Summary
The education-mortality gradient in the U.S. widened substantially at the end of the 20th century — by 2000, college-educated 25-year-olds could expect to live 7 years longer than their non-college peers, a gap that had grown ~30% in the prior decade. Using the National Health and Nutrition Examination Survey (NHANES I, 1971–75) and the National Health Interview Survey (NHIS, 1987–2000) with proportional hazard models and Oaxaca-style decompositions, the paper asks how much of this widening is explained by differential trends in smoking, obesity, hypertension, and cholesterol. The answer: essentially none. Behavioral risk factors explain almost none of the widening for men and no more than 8% for women. Instead, the widening reflects rising returns to education conditional on behaviors, and growing mortality consequences of the same risk factors — not changes in who has those risk factors.
Key Claims
- The gradient widened dramatically: 5-year mortality ratio (less-educated/college) rose from 1.31 to 1.52 (+22 percentage points, pp) for men and from 0.92 to 1.34 (+42 pp) for women, between NHANES I (1971–75) and the NHIS (1987–96).
- Behavioral risk factors explain essentially nothing of the widening trend: For men, changes in smoking and obesity predict a 4 pp decrease in the mortality ratio while the actual ratio increased 22 pp. For women, risk factors explain ≤3 pp of a 41 pp increase (≤8%). Adding richer smoking controls (intensity, quit timing) makes no material difference for men; for women it raises the explained fraction to 20–40%, but not statistically significant.
- Smoking differential is age-stratified: Smoking declined more for college-educated adults overall, but the differential is concentrated among younger adults (under 60). Among adults aged 60+, who have the highest death rates, the educational differential in smoking is absent or reversed. This is why aggregate smoking trends don't translate into mortality gradient changes in the period studied.
- Obesity growth was parallel across education groups: Between 1971–75 and 1999–2004, obesity increased 17–20 pp for less-educated men and women and 18–20 pp for college-educated men and women — no differential.
- Hypertension and cholesterol control were similar across education groups: Hypertension (Stage I+II) fell 26–29 pp for both groups; high cholesterol fell 10–13 pp for both. The less educated remained more likely to have these conditions at each point in time, but the secular changes were not differentially favorable to the educated.
- What does explain the widening: (1) The mortality hazard of smoking grew over time (men: hazard ratio [HR] ~1.88 in 1971–75 → 2.50 in 1987–2000), meaning the same smoking prevalence translates into more deaths; (2) the residual education coefficient grew (men: HR 0.88 → 0.75 controlling for behaviors), meaning education increasingly protects against mortality through channels beyond measured behaviors. These coefficient effects, not changes in risk factor levels, account for more than 90% of the widening.
- Compositional robustness: College attendance rose from 33% to 54% for men and 24% to 49% for women over the study period. Propensity-score reweighting to hold education group shares constant produces virtually identical results.
- By cause of death: Among men, the education gradient widened most for cancer (5-year cancer death rate rose 1.7% → 1.9% for less-educated; fell 1.4% → 1.3% for college). Cardiovascular disease (CVD) disparities actually narrowed for men. Among women, CVD education gradient widened significantly (college female CVD hazard ratio 0.99 → 0.76). Changes in behavioral risk profiles explain nothing for either cause.
- Cholesterol and hypertension control did not differ by education: Supplementary NHANES analysis finds education's protective effect against high cholesterol was nearly identical in all four NHANES waves. This rules out differential medical care access for these CVD risk factors specifically.
- The strong policy implication: "Even the complete elimination of disparities in behavioral risks across education groups would be unlikely to substantially reduce education-related differentials in mortality."
Concepts Introduced or Extended
- Education-Mortality Gradient — provides the foundational decomposition showing behavioral risk factors explain levels but not trends; defines the methodological distinction between level cross-sections and trend decompositions
Entities Mentioned
Quotes
"The widening educational gaps in death rates are not explained by secular changes in key behavioral risk factors."
"Even the complete elimination of disparities in behavioral risks across education groups would be unlikely to substantially reduce education-related differentials in mortality."
"It is the return to education (conditional on health behaviors) and changes in returns to behaviors that are important."
"By 2000, college-educated 25-year olds could expect to live 7 years longer than their peers with less schooling."
My Take
The core finding is clean and well-identified: the Oaxaca-style decomposition is the right approach, the data (NHANES + NHIS) are the best available for this question, and the result is robust to multiple extensions. The paper's chief contribution is negative — ruling out the most plausible mechanism (behavioral convergence) for why the gradient widened — while pointing toward an unresolved positive mechanism (growing returns to education and to risk factors).
The age-stratification of the smoking result deserves special attention: smokers at the relevant ages for cancer mortality don't show differential declines by education during 1971–2000. This is the key reason the aggregate smoking differential doesn't show up in mortality trends. The widening smoking gap at younger ages may explain future gradient widening (as these cohorts age), but not what was observed in the study window.
Three hypotheses for the growing return to education are proposed (better medical care access, environmental differences, complex treatment adherence) but none are tested. Case and Deaton's Deaths of Despair work, published after this paper, adds a fourth mechanism post-2000: the opioid epidemic concentrated among less-educated whites. The Cutler et al. finding that behavioral factors don't explain the pre-2000 widening makes the post-2000 opioid period analytically distinct — opioids may be the behavioral mechanism that finally does show up as gradient-widening among older adults.
Limitation: restricted to non-Hispanic whites; the education-mortality dynamics for Black, Hispanic, and other populations may differ substantially, especially given evidence of smaller education gradients in mortality for women and Hispanics (Olshansky et al. 2012).