Professional Documents
Culture Documents
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 The
American Economic Review.
http://www.jstor.org
Aging and Inequality in Income and Health
In our previous work, Deaton and Paxson independent of economic status. T'here is a
(1994, 1997), we showed that, in a large well-documented bujtpoorly understood "gra-
group of countries, inequality in consumption dient" lin ng socioeconomic status to a wide
increases with age within cohorts of indi- range of health outcomes (see Nancy E. Adler
viduals. This finding was motivated by a et al. [1994] and Sally Maclntyre [1997] for
well-known feature of standardautarkicinter- reviews). The gradient has both a life-cycle
temporal choice models, that under appropri- and a temporalcomponent;differences in mor-
ate assumptions consumption follows a tality across socioeconomic groups are widest
martingale (see Robert E. Hall, 1978). The in late middle age Evelyn M. Kitigawa and
theory implies that within-cohortconsumption Philip M. Hauser, 1973; Harriet Orcutt
inequality should rise over time as cohorts age, Duleep, 1995; Inna T. Elo and Samuel H.
provided that shocks to consumption are not Preston, 1996) and are increasing over time
perfectly correlated across individuals. The (Jacob J. Feldman et al., 1989; G. Pappas et
same should be true of income, at least up to al., 1993; Preston and Elo, 1995).
the date of retirement,and of earnings, if em- In our earlier work, we used data from the
ployers pay workers their expected marginal NationaalHealth Interview Survey (NHIS) to
product (see Henry S. Farber and Robert examine life-cycle patternsin health status and
Gibbons, 1996). in the joint distribution of health status and
More recently we have examined whether income (Deaton and Paxson, 1998). In this
inequality in health status also increases with paper we summarize and extend those results
age, and how the joint distribution of health and provide new evidence fromi the Panel
and income evolve over the life cycle. It is Study of Income Dynamics (PSID). Both sur-
plausible that health shocks have both per- veys contain a measure of household income
manent and transitorycomponents. The pres- and collect informationon an ordinal measure
ence of the former implies that health status of self-reported health status (SRHS) that
will be nonstationaryso that, provided health ranges from 1 (excellent) to 5 (poor).
shocks are not perfectly correlated across in-
dividuals, the dispersion of health status will 1. MeasurementIssues
grow with age. This view of health status as a
nonstationary random variable is consistent The measurementof health inequalityraises
with stress models in which poor health is the two importantissues. The first is the difficulty
result of "the piling up of adverse life expe- of identifying a measure of health status that
riences" (Carol D. Ryff and Burton Singer, is useful over the complete adult life cycle. For
1997 p. 90). example, measures of the inability to complete
Health status, along with income and con- 66activities of daily living" (ADL's), such as
sumption, is an importantdeterminantof wel- dressing or bathing, have been fruitfully used
fare, so that our interest in health inequality to assess the health of the elderly. However,
stems from a more general interest in the dis- these measures do not adequately capture
tributionof welfare. Furthermore,health is not health differences among younger people.
Self-reported "days of illness" or "doctor
visits" are themselves conditioned by socio-
* Research Program in Development Studies, Prince-
economic status and sometimes show perverse
correlations with income, with better-off peo-
ton University, PrincetonNJ 08544. The researchwas sup-
ported by the National Institute of Aging through grants ple apparentlyperceiving and treatingtheirill-
PO1-AG05842 and ROI-AGI 1957. We thank our discuss- nesses more seliously. The properties of the
ant. James Poterba. for useful commenits. measure of self-reDortedhealth status used in
248
VOL. 88 NO. 2 DEMOGRAPHICTRENDSAND ECONOMIC CONSEQUENCES 249
this paperhave been studied extensively. First, life expectancy, for example, one that prefers
it predicts subsequent mortality. A large num- a decrease in infant mortality to an increase in
ber of studies that use data from a variety of longevity at older ages. Of course, to focus
countries indicate that reports of poor health solely on life expectancy ignores the quality
are significantly related to subsequent mortal- of life. SRHS may well give some indication
ity (see E. L. Idler and S. V. Kasl [1995] for of quality as well as the likely length of life,
summary of this research). The correlation so that changes in the distiibution of SRHS
between SRHS and subsequent mortality could still have adverse welfare consequences
remains strong even after controlling for ob- even in the absence of a relationshipbetween
jective measures of health status (obtained SRHS and mortality. We also note that much
from doctors' examinations) and life-style fac- of the literatureon health inequalityis inotcon-
tors such as smoking. This fact has led some cerned with inequality in years lived, but with
to argue that SRHS is itself an independent the inequalities in health outcomes across so-
determinant of longevity: individuals with cioeconomic groups. That these are quite dif-
healthier self-images live longer. An alterna- ferent has been emphasized by Richard G.
tive to these psychosocial explanations is that Wilkinson (1986) who points out that, over
individuals have information about their the 20th century in Britain, socioeconomic dif-
health that is unobserved by others, including ferences in mortality have increased while the
physicians. For our purposes, it is the raw cor- inequality in years lived has decreased, es-
relations between self-rated health and mor- sentially because of the decline in infant
tality that are of interest, since we want to mortality.
identify a variable that can serve as a single
summary measure of health status. Other re- II. Evidenceon Life-CyclePatternsin the
search has found that those with low SRHS Distributionof Healthand Income
are more likely to develop problems with
ADL's (Idler and Kasl, 1995) and miss more The NHIS is an annual survey of approxi-
work due to illness (M. Marmot et al., 1995). mately 50,000 adults (plus children) that col-
Once a measure of health is identified, the lects information on health, illnesses, doctor
second issue is how to measure inequality in visits, spells of hospitalization, and basic so-
health status. Although it is straightforwardto cioeconomic characteristics. We use data on
compute measures of dispersion in SRHS, it is all adults between the ages of 20 and 70, in-
not clear how we should judge such measures clusive, interviewed from 1983 through 1994.
in terms of social welfare. Consider, for ex- The survey provides sampling weights, which
ample, the familiar result that, if a distribution we use, so that the results should be represen-
F, (second-order) stochastically dominates a tative of the whole U.S. population. The PSID
distribution F2, then F1 will result in higher is a panel survey of households that has been
social welfare, when social welfare is repre- in existence since 1968, and since 1984 it has
sented as the integral over the population of a collected information on the self-reported
monotone increasing and concave function of health status of household heads and their
the variable in question. Although we are used spouses. We use a sample of 3,435 men and
to the assumption that social welfare is in- 4,561 women who were either heads of house-
creasing and concave in income or consump- holds or their spouses in all years between
tion, it is much less clear why it should be 1984 and 1992. Unlike the NHIS, this is not a
increasing and concave in an ordinal self- nationally representative sample of all adults,
reported measure of health status. Neverthe- both because it is only households heads and
less, the literature on SRHS provides some spouses, and because the PSID oversampled
supportfor the idea thatchanges in SRHS have poor households in 1968. Given these circum-
a larger effect on mortality when SRHS is stances, we did not use sampling weights with
"poor" than when it is "excellent." If so, a the PSID.
mean-preserving spread in SRHS will lower Although the NHIS has much larger sam-
average life expectancy and will lower the ex- ple sizes and more extensive health infor-
pected value of any function that is concave in mation than the PSID, it has far less
250 AEA PAPERS AND PROCEEDINGS MAY 1998
Females Males
3.5 - 0.2 -
= 3.0-
V 2.5 -
= O~S0
-0.4-
2.0-
Q 1.5 - -0.6 I _
20 40 60 80
0.5 -
Females
0.2-
20 40 60 80
O-
~0-
-0.2
-
0.2 A
?
-0.4 - -0.4
20 40 60 80 -0.6 -
Age
20 40 60 80
FIGURE 2.
HEALTH STATUS, THE VARIANCE OF HEALTH
Age
STATUS, AND THE CORRELATION BETWEEN HEALTH
PSID
AND INCOME FOR COHORTS IN THE PSID, FEMALES
NHIS -
There are several possible interpretationsof per year of birth. However, a visual inspec-
these results, none of which necessarily tion of the cohort effects indicates that they
excludes any other. One is that labor supply are not linear. There has been no improve-
and eamings ability are adversely and cumu- ment, and possibly some deterioration, in
latively affected by health shocks, so thatpoor health status across cohorts born after 1945,
health and low income are increasingly cor- and there were larger improvements across
related with age. The correlationmay weaken those born before 1945. The results for the
in old age, since health shocks received after males in the PSID are at odds with the other
retirementwill not affect pensions and Social groups. The estimate of the cohort trend in-
Security (although they could affect asset in- dicates that more recently born groups are
come if sick people run down assets to pay for significantly less healthy, by about 0.009
care). There is also undoubtedly some causal- units per year of birth. This is largely due to
ity runningfrom income to health. Poorerpeo- declines in reported health, controlling for
ple are more prone to lifestyles with enhanced age, of the youngest four cohorts. These de-
risk factors (e.g., obesity or cigarette smok- clines, which can be seen in the raw data
ing), have less access to health care, including graphed in the top panels of Figures 1 and 2,
preventative health care, and live and work in warrantfurtheranalysis. Second, for all of the
less healthy environments. There is also a lit- samples, younger cohorts have a lower vari-
erature documenting the adverse health con- ance of health status.
sequences of unemployment. The provision of Third, the results provide some support for
Medicare at older ages could reduce the cor- the findings cited above that the gradient be-
relation by making one determinantof health, tween income and health is becoming stronger
medical care, available to everyone. Sorting over time. The coefficient on the cohort trend
out the respective contributions of these vari- ranges from 0.001 (for females in the NHIS)
ous mechanisms remains an importanttask for to 0.003 per year (for males in the PSID), a
future research. positive sign indicating that for more recently
Perhaps even more important than life- born cohorts there is a larger correlation be-
cycle patterns is the question of changes over tween income and health. For example, the ac-
time in the relationship between income and tual correlation between the logarithm of
health status. It is difficult to discern cohort income and SRHS is -0.40 for PSID males
effects, represented by upward or downward born between 1940 and 1944, when they
shifts in the traces for different cohorts, from reached the age of around 50 in 1992. Our re-
a visual inspection of Figures 1 and 2. How- sults imply that the correlation for the cohort
ever, a more systematic approach shows that born ten years later, between 1950 and 1954,
there are significant differences across co- will equal -0.43 when this group reaches the
horts. We first examined the cohort effects age of 50.
from regressions of each of the variables
(mean health, the variance of health, and the REFERENCES
correlation of health and income) on sets of
age and cohort dummy variables. This was Adler, N. E.; Boyce, T.; Cohen,S.; Folkman,S.;
done separately for men and women, and for Kahn,R. L. and Syme,S. L. "Socioeconomic
the PSID and the NHIS. The estimated cohort Status and Health: The Challenge of the
effects are jointly significant at the 5-percent Gradient."American Psychologist, January
level or better for each of the variables and 1994, 49(1), pp. 15-24.
samples. To summarize the size and sign of Deaton,AngusS. and Paxson,ChristinaH. "In-
these cohort effects, we then regressed each tertemporalChoice and Inequality." Jour-
of the variables on a complete set of age dum- nal of Political Economy, June 1994,
mies and a linear cohort trend. The results are 102(3), pp. 437-67.
as follows. First, for females from both data . "'TheEffects of Economic and Pop-
sets and males in the NHIS, average health ulation Growth on National Saving and In-
status has improved over time across cohorts. equality." Demography, February 1997,
The effect is small: approximately0.004 units 34(1), pp. 97-114.
VOL. 88 NO. 2 DEMOGRAPHICTRENDSAND ECONOMIC CONSEQUENCES 253