Use the Table of Summaries to link to the summaries either by author or by risk factor.
Abstract | Bibliography | Introduction | Table of
Summaries
Predictors of Mortality Among Elderly
Hurd, Michael, Daniel McFadden, and Angela Merrill, 1999. NBER Working Paper Series,
Working Paper 7440, National Bureau of Economic Research, December 1999. †
In this paper, the authors propose that individuals are likely to have additional
subjective information about their own life chances. Some of this information may be related to observable
characteristics such as health status and socioeconomic status but some may be a result of behavior and
personal characteristics that are not easily measured. The objective of this study was to determine how
important socioeconomic indicators, health indicators and subjective probabilities of survival are as
predictors of mortality among the population aged 70 or over. In particular, the authors were interested in
evaluating if subjective perception has predictive power for mortality both unconditionally and conditionally
on the observable characteristics of SES and health indicators.
The socioeconomic indicators considered were income, wealth and education. Wealth was the
total of that from financial, business and housing/real estate wealth, but not pension wealth. Income
included all financial income, including pension income, but no flow from owner occupied housing. Education
was measured by years of schooling. Thirteen health indicators were considered: heart disease/attack,
cancer, stroke, high blood pressure, diabetes, lung disease, arthritis, incontinence, hip fracture, a fall
requiring treatment, restrictions in activities of daily living (ADL), cognitive impairment and depression.
Respondents were questioned if they had ever suffered from any of the first 10 of these. Limitations of
activities of daily living (ADL), cognitive impairment and depression were assessed using a battery of
questions. Subjective perceptions were measured by asking respondents to give an estimate of their survival
chances to a target age, approximately 12 years in the future. Zero indicated absolutely no chance of
survival and 100 indicated absolute certainty.
The results are based on mortality between waves 1 and 2 of the Asset and Health Dynamics
among the Oldest–Old study (AHEAD). This was a biennial panel study of individuals born in 1923 or
earlier. At baseline in 1993, the study was representative of the community–based population, except
for oversamples of blacks, Hispanics and Floridians. The age group under study (those aged 70 or over) has
not been studied to the same extent as younger populations. As the population is almost completely retired,
a confounding effect of health on income via work status is practically eliminated. In addition, almost the
entire population is covered by Medicare, therefore significantly reducing the causal pathway linking SES to
mortality via access to health care services.
Wealth, income and educational attainment were all tested individually for an effect on
mortality. In each case, there was evidence of differential mortality but the differences decreased with age.
Thus verifying that SES is related to mortality and that the relationship declines with age. Controlling for
age is important as wealth and income fall with age. The health indicators were found to be strong
predictors of mortality and the subjective survival probabilities predicted mortality even after
controlling for the socioeconomic and health indicators. The relationship between SES and mortality
disappeared when health status was controlled for suggesting that any differential relation between access
to health care services and SES is small.
Eight of the thirteen health variables were significantly associated with mortality and
also a reduction in the subjective survival probabilities. Therefore, as subjective survival probabilities
are associated with health conditions, which are in turn associated with mortality, this would imply that
subjective survival probabilities do predict actual mortality. It was also found that in all age bands, the
subjective survival probabilities (to the same target age) increased between the waves, as would be expected
and these increases were reasonably close to those in the life–table probabilities, except in the
oldest age group. This was a reflection of the overestimation of subjective survival probabilities among the
oldest age group.
As mentioned, the average survival probability declined with age, but there was little
systematic variation in the survival probability as a function of wealth, income or education. Also, few
differences were found between males and females for the effects of SES or health conditions on mortality.
However, males were more likely to overstate their survival chances.
The onset or incidence of a new health condition was found to reduce the subjective
survival probability, but the effect was still substantial. This was indicated by a number of conditions, in
particular, cancer, high blood pressure, diabetes and depression. However, the onset of ADL limitations
increased subjective survival probabilities. The authors provide no explanation for this increase.
The authors note that individuals may misperceive their survival chances and prior work
had indicated that unrealistic stated subjective survival probabilities were associated with low cognitive
performance. Subjective survival probabilities were also overstated at older ages. Despite these recognized
errors in responses, there remains considerable heterogeneity.
As more detailed cross–tabulations of the mortality correlates was not practical,
data–descriptive probit estimation was used. This provided broadly similar results to the cross–
tabulations; again, showing the subjective survival probability is a powerful predictor of mortality. In
addition, it was identified that married respondents had lower mortality rates than singles and that there
was no differential effect of marital status for men compared with women; that is, there was no additional
mortality protection for men.
A significant problem in this study was the rather high rate of non–response to the
survival probabilities. A substantial number of interviews were by proxy and so the question of subjective
survival probability was not asked. This could have important effect on the level and variation in the
subjective survival probabilities. Interviews by proxy were due to the frailty of cognitive impairment of
the subject. This is likely to explain the elevated mortality experienced by this group.
The results of this study suggested that subjective survival probabilities have
considerable explanatory power for mortality. An earlier study, The Health and Retirement Study (HRS) also
found that a subjective perception of health is a significant predictor of mortality and that it varied
appropriately with known risk factors. However, there remains considerable heterogeneity in subjective
survival probabilities, and although this study identifies effects, it makes no attempt to determine
causality. This would be the next step in developing this research, but the authors acknowledge that there
are problems in identifying the subjective variables that individuals use in making their choices and also
some personal characteristics are not easily measurable.
† This paper is based on data from the United States.