Sanzenbacher et al 2017 — Rising Inequality in Life Expectancy by Socioeconomic Status

Summary

Using National Longitudinal Mortality Study (NLMS) data on 1.5 million individuals observed 1979–2011, Sanzenbacher et al. address the contested question of whether low-socioeconomic-status (SES) Americans have experienced absolute declines in life expectancy (LE). The paper's innovation is assigning individuals to relative education quartiles via an ordered probit — holding each quartile at exactly 25% within each birth cohort, preventing the rising educational attainment of later cohorts from artificially concentrating disadvantage in the "high school dropout" category. Under this relative-rank measure, all four SES quartiles gained life expectancy between 1979 and 2011, but gains were sharply concentrated at the top: men in Q4 gained 5.9 years at age 65 while Q1 gained only 4.1 years; the corresponding figures for women are 3.1 and 1.3 years. The apparent LE declines documented by Olshansky et al. (2012) using absolute education are largely an artifact of the compositional change in who counts as "low education" as college attendance rose from ~40% to ~60%.

Key Claims

• All quartiles gained LE under relative-rank measure: Men Q1–Q4 gained +4.1, +5.0, +5.4, +5.9 years at age 65 (1979→2011); women gained +1.3, +2.3, +2.5, +3.1 years. No absolute declines.
• Annual mortality improvement by quartile: Men: 1.6%, 2.0%, 2.2%, 2.4% per year; Women: 0.5%, 0.9%, 1.0%, 1.2% per year. Q4/Q1 improvement ratio ≈ 1.5 for men, ≈ 2.4 for women.
• White women Q1: near-zero improvement: Only 0.1%/year annual mortality improvement 1979–2011 — the worst outcome of any sex × quartile cell. Survival curves for white women bottom-quartile are nearly identical in 1979 and 2011.
• Composition change explains Olshansky: Absolute LE declines found by Olshansky et al. (2012) reflect two simultaneous forces — rising SES-mortality inequality AND the rising disadvantage of the high-school-dropout residual group as college became near-universal. The relative-rank approach isolates the first.
• Case-Deaton corroborated for the specific subset: Restricting to whites ages 45–54, 1999–2011, and combining the bottom two quartiles (equivalent to high school [HS] or less), the paper confirms a mortality increase for both white men and women — fully consistent with Case and Deaton (2015). This is not contradicted by the no-decline finding; Case-Deaton describes a specific age-race-credential group, not the general population.
• Policy: raising retirement age is regressive: Because low-SES workers gained far fewer years of LE, a uniform increase in the Social Security retirement age cuts real lifetime benefits more for Q1 than Q4.
• Method: Ordered probit assigns relative education quartiles; Gompertz log-mortality regression: $\ln(m_{a,t}) = \beta_0 + \beta_1(a-52) + \beta_2(t-1979)$ fit separately by sex and quartile; death-weighted regression; period LE computed from 1979 and 2011 mortality schedules held fixed.

Concepts Introduced or Extended

• Income-Mortality Gradient
• Education-Mortality Gradient
• Period vs. Cohort Life Expectancy
• Social Security Progressivity
• Deaths of Despair

Entities Mentioned

• Geoffrey T. Sanzenbacher
• Anthony Webb
• Candace M. Cosgrove
• Natalia S. Orlova
• John Bound
• Hilary Waldron
• S. Jay Olshansky
• Anne Case
• Angus Deaton

Quotes

"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 reason some studies have found that individuals with low levels of education have seen decreases in longevity — for example, Olshansky et al. (2012) — results from two changes occurring over time: 1) the rising inequality documented in this study; and 2) the fact that high school dropouts have become a more disadvantaged economic minority."

"Efforts to improve the finances of Social Security through higher retirement ages will have to reckon with the distributional effects of increasing inequality in mortality."

My Take

A methodologically clean contribution to a messy literature. The ordered-probit reassignment is more principled than Bound et al.'s simpler quartile cutoff — both arrive at the same conclusion (no absolute decline), which strengthens the case against the Olshansky narrative. The finding that white women in Q1 have experienced essentially zero mortality improvement over three decades is alarming and deserves more attention; it is a subgroup-specific crisis masked by aggregate gains. The NLMS-based 2.0%/year men's mortality improvement is suspiciously high versus the Social Security Administration's (SSA) 1.3%, which the authors note and address by using period LE to avoid projecting the trend forward. The paper is cautious and credible.