Marginal Treatment Effects and the External Validity of the Oregon Health Insurance Experiment

Add to Calendar
Date and Time: 
Monday, February 8, 2016
2:00 pm – 3:30 pm
Amanda Kowalski
Yale University

I examine the external validity of the Oregon Health Insurance Experiment (OHIE) usingmarginal treatment e effect (MTE) methods. A central nding from the OHIE is that individuals who gained health insurance through the OHIE lottery increased their emergency room utilization (Taubman et al. [2014]). Using data from the OHIE, I find that that the marginal treatment effect of health insurance on emergency room utilization was positive for some types of individuals and negative for others. Thus, the external validity of the finding that emergency room utilization increases when health insurance expands depends on the types of individuals who gain coverage. Using data from Kolstad and Kowalski [2012], I reexamine the types of individuals who gained coverage through the Massachusetts health reform, which established a mandate for uninsured individuals to purchase coverage or pay a penalty, among other interventions. I find that the Massachusetts individuals induced to gain health insurance appear more similar to the Oregon individuals that decreased their emergency room utilization than they do to the Oregon individuals that increased their emergency room utilization. Furthermore, on the whole, individuals who entered a lottery for health insurance coverage in Oregon likely had a higher desire to use the emergency room than individuals who gained coverage when a mandate required them to do so in Massachusetts. Therefore, it is not surprising that that literature finds decreases in emergency room utilization in Massachusetts despite increases in Oregon because the two state policy interventions expanded coverage to different types of individuals. Given that my findings deepen our understanding of the external validity of the OHIE, I conclude that MTE methods are a valuable addition to the standard toolkit for analysis of experiments and quasi-experiments with binary treatments. Based on my analysis, I highlight several research design considerations that can maximize the ability of MTE methods to shed light on external validity in future applications.