Medicaid and Mortality refers to the causal question of whether public health insurance coverage through Medicaid reduces mortality among low-income adults — and by how much, for whom, and through what mechanisms. The question is contested because establishing causality requires solving adverse selection (sicker people are more likely to enroll), and because prior experimental evidence (the Oregon Health Insurance Experiment, N ≈ 12,000) was underpowered for mortality. Goldin, Lurie, and McCubbin (2021) resolved the experimental gap with a large-scale Internal Revenue Service (IRS) outreach randomized controlled trial (RCT) (N = 4.5M households), finding the first statistically significant experimental evidence that health insurance reduces mortality (−6.1 basis points 2-year all-cause mortality for ages 45–64). The quasi-experimental literature exploiting the Affordable Care Act (ACA) Medicaid expansion has also converged on a positive finding. The two largest quasi-experimental studies are Miller, Johnson, and Wherry (2021) — near-elderly (55–64) low-income adults, −9.4% annual mortality — and Wyse and Meyer (2023) — all non-elderly (19–59) low-income adults, −2.5% intent-to-treat (ITT) / −21% treatment-on-the-treated (TOT) — with the difference in ITT magnitude driven primarily by the much younger age distribution of the Wyse-Meyer sample.
Medicaid enrollees are systematically sicker than the uninsured who remain unenrolled, so naive comparisons understate mortality if comparing enrollees vs. non-enrollees, and overstate if comparing expansion states to non-expansion states without controlling for income composition.
The ACA (2010) extended Medicaid eligibility to all adults with income ≤138% Federal Poverty Level (FPL); the Supreme Court's 2012 ruling made expansion optional; 29 states + DC expanded in 2014, 7 more over subsequent years — creating staggered quasi-experimental variation. Pre-ACA waiver programs in some states provided earlier variation.
Baicker et al. (2013) and Finkelstein et al. (2012); first randomized experiment; ~6,000 Medicaid lottery winners vs. ~5,800 controls; found large improvements in financial protection and subjective well-being but no statistically significant mortality effects at 16 months. Almost certainly underpowered — confidence intervals (CIs) compatible with a ~16% mortality reduction. The OHIE null was a statistical artifact of small sample size, not a true zero.
The IRS randomized 4.5 million households that had paid the ACA individual mandate penalty in 2015 to receive (86% treatment) or not receive (14% control) an informational letter in January 2017 explaining available health insurance options and subsidy eligibility. N = 8.9 million individuals; block-randomized at the household level.
Coverage: +1.3 pp any 2017 coverage among partially-uninsured households (control mean 47.6%); Exchange +1.02 pp (30% relative increase), Medicaid +0.45 pp (15% relative). Coverage peaked at ages 45–64 (+2.06 pp) and incomes 100–138% FPL (+2.29 pp). Effects persisted into 2018.
Mortality: −6.1 basis points 2-year all-cause mortality for ages 45–64 (p = 0.01); 1 fewer death per 1,648 letters. The first statistically significant mortality reduction from a health insurance RCT. No significant mortality effect for ages 18–44. Pre-period placebo checks pass; prior-insured households (unaffected by the letter) show no mortality effect.
Two-stage least squares (2SLS) Average Causal Response (ACR): −0.166 to −0.178 pp per month of coverage (95% CI: −0.04 to −0.31). The ACR is defined under the Angrist-Imbens (1995) variable-intensity framework: the effect of one additional month of insurance coverage on cumulative mortality probability. Authors explicitly warn against multiplying by 12 to get an annual rate (concavity from survival selection and diminishing returns means annualization severely overstates the expected mortality reduction from a full year of coverage).
Mechanism: Effect concentrated above the Medicaid threshold (FPL > 138%), implying Exchange coverage — not Medicaid — drives the mortality reduction. Consistent with fast-onset channels: reduced delays to acute care for cardiovascular events and rapid initiation of cardiovascular and chronic disease medications in the 1–2 years post-enrollment.
Adverse selection inversion: Among penalty-paying non-enrollees, behavioral frictions (salience, complexity, timing mismatch) rather than rational low-benefit calculation explain non-enrollment. Consequently, the friction-impeded marginal enrollees are high-benefit, not low-benefit individuals. Ordinary least squares (OLS) estimates of coverage's mortality effect are attenuated relative to 2SLS — confirming that standard adverse selection (uninsured = low-benefit) does not characterize this population. See Information Frictions.
Comparison to quasi-experimental studies: The Goldin ACR (−0.178 pp/month) is broadly consistent with Miller et al. (2021) when translated to the same age group and unit of analysis. The age-weighted Oregon OHIE confidence interval (−0.032 to −0.101) also overlaps with the Goldin estimate at comparable ages.
Individual-level American Community Survey (ACS) (566,000 near-elderly low-income U.S. citizens ages 55–64) linked to Census Numident death records and Centers for Medicare & Medicaid Services (CMS) Medicaid enrollment; difference-in-differences (DiD)/event study exploiting ACA expansion stagger. Main estimate: −0.132 pp annual mortality = 9.4% reduction; growing to −0.208 pp (11.9%) by Year 3; ~4,800 fewer deaths/year in expansion states. Mechanism: reductions concentrated in disease-related (internal) deaths — cardiovascular/circulatory (38% of the internal reduction) and endocrine/metabolic/diabetes (~18%). No effect on external causes (accidents, overdoses) — consistent with the healthcare access channel (preventive care, medication management) rather than income or behavioral channels.
Population-level linkage: 37.5 million non-disabled adults ages 19–59, income <138% FPL, 2010 Census × IRS Modified Adjusted Gross Income (MAGI) income × CMS Medicaid enrollment × Social Security Administration (SSA) Numident mortality through April 2022. 60× larger than Miller et al. Staggered DiD; discrete-time proportional hazard model (vs. linear probability model [LPM] in prior work); preregistered analysis plan (only second nonexperimental economics study to do so). State-clustered standard errors.
First stage: +11.7 pp Medicaid enrollment from 24% pre-expansion baseline.
ITT mortality: −2.5% annual mortality hazard (95% CI: −0.43% to −4.5%); significant at 5%. The smaller ITT relative to Miller et al. reflects (a) coverage of all ages 19–59 including young adults with very low baseline mortality, and (b) the proportional model specification.
TOT mortality: −21% reduction in annual mortality hazard for new enrollees (95% CI: −3.7% to −38%). Triple-difference using 4–6x FPL comparison group yields −17.7% (CI: −3.4% to −32%), consistent.
Lives and life-years saved, 2010–2022:
Age heterogeneity: Proportional TOT effects are consistently negative across all age groups: ages 19–29 (−25.8%), 30–39 (−27.1%), 40–49 (−15.6%), 50–59 (−17.4%). Only ages 50–59 is significant at 95%; ages 30–39 and 40–49 at 90%. Signs are consistently negative across all subgroups (race, gender, income level, family structure, employment status).
Comparison to prior estimates: Wyse-Meyer CI falls within Miller et al.'s CI but at the lower end. Goldin, Lurie, and McCubbin's (2021) 2SLS ACR of −0.178 pp/month uses a different estimand than Wyse-Meyer's proportional hazard ratio and is not directly comparable; annualizing the ACR by multiplying by 12 produces a physically impossible mortality reduction that Goldin themselves explicitly caution against. Prior studies' large point estimates likely reflect upward bias from small samples, use of linear probability models, and comparison group misidentification.
Income-mortality gradient closure: Universal Medicaid enrollment would close only 5–20% (middle estimate ~12%) of the mortality gap between the highest and lowest income quintiles (Chetty et al. 2016). The remaining 80–95% of the gradient operates through channels other than insurance access — human capital, behavioral pathways, neighborhood effects.
External causes dominate young adult mortality: >80% of deaths among ages 19–29 are from accidents (predominantly drug overdose/poisoning), suicide, and homicide — vs. ~11% for ages 50–59. If Medicaid reduced mortality primarily through management of chronic diseases (the Miller et al. mechanism for near-elderly adults), we would not expect meaningful mortality effects among young adults where chronic disease is rare. The ACA Medicaid expansion significantly expanded substance use disorder (SUD) and mental health treatment coverage; many expansion states used waiver programs to further extend residential SUD treatment access. The hypothesis: Medicaid → SUD and mental health treatment → reduced overdose deaths, suicide, homicide mortality among young adults. Wyse and Meyer (2023) lack cause-of-death and utilization data to test this directly; it remains the leading explanation for the young adult mortality effects.
Based on Wyse and Meyer (2023):