant
Nj = Nj(pn,pc, A,S,B,V,D,I,[t,g,) (6) cost is often travel time to service
providers and waiting time once there
which depends on health input prices,
(Jan Acton 1975). Potential instruments
pn; consumption prices, Pc; demographic therefore include time costs and their
characteristics, A; schooling human capi-
proxies such as distance to providers or
tal, S; family background, B; nonlabor in-
the availability of health services in the
come, V; the disease and health environ-
community. Along the same lines, the
ment, D; and underlying nonhealth
quality of health services may also be
determinants of wages, L. All of these
valid instruments for health. The instru-
are observable. Unobservables also affect
ments, D, may include the local disease
demand for inputs. These include tastes,
environment and health infrastructure
4, innate healthiness, g, and unob-
such as water quality or sanitation ser-
servables such as ability, oc,which affect
vices. One might expect measures of
wages.14
underlying community disease inci-
The presence of cc in both the wage
dence to be potential instruments. For
function (2) and the health input de-
example, in tropical areas, rainfall is
mand function (6) captures the simulta-
often related to the incidence and se-
neity problem that is central to the dif-
verity of malarial diseases. However,
ficulty in disentangling the causal
rainfall (or the lack of it) may also af-
effects of health on productivity. A
fect labor demand, and thus wages, in
comparison of the functions does sug-
which case it is not a valid identifying
gest instrumental variables. Conditional
instrument. More generally, if instru-
on health outcomes, H, and levels of lo-
mental variable (IV) estimates of the ef-
cal infrastructure, I, productivity should
fect of health on wages are going to be
not be affected by health input prices,
consistent, it is key that local labor de-
p, or the disease environment, D. Both mand conditions and infrastructure, I,
sets of characteristics are, therefore,
be included in the wage function. If
potential instruments for health in wage
they are not, and if they are correlated
functions (2).
with health prices or health infrastruc-
Studies have used food prices as in-
ture, which seems plausible, then the
struments for nutritional inputs, such as
instruments will be correlated with the
calorie intakes, and health outcomes,
unobservables, a, in the wage function
such as body mass (Strauss 1986; David
and will be invalid.16
Sahn and Harold Alderman 1988).15 In
There are, however, costs associated
many poor countries, some health in-
with using instruments measured at the
14 There is generally no reason to expect that community level. First, they may be
measurement error in wages, e, will be correlated
with health input demand, thus it does not appear
in (6). 16 It is important to recognize that the appropri-
15 Because the cost of living varies across space ate instruments using prices or infrastructure are
and time, it is important to use real wages (or con- those available in the community and not those
trol for cost of living in the wage function) so that actually paid or used, since the latter may be cho-
it is relative health prices that provide identifica- sen by the individual and hence are correlated
tion in the model. with a.

only weakly correlated with individual Comparing (2) with (6) indicates that
health input utilization or health out- the value of assets, such as land owned
comes, which can result in biased infer- or nonlabor income, V, are also poten-
ences in small samples, the bias being tial instruments for health. However,
in the same direction as OLS (Charles assets may not perform well empiri-
Nelson and Richard Startz 1990; cally, since in many surveys a large
Douglas Staiger and James Stock fraction of households reports having
1997).17 Second, several studies have none, and these households tend to be
argued that local infrastructure may be the poorest. Furthermore, as we shall
endogenous. This could occur for two see below, there are theoretical reasolns
reasons: individuals may choose their to be wary of relying on this restriction
location partly based on public health for identification. First, assets and
infrastructure (Mark Rosenzweig and wealth may be systematically correlated
Kenneth Wolpin 1988) or the infra- with measurement error in health.
structure itself may be placed selec- Second, in a dynamic setting, the evolu-
tively by public policy, perhaps in tion of assets is a choice and becomes
response to local health conditions endogenous. To the extent that assets
(Rosenzweig and Wolpin 1986). Selec- reflect saving from prior earnings, it
tive migration may not be a major issue is likely that they will be correlated
in this context, because migration with unobservables in the wage func-
would have to be correlated with unob- tion.
served factors that are correlated with
3.2 Effects of Health on Labor Supply
health in a location, such as the avail-
ability of a clinic, over and above other Our primary focus thus far has been
measures of infrastructure included in on the relationship between health and
the wage function. This does not seem market wages; we turn now to examine
to be the most likely cause of migration the impact of health on labor supply,
in poor economies, compared to the which includes the decision to work,
pull of wage differentials, for example. sectoral choice, and hours of work. In
Selective program placement, however, low-income economies, analyzing time
may be a more serious issue for health allocation choices between self-employ-
programs (Mark Pitt, Rosenzweig, and ment and market wage work and self-
Donna Gibbons 1993; Elizabeth employment productivity may be of spe--
Frankenberg 1995). One strategy for cial interest, since self-employment is
addressing this concern may be to overwhelmingly dominant, particularly
model the process underlying public in rural areas.
investments in infrastructure, although For households without productive
to date there has been little empirical (nonlabor) assets such as land, labor
experience with this strategy (but force participation decisions are made
see Timothy Besley and Anne Case in the usual way, by comparing the mar-
1995). ginal rate of substitution (MRS) be-
tween goods consumption and leisure
17This problem may be alleviated somewhat if when no work is performed to the po-
"price" elasticities vary with individual charac-
teristics, in which case interactions between those tential market wage and choosing to
characteristics and "price" proxies may yield im- work if the wage is greater. Both the
proved predictions in the first stage regression. market wage and the marginal rate of
See, for example, Paul Gertler and Jacques van
der Gaag 1990, who argue that demand for health substitution depend on health out-
care price elasticities are greatest for the poorest.) comes, provided that health affects pro-

ductivity and that health is also valued ences from the productivity-enhancing
directly. Better health may raise the effects of health. Estimation of the
MRS between goods and leisure if equation, however, poses the same
health and leisure are complements. problems of endogeneity and measure-
Thus, even if health raises market pro- ment discussed with regard to market
ductivity, it is an empirical question wage equations. The prices of health in-
whether better health raises labor force puts that do not appear in the utility
participation rates on the margin. function, health infrastructure, and the
To see how health affects labor sup- disease environment do not belong in
ply, it is useful to define a "structural" the structural equation. They are poten-
labor supply function that is conditional tial identifying instruments and are sub-
on health outcomes and wages: ject to all the caveats raised above (Pitt
and Rosenzweig 1986).
L = L(H,pc,w(H;S, A,B,I,5c,ew),S,A,B,V,4) (7) Thus far, the discussion of identifica-
Health, like schooling, will affect labor tion of health effects has focused exclu-
supply by influencing offered wages, sively on analyses conducted at the level
with resulting substitution and income of a single individual. In many contexts,
effects, and also by affecting the mar- studies examine data on households
ginal rate of substitution between goods with multiple members, and identifica-
and leisure. This latter effect, which is tion is substantially more complicated.
independent of any health effects on The essential nature of the problem is
wages, results from our assumption that that, in general, prices of health inputs
health is also directly valued (appears in will not vary across family members,
the utility function). but levels of health outcomes or inputs
The reduced form market labor par- will (Pitt and Rosenzweig 1990; Pitt
ticipation and labor supply equations 1995). We can see the issue clearly by
that would be derived from substituting considering labor supply of an individ-
for w and H in (7) will depend on the ual conditional on health, H. It is not
same factors that appear in the health just the individual's own health that will
input demands (6). These include affect time allocation; the health status
health input prices, pn, and the local of all other household members will
disease and public health environnAent, also, and instruments are then needed
D. Since these factors will appear in the for each element of this vector H.
reduced form if there is a direct effect A natural solution might be to use indi-
of health operating through preferences vidual-specific health input prices or
on the MRS between goods and leisure, health infrastructure. But market prices
or if health affects wages, which in turn of purchased inputs and infrastruc-
affect labor supply, the reduced form ture-such as the distance to facili-
cannot provide an unambiguous deter- ties-do not typically vary among indi-
mination of whether health does influ- viduals within a household, and so it is
ence productivity. It can only give us an easy to imagine quickly running out of
assessment of the (total) effect of instruments, particularly in large, com-
health prices and the health environ- plex households as is the case in many
ment on labor supply. low-income settings. Including the
In contrast, estimation of the "struc- health of only a subset of household
tural" equation (7) (and its analogue for members in the labor supply function
participation) provides an opportunity will not solve the problem, since health
to identify separately the role of prefer- prices will then belong in the function

(except under very strong separability pared to how it affects the rate of sub-
assumptions). 18 stitution between leisure and income.
Both total and market labor supply
3.3 Effects of Health on Sectoral Time functions can be defined that are condi-
Allocation and on Self-Employment tional on health and wages, similar to (7);
Input Use, Outputs, Costs and they will also depend on self-employ-
Returns ment input and output prices and tech-
nology factors. Identification of health
In order to model time allocation be-
in these functions is no different from the
tween self-employment and market la-
case where there is no self-employment.
bor, and the effect of health on self-
Time allocated to self-employment
enmployment productivity, it is useful to
should be viewed as being derived from
extend the discussion and include home
demand for labor in self-employment.
production of a good that is both con-
Thus, modeling health effects on labor
sumed and sold by the individual. With
allocated to self-employment is analo-
the addition of self-employment, a dis-
gous to modeling health effects on self-
tinction needs to be drawn between
employment output, costs, or profits.
market and total (including both market
There are two alternative strategies: us-
and self-employment) labor supply. The
ing primal or dual functions. One could
individual will work exclusively in the
estimate either production functions or
market sector if the market wage is at
input demand, cost or profit func-
least as great as the marginal rate of
tions.19 Production functions have the
substitution (MRS) between goods and
obvious advantage of being able to esti-
leisure and the value marginal product
mate the production effect directly, al-
(VMP) of labor in self-employment,
though not allocative efficiency effects,
both evaluated at zero hours of work.
if any. As will become clear, production
Exclusive self-employment will be cho-
effects may exist without there being
sen if the value marginal product is at
any consequent effects on input de-
least. as great as the MRS between
mands, costs, or profits.
goods and leisure, at zero hours of
We assume, first, that hired and fam-
work, and greater than the market wage
ily labor are imperfect substitutes, per-
at the point at which labor's VMP
haps because of greater moral hazard
equals the MRS. Finally, the individual
for hired labor. If the individual spends
may choose to work in both the market
time working in both the market and
and self-employment sectors, in which
self-employment sectors, and if health
case there will be equality of the market
does not affect labor efficiency, the
wage, MRS and VMP in equilibrium. The
household's optimization problem is
impact of improved health on time allo-
recursive (or separable) in production
cation will clearly depend on how health
and consumption (Inderjit Singh, Lyn
differentially affects productivity in the
Squire, and Strauss 1986).20 That is,
market and self-employment work, com-
the household acts as though its deci-
18 Incorporating multiple household members in
a model of health and wages does not coinplicate 19 See, for example, Strauss (1986), Pitt and
estimation if the wage of one member, con itional Rosenzweig (1986), Deolalikar (1988), John Antle
on own health, is not affected by the health of and Prabhu Pingali (1994); and Behrman, Andrew
another household member. This rules out exter- Foster, and Rosenzweig (1997).
nalities (such as the ill health of one member be- 20 It is also necessary for co petitive markets to
ing transmitted to another) and assumes wages do exist for other commodities thiat are both pro-
not vary with hours (if the ill health of one mem- duced and consumed (e.g., foods) and that no cor-
ber alters time allocation of others). ner solutions are chosen for them.

sions can be divided into two steps: the illness. Hence health has no impact
first, maximize self-employment profits on inputs, costs, or profits (Pitt and
subject to the available technology, and Rosenzweig 1986). What does change, if
second, maximize utility subject to the a family member falls ill, is the house-
budget constraint augmented by the hold's full income, which declines by
value of profits. the value of family labor time lost.
In this case, reduced form solutions However, if it is optimal to allocate
for self-employment family labor (as no time to market work, the model is
well as other input demands, output not recursive and health input prices
supplies, cost and profits) depend only will affect reduced form self-employ-
on prices of self-employment outputs ment profit, cost, and input demand
and inputs, on quasi-fixed production functions. In this case, there is a
factors (including human capital), and shadow wage for family labor that de-
on any relevant community infrastruc- pends on all prices and infrastructure,
ture (Dwayne Benjamin 1992). Taking as well as on general taste parameters,
the profit function as an example: 4, in the utility function (3). Assuming
that the shadow family labor wage is not
It = l(Pf,pm,Wh,wf,I,F,S,O,en) (8) observed, or cannot be solved analyti-
where pf represents price of the output cally, then reduced form self-employ-
(say food); pm' the price of nonlabor in- ment profit, cost, and input demands
puts; W1h,the wage of hired labor; twf, the will be a function of all the determi-
wage of family labor; I, local infrastruc- nants of the shadow family wage, in-
ture; F, fixed factors, if any; 0, unob- cluding health input prices and the dis-
served factors affecting the production ease environment. Profit functions that
of the self-employment output; and e7, are conditional on health and wages are
measurement error in profits. Note that similar in structure to those for labor
these functions do not depend on health supply (7), and do not depend on health
input prices (except insofar as they are input prices, or on the disease environ-
also prices of self-employment inputs or ment. Thus, these latter variables are
outputs, such as food prices), nor can potential identifying instruments. As is
functions conditional on health be derived. the case for labor supply, discussed
Even if health does affect labor pro- above, if health enters the utility func-
ductivity, production and consumption tion, the effects of health prices in re-
decisions may still be recursive, and duced forms, or health outcomes in
dual functions equivalenit to (8) will be conditional functions, do not allow any-
valid, provided wages are paid per effi- thing to be necessarily inferred about
ciency unit. This would be the case if the effects of health on productivity.
productivity were observable and wages Even in this case, however, the possi-
were paid on a piece-rate basis. (Effi- bility that profits and other dual func-
ciency units of labor may, of course, de- tions are influenced by health is depen-
pend on health as well as other human dent on model assumptions. As pointed
capital factors.) Intuitively, with a mar- out by Pitt and Rosenzweig (1986), if
ket for effective labor, if a family mem- the model is changed so that effective
ber falls ill, time can be reduced from family and hired labor are perfect sub-
market work. As long as market work is stitutes in the production technology,
not reduced to zero, the opportunity then even if no family labor is sold on
cost of family labor remains the market the wage labor market, health will not
piece-rate wage, which is unchanged by affect self-employment inputs, outputs,

and profits. This follows since the op- are often thought to be at particular
portunity cost of effective family time is risk of deprivation in health-input allo-
the market piece-rate wage, which is in- cations. Economists have tended to at-
variant to any individual's health. tribute different investments to differ-
ential economic returns (Rosenzweig
3.4 Context of a Household and T. Paul Schultz 1982). For example,
if households maximize income and pay
As noted above, if we move from
no attention to consumption levels of
thinking about individual behavior to
individuals, then better-endowed mem-
household choices, the analytics are
bers would be allocated more inputs if
more complicated, particularly in the
the returns to those inputs in terms of
case of labor supply. Because of its im-
their wages were greater than returns to
portance, we return to the general issue
increased levels of health inputs on the
of treating households as collections of
wages of the less well-endowed in the
individuals. Although this raises a rich
household (see Pitt, Rosenzweig, and
set of questions regarding the allocation
Md. Nazmul Hassan 1990). Under more
of consumption across different house-
general models of household welfare
hold members or member types, it
maximization, other considerations
comes at a price since it is also neces-
come into play, including the weights
sary to grapple with the question of how
associated with the well-being of each
allocations are made within households.
individual and aversion to unequal out-
The nutrition-wage literature sug-
comes. In that case, health investments
gests a good, simple example. Suppose
will depend on health endowments
there is a minimum health threshold be-
(Behrmnan1988).
low which a person cannot find work
Differences in health input alloca-
(but can stay alive). Consider an ex-
tions among household members may
tremely poor household whose income
also reflect different preferences and
is so low that it would always be insuffi-
degrees of influence over decision mak-
cient to allow everyone to consume at
ing between household members
this threshold level, even when all con-
(Nancy Folbre 1984).22 This suggests a
sumption is devoted to health inputs.
model that falls outside the domain of
Even if everyone in the household has
the usual assumption in the economics
the same endowment, it will be optimal
of the family that the household may be
to have unequal allocations among
treated as a unit with a set of well-be-
members so that at least one person can
haved preferences. A discussion of al-
earn an income.21
ternative models of the household that
A large literature has now emerged
allows heterogeneity in preferences of
regarding differential (and possibly un-
members would take us far afield and
equal, in some sense) allocations of
well outside the scope of this paper.
health outcomes across different types
What is important here, however, is to
of individuals: men and women, adults
note that theoretical work that explores
and children (see Behrman 1992 and
applications to the household setting of
1997 for good summaries). Young chil-
cooperative and noncooperative game
dren and pregnant and lactating women
21 The threshold is not critical; the argument 22 Additional evidence in support of this inter-
carries through if returns to health are nonconcave pretation is provided by, for example, the finding
so that they increase at very low levels of health that income in the hands of a mother has a bigger
and decrease at higher levels. See James Mirrlees impact on the health of her children than income
(1975) and Joseph Stiglitz (1976). in the hands of a father (Thomas 1990).

theory (Marjorie McElroy and Jean come settings is an interesting but rela-
Horney 1981; Shelly Lundberg and tively unexplored issue. One might ex-
Robert Pollak 1993) or that examines pect health considerations to affect
the implications of Pareto-efficiency in consumption-smoothing strategies in
household allocations (Pierre-Andre low-income economies because the lev-
Chiappori 1988, 1992) has not yet been els of income and health risk are so
adopted in a sophisticated way in this high. Further, because consumption-
literature. There are several obvious im- smoothing mechanisms are often less
plications of these models. For exam- adequate for the very poor, the poten-
ple, investments in health will depend tial consequences of income shortfalls
not only on household resources but on the poor can be grave (see Partha
also on the distribution of those re- Dasgupta 1993, and Robert Townsend
sources. Moreover, a better under- 1995, for reviews).
standing of the determinants of power As an example of the potential mech-
within the household has the potential anisms, consider a simple two-period,
of yielding useful insights into possible perfect foresight model, in which future
instruments for individual health status productivity depends on consumption
in labor outcome functions of individu- today. It can be shown (Mark Gersovitz
als in multiple-member households. 1983) that very-low-income persons may
have little incentive to invest in finan-
3.5 Dynamic Issues
cial savings, because the loss of future
The discussion so far has been in wages due to lower labor efficiency may
terms of static models and cross-sec- more than offset any returns to savings.
tional data. Yet as Grossman (1972) This generates financial saving rates
pointed out, health has critical dynamic that depend positively on income. One
components, which provide additional can build on this, allowing current con-
important insights. After reviewing sumption to affect survival prob-
some of the theoretical issues, we dis- abilities, which will also provide incen-
cuss empirical difficulties and opportu- tives for the poor to consume more
nities that arise in this context. today, rather than save (Gersovitz
Health evolves over the life course. 1983), and to distinguish consumption
While some health indicators are fixed from health investments so that port-
during adulthood (such as height), most folio choice between investments in
change over time and thus contain both health and in financial assets can be
a stock and flow component. These considered (Gerhard Glomm and Mi-
flows, or changes in health, may reflect chael Palumbo 1993).23
investments in health or they may be
the result of unexpected shocks; thus, 23 Glomm and Palumbo have a more compli-
some changes will be anticipated cated model in which they build a dynamic, but
whereas others will not. deterministic, health production function and al-
low for fluctuating but exogenous income. They
In this context, health can be thought argue that if income is fixed over the life course, it
of as a durable good: investments now is optimal for poor persons to diversify wealth into
can reap benefits in the future. These health and to increase consumption early in life
(relative to a simple life-cycle model) in order to
investments in health-augmenting indo so. With income uncertainty added, it is still
puts can, therefore, be thought of as an optimal to invest in health early in life, but finan-
alternative form of savings or as a con- cial savings become far more important as a
method of smoothing consumption later in life,
sumption-smoothing device. Whether and thus reducing the probability of a disaster that
health does serve this role in low-in- results in death.

Incorporating health dynamics in the time invariant and known, and a part that
model has important empirical as well varies over time and is only known when
as theoretical implications. It is natural wages are revealed.24
to start with a dynamic health produc- Inspection of (10) immediately yields
tion function in which current health several important new questions about
status, Ht, depends on current and past the impact of health and health changes
health inputs, N, labor supplies, L, and on labor outcomes. For example, to
the disease or public health environ- what extent do current episodes of ill-
ment, D (Grossman 1972): health affect labor market outcomes in
the future? Does it matter whether
Ht = H(Nt,Nt_i,.... No,Lt,Lt_j,. .,Lo;Dt, these episodes are expected or unex-
Dt.l,...,Do,At,B'4,L4t,t,t-l,...,JIo,eht). (9) pected? How important is previous
We have assumed that individual health, conditional on current health, in
background variables, B', are time in- determining labor market outcomes
variant; it is, in principle, straightfor- over the life course? By allowing for
ward to allow them to change over time, separate effects of current and past
as we have for age, At. We have also as- health on wages in (10), it is possible to
sumed that the unobserved, individual- examine the impact of changes in health
level health endowment comprises two status, or of the joint effects of levels
components: one that is time invariant, and changes in health, on labor market
x, and is known to the individual, and outcomes. To date, however, very little
another that varies over time, At. evidence has been amassed to answer
Whereas past realizations of jt will also these sorts of questions.
be known to the individual, future reali- There are several possible avenues
zations can only be forecast with error. that might be followed to estimate the
We assume that only concurrent mea- wage function (10). The decomposi-
surement error, eht, affects the observed tion of unobservables into time-varying
realization of current health. and time-invariant components sug-
If health contains both a stock and a gests the use of a fixed effects proce-
flow component, then it is conceivable dure (Deolalikar 1988; Lawrence
that labor productivity will depend on Haddad and Howarth Bouis 1991). Use
lagged as well as current health status of an individual fixed effect will sweep
in addition to schooling, S, and infra- out the individual unobservable time-
structure, I: invariant component, cc, so long as it
enters the model linearly. The fixed ef-
Wt = w(Ht,Ht-l...,Ho,S; At,B,I,ca,at,ewt). (10) fects transformation will, however, also
As an example, productivity may be af- sweep out observable time-invariant
fected by the duration of an illness, in health indicators (like height), whose
addition to its incidence and severity. Al- impact may be of considerable interest.
ternatively, levels of health and changes Although it is very convenient, the
in health may interact in their influence linearity assumption is both ad hoc
on productivity; for example, the effect and not necessarily innocuous. For
of a loss in body mass due to a crippling
bout of diarrhea is likely to differ de- 24 The potential effects of worker health invest-
pending on the initial level of mass ments on worker or firm investments in on-the-job
(health) of the individual. Once again, it training or related human capital are ignored here
because, to date, they have not been considered in
is useful to decompose unobserved het- the empirical literature. Incorporatingthem would
erogeneity, a, into a component that is add both richness and complexity to the analysis.

example, in a conditional self-employ- namic, health-conditioned profit func-

ment profit function, the time-invariant tion.
component of alpha may represent
3.6 Observability of Effort and Health
managerial ability (Yair Mundlak 1963).
If it does, and if the profit function is Another important issue relevant to
represented by a flexible form approxi- our ability to detect effects of health on
mation such as translog, there is no market wages is the role of imperfect
theoretical reason why managerial abil- information by the employer. We have
ity, just because it is not observed in assumed that real wages equal marginal
the data, should not interact with other productivity. This may be a reasonable
quasi-fixed inputs or with input or out- characterization for self-employed and
put prices. Other observable quasi-fixed piece-rate workers; it may also ade-
factors would be interacted in this way, quately characterize workers on time
which would violate the linearity as- rates whose productivity is observed at
sumption. little or no cost. However for many
In order to consistently estimate the workers paid on a time basis, produc-
impact of (changes in) health on wages, tivity may not be easily observed if
differences in the time-varying errors monitoring is costly.
(?Ct-oct_i,ewt-ewt_0) must be uncorre- Assume, for the moment, that health-
lated with differences in time-varying ier workers are more productive, but
health characteristics. If the time-vary- their productivity is not observable.
ing errors only contain random mea- Health will then be reflected in the
surement error (ewt), the assumption wages of these workers only if it (or
seems plausible. It is less appealing if some factor known to be correlated
the errors also comprise omitted char- with it) is observable to the employer.
acteristics, OCt,which affect wages since If health is not rewarded in the labor
part of these wage shocks may be in- market, because of costly observability
vested in health. This will arise, for ex- and monitoring, say, then in the long
ample, if a worker experiences a sur- run one would expect the development
prisingly good year and spends some of of alternative labor or land contractual
the unanticipated income on health- forms (such as sharecropping or longer-
augmenting inputs (perhaps by eating term labor contracts) that minimize
more nutritious foods). problems associated with observability
A natural solution to this problem is of productivity. One would also expect
to combine fixed effects with instru- healthier (and more productive) work-
mental variables. In a dynamic optimi- ers to selectively move into sectors that
zation framework, in which households do reward health, such as self-employ-
maximize expected utility, reduced ment or piece-rate work (Foster and
form health input demands will be a Rosenzweig 1996).
function of all lagged and expected fu- Thus, a key factor determining
ture health input prices (and possibly whether better health is translated into
their interactions with various exoge- higher market wages is the observability
nous household-level variables). These of both productivity and health. As we
prices are, therefore, additional instru- will explore below, there are many po-
ments, provided they can be measured. tential indicators of health and they
As an example, Behrman, Foster, and may have different effects on produc-
Rosenzweig (1997) use lagged food tivity. These measures also differ in
prices to estimate health effects in a dy- their degree of observability: for exam-

ple, it may be difficult for employers to firms will set the value marginal prod-
monitor total nutrient intakes, even uct of labor in efficiency units to the
when they feed their workers (because daily wage paid per efficiency unit of
workers may substitute food to other labor, the efficiency wage, where effi-
family members by eating less at home). ciency units per hour worked can be de-
So it is not obvious that higher nutrient fined as y(w) and the efficiency wage as
intakes will be rewarded in higher w/y. Equivalently, firms will set daily
wages in an environment with costly wages so as to minimize the efficiency
monitoring (Foster and Rosenzweig wage, subject to the daily wage being at
1994). But outputs related to nutrient least as great as the reservation wage of
intakes, such as body mass, are readily workers. As long as the reservation
observed and so may be rewarded, wage constraint is not binding, the equi-
either because they are correlated with librium condition for minimizing effi-
nutrient intakes, or because they have ciency wages will involve equating the
independent effects on wages, or both. average efficiency wage to the marginal
wage cost of increasing labor by an effi-
3.7 General EquilibriumnImnplications: ciency unit. The latter is simply the re-
Nutrition Efficiency Wages ciprocal of the marginal efficiency unit
gains of an increase in wages, y7, Hence
Workers in poor health may be ra-
in equilibrium we have the familiar con-
tioned out of the job market because
dition: '/yty =tV/y.Since this interior solu-
they are too expensive to hire. This is
tion assumes that the offered time wage
the key insight underlying nutrition-
is above the reservation wage of workers,
based efficiency wage models.25 In all
some will be unemployed. Indeed, with-
of these models, a healthier worker is a
out transfers or nonlabor income of some
more efficient worker, and this relation-
form, those who are unemployed will
ship is not concave, being characterized
starve. This is a striking, if implausible,
either by a region of increasing returns
implication of the theory. As mentioned
before decreasing returns set in, or by a
above in the context of intrahousehold
minimum health threshold that is nec-
allocations, if there are multiple mem-
essary in order to work at all. This hy-
bers in a household, consumption will
pothesized nonconcavity has critical im-
differ among members, with some
plications for the theory. In particular,
if the reservation (time) wage faced by members having first claims on con-
firms is sufficiently low, it pays all firms sumption and others possibly starving.26
There are several additional problems
to raise wages above that level, because
with the theory. For example, if invest-
the gain in profits from increased
ments in health now reap productivity
worker productivity outweighs the costs.
benefits in the future, then this exter-
Assume that all workers are homoge-
nality can only be captured by writing
neous, that hours of labor supply are
long-term labor contracts. We would
fixed for those who get jobs, and that all
therefore expect these contracts to exist
consumption is spent on health inputs,
in very-low-income settings, but in fact
so that health varies directly with the
in many rural areas of developing
wage paid. In order to maximize profits,
25 The major papers that exposit this theory 26 Leibenstein did allow for a nonstarvation so-
include Leibenstein (1957), Dipak Mazumdar lution, specifically, social norms that would pre-
(1959), Mirrlees (1975), Stiglitz (1976), Christo- vent starvationby inducing firms to give up profits
pher Bliss and Nicholas Stern (1978a), and Das- by lowering their wages and increasing employ-
gupta and Debraj Ray (1986, 1987). ment.

economies, short-term, even daily, con- employed.27 If healthier people also

tracts are the norm (Rosenzweig 1988). have higher reservation wages (because
As noted by T.N. Srinivasan (1994), they have higher total incomes and
Dasgupta and Ray's (1986, 1987) model value leisure more), then some may
offers a solution to the appropriability choose not to work in the labor market,
problem, while also deriving the gen- and so there will be voluntary employ-
eral equilibrium properties more comment among them. In a competitive
pletely than other models. In their equilibrium, efficiency wages for the
model, piece-rate (or efficiency) wages employed will be bid up to the equilib-
are paid, where the piece-rate wage is rium level, even though workers with
determined in a competitive equilib- more assets than the marginal worker
rium in which the amount of efficient could have been hired for less. Thus
units of labor demanded (derived from time wages will not be equal across
an aggregate production function as the workers, but will be higher for persons
marginal product of efficient labor) is with greater assets.28
equated to the number of efficient units One of the more interesting empiri-
supplied. People in the model are al- cal possibilities derived from the model
lowed to be heterogeneous in their asis that some persons who are alike in all
set holdings, but are otherwise identi- respects (including being landless) may
cal. There are two sources of income: differ in their employment experiences.
returns on assets (for landowners) and One of the possible equilibria entails
wages (for the employed). As in the that the minimum efficiency wage of
other nutrition efficiency wage models, the landless just equals the market
all income is consumed in the form of equilibrium efficiency wage. If so, it is
health inputs. Because of the heteroge- possible that the demand for efficient
neity in asset income, one can deter- units of labor is less than the supply,
mine for each person (or persons having requiring rationing of jobs. Thus not
a unique level of assets) a distinct effi- only does one's asset position matter for
ciency wage: that is, a wage rate per ef- employability, but so does luck.
ficiency unit that minimizes the cost per However, contrary to some views,
efficiency unit, w/y(w + pZ)), of hiring one implication that is not necessarily
themn; subject to their reservation wage implied by the Dasgupta-Ray model is
constraint. Here p is the rate of return fixity of the market equilibrium effi-
on assets, Z, equal in competitive eqiui- ciency, or piece-rate, wage. In the case
libriumn to the marginal product of ag- when the market efficiency wage is so
gregate capital. high that no involuntary labor exists,
If the market equilibrium piece-rate this implication is not surprising. How-
wage, wP, is lower than any individ- ever, it also holds when the market
ual's cost-minimizing efficiency wage, piece-rate is below the efficiency wage
w*/y(w*+ pZ), that worker will not be of the landless. In this case, too, if de-
hired, even if the time-wage equivalent mand for labor shifts out, the market
is above the worker's reservation wage, piece-rate wage will rise, and the more
since it is not profitable for any em- healthy of the unhealthy unemployed
ployer to do so. The model thus pre-
dicts there mnay be involuntary unem- 27 Of course, if equilibrium piece-rate wages are
ployment, and if there is, those who sufficiently high, no involuntary unemployment
need ensue.
have lower levels of assets and are thus 28 If the labor market equilibrium were monop-
in poorer health are less likely to be sonistic, this implication would be reversed.

Stratussand Thomas: Health, Nutrition, and Economic Developmnent 789
will now get jobs. So long as the new set it apart from other indicators of
market piece-rate continues to fall be- human capital like education: first, it is
low the efficiency wage of landless multidimensional and second, measure-
workers, the least healthy will remain ment error in health is likely to be
unemployed. It is only in the rationing related to income and labor market
equilibrium described above, in which outcomes.
the market piece-rate equals the effi- First, there is a consensus in the lit-
ciency wage of the landless, that market erature that the number of years of
piece-rates will not respond to small schooling is a reasonably good indicator
shifts in labor demand. of educational attainment. No similar
The Dasgupta-Ray representation, agreement exists for health, in part be-
then, provides predictions regarding cause health is fundamentally multi-
who will be employed and who will be dimensional (John Ware, Allyson
destitute. Given the historical impor- Davies-Avery, and Robert Brook
tance of the nutrition efficiency wage 1980).31 Moreover, different dimensions
models in the economic development of health are likely to have different
literature, it is of considerable interest effects on one's productivity or labor
to empirically test their imiplications.29 supply, and these effects may well vary
In order to consider the empirical evi- over the life course or wage distri-
dence, we must consider both how to bution. We argue therefore that it
measure health and the effect mismeas- makes good sense to examine the rela-
urement may have on interpretation of tionship between labor outcomes and
empirical work. multiple health indicators simultane-
ously, wherever possible.
4. Measurement of Health Second, it is typically assumed that
measurement error in schooling is ran-
Whereas economists have largely fo- dom (Griliches 1977). In contrast, many
cused on understanding the behaviors health indicators are measured with er-
that underlie associations between ror that is systematically related to de-
health and labor market outcomes, phy- mand for health and other behaviors
sicians and epidemiologists have de- which are in turn related to wages, la-
voted considerable effort to measur- bor supply, and other socioeconomic
ing health. This section highlights characteristics. This complicates inter-
issues revolving around mneasurement, pretation of empirical relationships be-
both because it is important for inter- tween health and labor outcomes and
preting empirical relationships with la- seriously compromises the value of
bor outcomes and because of its relative standard fix-ups, such as instrumental
neglect by economists .30 Specifically, variables. In addition, the extent and
we highlight two aspects of health that nature of errors are likely to vary from
29 Dasgupta (1997) argues that the theory
measure to measure. (See Anita Stewart
should not be judged solely by its sharp predic- and Ware 1992, for a comprehensive
tions, such as the potential starvationof the unem- and thoughtful discussion.)
ployed landless, but rather should be interpreted
more loosely as implying a cycle of destitution that 31 In fact, education is more than the highest
may be initiated by low assets and bad luck. grade completed, but may also be characterized by
30 One may well argue there is a need to discuss years in school (or grade repetition), test scores,
measurement of labor market success; since much quality of schooling, post-school training and expe-
has been written about it in the economics litera- rience. Several studies have included these mea-
ture, this does not seem like an appropriate place sures when calculating returns to education (for
to repeat the arguments. example, Joseph Altonji 1995).

In this section, a series of different GHS, however, suffers from two key
health indicators that have been used in drawbacks. First, a respondent is typi-
the empirical literature are discussed. cally asked to rate his health in one of
We begin with self-reported general four or five discrete categories ranging
health status, followed by self-reported from excellent to poor health. Relying
morbidities, limitations to normal ac- on such a small number of discrete
tivities, and measures of physical func- categories cannot possibly do justice to
tioning. We then turn to nutrition- the complexity and diversity of health
based health measures and include both status of individuals; this is one reason
health outputs (anthropometrics) and we argue for using multiple indicators
inputs (nutrient intakes). For each group simultaneously.
of indicators, we discuss the nature of The second problem has to do with
the information they are likely to contain measurement. "Good" health may not
along with the kinds of measurement mean the same thing to all people, and
error that are likely to be important. The respondents are not provided with an
section ends with evidence on relation- established metric against which to
ships among different health measures compare their own health. Indeed, few
to highlight the importance of multi- surveys provide a clear definition of a
dimensionality and a discussion of the reference health status, and there is sel-
implications of random and systematic dom an explicit reference group. Be-
measurement error for empirical work. cause questions about GHS are typically
vague, we have no idea whether the re-
4.1 General Health Status
spondent is rating his health relative to
Comprehensive clinical evaluations of the national average, to his neighbor, or
health status, for example as under- to whom.
taken in the United States National Moreover, self-evaluations reflect
Health and Nutrition Examination Sur- perceptions of health. While important,
veys, are far too expensive to be in- perceptions are likely related to values,
cluded in a typical socioeconomic sur- background, beliefs, and information,
vey, except on a small scale or with very all of which are systematically related to
selected samples; a fortiori in lower in- socioeconomic characteristics, including
come settings where health service in- wages and income. For example, infor-
frastructure is relatively weak. Most mation about own-health status is al-
household surveys have, therefore, re- most surely correlated with the extent
lied on intetviews with respondents who of concurrent and prior use of health
provide an assessment of their own care, because people who have used the
health. health care system are likely to be bet-
Within the class of self-evaluations, ter informed. Since most people assume
general health status (GHS) is probably they are in good health unless they have
the most widely used indicator in the information to the contrary, it is plausi-
empirical literature in the United ble to suppose that, conditional on a
States. Some advocates have argued level of health status, those who have
that it is the best single health index little exposure to the health system are
available, citing, for example, the fact likely to report themselves as being in
that GHS is correlated with subsequent better health. If this is true, given that
morbidity and mortality (see, for exam- lower-income people are less likely to
ple, Ware, Davies-Avery, and Cathy use health care, particularly in poor so-
Donald 1978). cieties, measurement error in GHS will

Strauss and Thomas: Health, Nutrition, and EconomnicDevelopment 791
be systematically related to income (and functional consequences of ill health

wages). There is evidence in the RAND that are of primary interest, and those
Health Insurance Experiment (HIE) consequences typically cut across dis-
that suggests this is a legitimate con- eases and are exacerbated by interac-
cern, at least for those people who, in tions among different diseases.
the baseline, were in poor health and in As a low-cost alternative to clinical
the bottom quintile of the income dis- evaluations of health status in a house-
tribution. Among these people, those hold survey, some studies have drawn
randomly assigned to receive free care data from health facility records. But in
used more health care and were, at the low-income countries, a significant pro-
end of the experiment, in better health portion of the population does not use
as measured by clinically evaluated out- these facilities, and those that do tend
comes (such as blood pressure) and to be a select group that is not necessar-
risk of subsequent death. But, accord- ily those in the worst health. In fact,
ing to their own evaluation of their they tend to be higher-income people,
health, measured by GHS, it actually and so it is very hard to infer anything
worsened (Joseph Newhouse et al. about the relationship between health
1993). Similar evidence is reported and well-being in the whole population
from a health price experiment in Indo- without at least information on the
nesia, where people used more care in mechanisms underlying the choice to
those places where prices were lower; visit a health care facility.33
their health improved, but their self-re- Several surveys have asked questions
ported GHS was worse (William Dow et about illness or specific symptoms (such
al. 1997). as fevers, diarrhea, respiratory prob-
We conclude that GHS does provide lems) during a reference period. As
information in that it predicts later with GHS, these self-evaluations are
health problems. But, in the context of difficult to interpret if what is deemed
models of labor outcomes, there are an illness or a symptom is not the same
good reasons to be concerned that it is thing for all respondents. In fact, in sur-
contaminated by measurement error veys from low-income countries, it is
which is correlated with socioeconomic not unusual for the poorest to appear
characteristics, including income. to be the most healthy by this metric!
For example, in Ghana and the Cote
4.2 Self-Reported Morbidity, Illness d'Ivoire, the propensity for adults to re-
and "Normal"Activity port being ill in the last four weeks is
positively associated with own educa-
Disease-oriented definitions of health tion (Schultz and Aysit Tansel 1997)
status, favored by many clinicians and and with per capita household expendi-
some epidemiologists,32 have the ad- tures (Mead Over et al. 1992). As with
vantage of a foundation in medical prac- GHS, these indicators are likely to be
tice. However, even in a clinical setting, measured with error that is correlated
it is difficult to diagnose all problems with use of the health system (and thus
unambiguously, especially in the pres-
ence of multiple, interrelated problems.
33 For example, Gertler and van der Gaag (1990)
Furthermore, from a social science and report that in the Cote d'Ivoire only 40 percent of
public health perspective, it is often the persons who said they were ill in the last month
sought medical care at a modern facility, and they
32 See the discussion, for instance, in Jamison et tended to be from higher-income households. A
al. (1993) and World Bank (1993). similar pattern is observed in many countries.

with income and the price of health labor force participation, productivity,
services). In addition, other price incen- and wages.
tives may influence self-reported mor-
4.3 Self-Reported Physical Functioning
bidity. For instance, an individual may
claim to suffer from an illness in order Several household surveys have col-
to become eligible for health-related lected information on difficulties with
benefits (Donald Parsons 1982; John physical functioning that are considered
Bound 1991). normal activities for people in good
Another commonly used variant of health; these might include walking a
self-reported illness is to ask whether specified distance, lifting a particular
any days of "normal" activity were lost weight, bending, or climbing stairs.
to ill health. Some argue that this mea- While the notion of "difficulty" is sub-
sure is less likely to be contaminated by jective, questions about specific activi-
systematic respondent error. Apart from ties of daily living (ADLs) are more
the fact that "normal" is not well- precisely defined than "being ill"
defined, people whose opportunity cost or "normal activities," and there is
of time is high (the better. educated, say) some evidence that, relative to GHS
will have less incentive to miss activi- and morbidity, ADLs are less prone to
ties. By this metric, they will appear to the type of measurement error dis-
be in better health than people with a cussed above. (See the discussion in
lower value of time (conditional on a Strauss et al. 1993; Michael Schoen-
particular "true" health status). This is baum 1995, and Dow et al. 1997, pre-
in contrast with the argument above sent corroborating evidence.) This is an
whereby the better educated report important advantage of ADLs in a sur-
themselves as being in worse health be- vey setting. However, they have one key
cause of better knowledge or greater ex- drawback: the limitations in physical ac-
posure to health services. While the net tivities that ADLs typically capture are
impact is unclear, there is no reason to frequently due to physical health prob-
expect a closer correspondence be- lems such as shortness of breath, joint
tween normal days lost with "true" problems, or back problems (Stewart et
health than with days ill or some other al. 1978). Few prime-age adults have
indicator. As it turns out, reported cor- difficulty with these activities, and so
relations between days lost and educa- ADLs may not be as useful in studies of
tion have both been negative (Allan Hill health and labor outcomes as they have
and Masuma Mamdani 1989), and posi- proved to be in the gerontological lit-
tive (Schultz and Tansel 1996). More- erature.
over, people with acute health problems
4.4 Nutrition-Based Indicators:
are likely to make lifestyle and occupa-
Nutrient Intakes
tion choices in response to these prob-
lems, making it very difficult to inter- Nutrient intakes play a central role in
pret normal. the efficiency wage literature, and so it
In general, for all of these measures is not surprising that nutritional status
and for GHS, it will be very difficult to has been prominent in the associated
separate true health status from mea- empirical literature. It is presumably
surement error. The key point for our net energy intake that is related to pro-
purposes is that this measurement error ductivity, but since energy expenditures
is likely to be correlated with socioeco- are difficult to measure, studies have
nomic behaviors and outcomes, such as tended to focus on nutrient intakes.

Strauss and Thomas: Health, Nutrition, and Economic Developmnent 793
Biomedical evidence indicates that calo- actually did consume.34 If the (net) re-
rie intake is associated with increases in ceipt of in-kind food (including trans-
maximum oxygen uptake (Spurr 1983, fers and gifts) declines with income and
1988), which suggests there may be a the probability of having guests rises
link between nutrition and productivity. with income, then nutrient intakes will
Interpretation, however, is made some- be biased and the bias will, also, be
what ambiguous by different body positively correlated with income and
maintenance needs, which vary by wages. As a third example, it is very dif-
weight and height. In the absence of ficult to measure nutrient intakes for
measures of energy output, this sug- meals eaten away from home; typically,
gests examining the link between nutri- it is assumed that those meals have the
ent intakes and labor outcomes condi- same calorie content as meals at home.
tional on, say, height and BMI. This Clearly, this need not be true. For ex-
provides one example of the usefulness ample, low-income workers are often
of examining health measures in combi- given food at work; if it is more nutri-
nation. tious than the food eaten at home, mea-
Measurement of energy consumption surement error will be negatively corre-
is not nearly as straightforward as might lated with income.
be thought. Several different methods There is prima facie evidence that
have been adopted and each involves its these sorts of leakages (from wastage,
own strengths and weaknesses. Calorie food given or received in kind, and
availability is computed by converting meals away from home) are important
food quantities (purchases and con- and result in systematic errors in esti-
sumption from own production) into mated nutrient intakes. In most studies
nutrient intakes, using standard food that use estimates based on availability,
composition tables. This has the advan- a substantial fraction of the population
tage of being relatively easily calculated appears to be consuming at unrealisti-
using data commonly collected in many cally low or high levels. Srinivasan
household expenditure and farm pro- (1992), for example, reports that in the
duction surveys. But the method suffers 1976 National Sample Survey of India,
from several potentially important over 5 percent of rural households con-
sources of systematic bias. sume less than 1,500 calories per capita
First, it assumes no food is wasted: per day and almost 20 percent consume
everything that is available is converted over 4,000 calories per capita daily!
into nutrients. It is plausible that very An alternative method of collecting
low-income (or low-wage) households nutrient consumption is to obtain infor-
waste less than those that are better off, mation on intakes rather than on avail-
in which case nutrient intakes will tend ability. One method weighs ingredients
to be upward biased and the bias will prior to each meal, and wastage after it,
increase with income. Second, it is very and then converts measured consump-
difficult in consumption and production tion into nutrients. Although meals
surveys to take into account all meals eaten away from home are not cap-
that are given to guests or employees tured, this method is probably the most
and all meals that are received in-kind.
For example, the members of a house- 34While uncommon, it is possible to draw up an
hold that entertained many guests dur- entire list of all people at all meals for a week (in a
consumption survey), but that is not a realistic op-
ing the survey period will appear to tion over a longer survey period (such as in a farm
have consumed more calories than they household survey).

accurate one used to date. But the household averages of reported daily in-
method is prohibitively expensive to takes over the four rounds are very
field in large-scale household surveys similar: 2,439 per adult equivalent for
and has therefore been employed very availability and 2,358 for 24-hour re-
infrequently.35 call, although the availability measures
Thus, the most common strategy used have somewhat fatter tails. Key for our
to collect nutrient intakes has been to purposes is that the difference between
ask respondents to recall ingredients availability and intakes for the same
that went into meals consumed, usually household is positively correlated with
over the previous 24 hours. While this household characteristics, including in-
approach has the advantage of poten- come, expenditure, and education of
tially excluding leakages (such as meals the head. Among low-income house-
for guests, transfers of food, and wast- holds, availability tends to be less than
age of food), it comes at a substantial intake, but higher-income households
cost in terms of survey time. Moreover, report higher availability than intakes.
since there is considerable variation in This is all consistent with the idea that
eating habits, 24-hour recalls are likely leakages are missed in availability and
to be very noisy; extending the recall tend to be systematically associated
period raises concerns about recall bias with income.
(which is thought to rise rapidly in this It is possible to get some idea of the
context), and multiple visits further relative importance of random noise by
raise the cost of collecting these data. regressing intake on availability and
Systematic evaluation of the accuracy vice-versa. If 24-hour recalls are meas-
of these different methods of collecting ured without error, then the coefficient
nutrient data is scarce, and there are when it is the regressand should be 1: it
virtually no surveys that contain more is very close (0.97 with a standard error
than one measure. A survey in Bukid- of 0.01). But, as suggested by the fatter-
non, Philippines, conducted by Bouis tailed distribution, availability seems to
and Haddad (1992), is an important ex- be measured with more noise: when it
ception. It has two different measures: is the regressand, the slope is signifi-
intakes from 24-hour recalls, and calo- cantly smaller (0.86 with a standard er-
rie availability from a recall of pur- ror of 0.01). In these data, availability
chases for the month. The survey was appears to be a worse indicator of calo-
conducted over four rounds, and the rie intakes from the point of view of
both random and systematic errors.
35The method was used in the Brazilian EN- It is apparent, therefore, that all nu-
DEF survey. To account for the fact that there is a trient intakes are likely to be subject to
good deal of variation in daily intakes, each of
55,000 households was visited on a daily basis for a some random measurement error that is
week. All food that would be prepared in the fol- probably greatest for recall-based meth-
lowing 24 hours and wastage from the previous 24 ods. In addition, availability-based mea-
hours was measured at each visit. The enumerator
also listed at each visit all household members and sures suffer from systematic errors that
guests present at every meal during the previous are probably correlated with income,
24 hours. It is therefore possible to control for a wages, and productivity. Since the na-
good part of the leakages discussed above. Of in-
terest is the fact that among the poorest house- ture of the errors and the correlations
holds, there is evidence of a decline in intakes af- with labor outcomes are likely to differ
ter the first day or two; the survey staff attributed depending on how intakes are meas-
this to an attempt by the respondents to impress
the enumerators (or downplay their own poverty) ured, it seems safe to conjecture that
(Marcello Vasconcellos, personal communication). interpretation of the impact of nutrient

Strauss and Thomas: Health, Nutrition, and Econom-ticDevelopment 795
intakes on wages will not be indepen- been commonly used for adults is body
dent of the way the data have been col- mass index (BMI), the ratio of weight
lected. (in kilograms) to height (in meters)
squared. On average, a prime-age male
4.5 Nutrition-Based Indicators:
in the United States has a BMI of about
Anthropometrics
25; BMIs are considerably lower in poor
Having discussed nutrient intakes, countries (ranging, on average, between
which are inputs in a health-production 21 and 23 in the three countries in Fig-
function, we turn next to outputs, spe- ure 1).36 Extremely low (below 18) and
cifically, anthropometrics such as high (above 30) values have been associ-
height, weight, and BMI. It is reason- ated with higher adult mortality (Hans
able to wonder how much information Waaler 1984; Fogel 1994), but BMI and
on health can be embodied in anthro- mortality are essentially unrelated be-
pometry. As discussed in Section 2, tween that range. The relationship be-
adult stature has been profitably used tween adult BMI and subsequent mor-
as an indicator of well-being in both the tality, the so-called "Waaler curve," is
economic history and the development U-shaped with a long, flat bottom be-
literatures. Moreover, child height has tween approximately 20 and 30. The
proven to be an informative indicator of causal mechanism underlying this asso-
the nutritional status of children and is ciation has not been established, and
viewed as a longer-run indicator of nu- the precise definition of extreme values
tritional status (see, for example, John remains debated.
Waterlow et al. 1977; Frank Falkner Because they are relatively inexpen-
and James Tanner 1986; and Waterlow sive to collect, in many surveys, height,
1988). While height is clearly deter- weight, and sometimes arm circumfer-
mined by the time an individual reaches ence have been measured in the field
adulthood (apart from shrinking later in by an anthropometrist. While the mea-
life), there is some debate in the litera- sures may be subject to random error
ture about the extent to which adult (which can be reduced by good field
stature is completely determined by the procedures), a key virtue of these
time the child has progressed beyond. health indicators is the absence of mea-
early childhood. Nevertheless, to the suremnent error that is systematically
extent that height does affect labor out- correlated with respondent charac-
comes, it will clearly reflect returns to teristics (such as income). (But see
human capital investments made during Strauss and Thomas 1996 for a discus-
early childhood and, perhaps, returns to sion of the National Longitudinal Sur-
strength. vey in the United States, which relies
Whereas height is predetermined by
adulthood, weight varies in the short 36 Moreover, the fraction of the population at
run and so provides a nmorecurrent in- either extreme varies dramatically across coun-
tries. In the United States, for example, only 1
dicator of nutritional status. Since a percent of men have BMIs below 18, 6 percent
light person may also be small, and thus below 20, and 13 percent above 30. In Viet Nam,
not underweight given height (and, con- where the BMI of the average male is below 20,
22 percent are below 18, 66 percent are below 20,
versely, a heavy, tall person may not be and less than 1 percent have BMIs above 30. In all
overweight), nutritionists have found it four countries, the distributions for women are
convenient to analyze weight given more fat-tailed: 3 percent are below 18, and 18
percent are above 30 in the United States with the
height. There are many potential ways proportions being 24 percent and 1 percent, re-
of expressing this ratio; one that has spectively, in Viet Nam.

on self-reported height and weight, % able to carrya heavy load easily

measured with error that is related to -100
income.)
4.6 Multidimensionality of Health Status -~~~~~~~~~
~~-99
We have argued above that because * 98- 98

of the multidimensionality of health, it
is often useful to examine several indi-
97 - 97 E
cators simultaneously. Data from the ~~~~~~~~~~~~~
Indonesia Family Life Survey (IFLS) b.0
~96-
are used to illustrate this point: Figure Male
5 presents the relationship between Female ---
95 l_ l_ l_ l
height and the ability to carry a heavy
load easily, a commonly used ADL or 140 150 160 170
functional limitation. In addition, we
have included only those people who -96
report their GHS as being "fair" (ac-
counting for almost three-quarters of m90 - - 94-i
the adult respondents). The figure dem-
onstrates three points. First, there is . 88- 923
heterogeneity in both height and func-
tional limitations even within the group 86- - 90
of people in fair health; relying only on
GHS is throwing away information. Sec- 084 "
ond, the positive slopes in the figure in-
dicate that taller people have less diffi- l l l I
culty carrying a heavy load (a fact that 140 150 160 170
carries through if we do not control for Height (cm)
GHS). Third, comparing the two panels, Figure 5. Anthropometricsand PhysicalFunctioning,
younger adults (aged 25 to 40) have less Conditionalon General Health Status
difficulty than older adults (40-55), and
comparisons within panels indicate that trols, such as the effects of increased
men have less difficulty than women. nutrient intakes without controlling for
The evidence is suggestive that height body mass (height and weight, say). In
is capturing more than background (or other cases, the inclusion of multiple in-
investments in early childhood) but also dicators changes the interpretation of
reflects differences in strength and ro- the effects. For example, holding calo-
bustness. In any case, the fact that rie intakes constant, increasing protein
height predicts functional limitations intake implies an increase in the protein
(even after controlling for general density of the diet and hence an in-
health status, age, and gender) points to crease in diet quality.
the potential usefulness of employing
multiple indicators of health in addition 4.7 Empirical Implications of
to (ordinal) measures of general health MeasurementError in Health Status
status.
More generally, in some instances, it Throughout the discussion, we have
is not at all clear how to interpret taken care to distinguish random from
health effects without additional con- systematic errors in measurement of

health. This is important because esti- reason people report themselves as be-
mated effects of health on labor out- ing sick is because they are better in-
comes will vary with the type of error, formed, which is in turn a reflection of
as will strategies that seek to ameliorate the availability of health facilities in the
the influence of measurement error. community, health infrastructure will
We therefore discuss each in turn, al- be correlated with the measurement er-
though it is important to recognize that ror and so is not a valid instrument. As-
a particular health indicator may be af- suming that people's propensity to re-
fected by both types of error, albeit to port themselves as ill changes slowly
different degrees. over time, repeated observations on the
Random measurement error in health same individual may help, since examin-
will bias estimated effects towards zero; ing differences in health status (transi-
this is a case of classical errors-in-vari- tions) will reduce the influence of sys-
ables. It is difficult to think of any tematic measurement error-but at the
health indicator that is not subject to at expense of increasing the relative mag-
least some random measurement error. nitude of random error.
Even weight and height measured by a In general, measurement error in
trained anthropometrist are not im- health status or inputs is likely to be
mune to error. For example, subjects partly random and, in many cases,
may not be aligned properly on measur- partly systematic. Obviously, the best
ing boards, scales may not be recali- way to reduce measurement error is to
brated, and so forth. Moreover, many pay more attention to measurement. In
indicators vary over the course of a day some cases this involves taking averages
(like weight) or across days (like nutri- across repeated measures (to reduce
ent intakes or functional problems). In random noise), examining differences
all of these cases, repeated measures (to reduce systematic error), taking spe-
can ameliorate the impact of random cial care in fieldwork to avoid system-
error in regression models. atic errors (by carefully measuring
Systematic measurement error poses "leakages" when collecting nutrient in-
thornier problems. Consider the exam- take data, for example), or by avoiding
ple discussed above in which poor certain indicators that are especially
health is more likely to be reported if prone to error. Recent innovations in
the respondent has more contact with a survey methodology offer some poten-
modern health practitioner. If health tially exciting opportunities in this re-
care utilization rises with income, then gard. For example, a small number of
higher-wage individuals are more likely household surveys have experimented
to report themselves as ill, given a par- with greater reliance on direct observa-
ticular level of underlying health status: tion, such as assessing the incidence of
the impact of health on wages will be anemia (based on hemoglobin counts),
negatively biased. tuberculosis (using sputum), and hyper-
In contrast with random errors, aver- tension (based on blood pressure). In
aging multiple reports by the same re- addition, studies have sought to mea-
spondent is not likely to help reduce sure glucose levels (with saliva), net en-
the impact of systematic error. More- ergy intake (with labeled water), and
over, it is not easy to think of instru- lung capacity (using peak flow meters),
ments that are likely to be correlated and also to directly observe functional
with "true" health status, but uncorre- limitations through timed moves (such
lated with wages. For example, if the as walking a particular distance, stand-

ing from a sitting position, and so on). type of activity and type of contractual
Whether these measures will prove to arrangement in order to test hypotheses
be fruitful in studies of the links be- about the efficiency of markets, includ-
tween health and productivity remains ing labor markets; testing conjectures in
to be seen. the literature about the importance of
thresholds in health and their links with
5. The Relationship Between Health poverty and deprivation; and drawing
and Productivity inferences about how resources are al-
located within families.
Nutritionists, physicians, and a hand- To begin our review of the evidence,
ful of economists deserve the credit for we return to the association between
much of the early empirical work on height and wages presented in Section
health and productivity. Most of these 2. We next discuss the experimental and
studies were interested in documenting nonexperimental evidence on the na-
correlations between physical produc- ture of the effects of different health
tivity and various health indicators, par- indicators on productivity and labor
ticularly those that are nutrition-ori- supply. The discussion distinguishes
ented or that measure the incidence of morbidities, limitations, and ADLs from
specific diseases. By and large, these nutrition-related health indicators. We
studies were oblivious to problems asso- do not stand at one end of the spectrum
ciated with time-invariant and time- on the virtue of experimental relative to
varying unobserved heterogeneity. (See nonexperimental studies but see them
Ralph Andriano and Thomas Helminiak as complements with each contributing
1988; Behrman 1993; and Strauss 1993, value added to our understanding of the
for reviews.) Most of the nonexperi- relationships.37 On balance, our read-
mental studies that have attempted to ing of the literature suggests that some
sort out causality have also focused on dimensions of health status and some
documenting whether or not health af- health inputs do affect labor supply and
fects productivity and remarkably little worker productivity. In several cases,
attention has been paid in these studies the effects tend to be largest for the
to potential biases that might arise from poorest. With this in mind, the section
mismeasurement of health. A small continues with a discussion of the
number of experimental and quasi- shapes of the relationships and how
experimental studies are important be- these shapes vary by type of work, sea-
cause they have sought to address con- son, gender and also the observability of
cerns revolving around both causality health and work effort. Many of the re-
and measurement. sults provide insights into the function-
Over the last few years, however, ing of markets and asymmetries of
there have been substantial advances in information in low-income settings.
the field, and the literature has pro-
gressed beyond looking for relation- 37 Our goal is to summarize the evidence. In vir-
ships. Attention is now often focused on tually every nonexperimental study, one could rea-
more economically interesting, and sub- sonably quibble about specifications and identifi-
cation. Since these issues were covered in detail in
tle, questions. These include, for exam- the previous section, we prefer to focus here on
ple, drawing distinctions between the results and to discuss details only when new in-
effects of different health and nutrition sights emerge. However, because it has not yet
been discussed, experimental design is treated
indicators; comparing the effects of somewhat asymmetrically in this section and is de-
health depending on an individual's scribed in more detail.

Finally, we review the rather scant evi- cant (Thomas and Strauss 1997).
dence that relates to the potential gen- Schooling also provides a useful bench-
eral equilibrium implications of health- mark against which to evaluate the mag-
based efficiency wage models. nitude of the correlation between
height and wages. For example, in ur-
5.1 Height and Labor Outcomes
ban Brazil, for a literate man (who has
As Figure 2 indicates, taller men earn not completed elementary school) and
higher wages in Brazil and the United an illiterate man to earn the same wage,
States; this is not a new result. It has with their only difference being height,
been observed in a wide range of coun- then the illiterate man would have to be
tries across the world (Haddad and 30 cm taller than the literate man; this
Bouis 1991; Foster and Rosenzweig is 16 standard deviations in the data!
1993b; Schultz 1996; Thomas and
5.2 Morbidities, ADLs, Health
Strauss 1997; Peter Glick and Sahn
Limitations, GHS and Labor
1997). Although it has been demon-
Outcomes
strated less frequently, taller women
also earn higher wages, and further- Nutrition-related health indicators
more, both taller men and women are are discussed in the next subsection; we
more likely to participate in the labor focus on all other indicators in this sub-
force. The link between height and section. In an experimental study that
wages has also appeared in the histori- sought to measure the impact of a spe-
cal literature, where, for example, cific disease on output, sugarcane work-
Robert Margo and Steckel (1982) dem- ers on an irrigated estate in Tanzania
onstrate that sales values of slaves in were divided into two groups at the
Mississippi were influenced by height time of a baseline: those diagnosed with
(and also weight). While among slaves, schistosomiasis and those who were not.
part of the value of height reflects the Schistosomiasis is caused by parasitic
benefits of longevity (Gerald Friedman worms that live in slow-moving water,
1982; Meredith John 1988), today em- and repeated exposure can result in fe-
ployers are more likely to pay for better vers, aches, and often fatigue. Earnings
health and robustness, and, possibly, of workers on the estate were paid on
strength. the basis of the amount of sugarcane
More generally, height reflects in- cut in a day. This indicated that, at the
vestments made in the worker during baseline, the productivity of those suf-
childhood, and so we might interpret it fering from schistosomiasis was lower
as an indicator of human capital much than those who were not infected. In
along the lines of education. Indeed, as the experiment, those infected were
shown in Section 2, height and school- randomly divided into two groups. The
ing are positively correlated. In Brazil, treatment group members were given
adding height to a wage function for ur- chemotherapy, and their earnings in-
ban, self-employed males results in the creased but did not fully make up the
estimated returns to schooling falling by gap with the uninfected group. There
nearly 20 percent for all levels of was no change in the earnings of the
schooling and, in market wage equa- control group, the infected workers who
tions, cuts the magnitude of a signifi- were not given chemotherapy. The
cantly positive hazard term (controlling study clearly suggests that schistosomi-
for selection into market work) by 60 asis does affect productivity (A. Fen-
percent, causing it to become insignifi- wick and B. Figenschou 1972).

In contrast, however, a similar experi- survey data in which general health,

mental study of sugarcane workers in morbidities, and ADLS are all self-
the Cameroons found no effect of the reported; thus the measures carry with
same illness on output. In that study, them the problems of interpretation
however, there was no difference in the discussed in the previous section.
baseline output of workers infected The nonexperimental evidence on
with schistosomiasis and those who whether these health measures affect
were not infected. Moreover, treating wages or labor productivity is, at best,
the sick with chemotherapy had no im- also mixed. On one hand, a study of
pact on later productivity, and in the Ivorian males reported that those who
follow-up survey there was no differ- missed a day of work in the four weeks
ence in output of the treatment group, prior to the survey earned 18 percent
the control group, or the workers who lower wages than those who were not
were not sick at all .(C. Gateff et al. absent. But in a companion analysis us-
1971) 38 ing comparable data from Ghana, there
These studies highlight some of the is no link between wages of men and
strengths and weaknesses of experi- days ill. Moreover, women's wages are
ments in this literature. By design, both unaffected by days ill in both countries
studies address concerns regarding the (Schultz and Tansel 1997). A study of
endogeneity of health, since the evalu- farmers in Indonesia demonstrates that
ation is based on the impact on produc- farm profits are invariant to whether
tivity of changes in health status that re- the head of the household or spouse re-
sult from a randomized experiment. port themselves as being ill during the
Moreover, because they are very fo- previous week although, as we shall see,
cused, the studies are able to measure that does not necessarily imply the ill-
health status with great confidence: in ness did not affect the individual's pro-
these cases, they exploit the fact that it ductivity (Pitt and Rosenzweig 1986).
is straightforward to test for schis- In sharp contrast, these and other
tosomiasis with blood or urine samples. household surveys do indicate that poor
The downside is that experiments tend health reduces hours of labor supply. In
to be small, have a short time-frame, Pitt and Rosenzweig's Indonesia study,
and involve relatively homogenous the magnitude of the estimated effect is
populations, and therefore may not large: those men who reported being ill
have great power. It is difficult to know the previous week worked 70 hours less
whether low power explains the absence than other men. The authors explain
of an observed productivity effect in the the magnitude by arguing that only se-
Cameroons. verely sick respondents reported them-
In contrast with the experimental re- selves as being ill. This seems reason-
search, nonexperimental studies have able, given that less than 3 percent of
paid little attention to issues of mea- farmers report themselves as sick, al-
surement, and those that have tend to though it does raise the question of
focus on randomnmeasurement error in what being ill means in the survey.
health. The fact that systematic mea- It is worth taking a short detour to
surement error has been largely ignored discuss an important implication of
is troubling, since these studies use these results, since they speak to a
question broader than the link between
38 Other studies are reviewed in Nicholas health and labor outcomes. While poor
Prescott (1979). health is associated with reduced labor

supply, the fact that farm profits are not -ment error in self-reported health
affected does not imply the farmer's status is correlated with education or
own productivity was unaffected but, income, the estimated health effects
instead, that labor was available that will be biased. Moreover, it seems likely
could substitute for him during this that an individual who is inclined to re-
time of illness. Pitt and Rosenzweig port himself as being in poor general
conclude that since this labor was pro- health will also be inclined to report
vided by hired wage earners, labor mar- difficulties with some ADLs (condi-
kets work efficiently in rural Indonesia, tional on "true"health). Thus, we would
an argument that is corroborated by expect there to be common reporting
other evidence on efficiency of markets error across these multiple health indi-
in Indonesia (Benjamin 1992). This is a cators for the same individual. In that
good example of how studies in this lit- case, the ADLs will be poor instru-
erature have moved beyond document- ments for general health.39
ing relationships, providing empirical A social experiment in Indonesia, the
insights into questions that economists Indonesian Resource Mobilization
have been grappling with for a long Study (IRMS), provides an opportunity
time. to examine the importance of reporting
Returning to health and labor supply, error among these different indicators.
it would be remarkable if workers who In the experiment, user fees at public
reported missing a day of work during health centers were raised in randomly
the reference period did not also report selected "treatment" districts, while
lower hours. Schultz and Tansel present prices were held constant (in real
evidence in support of this intuition for terms) in neighboring "control" dis-
men and women in both Ghana and tricts. A baseline household survey was
Cote d'lvoire. It is, however, not en- conducted prior to the intervention,
tirely clear how to interpret the result and the same households were resur-
since, as discussed in Section 4, one veyed two years later. The goal of IRMS
might interpret work days lost or days was to assess the effect of the price in-
of limited activity as being time alloca- crease on utilization: it declined in the
tion choices, and thus intimately related treatment areas, relative to controls
to labor supply, rather than as being (Gertler and Jack Molyneaux 1996). La-
only reflections of health status. bor force participation also declined in
Other studies have used measures of the treatment areas, relative to controls,
general health to explain labor force with effects being particularly large
participation for prime-age adults in (and significant) for men and women at
Jamaica (Victor Lavy, Palumbo and the bottom of the education distri-
Steven Stern 1995) and for men and bution, those whom we would expect to
women over age 60 in Taiwan (Schoen- be the most vulnerable. Dow et al.
baum 1995). Both studies report that la- (1997) estimate a conditional labor sup-
bor force participation is significantly ply function that relates changes in
lower anmongthose in poorer health. Ar- health outcomes to changes in labor
guing that health is endogenous and force participation, thus differencing
measured with random error, both stud- out any time-invariant unobserved het-
ies predict an ordinal index of general
health status with education, income or 39 Schoenbaum also finds substantial negative
effects on labor force participation of difficulties
assets, and with ADLs. If, as suggested with ADLs when they are included directly, both
may be the case in Section 4, measure- in the form of an index and as separate covariates.

erogeneity. Changes in health status are sence of a consistent impact on wages

instrumented with experimental treat- in the nonexperimental literature may
ment status, which is, by virtue of the be because the health indicators used in
design, random. In addition to these ad- those studies tend to reflect shorter-
vantages, IRMS contains information on term health problems (morbidities or
all the self-reported health measures days of limited activities in the last few
discussed in this subsection. This turns weeks) but wages tend to adjust rela-
out to be key, since the estimated im- tively slowly. Hours of work, in con-
pact of health differs substantially de- trast, can vary in the short run. We
pending on the particular measure. suspect that the differences in the
Worsening of ADLs, such as greater dif- estimated impacts on wages and labor
ficulty in walking 5 kilometers, and supply also reflect differential impacts
those measures related to labor supply, of systematic reporting or measurement
such as more days in bed, significantly error that is correlated with information
depress male labor supply. However, and exposure to the health system.
reduced labor supply was also recorded Thus, reporting error will be more
for those men who reported their gen- highly correlated with wages and in-
eral health status had improved during come than with hours of labor supply.
the hiatus between the surveys. Looking
5.3 Nutrient Intakes and Labor Market
a little more closely at the data, ADLs
Outcomes
and days of limited activity changed for
the worse among individuals in treat- The relationship between nutrition-
ment areas, relative to controls. But, based health measures and labor out-
counterintuitively, GHS and morbidi- comes has received substantially more
ties actually improved for treatments attention, particularly in the develop-
relative to controls. Since control ment literature. There are several
households faced lower prices and used experimental and quasi-experimental
more health care during this period, a studies demonstrating that nutrition-
reasonable interpretation of the evi- based interventions improve health out-
dence is that self-reported health status comes, particularly among children. Al-
is likely to be worse among those peo- though it has been more difficult to
ple who have greater exposure to the make the connection with improve-
health system and, presumably, more ments in productivity, a small number
information about themselves. If this is of studies with experimental designs do
correct, then reporting error in GHS provide evidence that low nutrient in-
and morbidities will be systematically takes have a deleterious impact on pro-
related to use of health care and, there- ductivity.
fore, in most nonexperimental studies, Until fairly recently, it was thought
to income and education as well as that poor nutrition largely reflects in-
health-seeking behaviors. The evidence adequate energy, or protein, intakes.
suggests that this problem is less perni- Recent work suggests that this focus has
cious in the case of ADLs, perhaps be- resulted in a failure to fully recognize
cause they are more "objective." the key role"that may be played by sev-
In sum, poor health, as measured by eral micronutrients. The central prob-
the indicators discussed here, does ap- lem lies in disentangling the separate
pear to reduce labor supply. However, effects of different macro- and micro-
the evidence that it affects productivity nutrients, all of which are typically con-
and wages is more ambiguous. The ab- sumed in combination. Experimental

designs have a clear edge in this regard, more on leafy greens and other foods
as they have the luxury of implementing that provide higher amounts of iron.)
specific nutritional interventions. The importance of adequate iron in
A particularly compelling example is the diet is a robust result that has been
a longitudinal study of 302 male rubber reported in several experimental studies
tree tappers and weeders in Indonesia of both productivity of adults and cogni-
(Samir Basta et al. 1979). Baseline tive achievement of children (see
health measures indicated that about Ernesto Pollitt 1997 for a review).41
half of these men were anemic. Their Results from less specific food supple-
baseline productivity (measured by kilo- mentation interventions have been
grams of latex collected by tappers per more varied for several reasons.
day and the area of trenches dug by Maarten Imminck and Fernando Viteri
weeders) was about 20 percent lower (1981a,b) examine the effects of calorie
than the productivity of nonanemic supplementation among sugarcane cut-
workers. In the experiment, workers ters in Guatemala, where assignment to
were randomly assigned to one of two groups was by village and not random-
groups (irrespective of their anemia ized within villages. Productivity rose in
status). The treatments were given a both the treatment villages and the con-
special iron supplement for 60 days; the trol villages. Because of the design, it is-
controls were given a placebo. Workers difficult to know whether this is be-
received an incentive payment to take cause calorie supplementation had no
the pills as scheduled.40 At the end of effect or because of changes in the vil-
the period, blood hemoglobin, aerobic lages that differed between treatments
capacity (measured by a step test), and and controls; that is, we cannot sepa-
output of those who were initially ane- rately identify an experimental effect
mic and received the treatment in- from village-time effects.42 This pitfall
creased to nearly the levels of the was avoided in a study of 47 Kenyan
nonanemic workers (whose biological road construction workers who were
indices did not change). Among those in randomly assigned to calorie supple-
the control group who were anemic, mentation or placebo (June Wolgemuth
productivity and blood hemoglobin lev- et al. 1982); calorie supplementation
els also rose, although the increase was
substantially smaller than among those 41 Behrman (1996) provides an excellent review
in the treatment group. (This is attrib- of the more general issue of links between health
and education of children. He concludes that
uted by Basta et al. to the effect of the "good health and nutrition have more nuanced
incentive payment, a claim which is cor- and qualified effects on schooling success than is
roborated by a comparison of dietary in- often recognized."
42A much-cited intervention of pregnant and
takes before and after the experiment: lactating women and their children in Guatemala
they point out that, during the experi- suffers from the same design problem. Partici-
ment, anemics in the control group spent pants in treatment villages were given a high calo-
rie drink (Atole) and controls were given a pla-
cebo. Perhaps the best evidence that the food
40 Nevertheless, attrition was substantial. This is supplement resulted in improved health outcomes
a concern that plagues all experiments and longi- and enhanced cognitive achievement is derived
tudinal data collection more generally. In their from a "difference-in-difference" type of esti-
comparisons of productivity, Basta et al. use only mates. Children who were in the poorest house-
77 of the 302 males enrolled in the experiment. holds (and thus were most malnourished) benefit-
The key question, to which we do not know the ted most from Atole, substantially reducing the
answer, is whether the attrition is random with re- gap with their better-off peers in the treatment
spect to unobservables that also affect produc- villages. No similar effect was observed in the con-
tivity. trol villages (Pollitt et al. 1993).

has a small positive impact on dirt dug data), by individual 24-hour recalls (or
per hour. In contrast, in the Indonesian averages of recalls) or by weighing food
iron-supplement study, productivity consumed. We discuss the evidence,
does not vary by calorie intake. There starting with intakes of calories and pro-
are several interpretations of these re- tein, and then draw in the role of body
sults. First, it may be that what appears size.
to be a beneficial effect of additional Several studies indicate that per cap-
calories in the Kenyan study is, in fact, ita household calorie availability has a
the effect of iron or other micronutri- powerful impact on farm output and on
ents. Second, it is possible that energy wages. For example, Strauss (1986) esti-
supplements are more important for mates that among households in Sierra
road diggers whose tasks are physically Leone, the elasticity of farm output
more strenuous than rubber tappers. with respect to calorie consumption of
Third, levels of protein-energy malnu- all family workers is 0.33.43 Sahn and
trition may be higher among the Ken- Alderman (1988) report that while
yan road diggers, in which case they will wages of women in rural Sri Lanka are
benefit more from additional calories, not responsive to calories, the wages of
assuming the calorie-productivity rela- men are: the elasticity is 0.2. Although
tionship is not linear. Fourth, several of the populations of both countries are
the food supplement studies that have extremely poor, and levels of health are
focused on children suggest that the very low in Sierra Leone, the estimated
benefits of protein and energy interven- effects in both settings are very large.
tions tend to emerge only after an ex- Based on the discussions in the pre-
tended intervention period. Finally, it is vious sections, there are three reasons
difficult to know how much of a calorie why the estimates may be biased. First,
supplement actually translates into if higher-income households tend to
higher intakes among the treatments. It waste more food (give more food away
is reasonable to suppose that the treat- or have more guests at their meals),
ments will consume less calories at other calorie availability will be an error-
meals, resulting perhaps in an increase ridden measure of intakes, and the
in calorie intakes by other family mem- error will be positively correlated with
bers. This suggests that individual- income (and therefore output or
specific food supplementation interven- wages). The estimated calorie effects
tions may be difficult to implement in an will be upward biased. Second, calories
experimental setting and is one dimen- may be proxying for other health indica-
sion in which nonexperimental house- tors with which they are correlated,
hold survey data has a clear advantage. such as body size. Since wages rise with
Nonexperimental studies of the im- stature (and perhaps mass), the esti-
pact that higher nutrient intake has on mated calorie effects in these studies
labor outcomes can also do better than will be upward biased. Third, calorie
studies of self-reported illness or availability is measured at the house-
health, because there are multiple ways hold level, but wages are recorded for
of measuring intakes and something is individuals; if calories are not shared
known about measurement error in
each case. This is important since the 43 This implies, for example, that a worker who
results differ substantially, depending consumes 1,500 calories daily (at the third decile
in the distribution) would be only 60 percent as
on whether intakes are measured by productive as a worker consuming 2,400 calories,
availability (from food consumption the sample average.

Health, Nutrition, and Econ0omicDevelopminent 805
Strauss and Thoma-as:
equally among all household members, specification, Haddad and Bouis report
the calorie effects will be biased, al- that individual-specific calorie intakes
thougchthe direction is a priori ambigu- have a large, positive impact on market
ous and depends on how resources are wages. The evidence in both studies is,
allocated within the household. We will therefore, consistent with the interpre-
discuss issues revolving around mnea- tation that, after controlling for unob-
surement error first. served heterogeneity, there is no effect
Instead of calorie availability, som-e of calorie intakes on wages. But, as
studies have used 24-hour food recalls Haddad and Bouis conjecture is likely,
measured at the individual level to esti- the evidence is also consistent with
mate nutrient intakes. As noted in Sec- measurement error in 24-hour recalls
tion 4, 24-hour recalls may suffer less swamping the impact of intakes on
from bias caused by income-related wages.
waste and transfers, but may suffer Alok Bhargava (1997) also uses calo-
mnorefrom random measurement error rie intakes based on 24-hour food re-
because of day-to-day variation in diets. calls, but he takes the average intake
For example, Deolalikar (1988) uses over seven consecutive days for the en-
24-hour food recall data to examine the tire household. This is important be-
effect of calorie intakes on market wage cause averaging is likely to reduce ran-
and farm production functions in rural dom measurement error. His result is
south India with data collected by the strikingly different: calorie intakes are a
International Crops Research Institute significant determinant of the prob-
for the Semi-Arid Tropics (ICRISAT). ability that a man in rural Rwanda
The effect of calorie intakes on wages engages in "strenuous" (and usually
in rural Philippines is estimated by higher-wage) activities.
Haddad and Bouis (1991). Both studies Calorie intakes based on weighing
exploit the longitudinal nature of their and measuring food prepared and food
data and include an individual fixed ef- wasted over seven days are likely to also
fect to control for all time-invariant un- minimize random noise. In addition,
observed heterogeneity that affects these estimates of intakes should not be
both intakes and wages or farm produc- affected by the biases that may arise in
tion. In both studies, the effect of calo- availability measures. This class of esti-
ries in that specification is small and mates is therefore probably subject to
not significant. However, calorie intakes the least measurement error of all. Us-
measured in this way are known to be ing these sorts of measures, Thomas
very noisy and the fixed-effect estimates and Strauss (1997) report that wages of
may well be biased downward by mea- workers in urban Brazil are positively
surement error. In fact, in the Indian and significantly affected by calories at
study, the estimated effect of calories is low intake levels.
positive and sizable, if imprecise, when Moreover, additional protein is also
the model replaces the fixed effect with associated with higher wages until rela-
a random effect.44 Similarly, in an OLS tively high levels of intakes. If calorie
intakes are held constant, then in-
44The random effects estimates may also be bi-
creases in protein intakes imply that the
ased because no correction is miade for correla- price (or quality) of calories increases.
tions between covariates (calories) and time- This is because a greater fraction of
invariant unobserved factors (such as height). It is
difficult to draw firm conclusions regarding the calories consumed will be rich in pro-
nature of the bias. tein which would typically mean a diet

with more meat and other animal prod- 1997). Moreover, higher BMI appears
ucts. When calories and protein are to raise the propensity of a worker to
both included in the wage function, the engage in work that is more physi-
relationship between protein and wages cally demanding (Bhargava 1997). Pitt,
switches from being concave to convex. Rosenzweig, and Hassan (1990) go a
This suggests that better-quality diets step further and ask whether "weight-
are not only associated with higher for-height endowment" is associated
wages but that there are increasing re- with labor outcomes. They report that
turns to diet quality in the labor market men with higher endowments are more
in urban Brazil. likely to engage in strenuous work in ru-
We think, therefore, that measure- ral Bangladesh, and that the average en-
ment error in nutrient intakes has im- dowment of all men in the household is
portant implications for understanding positively related to income.45
the links between intakes and labor out- To summarize our conclusions thus
comes. Whereas the magnitudes of the far, height does seem to have an impact
effects of calorie availability on wages on market wages, although whether this
may be upward biased (because of sys- represents health, strength or family
tematic measurement error), estimates background is not clear. Body mass
based on intake recalls for the previous seems to positively impact the produc-
24 hours seem to be downward biased tivity of men, at least those active in
(because of random error). Taking this physically demanding jobs. The effect
into account, we conclude that the bal- of nutrient intake is more uncertain.
ance of evidence points to a positive Micronutrient deficiency (particularly
effect of elevated nutrient intakes on iron) has a deleterious impact on pro-
wages, at least among those who are ductivity, and there is some evidence
malnourished. pointing to a positive impact when calo-
rie intake is well measured. All of these
5.4 Body Size and Labor Outcomes
conclusions refer to linear effects of
It is not obvious how to interpret the health on labor outcomes. There are
result that additional calories are associ- important additional insights to be
ated with higher productivity if higher gleaned from examining nonlinearities
productivity workers are stronger and in the effects of health, including inter-
consume more calories because they actions with the type of job, gender,
have higher needs. In the regressions
45"Endowments" are measured by residuals
based on calorie intakes discussed from a weight-for-height production function. The
above, height and either weight or BMI idea is that, after controlling for all inputs, the
are included in the regressions as prox- residual is a noisy measure of health end owment,
t + ej, in (1). For each individual, production
ies for body needs, so that the interpre- functions are estimated for four different periods
tation of calorie consumption is closer within a year, and so in any period, endowment
to net energy intakes. In addition, the measures in other periods are valid instruments.
Under the assumption that measurement error, el7,
effects of body size on productivity are is not correlated across time, these IV estimates
of interest in and of themselves. will be purged of measurement error. Of course, it
This section began by noting that is critical that all inputs be controlled in the pro-
duction function, since otherwise the residual will
height has a positive effect on wages. reflect those as well as endowment. This is very
Conditional on height, weight is also as- demanding of data in the case of weight-for-
sociated with higher wages or output in height, since it depends on past as well as current
health inputs; few surveys contain detailed infor-
many, but not all, studies (such as Deo- mation on past health inputs, and the survey used
lalikar 1988 and Thomas and Strauss by Pitt, Rosenzweig, and Hassan is no exception.

time of year, and the nature of the labor dence, however, is thin. In his analysis
contract. We turn to these issues next. of farm production in Sierra Leone,
Strauss (1986) finds evidence for mild
5.5 Nonlinearities in Relationships
concavity in the relationship between
Between Nutrition and Labor
labor efficiency and per capita calorie
Outcomes
availability in the household although
Most empirical work in this literature the turning point is around 5,000 calo-
has focused on establishing relation- ries, which seems implausibly high. In
ships between health and productivity. contrast, Thomas and Strauss (1997)
Subtleties regarding the form of these find a positive relationship between
relationships, and in particular the role wages and calories for low per capita in-
of nonlinearities, have seldom been ad- takes, which flattens out around 2,000
dressed. This is remarkable in view of calories per day and remains flat at
their prominence in theories of nutri- higher intakes. The differences between
tion efficiency wages, and also in light these results may reflect differences in
of evidence in the biomedical literature the nature of work in the two samples,
which suggests that thresholds in health since the heavy hoe-oriented agricul-
indicators may be important. For exam- ture dominant in rural Sierra Leone is
ple, as discussed above, mortality risk is more energy intensive than the average
elevated only among those with very urban job in Brazil. We suspect, how-
low or very high BMI, and a similar pat- ever, that the different results reflect
tern is observed in the relationship be- differences in measurement of calories.
tween BMI and other health indicators. There is too much spread in calorie
Consider nutrient intakes as a second availability in the Sierra Leone sample
example. It seems plausible that among to be plausible, with too many people
very poorly nourished populations, ad- apparently consuming 4,000 calories or
ditional energy intake may be associ- higher, and too many consuming 1,500
ated with greater energy output and or lower; Srinivasan (1992) pointed out
higher productivity, but that gain will the same problem in comparable Indian
diminish as intakes rise and may even NSS data. In the Sierra Leone and In-
decline when intakes become very high. dian samples, it seems very likely that
An additional but controversial argu- poor households receive more food
ment posits that the body is able to transfers and waste less than higher-
adapt to moderately low (or high) in- income households. In contrast, the
takes, and so there is no link between Brazil survey took special care to mea-
intakes and output over some range (ex- sure these leakages and wastage. Un-
cept over the short term because the derestimation of calorie intakes of
body has not had time to adjust); how- lower-income households is likely to
ever, once a lower threshold has been lead to a downward bias in the esti-
passed, adaptation mechanisms are mated effect of calories on wages at the
likely to be incomplete. The extent to bottom of the calorie distribution; while
which the body is able to successfully overestimates of intakes among those
adapt to nutritional stress is clearly an with higher incomes would lead to up-
empirical matter. ward-biased calorie estimates at the
We would argue, therefore, that top. The net effect would be to twist
there are good reasons to investigate the productivity-calorie relationship
the shape of relationships between counterclockwise.
health and labor outcomes. The evi- A number of other "nonlinearities"

808 Journal of EcononmicLiterature, Vol. XXXVI (June 1998)
have been explored in the literature and proxy for strength or aerobic capacity,
clarify our understanding of the rela- which are productive assets in the labor
tionship between dimensions of health market of men, and possibly women,
and productivity. First, the returns to when working in more menial jobs.46
health are likely to vary with the type of
work performed, with, for example, 5.6 Health and Constraints
strength being rewarded most in man- to Consumption Smnoothing
ual labor. To sidestep the fact that
wages and occupational choice are It has been possible to draw infer-
jointly determined, Thomas and Strauss ences regarding barriers to consump-
stratify the Brazilian sample by education smoothing by examining differ-
tion level under the presumption that ences in the impacts of health, and in
manual laborers have less education. consumption of health inputs, across
For the average urban Brazilian male, time (see Townsend 1995 for a general
body mass is associated with higher pro- discussion). The existence of such barri-
ductivity but that effect is significant ers would imply a socially inefficient
only among those with little education. level of investments in health inputs. To
For them, strength or aerobic capacity date, the stratification that has proved
is presumably more important at work most informative on this point, particu-
than for the better educated, who ap- larly in rural samples, distinguishes sea-
parently substitute brain for brawn as sons. For instance, using the ICRISAT
they climb the education ladder. Indian data, Behrman and Deolalikar
A second stratification that has (1989) find that calories have a signifi-
proved informative distinguishes health cant impact on wages during the peak
impacts by gender. Behrman and Deo- labor demand season (an elasticity of
lalikar (1989) find that body mass af- 0.29), but not during the slack season.
fects market wages of men in India, but While weight-for-height has a larger im-
that female wages are unrelated to body pact on wages in the off-peak season.
mass and calorie intake. Pitt, Rosen- This may reflect differences in tasks
zweig, and Hassan (1990) and Bhargava performed across seasons, but it may
(1997) find that while BMI positively also reflect the absence of perfect in-
influences the probability of engaging terseasonal consumption smoothing. If
in strenuous work for men, it has a body fat is used to store energy for later
much smaller effect on women. Pitt, use (rather than financial forms of sav-
Rosenzweig, and Hassan also find that a ing), then BMI should have a larger av-
higher average "health endowment" of erage impact during the off-peak sea-
males in the household is associated son.47 An interesting but unanswered
with increased household income, but question is whether using the body as a
not so for higher female "endowments." store of energy savings across seasons is
Finally, Thomas and Strauss (1997) 46 This gender difference in the link between
show that body mass has a positive ef- BMI and wages is in stark contrast with the asso-
fect on productivity only among those ciation between education and wages, which has as
women in Brazil who have no education high, or even higher, marginal returns for women
than men in niany developing economies (Schultz
at all (and are likely to be employed in 1993).
jobs such as domestic service); for Bra- 47 It is difficult to know whether this inference
zilian men, the impact of BMI declines is correct given the empirical result that peak la-
bor demand season does not correspond to a low-
with education. All of these results are income or "hungry"season (see Barbara Harriss
strongly suggestive that body mass is a 1995).

the most efficient method in the absence wages only if the employer can observe
of barriers to smooth consumption.48 either productivity or health (or some-
Other evidence based on drawing thing correlated with them that can be
comparisons across seasons provides used as a marker). The most convincing
even stronger evidence that consump- studies of this issue are of daily workers
tion smoothing is not complete. Using in Philippine agriculture (Foster and
data from Pakistan, Behrman, Foster, Rosenzweig 1993a, 1994). The impact
and Rosenzweig (1997) estimate an ex- of health on time wages is compared
plicitly dynamic model that distin- with its impact on piece rates, which
guishes harvest from planting seasons are presumably a good indicator of pro-
and takes into account the influence of ductivity. Exploiting the fact that both
assets and income on health, as well as payment schemes coexist for harvesting
the impact of health on farm profits. at the same time of year, and that some
Their harvest-season profit function is workers engage in both types of con-
conditioned on assets, harvest period tract simultaneously, Foster and Rosen-
prices, and planting period inputs. One zweig examine the effects of health on
of those inputs is planting period labor, differences in the implicit time wages
which is augmented by contemporane- between time and piece work for the
ous household per capita calorie intake. same worker. They are thereby able to
They find that while calories have a control for all individual unobserved
positive, concave impact on farm prof- heterogeneity and place the spotlight
its, the magnitude of the effect is small. on the role of observability of both pro-
However, an increase in per capita calo- ductivity and health.49 Calorie intakes
rie consumption throughout the plant- have a significantly larger impact on
ing season nearly pays for itself in terms piece rates than on time wages. Foster
of enhanced labor productivity. This and Rosenzweig (1993a) argue that em-
suggests that more calories ought to be ployers cannot directly observe nutrient
consumed during planting season. In- intakes on daily workers and so intakes
stead, calorie consumption is shown to are not fully rewarded in time-wage
rise during harvest season, which, to- contracts. Moreover, conditional on in-
gether with the first finding, strongly take, body mass has a significantly big-
suggests an absence of consumption ger effect on time wages than on piece
smoothing mechanisms that could be rates (Foster and Rosenzweig 1994),
used to raise calorie consumption dur- which is consistent with the interpreta-
ing planting season. tion that BMI serves in part as a signal
to employers for nutrient intakes.50
5.7 Observability of Productivity and
49One important difference is that time rates
of Health tend to be used in harvesting maize (mostly on
small farms), while piece rates tend to be used
We turn next to the issue of ob- more for sugarcane (on larger farms).
50 Consistent with this evidence, recall that
servability of health. It was argued Behrman and Deolalikar (1989) report that in In-
above that in the context of labor mar- dia calories have a significant impact on wages
kets with short-term contracts, even if during the peak season (harvest), but not at other
times. It is the case that piece rates dominate dur-
productivity is enhanced by good ing the harvest season in the ICRISAT villages,
health, it will be rewarded by higher while daily wages are more common at other
times. It should also be true that calorie intake
48 Given high food storage losses in some rural raises productivity among self-employed farmers
environments it is not so clear, see Philip Payne in the same villages; yet it does not (Deolalikar
(1989) for a detailed discussion and simulations. 1988).

These differences may, however, re- The ideas underlying this paper are
flect differences in the behavior of each clever, and provide excellent examples
worker as he or she chooses how much of how thinking carefully about the re-
effort to expend on time-rate and piece- lationship between health and labor
rate work. Unless monitoring is costless, outcomes yields novel empirical tests of
daily labor contracts have incentive ideas that have played a central role in
problems that may not arise in piece- many models of development and labor
rate schemes, when working one's own markets. However, as discussed at
land, or when contracts involve linkages length in Section 4, the credibility of
with other markets, such as sharecrop- results in this literature depends criti-
ping arrangements. Specifically, work- cally on the exclusion restrictions. One
ers are more likely to shirk and expend can certainly raise questions about
less effort when paid on a time basis. some of these. Height, season of the
Foster and Rosenzweig (1994) point out survey, and land owned are all included
that by comparing energy intakes and in the set of identifying instruments:
energy expenditures across different la- one may well argue that season (inde-
bor contracts, it is possible to assess the pendent of net energy intakes) may af-
extent to which there is shirking (or fect the biology of BMI. The latter ex-
moral hazard in the labor market). Ef- clusion is perhaps key, since, as Foster
fort is measured by energy expenditures and Rosenzweig tell us, piece rates are
which are deduced from changes in predominantly paid in the harvest sea-
body mass controlling for calorie intake. son. These concerns, however, may be
They estimate a body-mass production tempered by the fact that Foster and
function that depends on lagged body Rosenzweig find consistent results with
mass, on current individual energy in- different tests that make different as-
takes, and on whether the worker is ill. sumptions.
Controlling for these inputs, the impact In sum, the evidence may be viewed
on body mass of time worked in differ- as supporting the hypothesis that
ent labor contracts is examined.51 They worker effort and the dimensions of
report no significant differences in health that are hard to observe will have
worker effort between those who toil more impact on labor productivity when
more for piece rates and those working contracts are incentive compatible.
more time on their own land. But daily Costs that either give rise to imperfect
contract workers expend significantly information by employers or that im-
less effort than the others (that is, they pede the emergence of incentive-
lose less weight), and Foster and Rosen- compatible contracts thus induce social
zweig conclude that there is substantial inefficiencies regarding investments in
moral hazard associated with time-wage health, leading presumably to under-
employment. Corroborating evidence is investment in health inputs. This inter-
provided by the fact that calorie intake pretation, however, relies crucially on
is positively associated with piece-rate assumptions about the costs of monitor-
work or with working on one's own ing worker effort and observing health
farm, holding constant food prices, indicators. For example, contracts are
body mass, and illness. not immutable, and if day laborers are
not rewarded for good health, then one
51 BMI inputs, including the time allocation
would expect to see the development of
variables, are treated as endogenous, using height,
land owned, season, and age as identifying vari- sharecropping arrangements or longer-
ables. term wage contracts. This does not

Strauss and Thomas: Health, Nutrition, and EconomnicDevelopment 811
seem to characterize the rural labor is no evidence is that, in equilibrium,

market in Bukidnon, the area of the there is widespread starvation, and pre-
Philippines wherein the data were col- sumably death, in low-income settings.
lected. What does exist are contracts in While starvation and death does occur
which meals served are part of the re- in times of famine due to natural, politi-
muneration.52 By serving meals at the cal, or economic shocks, this is not an
job site, employers can be sure that a equilibrium condition.
minimum nutrient intake is reached and Nevertheless, a number of results in
thus reduce the observability problem the literature are consistent with the ef-
with respect to nutrient intakes.53 Fos- ficiency wage hypothesis. For example,
ter (1995) presents results indicating as noted above, people in better health
that more productive Filipinos work on (measured by BMI, for example), are
jobs that provide meals, suggesting that more likely to undertake strenuous
meals lead to higher productivity. The tasks (Pitt, Rosenzweig, and Hassan
generality of these results remains to be 1990; Bhargava 1997). But that evi-
established. For example, in urban Bra- dence is also consistent with a simple
zil the impact of nutrient intakes is ac- model of investment in human capital
tually stronger in the market sector without involuntary equilibrium unem-
relative to self-employment, suggesting ployment, as discussed above.55 Simi-
that either observability of effort or of larly, the efficiency wage hypothesis
health is not a key issue there.54 predicts that poor households will allo-
cate health inputs unequally among
5.8 Efficiency Wage Hypothesis: members to ensure that at least one
Empirical Evidence member is fit for work. There is evi-
Finally, we turn to the empirical im- dence that household resources are al-
plications of the nutrition efficiency located unequally toward workers,
wage hypothesis which, as mentioned, which is consistent with the hypothesis;
posits that there is a nonlinear link be- but the evidence is not consistent with
tween health and productivity and that the additional implication that workers
the function is convex at very low levels should receive preferential treatment in
of health. This implies that there may allocations. For example, in Bangladesh
be some people who are so poor and so men are more likely to be workers, and
unhealthy that they are too costly to be those with better health "endowments"
employed. incur the heaviest health tax in terms of
The evidence in support of the hy- within-family allocations (Pitt, Rosen-
pothesis is thin (see Bliss and Stern zweig, and Hassan 1990).56
1978b, and Rosenzweig 1988 for discus- Several studies indicate that poor
sions). One implication for which there health reduces labor force participation.
According to the efficiency wage hy-
52 Serving meals is fairly common in many rural
pothesis, the effect should be greatest
areas in low-income economies.
53 However, serving meals may result from other
considerations: Saving on transaction costs of 55 Recall that such an equilibrium is possible in
workers traveling long distances to and from home the Dasgupta-Ray model.
is one; using meals to build team spirit when team 56 They show that men who have better "endow-
production is involved is another. ments" tend to be fed more, relative to other
54 Glick and Sahn (1997) find the opposite result household members. However, the effect on
for the wages of urban male workers in Guinea. health of those better-endowed men working in
However, they also find that body mass of males more energy-intensive jobs outweighs the effect of
has a larger impact on market than on self- having larger calorie intakes. This represents a
employment wages. "tax"on the health of better-endowed men.

among the poorest. In most studies, it is only 36 percent greater at 5,000 calories
not; see, for example, Schoenbaum's than it would be at 2,500. Hence it
(1995) examination of older men and would appear unlikely that these wage
women in Taiwan. Dow et al. (1997) levels, plus low assets, are responsible
find that changes in health have a big- for large-scale involuntary unemploy-
ger impact on changes in participation ment among the poor. Moreover, there
among the poor in Indonesia, although is little evidence that involuntary unem-
they argue that this is likely to be be- ployment is widespread in rural India
cause health care price increases took (see the discussion in Rosenzweig 1988,
their greatest toll on the budgets of the for example). Within the Dasgupta-Ray
poorest. Again, the evidence is more framework discussed earlier, it would
consistent with a human capital inter- appear that the state of the world is
pretation of the effect of health on la- normally one in which equilibrium
bor market outcomes than with an effi- piece-rate wages are above the mini-
ciency wage interpretation. mum wage per unit of work (subject to
One might argue that standards of being above the reservation wage) of
living in Taiwan and Indonesia are too landless (or low-income) workers.
high to detect a labor market equilib- The efficiency wage hypothesis
rium with rationing based on low nutri- speaks to a market adjustment in re-
tion or health. India is a more promis- sponse to specific nonlinearities in the
ing context for finding such an relationship between health and pro-
equilibrium. However, Anand Swamy ductivity. It is important to recognize
(1997) has documented that rural wage that failure to find empirical support for
levels in much of India are simply too that hypothesis says nothing about
large to be consistent with there being a whether or not health affects produc-
nutrition-based constraint to labor force tivity at the individual level. As we have
participation.57 He shows, using several discussed, there is abundant evidence
data sets covering both casual and per- that a range of dimensions of health do
manent workers, that daily wages would reap a reward in the labor market.
purchase roughly 5,000 calories, assum- Moreover, there is some evidence indi-
ing the income is spent in the same way cating that households allocate their re-
an average poor household allocates its sources in response to the fact that poor
expenditures. While these calories health may impede some people from
might well be fed to multiple family engaging in certain types of work, and
members, the key point is that an un- that there are complex interrelation-
employed worker could offer to work ships between the nature and ob-
for half the wage, given in the form of servability of health and of the work
meals (which are often part of the wage performed.
payment). As long as the resultant labor It has been difficult to establish that
efficiency is greater than half that of poor health results in catastrophic labor
current workers, it would be cheaper market outcomes. However, moderate
for employers to hire that worker. Us- to severe shortfalls in health do appear
ing calorie-wage estimates from Behr- to have negative impacts on produc-
man and Deolalikar (1989), Swamy esti- tivity and to reduce labor supply. Not
mates that worker efficiency would be only have these patterns emerged in the
empirical literature summarized above,
57 Shankar Subramanian and Angus Deaton they also appear in the historical litera-
(1996) note the same fact, also using Indian data. ture. Fogel (1992), for example, inter-

prets the decline of European mortality bustness promise to be very profitable.

in the 19th anid 20th centuries as result- Many studies have been tripped up by
ing primarily from improvements in nu- inadequate data. Investments in im-
trition among the moderately under- proved data, particularly longitudinal
nourished. surveys with good measures of health
status, are likely to yield large returns.
6. Conclusion How to improve our health measure-
ments, while not constituting economic
There is a general consensus that in- analysis in itself, has very important
come has a strong effect on demand for ramifications for economic research.
health. Until recently, there was less Work remains to be done on which di-
agreement on the reverse relationship: mensions of health matter, which popu-
the effect of health on income or, more lation groups are most affected, and
generally, labor outcomes. Even fairly under what labor market (or other) con-
recent reviews of this literature conditions health emerges as important.
clude that there is little reliable evi- Questions of the impact of health dy-
dence that health has an important im- namics, particularly in response to
pact on labor productivity or labor use negative economic shocks and aging,
(Rosenzweig 1988). Today that assess- have barely been touched in low-in-
ment would have to be amended: In re- come environments. The same is true of
cent years, substantial progress has the important issue of whether ineffi-
been made in documenting the exist- ciencies exist in household and individ-
ence of a causal impact of health on ual health input allocations, given the
wages and productivity in low-income estimated productive impacts. Thus, we
settings using both experimental and conclude that, while considerable pro-
nonexperimental methods. A small gress has been made in recent years, re-
number of studies suggest that health search on the interactions between la-
has a larger return at very low levels of bor markets and health remains a very
health and (perhaps) in jobs requiring promising area in the study of develop-
more strength. With economic developing economies.
ment, these types of jobs will shrink,
and one might expect the labor market REFERENCES
impact of improved health to decline, ACTON, JAN. "Nonmonetary Factors in the De-
especially relative to the impact of edu- mand for Medical Service: Some Empirical Evi-
dence," J. Polit. Econ., June 1975, 83(3), pp.
cation and skill acquisition. 595-614.
However, the critical reader will ALTONJI, JOSEPH. "The Effects of High School
agree that teasing these relationships Curriculum on Education and Labor Market
Outcomes," J. Human Res., Summer 1995,
out of data has proved to be very diffi- 30(3) pp. 409-38.
cult. Health and income clearly affect ANDRIANO, RALPH AND HELMINIAK, THOMAS.
each other and are related to many fac- "Economics, Health and Tropical Diseases: A
Review," in Economics, health and tropical dis-
tors that are hard to measure. In the eases. Eds.: ALEJANDRO HERRIN AND PATRI-
nonexperimental literature, interpreta- CIA ROSENFIELD. Manila, Philippines: U. of
tions of associations between health, the Philippines, School of Economics, 1988,
pp. 19-72.
wages, and income are not unambigu- ANTLE, JOHN AND PINGALI, PRABHU. "Pesticides,
ous. But analyses of these sorts of data Productivity and Farmer Health: A Philippine
that are judicious in their choice of as- Case Study," Amer. J. Agr. Econ., Aug. 1994,
76(3), pp.418-30.
sumptions, thoughtful in the choice of ASHENFELTER, ORLEY AND KRUEGER, ALAN.
estimation, and sensitive to issues of ro- "Estimates of the Economic Return to School-

ing From a New Sample of Twins,"Amer. Econ. Too High? A Recalibration of the Plausible
Rev., Dec. 1994, 84(5), pp. 1157-73. Range," J. Devel. Econ., Oct. 1992, 39(2), pp.
BASTA, SAMIR ET AL. "Iron Deficiency Anemia 333-64.
and Productivity of Adult Males in Indonesia," BOUND, JOHN. "Self-Reported Objective Measure
Amer. J. Clinical Nutrition, Apr. 1979, 32(4), of Health in Retirement Models," J. Human
pp. 916-25. Res., Winter 1991, 26(1), pp. 106-38.
BECKER, GARY. "A Theory of the Allocation of CHIAPPORI, PIERRE-ANDRE. "RationalHousehold
Time," Econ. J., Sept. 1965, 75(299), pp. 493- Labor Supply," Econometrica, Jan. 1988, 56(1),
517. pp. 63-89.
BEHRMAN, JERE R. "Intrahousehold Allocation - . "Collective Labor Supply and Welfare,"J.
of Nutrients in Rural India: Are Boys Fa- Polit. Econ., June 1992, 100(3), pp. 437-67.
vored?" Oxford Econ. Pap., Feb. 1988, 40(1), CLEVELAND, WILLIAM S. "Robust Locally-
pp. 32-54. Weighted Regression and Smoothing Scatter-
. "Intrahousehold Allocation of Nutrients plots," J. Amer. Statist. Assoc., Dec. 1979,
and Gender Effects: A Survey of Structural and 74(368), pp. 829-36.
Reduced Form Estimates," in SADDIQ R. Os- DASGUPTA, PARTHA. An inquiry into well-being
MANI 1992, pp.287-321. and destitution. Oxford:Oxford U. Press, 1993.
-. "The Economic Rationale for Investing in - . "Nutritional Status, the Capacity for Work
Nutrition in Developing Countries," World De- and Poverty Traps,"J.Econometrics, Mar. 1997,
vel., Nov. 1993, 21(11), pp. 1749-72. 77(1), pp.5-38.
- "The Impact of Health and Nutrition on DASGUPTA, PARTHA AND RAY, DEBRAJ. "Inequal-
Education," World Bank Res. Observer, Feb. ity as a Determinant of Malnutrition and Un-
1996, 11(1), pp. 23-37. employment: Theory," Econ. J., Dec. 1986,
. "Intrahousehold Distribution and the 96(384), pp. 1011- 34.
Family," in Handbook of population and family - . "Inequality as a Determinant of Malnutri-
economics, Vol. IA, Eds.: MARK R. ROSEN- tion and Unemployment: Policy," Econ. J., Mar.
ZWEIG AND ODED STARK. Amsterdam: North 1997, 97(385), pp.177-88.
Holland Press, 1997, pp. 125-87. DEOLALIKAR, ANIL B. "Nutrition and Labor Pro-
BEHRMAN, JERE R. AND DEOLALIKAR, ANIL B. ductivity in Agriculture: Estimates for Rural
"Health and Nutrition," in Handbook of devel- South India," Rev. Econt. Statist., Aug. 1988,
opment economics, Vol. 1. Eds.: HOLLIS CHEN- 70(3), pp. 406-13.
ERYAND T. N. SRINIVASAN.Amsterdam: North Dow, WILLIAM ET AL. "Health Care Prices,
Holland Press, 1988, pp. 631-711. Health and Labor Outcomes: Experimental Evi-
. "AgriculturalWages in India: The Role of dence." Labor and Population Program Work-
Health, Nutrition and Seasonality,"in Seasonal ing Paper Series 97-01, RAND, Santa Monica,
variability in third world agriculture. Ed.: CA, Jan 1997.
DAVID SAHN. Baltimore: Johns Hopkins U. FALKNER, FRANK AND TANNER, JAMES M. Hu-
Press, 1989, pp. 107-17. man growth: A comprehensive treatise, Vol. 3.
BEHRMAN, JERE R., FOSTER, ANDREW AND New York:Plenum Press, 1986.
ROSENZWEIG,MARK R. "The Dynamics of Ag- FENWICK, A. AND FIGENSCHOU, B. M. "The Ef-
ricultural Production and the Calorie-Income fect of Schistosoma Mansoni on the Produc-
Relationship," J. Econometrics, Mar. 1997, tivity of Cane Cutters on a Sugar Estate in Tan-
77(1), pp. 187-208. zania," Bull. World Health Organ., Sept. 1972,
BENJAMIN, DWAYNE. "Household Composition, 47(5), pp. 567-72.
Labor Markets, and Labor Demand: Testing FOGEL, ROBERT "Second Thoughts on the Euro-
for Separation in Agricultural Household Mod- pean Escape From Hunger: Famines, Chronic
els," Econometrica, Mar. 1992, 60(2), pp. 287- Malnutrition and Mortality Rates," in SADDIQ
322. R. OSMANI 1992, pp. 243-86.
BESLEY, TIMOTHY AND CASE, ANNE. "Incumbent FOGEL, ROBERT. "Economic Growth, Population
Behavior: Vote Seeking, Tax Setting and Yard- Theory and Physiology: The Bearing of Long-
stick Competition," Amer. Econ. Rev., Mar. Term Processes on the Making of Economic
1995, 85(1), pp. 25- 45. Policy,"Amer. Econ. Rev., June 1994, 84(3), pp.
BHARGAVA, ALOK. "Nutritional Status and the 369-95.
Modeling of Time in Rwandese Households,"J. FOLBRE, NANCY. "Market Opportunities, Genetic
Econometrics, Mar. 1997, 77(1), pp. 277-95. Endowments and Intrafamily Resource Distri-
BLISS, CHRISTOPHER. AND STERN, NICHOLAS. bution: Comment," Amer. Econ. Rev., June
"Productivity,Wages and Nutrition: Part I: The 1984, 74(3), pp. 518-20.
Theory,"J. Devel. Econ., Dec. 1978a, 5(4), pp. FOSTER, ANDREW. "Nutrition and Health Invest-
331-62. ment," Amer. Econ. Rev., May 1995, 85(2), pp.
- -. "Productivity,Wages and Nutrition: Part 148-52.
II: Some Observations,"J. Devel. Econ., Dec. FOSTER, ANDREW AND ROSENZWEIG, MARK R.
1978b, 5(4), pp. 363-98. "Information Flows and Discrimination in La-
BouIs, HOWARTH E. AND HADDAD, LAWRENCE bor Markets in Rural Areas in Developing
J. "Are Estimates of Calorie-Income Elasticities Countries," Proceedings of the WVorld
Bank An-

nual Conference oni Development Economics Through Household Surveys." Mimeo, Centre
1992, Mar. 1993a, pp. 173-204. for Population Studies, School of Hygiene and
-. "Information, Learning and Wage Rates in Tropical Medicine, U. of London, 1989.
Low-Income Rural Areas," J. Human Res., Fall HIRSCHMAN, CHARLES; PRESTON, SAMUEL AND
1993b, 28(4), pp. 759-90. Loi, Vu MANH. "Vietnamese Casualties During
---. "A Test for Moral Hazard in the Labor the American War," Population Devel. Rev.,
Market: Effort, Health and Calorie Consump- Dec. 1995, 21(4), pp. 783-812.
tion," Rev. Econ. Statist., May 1994, 76(2), pp. IMMINCK, MAARTEN AND VITERI, FERNANDO.
213-27. "Energy Intake and Productivity of Guatemalan
. "Comparative Advantage, Information and Su garcane Cutters: An Empirical Test of the
the Allocation of Workers to Tasks: Evidence Efficiency Wage Hypothesis-Part 1," J. Devel.
From the Agricultural Labour Market," Rev. Econ., Oct. 1981a, 9(2), pp. 251-71.
Econ. Stud., July 1996, 63(3), pp. 347-74. - . "Energy Intake and Productivity of Guate-
FRANKENBERG, ELIZABETH. "The effects of ac- malan Sugarcane Cutters: An Empirical Test of
cess to health care on infant mortality in Indo- the Efficiency Wage Hypothesis-Part 2," J.
nesia," Health Transitions Review, 1995, 5(2), Devel. Econ., Oct. 1981b, 9(2), pp. 273-87.
pp. 143-62. JAMISON, DEAN ET AL., eds. Disease control pri-
FRIEDMAN, GERALD. "The Heights of Slaves in orities in developing countries. Oxford: Oxford
Trinidad," Soc. Sci. Hist., Fall 1982, 6(4), pp. U. Press, 1993.
482-515. JOHN, MEREDITH. "Plantation Slave Mortality in
GATEFF, C. ET AL. "Influence de la Bilharziose Trinidad," Population Stud., July 1988, 42(2),
Vesicale sur la Rentabilite Economique d'une pp. 161-82.
Population Adult Jeune du Cameroon," Annales LAVY, VICTOR; PALUMBO, MICHAEL AND STERN,
de la Societe Belge de Medicin Tropical, 1971, STEVEN. "Health Care in Jamaica: Quality, Out-
51(3), pp. 309-24. comes, and Labor Supply." Living Standards
GERSOVITZ,MARK. "Savings and Nutrition at Low Measurement Study Working Paper No. 116,
Incomes," J. Polit. Econ., Oct. 1983, 91(5), pp. World Bank, 1995.
841-55. LEIBENSTEIN, HARVEY. Economic backwardness
GERTLER, PAUL AND MOLYNEAUX, JACK. "Pricing and economic growth: Studies in the theory of
Public Health Services: Results From a Social economic development. New York: Wiley &
Experiment in Indonesia." Mimeo, RAND, Sons, 1957.
Santa Monica, CA, 1996. LoPEZ, ALAN. "Causes of Death in Industrial and
GERTLER, PAUL AND VAN DER GAAG, JACQUES. Developing Countries: Estimates for 1985-
The tvillingness to pay for medical care: Evi- 1990," in DEAN JAMISONET AL. 1993, pp. 35-
dence from two developing countries.. Balti- 50.
more: Johns Hopkins U. Press, 1990. LUNDBERG, SHELLY AND POLLAK, ROBERT.
GLICK, PETER. AND SAHN, DAVID. "Health and "Separate Spheres Bargaining and the Marriage
Productivity in a Heterogeneous Urban Labor Market,"J. Polit. Econ., Dec. 1993, 101(6), pp.
Market." Mimeo, Cornell U., Mar. 1997. 988-1010.
GLOMM, GERHARD AND PALUMBO, MICHAEL. McELROY, MARJORIE AND HORNEY, JEAN.
"Optimal Intertemporal Consumption Deci- "Nash-bargained Household Decisions: Toward
sions Under the Threat of Starvation,"J. Devel. a Generalization of the Theory of Demand,"
Econ., Dec. 1993, 42(2), pp. 271-92. Int. Econ. Rev., June 1981, 22(2), pp. 333-49.
GRILICHES, ZVI. "Estimating the Returns to MARGO, ROBERT AND STECKEL, RICHARD. "The
Schooling: Some Econometric Problems," Heights of American Slaves: New Evidence on
Econometrica, Jan. 1977, 45(1), pp. 1-22. Slave Nutrition and Health," Soc. Sci. Hist.,
GROSSMAN, MICHAEL. "On the of
Concept Fall 1982, 6(4), pp. 516- 38.
Health Capital and the Demand for Health,"J. MARTORELL, REYNALDO AND ARROYAVE,
Polit. Econ., Apr. 1972, 80(2), pp. 223-55. GUILLERMO. "Malnutrition, Work Output and
HADDAD, LAWRENCE J. AND BouIs, HOWARTH Energy Needs," in Capacity for work in the
E. "The Impactof Nutritional Status on Agri- tropics. Society for the Study of Human Biology
cultural Productivity:Wage Evidence From the Symposium 26. Eds.: K. COLLINS AND F.
Philippines," Oxford Bull. Econ. Statist., Feb. ROBERTS. Cambridge: Cambridge U. Press,
1991, 53(1), pp. 45-68. 1988, pp. 57-76.
HARRISS, BARBARA. "Gender Bias in Intrahouse- MAZUMDAR, DIPAK. "The Marginal Productivity
hold Nutrition in South India: Unpacking Theory of Wages and Unemployment," Rev.
Households and the Policy Process," in Intra- Econ. Stud., June 1959, 26(3), pp. 190-97.
household resource allocation in developing MIRRLEES, JAMES. "A Pure Theory of Underde-
countries: Methods, models and policy. Eds.: veloped Economies," in Agriculture in develop-
LAWRENCE HADDAD, JOHN HODDINOTT, AND ment theory. Ed.: LLOYD G. REYNOLDS. New
HAROLD ALDERMAN. Baltimore: Johns Hop- Haven: Yale U. Press, 1975, pp. 84-108.
kins U. Press, 1997. MUNDLAK, YAIR. "Estimation of Production and
HILL, ALLAN AND MAMDANI, MASUMA. "Opera- Behavioral Functions From a Combination of
tional Guidelines for Measuring Health Time Series and Cross Section Data," in Mea-

suremnentin economics: Studies in mathematical tional Deficiency," Nutritional Reviews, 1997,
economics and econometrics. Ed.: CARL CHRIST 55(4), pp. 133-40.
ET AL. Stanford: Stanford U. Press, 1963, pp. PRESCOTT, NICHOLAS. "Schistosomiasis and De-
138-66. velopment," World Devel., Jan. 1979, 7(1), pp.
NELSON, CHARLES R. AND STARTZ, RICHARD. 1-14.
"The Distribution of the Instrumental Variables ROSENZWEIG, MARK R. "Labor Markets in Low
Estimator and Its t-ratio When the Instrument Income Countries," in Handbook of develop-
is a Poor One," J. Bus., Jan. 1990, 63(1) Part 2, ment economics, Vol. 1. Eds.: HOLLIS CHEN-
pp. 125-40. ERY AND T. N. SRINIVASAN.Amsterdam: North
NEWHOUSE, JOSEPH AND THE INSURANCE EX- Holland Press, 1988, pp. 713-62.
PERIMENT GROUP. Free for all? Lessons from ROSENZWEIG, MARK R. AND SCHULTZ, T. PAUL.
the RAND Health Insurance Experiment. Cam- "Market Opportunities, Genetic Endowments
bridge: Harvard U. Press, 1993. and Intrafamily Resource Distribution: Child
OSMANI, SIDDIQ R., ed. Nutrition and poverty. Survival in Rural India," Amer. Econ. Rev.,
Oxford: Clarendon Press, 1992. Sept. 1982, 72(4), pp. 803-15.
OVER, MEAD, ET AL. "The Consequences of Adult ROSENZWEIG, MARK R. AND WOLPIN, KENNETH
Ill-Health," in The health of adults in the devel- I. "Evaluating the Effects of Optimally Distri-
oping world. Eds.: RICHARD FEACHEM ET AL. buted Public Programs:Child Health and Fam-
Oxford: Oxford U. Press, 1992, PP. 161-208. ily Planning Interventions," Amer. Econ. Rev.,
PARSONS, DONALD. "The Male La or Force Par- June 1986, 76(3), pp. 470-82.
ticipation Decision: Health, Reported Health - . "Migration Selectivity and the Effects of
and Economic Incentives," Economica, Feb. Public Programs,"J. Public Econ., Dec. 1988,
1982, 49(193), pp. 81-91. 37(3), pp. 265-89.
PAYNE, PHILIP. "Public Health and Functional SAHN, DAVID AND ALDERMAN, HAROLD. "The
Consequences of Seasonal Hunger and Malnu- Effects of Human Capital on Wages, and the
trition," in Seasonal variability in Third World Determinants of Labor Supply in a Developing
agriculture: The consequencesfor food security. Country,"J. Devel. Econ., Sept. 1988, 29(2), pp.
Ed.: DAVID SAHN. Baltimore: Johns Hopkins U. 157-83.
Press, 1989, pp. 19-46. SCHOENBAUM,MICHAEL. "The Health Status and
PINSTRUP-ANDERSEN, PER ET AL. "Protein- Labor Force Behavior of the Elderly in Tai-
energy Malnutrition," in DEAN JAMISONET AL. wan." Comparative Study of the ElderTyin Asia,
1993, pp. 391-420. Research Report No. 95-32, Population Studies
PITT, MARK, M. "Specification and Estimation of Center, U. o Michigan, 1995.
the Demand for Goods Within the Household," SCHULTZ, T. PAUL. "Returns to Women's Educa-
in Intrahousehold resource allocation in devel- tion," in Women's education in developing
oping countries: methods, models and policy. countries: Barriers, benefits and policies. Eds.:
Eds.: LAWRENCE HADDAD, JOHN HODDINOTT, ELIZABETH M. KING AND M. ANNE HILL. Bal-
AND HAROLD ALDERMAN. Baltimore: Johns timore: Johns Hopkins Press, 1993, pp. 51-99.
Hopkins U. Press, 1997. - . "Human Capital and Economic Develop-
PITT, MARK M. AND ROSENZWEIG, MARK R. "Ag- ment," in Agricultural competitiveness: Market
ricultural Prices, Food Consumption and the forces and policy choices. Proceedings of the
Health and Productivity of Indonesian Farm- 22nd International Conference of Agricultural
ers," in INDERJIT SINGH, LYN SQUIRE, AND Economists. Eds.: GEORGE H. PETERS AND
JOHN STRAUSS 1986, pp. 153-82. DOUGLAS D. HEDLEY. Aldershot: Dartmouth,
. "Estimating the Intrahousehold Incidence 1995, pp. 523-39.
of Illness: Child Health and Gender-Inequality --. "Wage Rentals for Reproducible Human
in the Allocation of Time," Int. Econ. Rev., Capital: Evidence from Two West African
Nov. 1990, 31(4), pp. 969-80. Countries." Mimeo, Department of Economics,
PITT, MARK M., ROSENZWEIG, MARK R. AND Yale U., 1996.
HASSAN, MD. NAZMUL. "Productivity, Health SCHULTZ, T. PAUL AND TANSEL, AYSIT. "Wage
and Inequality in the Intrahousehold Distri- and Labor Supply Effects of Illness in Cte
bution of Food in Low-Income Countries," d'Ivoire and Ghana,"J. Devel. Econ., Aug. 1997,
Amer. Econ. Rev., Dec. 1990, 80(5), pp. 1139- 53(2), pp. 251-86.
56. SINGH, INDERJIT, SQUIRE, LYN AND STRAUSS,
PITT, MARK M., ROSENZWEIG, MARK R. AND JOHN, eds. Agricultural household models: Ex-
GIBBONS, DONNA. "The Determinants and tensions, applications and policy. Baltimore:
Consequences of the Placemnent of Government Johns Hopkins U. Press, 1986.
Programs in Indonesia," World Bank Econ. SPURR, G. B. "Nutritional Status and Physical
Rev., Sept. 1993, 7(3), pp. 319-48. Work Capacity," Yearbook of Physical Anthro-
POLLITT, ERNESTO ET AL. Early supplementary pology, 1983, 26, pp. 1-35.
feeding and cognition. Monographs of the Soci- - . "Body Size, Physical Work Capacity and
ety for Research on Children, Vol. 58, No. 7, Productivity in Hard Work: Is Bigger Better?"
Chicago: U. of Chicago Press, 1993. in JOHN C. WATERLOW1988, pp. 215-44.
POLLITT, ERNESTO. "Iron Deficiency and Educa- SRINIVASAN, T.N. "Undernutrition: Concepts,

Measurements and Policy Implications," in SID- of Adult Health," J. Hu.man Res., Fall 1993,
DIQ R. OSMANI 1992, pp. 97-120. 28(4), pp. 791-837.
-. "Destitution: A Discourse," J. Econ. Lit., SUBRAMANIAN,SHANKARAND DEATON, ANGUS.
Dec. 1994, 32(4), pp.1842-55. "The Demand for Food and Calories,"J. Polit.
STAIGER, DOUGLAS AND STOCK, JAMES. "Instru- Econ., Feb. 1996,104(1), pp. 133-62.
mental Variables Regression With Weak Instru- SWAMY,ANAND. "A Simple Test of the Nutrition-
ments," Econometrica, May 1997, 65(3), pp. Based Efficiency Wage Model,"J. Devel. Econ.,
557-86. June 1997, 53(1), pp. 85-98.
STECKEL, RICHARD. "Stature and the Standard of THOMAS, DUNCAN. "Intra-Household Resource
Living," J. Econ. Lit., Dec. 1995, 33(4), pp. Allocation: An Inferential Approach,"J. Human
1903-40. Res., Fall, 1990, 25(4), pp. 635-64.
STEWART, ANITA AND WARE, JOHN. Measuring THOMAS, DUNCAN AND STRAUSS, JOHN. "Health
functioning and well-being: The Medical Out- and Wages: Evidence on Men and Women in
comnes Study. Durham, NC: Duke U. Press, Urban Brazil," J. Econometrics, Mar. 1997,
1992. 77(1), pp. 159-86.
STEWART, ANITA ET AL. Conceptualization and TOWNSEND, ROBERT. "Consumption Insurance:
measurement of health status for adults in the An Evaluation of Risk-Bearing Systems in Low-
Health Insurance Study: Vol. II. Physical health Incoim-e Economies," J. Econ. Perspectives,
in terms of functioning. R-1987/2-HEW, Summer 1995, 9(3), pp. 83-102.
RAND, Santa Monica, CA, 1978. WAALER, HANS. "Height, Weight and Mortality:
STIGLITZ, JOSEPH. "The Efficiency Wage Hy- The Noiwegian Experience," Acta Medica
pothesis, Surplus Labor and the Distribution of Scandi.navia, 1984, supplement no. 679.
Income in LDCs," Oxford Econ. Pap., July WARE, JOHN, DAVIEs-AVERY, ALLYSON AND
1976, 28(2), pp. 185- 207. BROOK, ROBERT. Conceptualitation and mea-
STRAUSS, JOHN. "Does Better Nutrition Raise surement of' health status for adults in the
Farm Productivity?" J. Polit. Econ., Apr. 1986, Health Insurance Study: Vol. VI. Analysis of re-
94(2), pp. 297-320. lationships among health status measures. R-
-- . "The Impact of Improved Nutrition on 1987/6-HEW, RAND, Santa Monica, CA, 1980.
Labor Productivity and Human Resource De- WARE, JOHN; DAVIES-AVERY, ALLYSON AND
velopment," in The political economy of food DONALD, CATHY. General health. perceptions.
and nutrition policies. Ed.: PER PINSTRUp-AN- R-1987/5-HEW, RAND, Santa Monica, CA,
DERSEN. Baltimore: Johns Hopkins Press, 1993, 1978.
pp. 149-72. WATERLOW,JOHN C., ed. Linear growth retarda-
STRAUSS, JOHN AND THOMAS, DUNCAN. "Human tion in less developed countries. Nestle Nutri-
Resources: Empirical Modeling of Household tion Workshop Series Vol. 14, New York: Raven
and Family Decisions," in Handbook of develop- Press, 1988.
ment economics, Vol. 3A. Eds.: T. N. SRINI- WATERLOW, JOHN C. ET AL. "The Presentation
VASAN AND JERE BEHRMAN. Amsterdam: and Use of Height and Weiglht Data for Com-
North Holland, Press, 1995a, pp. 1883-2023. paring the Nutritional Status of Groups of Chil-
. "Health, Nutrition and Economic Devel- dren Under the Age of 10 Years," Bulletin of
opment." Labor and Population Program Work- the World Health OrganiZation, July 1977,
ing Paper Series 95-23, RAND, Santa Monica, 55(4), pp. 489-98.
CA, 1995b. WOLGEMUTH, JUNE ET AL. "Worker Productivity
-- . "Measurement and Mismeasurement of and the Nutritional Status of Kenyan Road Con-
Social Indicators," Amner.Econ. Rev., May 1996, struction Laborers,"Amiier.
J. Clinical Nutrition,
86(2), pp. 30-34. July 1982, 36(7), pp. 68-78.
STRAUSS, JOHN ET AL. "Gender and Life-Cycle WORLD BANK. World Developmnent Report 1993,
Differentials in the Patterns and Determinants Oxford: Oxford U. Press, 1993.


American Economic Association

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

American Economic Association

Uploaded by

Copyright:

Available Formats

American Economic Association

Health, Nutrition, and Economic Development

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

Health, Nutrition, and Economic

1. Introduction opment. This paper reviews some of the

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

tence of health-based efficiency wages. problems, some of which stem from

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

177 UNITED STATES 164 170 COTE D'IVOIRE

BRAZIL 1. AdultStatur VIET

Figure 1. Adult Statureby BirthCohort

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

Birth Cohort: 1925-55 1956-70

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

1).C SEu ,,2 4 Age 55-64

160 170 180 190 160 170 180 190

160 1710 180 190 150 160 1710 180

Figuire 2. Wages, Education, anidHeight of Males in Braziland the United States

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

in the body and expended when A

ments in incomes or prices, so that the :4 1- United States 2 i

dynamics linking them, BMI and pro-

Furthermore, current BMI partly re- = .5 1.5 a

by the magnitude in Brazil, where the BMI kg/rn,2

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

H = H(N,L; A,B',D,t,eh) (1) puts (such as body mass) may affect

teristics, A, such as gender; in addition Discussions of how to interpret esti-

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

distributional consequences if, as is puts, such as health care in the public

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

example, in a conditional self-employ- namic, health-conditioned profit func-

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

economies, short-term, even daily, con- employed.27 If healthier people also

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

be systematically related to income (and functional consequences of ill health

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

on self-reported height and weight, % able to carrya heavy load easily

We have argued above that because * 98- 98

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

This content downloaded from 192.231.202.205 on Wed, 3 Dec 2014 11:17:36 AM

In contrast, however, a similar experi- survey data in which general health,