All Times EDT
Keywords: Medicaid, Causal Machine Learning, Cluster Robust Random Forest
The optional provision of Medicaid expansion, through the Affordable Care Act (ACA), has triggered a national debate among diverse stakeholders regarding the impacts of Medicaid coverage on various dimensions of public health, costs, and benefits. Randomized experiments like the Rand Health Insurance Experiment and the Oregon Health Insurance Experiment have generated some credible estimates of the average treatment effects of access to insurance. However, identical policy interventions can have heterogeneous effects on different subpopulations. This paper uses data from the Oregon Health Insurance Experiment to estimate the heterogeneous treatment effects of access to Medicaid on health care utilization, preventive care utilization, financial strain, and self-reported physical and mental health. I detect heterogeneous treatment effects using a cluster-robust generalized random forest, a causal machine learning approach. I find that the impact of Medicaid is more pronounced among relatively older non-elderly and poorer households, which is consistent with standard adverse selection theory. Furthermore, I implement the ``efficient policy learning,” another machine learning strategy, to identify policy changes that prioritize providing Medicaid coverage to the subgroups that are likely to benefit the most. On average, the proposed reforms would improve the average probability of outpatient visits, preventive care use, overall health outcomes, having a personal doctor and clinic, and happiness by a range of 2% to 9% over a random assignment baseline.