Goldin Lurie and McCubbin 2021 — Health Insurance and Mortality Experimental Evidence from Taxpayer Outreach

Summary

The authors randomize whether the Internal Revenue Service (IRS) sends informational letters to 3.9 million households — out of 4.5 million eligible — that had paid the Affordable Care Act (ACA) individual mandate penalty in 2015, informing them about available health insurance marketplace options and subsidy eligibility. Treated households increased coverage by 1.3 percentage points (pp) relative to controls, primarily through exchange and Medicaid enrollment. Among adults ages 45–64, two-year all-cause mortality fell by 6.1 basis points (p = 0.01) — the first statistically significant mortality reduction from a health insurance randomized controlled trial (RCT). A two-stage least squares (2SLS) average causal response (ACR) of −0.17 pp per month of coverage is consistent with quasi-experimental Medicaid estimates and identifies behavioral frictions, not rational selection against coverage, as the driver of mandate-penalty non-compliance.

Key Claims

• First experimental mortality evidence: the Oregon Health Insurance Experiment (OHIE; ~12,000 participants) was underpowered for mortality. This RCT (N = 8.9 million; 3.9M treated, 0.6M control) provides the first statistically significant experimental evidence that health insurance reduces all-cause mortality.
• Letter-induced coverage: +1.3 pp any 2017 coverage for partially-uninsured households (control mean 47.6%); Exchange +1.02 pp (30% relative increase), Medicaid +0.45 pp (15% relative). Coverage effect peaks at ages 45–64 (+2.06 pp) and incomes 100–138% of the federal poverty level (FPL) (+2.29 pp). Effect persists into 2018.
• Mortality effect: −6.1 basis points 2-year all-cause mortality for ages 45–64 (p = 0.01); 1 fewer death per 1,648 letters. No statistically significant effect for ages 18–44. Pre-period and prior-insured placebo checks pass.
• 2SLS ACR: −0.166 to −0.178 pp/month reduction in cumulative mortality per additional month of coverage (95% confidence interval [CI]: −0.04 to −0.31). Authors explicitly caution against multiplying by 12 to annualize (concavity from survival selection and health capital diminishing returns).
• Mechanism: Effect concentrated above Medicaid threshold (FPL > 138%), so Exchange coverage drives the mortality reduction — not Medicaid. Fast-onset channels: reduced delays to acute care (emergency room [ER] presentation for heart attack/stroke) and rapid initiation of cardiovascular medications in the first 1–2 years.
• Adverse selection inversion: Individuals who paid the penalty but did not enroll are high-benefit, not low-benefit. Behavioral frictions (salience, complexity, timing mismatch) concentrate among those who would gain most from coverage. ordinary least squares (OLS) coverage–mortality estimate is attenuated relative to 2SLS, confirming that standard adverse selection — in which the uninsured are the low-benefit individuals — does not characterize this population.
• IRS Form 1095 administrative data: First study to use Form 1095 A/B/C administrative data for individual-level monthly coverage records across all insurers; superior to survey-based coverage for identifying month-by-month enrollment.
• Cost: $0.49 per letter; 1,648 letters per death averted; ~$43 per additional year of coverage created per letter. Letter program is among the most cost-effective health insurance take-up interventions documented.
• Placebo checks: No differential mortality in pre-letter years (2015–2016); no effect among households fully insured prior to 2017 (unaffected by the letter).

Concepts Introduced or Extended

• Medicaid and Mortality
• Information Frictions
• Adverse Selection
• Instrumental Variables

Entities Mentioned

• Jacob Goldin
• Ithai Lurie
• Janet McCubbin
• Amy Finkelstein
• Jonathan Gruber

Quotes

"Our results suggest that health insurance has an economically large and statistically significant effect on mortality… These results provide the first experimental evidence that health insurance reduces mortality." (p. 1)

"Individuals who paid the mandate penalty but did not enroll appear to be high-benefit individuals who face frictions rather than low-benefit individuals who optimally chose to go uninsured." (p. 35)

"Absent frictions, we would expect the marginal enrollee to be a low-benefit individual (adverse selection into insurance). The opposite appears to be true in our setting." (p. 36)

My Take

Methodologically the cleanest identification in the health insurance mortality literature: the letter instrument varies salience and information without varying financial incentives, coverage quality, or eligibility — isolating the behavioral-friction channel. The experimental design also reveals who the marginal enroller is (penalty-paying, previously uninsured households that needed a nudge), and shows these are high-benefit individuals, inverting the standard adverse selection logic.

Key caveats: (1) The 95% CI for the ACR is wide (−0.04 to −0.31 pp/month), reflecting imprecision even at this scale. (2) Annualizing by multiplying by 12 produces a physically impossible mortality reduction and is explicitly cautioned against. (3) Exchange coverage (above Medicaid threshold) drives the effect; comparison to Medicaid-expansion quasi-experimental studies assumes comparability of coverage types. (4) The 1–2 year window captures acute-onset benefits; lifetime coverage benefits are not captured. (5) The letters operate through information/salience, implying that the implied cost of a "death averted" from the letters understates what full auto-enrollment would cost or achieve.