Health Insurance and Mortality
*Names in bold indicate Presenter
We use a "differences in differences" (DiD) study design that compares trends in mortality in Medicaid expansion states relative to non-expansion states, after versus before Medicaid expansion. We study deaths from a broad set of healthcare amenable causes, as well as specific major healthcare-amenable causes of death (cancer, heart disease, diabetes, respiratory disease), for persons under age 65. We also use a triple-difference specification, for which the third difference is deaths at ages 55-64 (hence affected by Medicaid expansion) to death at ages 65-74 (hence already Medicare insured). We also assess whether county-level decreases in uninsurance from before to after Medicaid expansion predict mortality, based on the any effects are more likely for counties with larger drops in uninsurance rates. Alternatively, we substitute baseline 2013 county uninsurance rates for the change in uninsurance rates. We also study treatment effects for subsamples stratified on gender, race, ethnicity, and education.
We do not find evidence of significant reductions in healthcare-amenable mortality following Medicaid expansion. While we find statistically significant results in some DiD specifications, most of these vanish in the triple difference specification; the DiD results thus appear to reflect non-parallel trends between expansion and control states, that are due to causes other than Medicaid treated.
We then use simulation methods to investigate the power of our research design to detect plausible effect sizes. We conclude that even a national-scale DiD study, with microdata on all deaths, is severely underpowered. Effects that are large enough to be detectable with reasonable power are implausibly large. Put differently, the confidence intervals around our estimates do not rule out fairly large effects. We conclude that it would be extremely challenging for a study, relying on death certificate data with mortality as the outcome, to have sufficient power to enable detecting effects of Medicaid expansion on population mortality.