Observational Studies of the Effect of Medicaid on Health: Controls Are Not Enough

Studies that use a covariate-adjusted, cross-sectional research design find that Medicaid patients have worse health outcomes than other patients. Policymakers frequently treat these results as causal and infer that Medicaid causally harms health. Using both hospital discharge data and health survey data, we evaluate the validity of this research design for estimating the causal effect of Medicaid coverage on health. We replicate an influential finding that Medicaid increases post-surgical mortality, and show directly that even after controlling for covariates, Medicaid patients have worse pre-operative health and lower SES than privately insured patients, and that these omitted variables bias the estimated mortality differences. Controlling for additional variables does not eliminate imbalance between Medicaid and privately insured patients in predetermined variables such as height, history of homelessness, mental health difficulties, age at smoking initiation, or a childhood diabetes diagnosis. We find no or even negative mortality differences under weak assumptions about the extent of unmeasured confounders. We conclude that existing cross-sectional estimates of Medicaid’s health effects are not valid causal estimates and that there is little scope for this research design to produce causal evidence on this question.