Education-Mortality Gradient

Definition

The education-mortality gradient is the empirical relationship between educational attainment and life expectancy: higher education is associated with longer life. The gradient is steep, has existed in the United States since at least the 1960s, and widened substantially at the end of the 20th century. By 2000, college-educated 25-year-olds could expect to live approximately 7 years longer than their non-college peers. The education gradient is a distinct — though correlated — dimension of the broader Income-Mortality Gradient; education and income each carry independent mortality effects once the other is controlled for.

Key Ideas

• Causal estimate — compulsory schooling IV (Lleras-Muney 2005): Using variation in state compulsory schooling laws (1915–1939) as an instrumental variable (IV), Lleras-Muney (2005) finds that one additional year of schooling reduces the probability of dying in the next 10 years by approximately 3.6 percentage points. This is one of the few credible causal estimates of the education-mortality link, distinct from the vast correlational literature.
• Fundamental causes mechanism (Link and Phelan 1995): Education is a "fundamental cause" of health because it enables people to adopt whatever health-enhancing behaviors and technologies are currently available. As cardiovascular disease (CVD) medications, cancer screening protocols, and other innovations diffuse, educated individuals adopt them faster. The gradient therefore persists and adapts across successive health technology regimes rather than being eroded by any single innovation. This is the primary theoretical explanation for why the gradient widened even as specific behavioral risk factors (smoking, hypertension control) converged across education groups (Cutler, Deaton, and Lleras-Muney 2006).
• The gradient widened sharply from the 1970s to 1990s: 5-year mortality ratio (less-educated/college) rose $22$ pp for men ($1.31 \to 1.52$) and $42$ pp for women ($0.92 \to 1.34$) between National Health and Nutrition Examination Survey (NHANES) I (1971–75) and National Health Interview Survey (NHIS) (1987–96) — Cutler et al. (2011).
• Behavioral risk factors explain the level of the gradient but not its trend: Smoking, obesity, hypertension, and cholesterol together explain some of the cross-sectional education gap in mortality at any point in time, but changes in these risk factors explain $<10\%$ of the widening for men and at most $20$–$40\%$ for women (statistically insignificant).
• The key paradox: Smoking declined more for the college-educated overall, yet this differential is concentrated among adults under 60, not among those 60+ who drive most mortality. Hence aggregate smoking trends don't translate into mortality gradient trends in the data.
• The "returns" mechanism: Both the mortality hazard of given risk factors (e.g., smoking hazard ratio [HR] $\approx 1.88 \to \approx 2.50$ for men) and the residual education coefficient (HR $0.88 \to 0.75$ for men, controlling for behaviors) grew over time. Coefficient changes, not risk factor level changes, explain $>90\%$ of the widening.
• Cancer is the main driver for men; cardiovascular disease for women: Male cancer death rates rose for the less-educated and fell slightly for college men; CVD disparities actually narrowed for men. Female CVD hazard ratio for college-educated fell from $0.99$ to $0.76$.
• Education has no direct effect on mortality conditional on health (Pijoan-Mas and Ríos-Rull 2014): In a Health and Retirement Study (HRS) hazard model with time-varying endogenous health, the conditional mortality channel for education is zero — the entire expected longevity gap ($6.6$ yrs for white men) is mediated through initial health stock at age 50 ($1.7$ yrs, $26\%$) and differential health trajectory after 50 ($5.0$ yrs, $76\%$). This is the most direct structural decomposition of the gradient available and pinpoints the mechanism: education works through health investment, not through a residual biological survival advantage.

How It Works

Measurement: Credential vs. Rank

Two approaches to measuring the education-mortality gradient have different properties for tracking trends:

Absolute credential (e.g., "less than high school diploma"): Straightforward cross-sectionally, but vulnerable to compositional bias when attainment is rising. As college attendance rose from $\approx 25$–$33\%$ in the early 1970s to $\approx 50$–$60\%$ by 2000, the non-college group became an increasingly adversely selected residual. Comparing the "less than high school" group in 1990 to 2010 means comparing different population fractions with different selection profiles. Bound, Geronimus, Rodriguez, and Waidmann (2015) show this alone can reproduce the dramatic life expectancy (LE) declines Olshansky et al. (2012) reported for low-education whites. When education is reclassified by rank within cohort, the apparent $3.9$-year LE decline for white women collapses to $1.2$ years; for white men, a $2.2$-year decline becomes a $0.4$-year gain.

College-attendance binary (Cutler et al. 2011) or earnings rank (Waldron 2007; Chetty et al. 2016): Both mitigate compositional bias by treating the threshold as a relative rather than absolute marker. Cutler et al. additionally test robustness by propensity-score reweighting to hold education group shares constant, confirming their main findings hold even allowing for compositional change.

Ordered probit reassignment (Sanzenbacher et al. 2017): The most methodologically rigorous relative-rank approach. An ordered probit predicts each individual's probability of completing each educational level conditional on family income, occupation, and birth cohort, then assigns individuals to quartiles until each represents exactly 25% of the cohort. Applied to National Longitudinal Mortality Study (NLMS) data (1.5M individuals, 1979–2011), this confirms all four quartiles gained LE — men Q1 +4.1 yrs, Q4 +5.9 yrs at age 65; women Q1 +1.3 yrs, Q4 +3.1 yrs — while the Q4/Q1 improvement ratio is 1.5 (men) and 2.4 (women). The near-zero improvement for white women Q1 (0.1%/year) is the most alarming subgroup finding. This corroborates Bound et al. (2015) on a different dataset and with a more principled reassignment procedure, effectively closing the methodological debate: when socioeconomic status (SES) composition is held constant, absolute LE declines for large groups disappear.

The Decomposition Framework

Cutler et al. (2011) use a proportional hazard model and counterfactual simulation closely related to the Oaxaca-Blinder decomposition:

• Estimate mortality hazard as function of education, smoking, and obesity from NHANES I data (1971–75)
• Apply those parameters to the risk factor distribution in the NHIS (1987–2000) — "what if 1990s people faced 1970s mortality equations?"
• Calculate implied change in education-mortality gradient due to risk factor trends alone
• The residual is attributed to changes in the mortality function itself (coefficient effects)

The key finding: holding mortality coefficients fixed and varying only risk factor distributions produces negligible gradient change. Holding risk factor distributions fixed and varying only coefficients reproduces essentially the full gradient widening.

Why Behavioral Factors Don't Explain the Trend

1. Age-stratification of smoking: The college-educated quit smoking at higher rates, but among adults 60+, who dominate mortality statistics, the smoking differential is absent or reversed. The widening smoking differential exists at younger ages where death rates are low.
2. Parallel obesity growth: Both education groups experienced nearly identical obesity growth ($17$–$20$ pp over 30 years), generating no differential.
3. Parallel hypertension/cholesterol trends: Both groups improved, with slight advantages for the less-educated in the rate of improvement in hypertension.

What Does Explain the Trend (Unresolved)

Three mechanisms proposed by Cutler et al. but not directly tested:

1. Differential medical care access and productive efficiency: Higher-educated individuals better access, comply with, and benefit from increasingly sophisticated medical technologies. Goldman and Smith (2002): better treatment adherence for HIV and diabetes among the better educated.
2. Environmental risk divergence: Occupational hazards (manual vs. professional work), residential pollution exposure, and neighborhood built environment may have improved more for the highly educated.
3. Complex treatment regimen management: As management of chronic conditions (cancer screening, multi-drug hypertension protocols) becomes more important and more complex, education's returns to adherence grow.

Post-2000 addition (not in Cutler et al.): The Deaths of Despair literature (Case and Deaton 2015, 2017) documents the opioid/suicide/alcohol cluster concentrated among less-educated whites after 2000. This may be the behavioral mechanism that finally does show up as gradient-widening among older adults — but as a new phenomenon, not as a continuation of the 1970–2000 behavioral trends studied by Cutler et al.

Why It Matters

Policy Implication: Behavioral Interventions Alone Will Not Close the Gap

The Cutler et al. finding directly challenges the dominant public health frame that health disparities by education (and SES) can be substantially closed by targeting health behaviors. Their conclusion is explicit: "even the complete elimination of disparities in behavioral risks across education groups would be unlikely to substantially reduce education-related differentials in mortality." The mechanism driving the gradient is structural — the returns to education — not behavioral differences in smoking or obesity.

Relation to DI and Social Insurance

The education-mortality gradient has direct implications for disability insurance (DI) policy:

• Vocational Grid age thresholds: The Vocational Grid uses age 50 and 55 as thresholds, and the populations near those thresholds are overwhelmingly non-college. Because the education-mortality gradient is steep, LE indexing that uses population averages (which include college-educated individuals) systematically overestimates the LE gains available to this DI-proximate population. See Income-Mortality Gradient.
• Continuing disability review (CDR) timing and the returning health of marginal workers: Moore (2015) finds that DI termination's employment effect peaks at $\approx 2.7$ years tenure, before declining. If the education gradient in mortality reflects structural disadvantage rather than correctable behaviors, then the pool of DI beneficiaries (disproportionately non-college) faces compound risks: both higher underlying disability and lower LE conditional on returning to the labor market.
• DI as targeted redistribution: The concentration of DI awards among low-education, low-income individuals means DI transfers flow to the worst-LE tier of the income-mortality gradient. Gelber et al. (2018): $1,000/year in DI income reduces annual mortality $\approx 0.26$ pp — mechanically coherent with the income-LE gradient at the bottom.

Retirement Age Indexing Is Regressive

Both the income-mortality gradient and the education-mortality gradient establish that LE gains have been concentrated at the top of the SES distribution. Chetty et al. (2016) find top-5% men gained $2.34$ years of LE from 2001–2014 while bottom-5% men gained only $0.32$ years. Proposals to raise Social Security eligibility ages based on aggregate LE gains thus impose equal cost on populations with unequal LE improvement — a regressive distributional effect concentrated on exactly the non-college populations with the highest DI incidence.

The widening education-based LE gap also erodes the progressivity of Social Security's benefit formula directly: lower-educated workers receive higher replacement rates by design, but collect benefits for fewer years. Bosworth and Burke (2014) estimate that differential mortality offsets approximately half of overall Social Security system progressivity. Rutledge (2018) synthesizes this evidence: between the 1930 and 1960 birth cohorts, bottom-quintile men's lifetime Social Security benefits fell in absolute real terms while top-quintile men gained $66k. See Social Security Progressivity.

Open Questions

• Why have the returns to education in the mortality function grown? Three mechanisms are proposed but no study has cleanly identified which dominates.
• Does the widening gradient documented through 2000 accelerate post-2000 once deaths of despair are included? Preliminary evidence from Case/Deaton suggests yes.
• The restriction to non-Hispanic whites limits generalizability. How does the education-mortality gradient trend among Black, Hispanic, and Asian/Pacific Islander (API) populations? Evidence from Olshansky et al. (2012) suggests smaller education gradients for women and for Hispanics.
• What is the interaction between the education-mortality gradient and the age-specific mortality improvements documented in Global Burden of Disease (GBD) studies? If morbidity expansion (rising years lived with disability [YLDs]) is also education-stratified, health-adjusted life expectancy (HALE) gaps by education may be widening even faster than mortality gaps.

Related

• Income-Mortality Gradient
• Period vs. Cohort Life Expectancy
• Josep Pijoan-Mas
• José-Víctor Ríos-Rull
• Deaths of Despair
• Healthy Life Expectancy
• Morbidity-Mortality Distinction
• SES Health Behavior Gradient
• Vocational Grid
• DI Beneficiary Mortality
• David Cutler
• Ellen Meara
• Christopher Ruhm
• Raj Chetty
• Anne Case
• Angus Deaton
• John Bound
• Arline Geronimus
• S. Jay Olshansky
• Geoffrey T. Sanzenbacher
• Anthony Webb
• Social Security Progressivity
• Matthew S. Rutledge
• Adriana Lleras-Muney
• Preston Curve

Sources

• Cutler et al. 2011 — Rising Educational Gradients in Mortality The Role of Behavioral Risk Factors
• Bound et al 2015 — Measuring Recent Apparent Declines In Longevity The Role Of Increasing Educational Attainment
• Olshansky et al 2012 — Differences in Life Expectancy Due to Race and Educational Differences Are Widening and Many May Not Catch Up
• Chetty et al. 2016 — The Association Between Income and Life Expectancy in the United States, 2001–2014
• Case and Deaton 2017 — Mortality and Morbidity in the 21st Century
• Pijoan-Mas and Ríos-Rull 2014 — Heterogeneity in Expected Longevities
• Rutledge 2018 — How Is the Mortality Gap Affecting Social Security Progressivity
• Cutler Deaton and Lleras-Muney 2006 — The Determinants of Mortality
• Sanzenbacher et al 2017 — Rising Inequality in Life Expectancy by Socioeconomic Status
• Bor Cohen and Galea 2017 — Population Health in an Era of Rising Income Inequality USA 1980–2015
• Poterba Venti and Wise 2017 — The Long Reach of Education Health, Wealth, and DI Participation