Summary
Pijoan-Mas and Ríos-Rull develop a hazard model with time-varying stochastic endogenous covariates (health, wealth, income, labor market status, marital status, smoking) estimated on Health and Retirement Study (HRS) data (white Americans 50+, 1992–2008) to compute expected longevity — the true within-cohort gap in life expectancy — as distinct from period life expectancy (LE), which treats socioeconomic status (SES) as static. Period LE overstates the income gradient 7× and the wealth gradient 3× relative to expected longevity. A three-part decomposition (initial health differences at 50; health trajectory after 50; mortality conditional on health) shows that health trajectory is the dominant channel for education, wealth, and income differentials, while conditional mortality drives marital-status and smoking gaps. Expected longevity gradients widened substantially for education and wealth between 1992 and 2008.
Key Claims
- Period LE vs. expected longevity: Period LE at age 50 for college vs. no–high school (HS) white men = 6.3 years; expected longevity gap = not directly estimated conditional on education alone (table 4: 6.6 yrs including health, but 1.7 attributable to initial health + 5.0 to trajectory; table 2 does not isolate education directly). For wealth Q5–Q1, period LE gap = 10.7 yrs (men), expected longevity = 3.6 yrs — a 3× overstatement. For income Q5–Q1, period LE = 6.1 yrs, expected longevity = 0.8 yrs — a 7× overstatement.
- The three-channel decomposition for education (men): (a) initial health at 50 accounts for 1.7 yrs (26%), (b) differential health trajectory after 50 accounts for 5.0 yrs (76% — dominant), (c) mortality conditional on health = 0.0 yrs (null). Wealth: (a) 1.4 yrs (33%), (b) 2.0 yrs (47%), (c) 1.0 yrs (23%). Labor market: initial health dominates at 2.3 of 3.8 yrs total.
- Marital status and smoking differ: Conditional on health, married men live 1.2 yrs longer than nonmarried — conditional mortality accounts for between one-third and one-half of the marital gap. Smoking similarly: conditional mortality 1.2 yrs of total 2.2–2.9 yr gap. These two characteristics affect survival beyond their effect on health, consistent with behavioral/social mechanisms.
- Education: no direct effect on mortality conditional on health. The full education gap is mediated entirely through initial health stock and subsequent health deterioration; once health is controlled, college-educated and no-HS white men die at the same rate.
- Widening gradients 1992–2008: Expected longevity gap between college and no-HS white men grew +1.7 yrs; wealth Q5–Q1 grew +2.1 yrs (men). Marital-status and smoking gradients also widened. Income gradient was flat (−0.1 yrs for men, −0.2 for women).
- HRS sample, white Americans only: Results apply to white men and women aged 50+ in the HRS. The paper explicitly restricts to this group to maintain sample homogeneity; generalizability to non-white populations or younger ages is not established.
- Period life expectancy (Table 1, white men / women at 50): overall 78.8 / 82.9; education gap 6.3 / 5.8; wealth Q5–Q1 10.7 / 8.5; income Q5–Q1 6.1 / 3.6; labor market 9.4 / 5.0; marital 4.6 / 2.4; nonsmoker vs. smoker 7.2 / 6.2.
- Expected longevity (Table 2, white men / women at 50): wealth Q5–Q1 3.6 / 2.9; income Q5–Q1 0.8 / 0.8; labor market 2.0 / 0.8; marital 2.2 / 1.1; smoking 2.2 / 2.2; married nonsmoker vs. nonmarried smoker 4.6 / 2.7.
- Expected longevity with health included (Table 4, white men / women): education 6.6 / 5.8; wealth 4.3 / 3.7; income 1.5 / 1.3; labor market 3.8 / 1.4; marital 2.6 / 1.4; smoking 2.9 / 2.2.
Concepts Introduced or Extended
Entities Mentioned
Quotes
"Period life expectancy overstates the income gap in life expectancy by a factor of seven and the wealth gap by a factor of three."
"For education and income, differences in health trajectories account for most of the gap in expected longevity, while conditional mortality does not matter. For marital status and smoking, the mortality channel is also quantitatively important."
My Take
This is one of the cleaner demonstrations in the longevity-inequality literature that measurement method drives substantive conclusions. The 7× inflation of the income gradient under period LE is a significant result: it suggests that much of the dramatic income-mortality gradient reported in descriptive studies (including early Chetty et al. framing) reflects compositional dynamics rather than within-individual causal effects. The main limitation is sample restriction to white HRS respondents — it is unknown whether the decomposition structure holds for Black Americans or for younger cohorts. The null conditional-on-health mortality finding for education is particularly striking and bears on the debate about whether education has independent survival returns beyond its health investments.