Rising U.S. Incomes Stop Translating Into Longer Lives

For most of the twentieth century, richer U.S. places tended to become longer-lived places. That was the historical logic behind Preston curves: income did not mechanically cause longevity, but rising resources moved alongside the diffusion of health-protective institutions, technologies, and public-health capacity. Khanna, O'Brien, Stokes, Venkataramani, and Wrigley-Field argue that this relationship changed after 2010. The U.S. kept getting richer, but the income-life-expectancy curve stopped moving upward and became steeper.

Core Contribution

The paper's main contribution is diagnostic rather than causal. It reframes the recent U.S. mortality crisis as a change in the conversion rate between aggregate resources and longevity, not only as a rise in specific causes of death such as overdose, suicide, alcohol-related mortality, or stalled cardiovascular improvement.

The authors use Preston curves to separate two patterns. A vertical upward shift means that places with similar income reach higher life expectancy over time. A rightward shift without an upward shift means income rose without commensurate longevity gains. A steeper curve means that income became more predictive of longevity differences across places. In the authors' reading, the post-2010 U.S. shows both decoupling and divergence.

Figure 1 is the central visual evidence: state curves from 1980 through 2010 move up and right, while the 2019 curve moves right without a comparable upward shift; county curves show a similar break between 2010 and 2019.

Technical Approach

The analysis combines state- and county-level data on life expectancy, income, and socioeconomic conditions. State life expectancy covers 1980 to 2019, using observations for 1980, 1990, 2000, 2010, and 2019. County life expectancy covers 2000 to 2019, using 2000, 2010, and 2019 to avoid the Covid-19 period. State income and county income come from the Bureau of Economic Analysis and are adjusted to 2019 dollars.

The main Preston curves are population-weighted second-order polynomial regressions of life expectancy on real per-capita income. The authors also test whether the result depends on the 2010 anchor around the Great Recession. They repeat the state analysis with 2006 income and life expectancy, pair 2012 life expectancy with 2006 income, substitute college-educated population share for income, and use cost-of-living-adjusted income. These checks preserve the same qualitative pattern.

The paper then asks whether observable county covariates can account for the post-2010 shortfall. The decomposition estimates earlier relationships between county life expectancy and covariates, then applies those relationships to later covariate levels. The covariates include income per capita, college share, wealth per capita, poverty, inequality, labor-force measures, insurance coverage, government spending, social capital, marriage, migration, ethnic fragmentation, male labor-force participation, and air quality.

Results and Analysis

The evidence is strongest for the descriptive claim: the U.S. Preston curve changed after 2010. In the state plots, earlier decades show the expected pattern, with life expectancy rising as real per-capita income rises. By 2019, the fitted state curve reaches much higher income levels but does not continue the broad upward movement seen from 1980 to 2010. In the county plots, 2000 to 2010 looks conventional, while 2010 to 2019 shows income growth with stagnating or declining life expectancy.

The subgroup results make the pattern harder to dismiss as composition. Stratified curves show the same post-2010 shift for women and men, and for non-Hispanic Black and non-Hispanic White populations. Appendix age-specific mortality curves sharpen the interpretation: older groups, especially 75 and above, retain a more conventional mortality decline pattern, while younger and working-age groups show more visible reversal.

Figure 4 tests a different explanation: perhaps standard socioeconomic and institutional measures moved the wrong way. The decomposition argues against that as a complete account. In 2019, observed life expectancy sits near 79 years, while several 2019 counterfactuals based on 2010 relationships predict higher life expectancy, including income per capita, college share, wealth per capita, and insurance coverage. By contrast, some social and demographic measures predict lower values than observed, which the authors interpret as consistent with broader social deterioration limiting gains that would otherwise be expected from improved economic measures.

Limits of the Evidence

The paper does not identify a single mechanism. Its ecological design cannot estimate individual-level effects of income on mortality, and the decompositions are counterfactual accounting exercises rather than causal identification. That matters: the evidence supports the claim that the income-longevity relationship weakened, but it does not prove why. The most useful takeaway is therefore narrower and stronger than a broad theory of U.S. decline: any explanation of the mortality crisis has to explain why more aggregate income, education, wealth, and insurance did not translate into broad longevity gains after 2010.