The numbers don’t add up in new Medicaid paper

In a recently released working paper, authors Angela Wyse and Bruce D. Meyer purport to show that the ACA Medicaid expansion saved 27,400 lives from 2010 to 2022. However, their own work shows that the headline result is almost certainly an overestimate.

Previous studies have shown that insurance generally has little effect on mortality or outcomes, but advocates of greater government involvement have always been eager to show otherwise. The benchmark RAND Health Insurance Experiment found that cost sharing led to reduced utilization of health care without affecting outcomes. In the late 2000s, another experiment was run in Oregon, specifically on Medicaid. The researchers found that while providing Medicaid to people increased utilization, improvements in health were not significant.  

Wyse and Meyer consider this question using a different approach. Whereas the RAND and Oregon experiments used random assignments to determine whether Medicaid, by itself, was the cause of any differences in outcomes, Wyse and Meyer use a “differences in differences” approach to determine a causal link between insurance and mortality. That methodology is frequently used in analyzing state-based policies because it compares the outcomes in states that did not implement the policy to states that did. Because the Supreme Court declared that foregoing Medicaid expansion would not endanger pre-ACA Medicaid funding, states adopting the expansion at different times allowed researchers to compare results between states that expanded and those that did not.

The analysis found that over the 2010-2022 timespan, Medicaid expansion resulted in lower mortality rates, meaning 27,400 fewer deaths over that time. As in previous research, the authors found the strongest effects for older beneficiaries. The effects were statistically insignificant for younger ages.

However, as an additional check on their results, they compared their results to people who certainly would not have been eligible for Medicaid because their incomes were too high. Medicaid expansion was only available for people whose incomes were below 138 percent of the federal poverty level. As a comparison and a check, the authors compared their results for this population to those earning 400-600 percent the federal poverty level, who should have been unaffected by expansion. They found that people in the higher income range also had improvements in mortality, suggesting that their estimates of improvement for the Medicaid expansion population are too high because they included some external effect in their estimates.

This leads to two conclusions. First, that the difference in difference framework may not be accurate for this type of study. This methodology works best when differences in the outcomes in the states can be attributed entirely to the difference in the policy under consideration. However, if there are mitigating issues, for example, differences in populations or policies that affect health outcomes, then its usefulness diminishes. The study attributes all of the reduction in mortality difference to the fact that some states expanded Medicaid and some did not, but the additional check on the higher income population suggests that mortality went down more in expansion states for reasons other than Medicaid expansion. Without knowing those reasons, or more importantly, without knowing the extent to which the estimate is driven by Medicaid expansion compared to other factors, the results here cannot be considered conclusive.

Secondly, even with that uncertainty, the authors should have incorporated the effect for the higher income people into their own results, which would have reduced the estimates of lives saved and increased the cost per life saved, suggesting Medicaid expansion has been less beneficial than the authors state.