Barry Bosworth Gary Burtless and Kan Zhang The Brookings Institution 16th Annual Joint Conference of the Retirement Research Consortium August 78 2014 Mortality differentials by social and economic status ID: 760035
Download Presentation The PPT/PDF document "Sources of Increasing Differential Morta..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Sources of Increasing Differential Mortality among the Aged by Socioeconomic Status
Barry Bosworth, Gary Burtless and Kan Zhang
The Brookings Institution
16th Annual Joint Conference of the Retirement Research Consortium
August 7-8, 2014
Slide2Mortality differentials by social and economic status
At a given age, death rates are higher for folks with low SES
SES measured by income, earnings, or education
Mounting evidence mortality gap is growing
Goal of study: Use HRS data to find reason
Evidence in HRS of growing SES differential?
Causes of death that explain growing gap?
Can growing differences in health-related behavior (smoking, exercise) account for the gap?
Slide3Expected age of death among white men and women attaining age 45 in the National Longitudinal Mortality Study, 1979-1985
Women
Men
Source:
Rogot
,
Sorlie
, and
Johnson (1992).
Slide4Health & Retirement Study
Cohorts spanning birth years 1890-1965
We examine deaths occurring in 1992-2010
Sample includes almost 32,000 aged & near-aged Americans
Of whom more than 11,500 died between 1992-2010
HRS data file contains range of info on SES
Educational attainment & current income
For almost 2/3 of sample, Social Security earnings record
Also: Health status, health-related behaviors, parents’ life spans
Slide5Our measures of SES
Educational attainment
Less than high school diploma
College degree or more
Actual
average nonzero earnings (ages 41-50)
Based on reported earnings in Social Security record
Combined husband-wife earnings adj. for family size
Predicted
average nonzero earnings (ages 41-50
)
Regression predictions explained by education, race/ethnicity, disability, marital status
In current paper, we use earnings estimate to predict R’s position in income distribution: Top or bottom half.
Slide6Age-specific mortality rates observed in RHS sample:
1992-2010
Born in 1925
Born in 1935
Born in 1945
Source:
Tabulations of HRS mortality data.
Slide7Age-specific mortality rates observed in RHS sample compared with SSA estimates (2005)
Born in 1925
Born in 1935
Born in 1945
SSA estimates
Source:
Tabulations of HRS and SSA (2005).
Slide8Age-specific mortality rates observed in RHS sample:
1992-2010
Born in 1925
Born in 1935
Born in 1945
Source:
Tabulations of HRS mortality data.
Slide9Age-specific mortality rates observed in RHS sample compared with SSA estimates (2005)
Born in 1925
Born in 1935
Born in 1945
SSA estimates
Source:
Tabulations of HRS and SSA (2005).
Slide10Does socioeconomic status affect mortality in the HRS?
To find out we use discrete-time logistic model to estimate the influence of risk factors linked to mortality:
Age
Race / ethnicity
Marital status
Alternative measures of SES
All measures of SES are linked in expected way with age-specific mortality rates
Low SES boosts mortality; High SES reduces it
.
Slide11Estimated mortality rates of low and high predicted earners, by age
Predicted low earner
Predicted high earner
Source:
Tabulations of HRS mortality data.
Slide12Estimated mortality rates of low and high predicted earners, by age
Predicted low earner
Predicted high earner
High earner born in 1935
Source:
Tabulations of HRS mortality data.
Slide13Does the impact of socioeconomic status on mortality grow in successive birth cohorts?
Using a simple discrete-time logistic model to estimate the influence of SES in “Early” and “Later” birth cohorts:
“Early cohorts” = Born between 1915-1930
“Later cohorts” = Born between 1931-1942
Restrict sample to respondents born in 1915-1942
Restrict sample to observations for these respondents when they were between ages 68-79
What is impact of predicted income in top half of income distribution in “Early” vs. “Later” cohorts?
Slide14Mortality rates of low and high predicted earners, by age in “Early” cohort
Predicted low earner born before 1931
Predicted high earner born before 1931
Source:
Tabulations of HRS mortality data.
Slide15Mortality rates of low and high predicted earners, by age in “Early” and “Later” cohorts
Predicted low earner born before 1931
Predicted high earner born before 1931
Predicted high earner born after 1930
Source:
Tabulations of HRS mortality data.
Slide16Mortality rate differential of low and high predicted earners, by age in “Early” cohorts
Mortality rate ratio for those born before 1931
Mortality rate difference for those born before 1931 (%)
Source:
Tabulations of HRS mortality data.
Difference
Ratio
Slide17Mortality rate differential of low and high predicted earners, by age
in “Early” & “Later” cohorts
Mortality rate ratio for those born before 1931
Mortality rate difference for those born before 1931 (%)
Mortality rate difference for those born after 1930 (%)
Source:
Tabulations of HRS mortality data.
Difference
Ratio
Slide18Mortality rate differential of low and high predicted earners, by age
in “Early” cohorts
Mortality rate ratio for those born before 1931
Mortality rate difference for those born before 1931 (%)
Source: Tabulations of HRS mortality data.
Difference
Ratio
Slide19Mortality rate differential of low and high predicted earners, by age
in “Early” & “Later” cohorts
Mortality rate ratio for those born before 1931
Mortality rate difference for those born before 1931 (%)
Mortality rate difference for those born after 1930 (%)
Source:
Tabulations of HRS mortality data.
Difference
Ratio
Slide20Does the impact of socioeconomic status on mortality grow in successive birth cohorts?
When we use our full sample we find meaningfully large and statistically significant increases in the SES mortality differential across successive birth cohorts
Using both actual and predicted Social-Security-earnings
Using indicators of low and high educational attainment
In specifications where we control for race/ethnicity, disability, and marital status:
We find worsening of mortality in low SES groups
Less than high school / Bottom half of actual or predicted income
Versus generally
declining
mortality in high SES groups
Slide21Changing impact of socioeconomic status on mortality by specific cause of death
We use discrete time logistic models to examine evolution mortality by 8 causes of death
Significant drop in age-specific mortality due to
heart disease
&
cancer
for top half of predicted income;
No significant decline in deaths due to these causes for people in bottom half of predicted income.
Similar pattern findings for male deaths due to
“
Allergies, hay fever, sinusitis and tonsillitis
” &
“
miscellaneous
”
Both among low-predicted-income men & women we find significant increases in mortality due
to
“
Digestive system issues
”
Slide22Can behavioral differences account for widening mortality differences by SES group?
To test, we added self-reported behaviors to specification:
Alcohol consumption (level)
Smoking (Sometime in past?
and
Currently?)
Vigorous physical activity at least once a week
We also examined impact of parental longevity
Finally, we tested the explanatory power and impact of self-reported health in the first HRS interview
Basic idea:
If the inclusion of the behavioral variables reduces the measured impact of SES on changes in mortality differentials, then changes in behavior by SES group may account for the growing difference in mortality by SES
Slide23Can behavioral differences account for widening mortality differences by SES group?
Self-reported, health-related behaviors have expected and highly significant impacts on risk of mortality --
Alcohol consumption and Smoking boost age-specific mortality
Vigorous physical activity reduces mortality
Inclusion of behavior variables in specification
increases
estimated mortality gradient
We find little effect of parental longevity on mortality
Inclusion of initial health status has little impact on the estimated size of change in mortality gradient
Slide24Conclusions
All our measures of SES show sizeable mortality differentials by SES group
All measures also show significant
increases
in magnitude of differentials in later cohorts compared with earlier ones
We find some causes of death--
heart disease
,
cancer
,
and (among men) “
Allergies
, hay fever, sinusitis and
tonsillitis
”—have declined among those with high predicted income but not among those with low SES
Mortality due to “
Digestive system issues
” has risen among low SES but not high SES groups
Inclusion of health-related behavioral variables does not reduce noticeably the estimated increase in mortality differentials by SES