Targeted Treatment Assignment Using Data from Randomized Experiments with Noncompliance

This paper considers randomized experiments with noncompliance where individuals in the treatment group become eligible for a treatment but some do not receive it. We study the estimation and evaluation of treatment assignment policies targeted to individuals on the basis of pretreatment characteristics. We consider a decision problem where the policy determines eligibility, which is costly, and compliance continues to be imperfect. Then optimal policies prioritize by a weighted average of the intent-to-treat effect of eligibility on outcomes and the treatment effect of eligibility on receiving the treatment. We illustrate the ideas using data from the Oregon Health Insurance Experiment.