If Medicaid Expansion Saves Lives, We Still Don't Know How
A famous study can easily be misinterpreted.
In a recent Open Health Policy post, my colleague Liam Sigaud discussed how emerging evidence casts doubt on previous studies that found that Medicaid expansion reduced mortality rates.
A few months ago, as part of our reading group, my colleagues and I delved into Medicaid and Mortality: New Evidence From Linked Survey and Administrative Data, a 2021 paper by Miller, Johnson, and Wherry published in the Quarterly Journal of Economics. It is likely one of the best analyses of the health impacts of health insurance outside the seminal studies of the RAND Health Insurance Experiment from the 1970s and the Oregon Health Insurance Experiment from 2008.
This made for a lively discussion, in part because the paper’s central finding that Medicaid expansion reduces mortality among near-elderly adults by 9.4% contradicts the results of the experiments mentioned above, which generally found little direct impact on health outcomes and no discernible effect on mortality. While there was widespread agreement about the paper’s many strengths, I expressed concern about how the authors interpreted their findings. I was unable to convince my colleagues that there was much to be concerned about. Now that the dust has settled, I wanted to give this another shot and try to express my thoughts more clearly; maybe this time I’ll be more persuasive.
Brief Recap of the Paper
The authors do a heroic job linking data from the 2008-2013 editions of the American Community Survey with administrative data on deaths from the Census Bureau and Medicaid enrollment records from the Centers for Medicare and Medicaid Services. Since not all states expanded Medicaid, and the expansions that did occur took place at different times, the authors use a difference-in-differences design to discern whether expanding Medicaid reduced death rates in expansion states among those most likely to gain coverage and for whom health insurance could plausibly impact mortality, relative to non-expansion states.
Figure II below, taken from the paper, plots their main finding. The coefficients on the left side of Figure II are all close to 0 and never statistically significant, showing death rates were on parallel trends in both groups of states leading up to expansion. But starting with the first year of expansion (labeled 0 in event-time), they find a reduction in annual mortality in the range of 0.1 and 0.2 percentage points, i.e., a 9.4% reduction relative to the sample mean.
MJW also presents a series of “placebo tests.” These tests repeat the same analysis for groups we wouldn’t expect Medicaid expansion to impact much, if at all – notably higher income people and folks 65 and over who are eligible for Medicare. These tests can be quite convincing when they don’t find an effect, as they provide evidence that there isn’t some unobserved factor occurring in expansion states at those exact times that is reducing mortality for everyone. The results show there doesn’t seem to be any effect among people 65 and over, but this is less clearly the case for higher income people.
A Problem of Interpretation
My problem with all of this is how it’s interpreted. The authors write, “Our analysis provides new evidence that expanded Medicaid coverage reduces mortality rates among low-income adults.”
This is technically a correct representation of the findings. But I think most people would interpret it to mean that health insurance reduces mortality. The distinction seems subtle but is actually important. The “health insurance reduces mortality” interpretation implies the reductions in mortality should primarily have been found among individuals who were uninsured prior to Medicaid expansion, which, as it turns out, is not really what the study found.
When, in the section after the placebo tests, the authors limit their analysis to the 30% of people who reported being uninsured prior to expansion, the results, quite surprisingly, were only slightly higher: a mortality reduction of 0.150 percentage points, compared to 0.132 for the full sample.
The authors note that “it is somewhat surprising that the differential in the reduction in mortality between those who were uninsured at the time of the ACS and the overall population of low-income older adults is not even larger, particularly because they experienced larger gains in Medicaid enrollment.” But they don’t go much beyond this, even though it actually suggests expanding Medicaid coverage reduces mortality, not primarily through insurance coverage, but some other channel.
Whether that channel is lower cost-sharing, the services covered by Medicaid vs. other types of insurance, or a diminution in rural hospital closures, my point is that these are all mechanisms that are quite different from expanding insurance coverage, which is what most people, including most economists, have in mind when they think about the health impacts of Medicaid expansion. As expanding Medicaid coverage is a blunt instrument, doing the most good with Medicaid funds requires understanding the mechanisms through which coverage impacts health. This ought to be the priority of more researchers – after all, lives depend on it.