Miller Johnson and Wherry 2021 — Medicaid and Mortality New Evidence from Linked Survey and Administrative Data

Summary

Miller, Johnson, and Wherry (2021) provide the first individual-level quasi-experimental evidence that Medicaid coverage substantially reduces mortality among low-income near-elderly adults. Linking ~566,000 American Community Survey (ACS) respondents (ages 55–64, income ≤138% of the federal poverty level [FPL]) to Census death records (Numident) and Centers for Medicare & Medicaid Services (CMS) Medicaid enrollment data, they exploit staggered Affordable Care Act (ACA) Medicaid expansion timing across states in a difference-in-differences (DiD) event study. They find a 9.4% average reduction in annual mortality (−0.132 pp) growing to 11.9% by year 3, driven entirely by disease-related deaths — cardiovascular and diabetes causes predominate. The finding resolves the prior null from the Oregon Health Insurance Experiment (OHIE), which was underpowered for mortality.

Key Claims

• ACA Medicaid expansion first stage: +49–51 percentage points (pp) Medicaid eligibility; +12.8 pp annual Medicaid enrollment probability; +0.375 cumulative Medicaid years; −4.4 pp uninsurance
• Annual mortality DiD: −0.132 pp = 9.4% of sample mean; growing from −0.089 pp (Year 0) to −0.208 pp (Year 3); significant at 5%
• Cause of death (Mortality Disparities in American Communities [MDAC] subsample): Internal/disease deaths fall significantly (−0.00235 pp, p<0.01); healthcare-amenable causes fall (p<0.10); external/accident deaths show no significant change — consistent with a healthcare access mechanism, not behavioral
• Lives saved: ~4,800 per year in expansion states; ~19,200 over 4 years; ~15,600 deaths attributable to non-expansion in holdout states
• Scaling: 1 additional year of Medicaid enrollment reduces annual mortality by ~0.35 pp; instrumental variables (IV) (enrolled-in-year) estimate: −1.03 pp per enrolled year
• Cause decomposition: Cardiovascular/circulatory diseases account for ~38% of the internal mortality reduction; endocrine/metabolic/diabetes ~18%
• Heterogeneity: Effects largest for ages 59–64; white non-Hispanic adults; males; no statistically significant effect for non-Hispanic Black adults despite higher baseline mortality
• Robustness: 12 specification checks all confirm baseline; Sun-Abraham (2021) staggered-treatment estimator replicates main result; Goodman-Bacon (2021) decomposition finds only 11% of identifying variation from different-timing comparisons; flat pre-trends in all event-study plots
• Placebo tests: No mortality reduction for ages 65+ (Medicare-covered); no effect in pre-ACA period; small consistent effect for 400%+ FPL group (who enrolled in Medicaid despite being above the threshold)

Concepts Introduced or Extended

• Medicaid and Mortality
• Difference-in-Differences
• Health-Poverty Trap

Entities Mentioned

• Sarah Miller
• Norman Johnson
• Laura R. Wherry
• Amy Finkelstein
• Raj Chetty
• Benjamin D. Sommers
• Katherine Baicker

Quotes

"We find that Medicaid expansion is associated with a 9.4 percent reduction in the annual mortality rate of near-elderly low-income adults."

"The effect grows over time, reaching 11.9 percent by the third year following the expansion."

"The reductions are driven primarily by reductions in disease-related deaths, including reductions in cardiovascular disease and diabetes mortality."

"Our estimates imply approximately 4,800 fewer deaths per year in states that expanded Medicaid."

My Take

This is the strongest quasi-experimental evidence to date that Medicaid saves lives. The methodological advance — individual-level ACS-Numident-CMS linkage — directly addresses the underpowering problem that sank the Oregon experiment's mortality analysis, enabling identification on a sample 35× larger than the OHIE. The growing effect over time (Year 0: 6.4% → Year 3: 11.9%) is theoretically coherent: preventive care and chronic disease management accrue benefits gradually. The null for Black adults is the paper's most puzzling finding — higher baseline mortality should if anything amplify absolute effects from equivalent relative risk reductions — and deserves more attention than the authors give it. The scaling estimate (1 year Medicaid = −0.35 pp annual mortality ≈ 11.9–21.5% relative reduction for compliers) sits at the upper end of the quasi-experimental literature (9–65% range) but well within it.