There has been substantial interest in how indicators of socioeconomic status are related to mortality and well-being, with prior papers documenting stark gradients along dimensions of income, wealth, and education. In the US, for example, Chetty et al. (2016) uses 2001 to 2014 period data to show that life expectancy differences can be as large as 15 years when comparing men in the top 1% versus bottom 1% of peak earnings. These gains are also accruing unequally over time: the same paper finds that from 2001 to 2014, the richest 5% of men saw their life span increase by 2.34 years, while the poorest 5% of men only gained 0.32 years. Echoing the importance of these results, Hudomiet et al. (2021) forecasts that these wealth-longevity inequalities will deepen over time based on a number of mortality risk factors in the US population. These concerns are not specific to the US: health equity is a concern globally, and O’Donnell et al. (2015) reviews its many sources and implications.

Our paper builds on these and other prior research by examining the composition of years lived after age 65 by quartiles of net total wealth, our metric of socioeconomic status. Specifically, we ask three questions: (1) How does wealth relate to the number of years post-65 with and without disability? (2) How does wealth relate to the number of years post-65 working versus not working? (3) How is the correlation of wealth with these outcomes changing across cohorts born a decade apart (turning 65 in 1996 versus 2006)? The motivation for these research questions is that most people look forward to enjoying a set of “golden years” in later life with good health and reduced or no work obligations. If socioeconomic indicators are related to how these years accrue and change over time, that is useful in numerous applications including the design of redistributive policies.

Our analysis combines the nationally representative Health and Retirement Study (HRS) data from 1996 to 2018 with life expectancy tables at older ages to compute the number of years lived (or expected to be lived) disabled, disability-free, working, and work-free. Within each pair of outcomes, such as disabled and disability-free years, the sum equals life expectancy at age 65. The analysis is conducted on two cohorts of individuals turning 65 a decade apart in 1996 and 2006. Wealth is measured in gender- and cohort-specific quartiles of net total wealth at age 65; the gender aspect helps account for the different wealth distributions among men and women who live alone.

We document several relationships to inform our research questions. Within each cohort of individuals turning 65, there is a strong and negative wealth gradient to the number of years after age 65 spent disability-free. The patterns also reveal that between the 1996 and 2006 cohorts of individuals turning 65, disability-free life expectancy (DFLE) gains accrued only to the top wealth quartile. For this group, DFLE gains were 0.66 years for males and 0.24 years for females. By contrast, for the bottom wealth quartile, there were DFLE reductions of 0.04 years for males and 0.13 years for females.

The results also show a relationship between an individual’s wealth and their propensity to engage in paid work after age 65. Within-cohort, our results show that wealthier individuals spend more years working and live more years work-free after age 65. When looking across cohorts, however, there is an increase in the propensity to work for the wealthier quartiles which results in a reduction of work-free life expectancy (WFLE) by 0.41 years for the top wealth quartile for males and by 0.66 years for females. For the bottom wealth quartile, the change is 0.09 years for males and −0.03 years for females.

Having established the baseline results, we turn our attention to robustness and mechanisms. The main robustness check we conduct is to also measure socioeconomic status via educational attainment. Importantly for our analysis, education is different from wealth because education cannot easily be “spent down” after a health shock which might otherwise affect both net total wealth and disability or work. We document consistent relationships using education in place of net total wealth. We also examine an alternative wealth metric, Social Security wealth, which follows Hudomiet et al. (2021) and captures lifetime earnings.

There are many possible mechanisms behind the within- and between-cohort correlations of net total wealth with disability and work. Examples include changes in healthcare access and private insurance (Dunlop et al., 2007), flexible work arrangements (Ameriks et al., 2020), and types of work (Breeze et al., 2001, Cambois et al., 2001). We cannot measure or disentangle all of these mechanisms, so we focus on three aspects: (1) whether hospitalizations or disability onset differentially affected out-of-pocket expenditures or wealth in the 2006 versus 1996 cohort, inspired by Dobkin et al. (2018); (2) whether there are also correlations of the wealth quartiles with a variety of health metrics at age 65, following the health factors studied in Hudomiet et al. (2021); and (3) whether substituting relative within-cohort wealth with relative wealth across cohorts affects the results. For this last mechanism, the motivation is to address the role of the wider wealth distribution in 2006 versus 1996. We find some suggestive evidence that health metrics and absolute wealth play a role in explaining the steeper wealth gradients in 2006 versus 1996.

To summarize, this paper shows that DFLE and WFLE are both related to net total wealth within the cohort of individuals turning age 65 in 1996 and 2006. This paper also shows that the gradient of wealth with respect to these outcomes is larger, or steeper, in 2006 versus 1996. There is some evidence that the changing relationship of wealth with respect to mortality risk factors such as extreme obesity can help explain these results, though we hesitate to highlight these too much given null results on many other mortality risk factors. Similarly, part of the reason for the steeper gradients in 2006 may be the increasingly dispersed wealth distribution; while we find some evidence for this explanation, we cannot rule out many other mechanisms.

In many ways, our paper can be viewed as a companion to Hudomiet et al. (2021), which also uses the HRS and documents strong relationships between Social Security wealth and mortality risk factors across a similar time period (that study uses data from 1992 to 2016, while we use data from 1996 to 2018). A key takeaway from that paper is that not only is Social Security wealth correlated with the mortality risk factors within each cohort, but these relationships are forecasted to steepen over time generating concerns about health inequalities. Our paper differs from Hudomiet et al. (2021) by using net total wealth instead of Social Security wealth as the primary proxy for socioeconomic status, a decision we discuss and explore further in Section 6. Also in contrast to that paper, our contribution is to document trends in the wealth gradient in the breakdown of longevity as measured by years after age 65 containing disability or work.

The next section contains a review of the related literature. We describe our data in Section 3, detail our methodology in Section 4, and present and discuss our results in Section 5. Sections 6 Alternative metrics for socioeconomic status, 7 Exploring mechanisms for the between-cohort wealth gradients present analysis related to robustness and possible mechanisms; we then conclude in Section 8.