52
The Effects of State Medicaid Expansions on Mortality: Evidence from a Synthetic Control Design

Tuesday, June 14, 2016
Lobby (Annenberg Center)

Author(s): Antonis Koumpias; Charles Courtemanche; Daniela Zapata

Discussant: Yanling Qi

This paper examines the causal effect of state Medicaid expansions in the early 2000s on state-level mortality. This is an important health policy question because these expansions served as preludes of the Affordable Care Act and provided part of the motivation for universal healthcare. Theoretically, Medicaid eligibility is expected to reduce mortality by improving access to medical care. However, very little evidence exists to date as to whether this is the case.  Sommers et al. (2012; NEJM) conduct a difference-in-difference analysis comparing counties in three states that expanded Medicaid in the 2000s to counties in adjacent control states, finding a significant reduction of 6.1% in adjusted all-cause mortality in the expansion states. However, Kaestner (2012) questions the validity of this study’s difference-in-differences design because pre-intervention mortality trends in the treatment and control states exhibited statistically significant differences.  Moreover, the levels of statistical significance may be overstated due to the well-known tendency for null hypotheses to be over-rejected when disaggregated data is used and the number of clusters is small.

We contribute to the literature on the effect of Medicaid on mortality by employing a synthetic control design, a data-driven procedure that uses as the control group the weighted average of non-expansion states that most closely resembles the pre-intervention mortality levels and trends of expansion states. The intuition is that a combination of units often provides a better comparison for the unit exposed to the intervention than any single unit alone, resulting in more reasonable counterfactuals of mortality rates in treated states in the absence of the expansion. An additional advantage of the synthetic control design is that p-values are computed through permutation tests that check for the possibility of obtaining large effects merely by chance.  We also contribute by studying a larger number of treated states than Sommers et al. (2012) (seven compared to three) and by distinguishing between short- and long-run effects. 

In contrast to Sommers et al. (2012), we do not conclude that Medicaid expansions reduce mortality.  In one of the seven treated states (New York), mortality drops by a qualitatively large amount relative to the synthetic control group after the Medicaid expansion, but the effect is not statistically significant at conventional levels when p-values are computed using the permutation tests.  In another state (New Mexico), we actually observe a statistically significant increase in mortality relative to the control group after the Medicaid expansion.  In the other five states, the effects are generally small and insignificant.  We test for impacts out to ten years post-expansion, so the lack of evidence that Medicaid reduces mortality does not appear to be due to an insufficient amount of time after treatment for deaths to be averted.