Rising Inequality in Life Expectancy by Socioeconomic Status

Abstract

Inequality in life expectancy is growing in the United States, but evidence is mixed regarding how much it has grown. Some studies have found that life expectancies have decreased for those with the lowest socioeconomic status (SES). Other studies have found that while inequality is rising, there have been life expectancy gains across the board. A primary difference in these studies is how SES is measured. Some studies use an absolute measure, such as years of school completed, while others use relative measures, such as a person’s ranking of years of school completed compared to others born at the same time. This study uses regression analysis to assign people a relative education ranking and, in doing so, attempts to isolate the changing relationship between SES and mortality from the fact that certain education-based groups, especially high school dropouts, actually have a lower SES level today than in the past. The study finds that when SES is defined in this way—relatively—inequality in mortality by SES is increasing but life expectancies have also increased across SES groups. The study also finds that white women in the bottom of the education distribution have experienced the least improvement of any group and that rectangularization of the mortality distribution has occurred much more in the top of the income distribution than at the bottom.

ACKNOWLEDGMENTS

The research reported herein was pursuant to a grant from the Alfred P. Sloan Foundation. The findings and conclusions expressed are solely those of the authors and do not represent the views of the Alfred P. Sloan Foundation, the U.S. Census Bureau, The New School for Social Research, or Boston College. This article is released to inform interested parties of research and to encourage discussion. Any views expressed on statistical, methodological, technical, or operational issues are those of the authors and not necessarily those of the U.S. Census Bureau. The authors thank participants at the Thirteenth International Longevity Risk and Capital Markets Solutions Conference for helpful comments.

Discussions on this article can be submitted until October 1, 2020. The authors reserve the right to reply to any discussion. Please see the Instructions for Authors found online at http://www.tandfonline.com/uaaj for submission instructions.

Notes

1 Other studies use earnings as the proxy for SES, e.g., Waldron (2013). Although we discuss these studies in the following, the focus of this article is on education.

2 Goldin (1998).

3 Described in more detail in the following.

4 For example, see Jemal et al. (2015).

5 Many studies of mortality trends, for example, Lee and Carter (1992) and Jemal et al. (2015), do not decompose trends by SES.

6 Information on year of birth and CPS sample year are not available in the public-use version of the NLMS. Because these variables are critical to the analysis presented here, all analyses were conducted on restricted-access data through employees at the U.S. Census Bureau.

7 The U.S. Social Security Administration (SSA) calculates mortality rates using different methods for different years of death. To calculate mortality rates for 1900–1967, SSA divides the number of deaths for each age and gender from the National Center for Health Statistics by the mid-year Census estimate of the population for that age and gender. After 1968, the SSA uses this same method for all individuals under age 65 years, but uses Medicare data to calculate the mortality rate for individuals over age 65 years. Medicare data have the advantage of containing the number of deaths and the number of individuals alive for the population within the same dataset.

8 In the NLMS, a small fraction of individuals included in the sample have not yet been matched to a death certificate, despite having obviously reached ages where they are deceased. If they are not accounted for, these so-called immortals result in mortality rates that are artificially low at high ages. To account for this data issue, we identify the number of individuals over 105 years old and alive in 2011 for each birth cohort and subtract these individuals from the denominator of Equation (1) for the years in which they were in the sample.

9 An alternative approach would have been to weight both the SSA probabilities and the NLMS probabilities by the share of NLMS deaths at a given age due a given birth cohort, i.e., weighting the calculations based on where the most data on death is available in the NLMS. The comparison between the SSA and NLMS death probabilities is similar under this approach.

10 We include only those age–year combinations for which we have observations and for which there are at least five NLMS observations.

11 We considered but rejected constructing a model in which mortality varied with a broader measure of socioeconomic status or occupational prestige. Our concern is occupation and other markers of socioeconomic status are endogenous to mortality, as in Waldron (2007; 2013).

12 In these regressions, each age and gender cell is weighted by the share of deaths in the year that occurred for that cell. In other words, the regression attempts to fit mortality rates within a year, placing the most weight on ages with the most deaths.

13 U.S. Social Security Administration, Long-Range Demographic Assumptions for the 2015 Trustees Report.

14 For example, Lu and Wong (2011) report yearly motality improvements for males ages 60 to 90 years ranging from 1.5 to 2.5% using the Human Mortality Database from 1995 to 2006.

15 Jemal et al. (2015).

16 In Table 4, the difference for men between the highest and third educational quartiles is visible in the third row and is a statistically significant 0.3%. For women, the corresponding difference is 0.4% and is also statistically significant.

17 This approach is in contrast to a “cohort”-based approach, which allows mortality to reflect age and birth year. For example, a person born in 1946 who reaches 65 years in 2011 experiences 2011 mortality rates for a 65-year-old in 2011 and 2012 mortality rates for a 66-year-old in 2012. In contrast, a 65-year-old born 1 year later, in 1947, reaches 2012 mortality rates for a 65-year-old in 2012 and 2013 mortality rates for a 66-year-old in 2013.

18 Our calculations based on Sasson (2016), Table 2.