Estimating population average treatment effects from experiments with noncompliance

This paper extends a method of estimating population average treatment effects to settings with noncompliance. Simulations show the proposed compliance-adjusted estimator performs better than its unadjusted counterpart when compliance is relatively low and can be predicted by observed covariates. We apply the proposed estimator to measure the effect of Medicaid coverage on health care use for a target population of adults who may benefit from expansions to the Medicaid program. We draw randomized control trial data from a large-scale health insurance experiment in which a small subset of those randomly selected to receive Medicaid benefits actually enrolled.