Medicaid Expansion and Insurance Coverage

The United States spends more per capita on healthcare than any other high-income country, yet coverage remains conditional rather than universal. Medicaid functions as a safety-net program for low-income Americans, with eligibility tied to income, household structure, and state-level policy decisions, not a guaranteed right. The Affordable Care Act's provision allowing states to expand Medicaid eligibility to 138% of the federal poverty level (FPL) created a sharp and exogenous divide in coverage rules across states, making it one of the most studied natural experiments in health policy.

This project asks a direct causal question: did expanding Medicaid eligibility change insurance coverage for low-income adults? State adoption decisions varied, with some states expanding in 2014 and others never expanding, providing the counterfactual needed to answer it. By comparing coverage trajectories in expansion and non-expansion states before and after 2014, we can isolate the effect of the policy from broader national trends affecting all states simultaneously.

Low-income households often experience economic volatility, making insurance stability particularly consequential. Understanding whether state expansion decisions translate to measurable differences in coverage, and how large those differences are, has direct implications for ongoing policy debates in the 10 states that have still not expanded.

This project uses the American Community Survey (ACS) Public Use Microdata Sample (PUMS) person-level files spanning 2010–2019 and 2021–2023. The ACS is an annual, nationally representative survey conducted by the U.S. Census Bureau, collecting detailed information on demographics, income, and health insurance coverage at the individual level. The 2020 survey year was excluded due to data unavailability.

Each year consists of two files (A and B), yielding 26 raw files across 13 survey years. Because each file contains millions of rows and hundreds of columns, column selection was applied at load time to retain only the variables required for analysis: state identifier, age, income-to-poverty ratio (POVPIP), insurance coverage indicators (PRIVCOV, PUBCOV, HICOV), and survey weights (PWGTP). All 26 files were combined into a single dataset.

The analysis was restricted to working-age adults (18–64) at or below 138% of the federal poverty level, the population directly affected by the Medicaid expansion threshold. Observations missing income-to-poverty ratio values or survey weights were dropped. Puerto Rico and the District of Columbia were excluded, limiting analysis to the 50 U.S. states. Each state was assigned its Medicaid expansion year based on published implementation dates, or flagged as a non-expanding state.

Uninsured rates were estimated for working-age adults at or below 138% FPL using ACS survey weights throughout. The primary causal analysis uses an event-study difference-in-differences (DiD) design estimated via two-way fixed effects (TWFE) using fixest::feols in R.

The DiD design compares how uninsured rates changed over time in states that adopted Medicaid expansion versus states that did not. Non-expansion states serve as the counterfactual, estimating what coverage trajectories in expansion states would have looked like absent the policy. This structure allows us to separate coverage changes attributable to expansion from concurrent national trends, such as other ACA provisions that affected all states simultaneously.

The model produces a separate treatment effect estimate for each year relative to 2014 (the expansion year), enabling visualization of both pre-trend parallelism and the dynamic post-expansion effect. A key design choice was restricting the DiD sample to states that expanded in 2014 and states that never expanded, excluding later adopters. This keeps treatment timing consistent, reduces complexity from staggered adoption, and improves interpretability of the event-study estimates.

State-specific counterfactuals were also constructed, estimating what uninsured rates in non-expansion states would have looked like had they expanded. The gap between observed and counterfactual rates, combined with state population estimates, was used to quantify the number of individuals left uninsured as a consequence of non-expansion.

Prior to 2014, uninsured rate trends in expansion and non-expansion states moved along similar trajectories, supporting the parallel trends assumption. Pre-trend coefficients were near zero with overlapping confidence intervals, followed by a sustained divergence immediately after expansion. The poster below summarizes the full set of estimates.

Figure 1 · Research Poster · Medicaid Expansion and Insurance Coverage

Visit WebsiteDownload Poster

Following the ACA's 2014 Medicaid expansion, expansion states experienced a larger and sustained drop in uninsured rates compared to non-expansion states. The gap remained persistent through 2023, the latest year included in the analysis.

The concentration of non-expansion states at the bottom of the national coverage ranking was striking. Of the 50 states, 40 have now adopted Medicaid expansion, and the states that have not continue to show markedly higher uninsured rates among low-income adults relative to the national average.

The cross-sectional pattern further reinforces the finding: in 2023, uninsured rates among low-income adults showed a clear gradient by income-to-poverty ratio, with the highest and most volatile rates occurring below 138% FPL, precisely the population targeted by expansion. Expansion states consistently fell below the national average uninsured rate across most income levels, while non-expansion states were generally above it.

Several limitations should be noted. The DiD design relies on the parallel trends assumption, which is supported by the pre-2014 event-study estimates but cannot be formally tested. The main analysis is restricted to 2014 adopters and never-expanders, meaning the results carry additional uncertainty when applied to late-adopting states, for whom estimates are approximated based on the 2014 cohort.

State-specific estimates of individuals left uninsured due to non-expansion (e.g., approximately 35,000 for Texas) are model-based and should be interpreted as approximations rather than direct counts. Finally, additional confounders may be present that simultaneously predict both the expansion decision and coverage trajectories. Further analysis is required to address these limitations fully.

Medicaid expansion has had a meaningful and sustained impact on insurance coverage among low-income adults. States that expanded consistently achieved lower uninsured rates than those that did not, and the effect has persisted for nearly a decade. The concentration of non-expansion states at the bottom of the national coverage ranking, and the estimated tens of thousands of uninsured individuals attributable to non-expansion in Texas alone, illustrate the scale of the coverage gap that remains.

With 10 states yet to expand, the findings here are not merely historical. The causal estimates and state-level counterfactuals developed in this project serve as evidence-based tools for understanding the coverage consequences of current policy decisions and for making visible the populations most directly affected by them.