Socioeconomic Status and Health

by James P. Smith
Citation
Title:
Socioeconomic Status and Health
Author:
James P. Smith
Year: 
1998
Publication: 
The American Economic Review
Volume: 
88
Issue: 
2
Start Page: 
192
End Page: 
196
Publisher: 
Language: 
English
URL: 
Select license: 
Select License
DOI: 
PMID: 
ISSN: 
Abstract:

Socioeconomic Status and Health 

The quantitatively large association be- tween many measures of socioeconomic status (SES) and a variety of health outcomes ap- pears pervasive over time and across countries at quite different levels of economic develop- ment (Evelyn Kitagawa and Philip Hauser, 1973; Richard G.Willunson, 1996). But many analytical difficulties exist in trying to under- stand its meaning, including the complex di- mensionality of health status which produces considerable heterogeneity in health out- comes, the two-way interaction between health and economic status, and the separation of anticipated from unanticipated health or economic shocks.

I present here new evidence on these issues using the first three waves of the Health and Retirement Survey (HRS), a representative national sample of 7,702 households (12,652 individuals) containing a person born between 193 1 and 1941. The baseline was fielded dur- ing 1992-1993 with follow-ups at two-year intervals. HRS collects excellent information on respondents' employment, income, and wealth (F. Thomas Juster and Smith, 1997). Many different aspects of respondents' health are also measured, including self reports of general health status, the prevalence and in- cidence of chronic conditions, the extent of functional limitations, and out-of-pocket and total health-care expenditures. The sample used here consists of those 10,236 respondents interviewed in each of the first three waves. All nominal values are expressed in October 1997 dollars.

I. Health and Wealth Transitions

There exists a strong positive association between levels of household income or wealth

" RAND, 1700 Main Street, Santa Monica, CA 90407. The expert programming assistance of Iva Maclennan and David Rumpe] is gratefully acknowledged, This was supported by grant ~01-~~08291

and grant ~01- AG12394 from the National Institute of Aging. 192

and self-reporred health status. For example, Smith and Raynard Kngton (1997b) report that HRS respondents in excellent health have

2.5 times as much household income and five tirnes as much household wealth as respon- dents in poor health. Table lA documents that levels and changes in wealth are also come- lated with reported changes in health. Respon- dents whose reported health status was worse in wave 3 generally had much lower levels of baseline household wealth. The average dif- ference in median wealth for each single threshold health change is $26,176, amounting to two-thirds of average household income.

While the patterns are not as smooth in panel B of Table 1, median wealth changes between waves 1 and 3 are also correlated with health transitions with an average wealth change of almost $4,400 per health transition. But this table says nothing about the principal direction of any causal pathway. An economic shock, such as a reduction in wealth, may well affect health outcomes over the long term, but the immediacy of the effects in Table 1 draws attention to the pathway from health to eco- nomic status.

One explanation for any cross-wave re- lationship is that respondents with the same self-reported wave-l health may be quite heterogeneous in their baseline health. If so, any initial heterogeneity becomes partly re- vealed in subsequent waves, so that those who report a lower health status by waves 2 and 3 were actually less healthy in wave 1than those who did not report any health change in sub- sequent waves. If such heterogeneity is im-- portant, part of the association of wealth with changing health status may reflect baseline heterogeneity in wealth and health.

Table 1C examines this possibility by list- ing a baseline index of mean functional limi- tations by the joint distribution of general health status in HRS waves 1 and 3. A value

in this index no lim

itations, and a higher score indicates worse function (Kington and Smith, 1997). The

TABLE1-WEALTH AND HEALTHSTATUS

Wave-3 health

Wave-1 health Excellent Very good Good Fair Poor

-

A. Median Baseline Wealth ($1,000'~):

Excellent 232 212 126 NA NA Very good 176 178 136 102 NA Good 140 143 109 74 36 Fair NA 81

76 57 31

Poor NA NA NA 30 24

B. Median Wealth Change ($1,000'~):

Excellent 31.7 17.2 10.9 
Very good 25.2 18.8 12.6 
Good 14.2 17.9 12.3 
Fair NA 0.6 2.9 
Poor N A NA NA 

C. Mean Wai~e-1 Functional Starus:

Excellent 5 6 6 
Very good 7 8 10 
Good 9 10 12 
Fair 14 20 19 
Poor 20 36 35 

hrote: NA = less than 1 percent of total sample

patterns are striking. Respondents whose re- ported health status declined between the waves had more functional limitations in wave 1.The range of variation is quite large. Among those in "fair" health in wave 1, baseline scores range from as low as 14 ("excellent" health in wave 3) to as high as 28 ("poor" health in wave 3). For comparison, there is less than a five-point spread between the bot- tom and top 20 percent of the income distri- bution. When stratified only by self-reported health status across waves, the remaining vari- ation in respondents' health is a serious ana- lytical problem. This argues that modeling must incorporate the multidimensionality of health status.

Another difficult modeling issue concerns how much of any health change is actually news to the respondent. Economists are now familiar with the conceptual necessity and em- pirical complexity inherent in separating out new information or "shocks" in time-series income processes. Similarly, based on currently available information about their current health stock and some prognosis about its fu-

TABLE2-PREVALENCI~

AND INCIDENCI? OF CHRONIC

CONDITIONS

-

-

New incidence

Baseline Waves Waves Condition mevalence 1-2 2-3

-

34.3 3.7 3.7

Hypertension 1.5 2.1

EEs

8.6

4.8 1.1 1.5 Lung disease 6.4 1.5 1.4 Heart condition 11.1 2.3 2.8 Stroke 2.1 0.6 0.9 Arthritis 35.4 6.5 6.7

ture, individuals make uncertain projections about their future health states. These health trajectories contain predictable age-related components, surrounded by consiiderable individual-level heterogeneity, all of which are updated with the realization of new, and often unpleasant, information about one's health.

One hope of isolating new health informa- tion lies in the onset of new chronic condi- tions. While people may anticipate some onset (e.g., smokers may think they will get cancer), the actual realization, and especially its timing, may be unanticipated. Table 2 presents base- line prevalence rates alongside the percentage of new incidence observed between each pair of successive waves. The pattern of disease prevalence is consistent with that revealed in other sources for this age group. While hyper- tension and arthritis are particularly common conditions, prevalence rates are nst trivial, even for more serious ailments. For example, one in nine HRS respondents has a heart con- dition, and 5 percent have experienced cancer. Given the two-year window between waves, incidence rates for most conditions are rela- tively small. While most respondents experi- ence only one new condition, more than one in three report at least one new chronic con- dition since wave 1. Twenty-three percent of HRS respondents had at least one mild new onset, and 12 percent had at least one severe new onset. Severe conditions were defined as cancer, heart condition, stroke, and diseases of the lung.

The four years spanned by the three HRS waves were marked by considerable health ac- tivity. During these years, one-third of all

TABLE~-OUT-OP'-POCKETMEDICAL

EXPEXDITURESBETWEEN WAVES1 AND 3

-

Medical expenditures ($)

Severe Mild new new No new chronic chronic chronic Percentile condition condition condition

10 32 49 22 30 793 434 358 50 1,985 1,072 868 70 4,399 2,255 1,833 90 11,659 6,324 4,774 95 17,108 9,489 7,983 98 31,601 18,322 15,452

respondents were hospitalized at least once, and 5 percent of those hospitalized spent at least one month there. Less than half of all hospitalizations were fully insured, so many respondents experienced some out-of-pocket expenses. Added on to these hospital trips were doctor and dentist visits, outpatient sur- gery, and drugs. Virtually all respondents vis- ited a doctor at least once, with most visits involving some or complete co-pay. Similarly, I6 percent had outpatient surgery, 60 percent of which involved some respondent payment.

One way health shocks can affect wealth ac- cumulation involves medical exDenses associ- ated with new health events. To explore this possibility, Table 3 lists distributions of out-of- pocket (OOP) medical expenses separately for those respondents with new chronic conditions. Compared to respondents who had no onset at all, the median increase in OOP expenses for a severe onset was only $1,117. ~hese incremental OOP costs were significantly higher in the tails ($7,000 at the 90th percentile and more than $16,000 at the 98th percentile). While the typical impact of these health events on 80P costs are modest, there are nontrivial probabil- ities that the impact might be much larger. Risk aversion and attitudes toward uncertainty then become key parameters in modeling behavioral responses tobOP costs.

Additional medical expenses are not the only way health shocks can affect wealth accumu- lation. Most directly, healthier people may be able to work more, leading to higher earnings. Reduced savings or the depletion of past asset

OOP Total Medical medical Net financial expenses expenses worth ($) assets ($) ($) ($1

Severe Onset

Wave 2 -20.927 -9,911 2,305 28,916

Wave 3 -22,973 --3,700 2,226 26,825

Miid Onset

Wave 2 -7,542 -10,228 476 2,312

Wave 3 --I7 --1,276 737 4,244

accumulations may be the preferred first step to cope with episodes of poor health. To estimate the impact of the onset of new chronic condi- tions, a parallel set of models were estimated predicting total and financial wealth accumu- lation, and total and OOP medical expenses all measured between waves I and 3. In light of the discussion above, in addition to a rather standard set of other covaiates, all models iacluded variables designed to capture baseline heterogeneity in health. Evaluated at wave 1, these health variables were self-reports of gen.- erd health status, the extent of functional lim- itations, the prevalence of chronic conditions, and whether the respondent and spouse had health insurance. New chronic conditions were separated into their severe and mild variants, and separate estimates were obtained for onset between each set of waves.

Table 4 summarizes estimated mean effects on these four outcomes. While there is legiti- mate concern about respondents5 ability to re- port medical expenses (especially those costs they do not pay), the ordering of estimated impacts on medical costs are reasonable. For. example, coefficients on total medical ex- penses are about nine times larger for severe compared to mild conditions. While severe on- sets impose nontrivial total medical costs (above $25,000), the impact is considerably muted in OOP expenses where even severe on- sets have only about a $2,250 effect. At least in this age group, the availability of private health insurance significantly softens the irn- mediate financial blow from a health shock.

In spite of these muted effects on OOP expenditures, the estimated effects on wealth ac- cumulation are not trivial. While new-onset coefficients exhibit much more variability as- sociated with the precise model specification, the average reduction in total wealth due to a severe onset is about $22,000 (8 percent of average baseline wealth). The estimated total- wealth effects associated with a new mild onset are less stable but considerably smaller. Table 4 also tentatively suggests that, while decrements in financial assets can account for the bulk of any adjustments required by the onset of mild conditions, new severe condi- tions may require adjustments in other parts of the household wealth portfolio. Since in- creased OOP medical expenses appear to ex- plain only a small part of the reason that new health shocks reduce wealth accumulation, the reasons must lie elsewhere. A plausible expla- nation is the labor-supply-induced decreases in household income due to these new onsets.

Table 4 indicates that there are quantitatively significant effects of health on at least one mea- sure of SES: household wealth. Earlier work (Smith and Kington, 1997a, b)based on HRS and AHEAD (respondents at least 70 years old) demonstrated that, at least in older populations, the correlation between health and current- period householdincome mostly reflected cau- sation from health to SES rather than the reverse. However, these strong feedbacks from health status to SES do not deny that there may also be a direct influence of SES on health. Good health is an outcome that people desire, and higher wealth or income enables them to purchase more of it. Similarly, a number of risk factors such as smoking and obesity are more prevalent among those in lower SES groups. However, research indicates that health-care utilization and behav- ioral risk factors can only explain a small part of the observed association of SES and health (Wilkinson, 1996; Smith and bgton, 1997b).

These findings have led some researchers (especially those associated with the British Whitehall studies) to suggest alternative ways through which SES may affect health. One in- triguing hypothesis is that societal-level in- equality has a direct influence on health outcomes. There exists a highly nonlinear re- lation between health and such measures of SES as income and wealth (Smith and Kington, 1997a, b) , strongest among those with relatively few resources, weakening among the middle-class, anti almost nonexist- ent among the affluent. While this nonlinearity would itself produce an association between resource inequality and heal-th, the hypothesis goes beyond individual-level nonlinearity. A common theme is that inequality in relative rank raises levels of psychosocial stress which negatively affects endocrine and immunolog- ical processes. In developed countries, it is not material deprivation thatmatters, but the stress associated with being at the bottom end of an unequal social pecking order. A frequent sup- porting citation is Robert Sapolsky's (1993) study which indicates that low-ranking male baboons have higher levels of glucocorticoids, apparently a reaction to the chronic stress they experience by their low status. Glucocorti- coids are steroid honnone:~ released during stress. Chronic elevated levels during pro- longed stress may negatively affect indivldu- als' return to normal functioning.

This hypothesis is important because it pro- vides a direct biological rationale for the rea- sons why SES may have significant long-run impacts on health status. If true, it raises con- cerns, given the growing levels of income in- equality experienced by many countries in the last few decades. However, the strength of the empirical evidence supporting it remains un- convincing, partly due to the frequent failure to control for the nonlinearity in the individual- level association between SES and health or the significant reverse causation between health and SES documented in this paper.

Another modeling complication is that health status at middle and older ages may reflect health at earlier stages of life, even back to childhood (David J. P. Barker, 1992), and may also be correlated across generations. Smith and Kington ( 1997a) demonstrate that health outcomes at quite old age (70f) are correlated with health attri- butes of past, concurrent, and fu.ture genera- tions of relatives. Whether this correlation reflects shared genetic endowments or the cu- mulative impact of common social, eco- nomic, and geographic endowments remains an unresolved research question. Similarly, Anders Forsdahl's (1997) study off regional differences in Norway indicates a stronger impact of childhood than adult poverty on the prevalence of coronary heart disease. Since early health outcomes can affect subsequent decisions such as schooling, marriage, and earnings, it is inappropriate simply to use these outcomes to explain individual varia- tion in current health. Instead, it is necessary to model these feedback mechanisms explic- itly and to isolate within-period innovations in the stock of health. While such a research agenda is easier said than done, progress on understanding the critically important rela- tion between SES and health requires it.

HI. Conclusions

There is unlikely to be a single winner in the continuing dispute regarding the dual pathways between SES and health. The direct influence of SES on health may be strongest during child- hood and early adulthood when levels and tra- jectories of health stocks become established. Moreover, economic "shocks" may dominate health "shocks" among those in their twenties or thirties as levels of lifetime earnings are de- termined. The dominant causal pathway may then reverse, as health largely affects SES among those age 50 and older. For example, this paper presents new evidence that new health events have quantitatively large effects on wealth accumulation among those in their fifties. After age 40, the big information people receive is not about their changing economic circumstances, but rather about their overall health and its implications for their eventual mortality and ability to function effectively in old age. Studies that ignore the large impacts that health status can have on SES are simply missing a major part of the story.

REFERENCES

Barker, David J. P. ' Tetal and Infant Origins of Adult Disease-The Womb May Be More Important than the Home." British Medical Journal, 17 November 1990, 301(6761), p. 1111.

Forsdahl, Anders. "Are Poor Living Gondi- tions in Childhood and Adolescence an Important Risk Factor for Arteriosclerotic Heart Disease?" British Journal of Pre- ventive Social Medicirze, June 1997, 31(2), pp. 91-95,

Juster, F. Thomas and Smith, James P. "'Improving the Quality of Economic Data: Lessons from HRS and AHEAD." Journal of the American Statistical Associai'ion,December 1997,92(440), pp. 1268-78.

Kington, Raynard and Smith, James Po "Socioeconomic Status and Racial and Ethnic Dif- ferences in Functional Status Associated with Chronic Diseases." American Journal of Public Health, May 1997, 87'(5), pp. 805-16.

Kitagawa, Evelyn and Hauser, Philip, Diferential nzortali~ in. the United States: A study in socioeconomic epidemiology. Cambridge, MA: Harvard University Press, 1973.

Sapolsky, Robert, '6Endocrinology Afresco: Psychoendocrine Studies of Wild Baboons." Recent Progress in Horraone Research, 1993,48, pp. 437.-68.

Smith, James P. and Mington, Raynard, "'Demographic and Economic Correlates of Health in Old Age." Demography. February 1997a, 34(1), pp. 159-70.

-. "Race, Socioeconomic Status and Health in Late Life,"' in Linda Martin and Beth Soldo, eds., Racial and ethnic di#eer- ences in the health of older Americans.Washington, DC: National Academy Press, 199'7b, pp. 106-62.

Wilkinson, Richard G. Unhealthy societies: The afflictions of inequality. London: Rout-ledge, 1996.

Comments
  • Recommend Us