Random Assignment Before Determining Treatment Eligibility: Power Considerations and Instrumental Variable Estimation

• In Event: Novel Data and Advances in Longitudinal Methods in Criminology

Abstract

Instrumental variable estimation (IVE) has been applied to the problem of treatment noncompliance in random assignment (RA) studies, which leads a portion of those assigned to treatment not to receive treatment. We extend this logic to another reason for not receiving treatment -- post-RA determination of treatment ineligibility. Although standard practice is to determine eligibility before random assignment (DEBRA), there may sometimes be advantages to random assignment before eligibility determination (RABED), and IVE methods can then produce unbiased estimates. However, RABED comes with statistical power costs. In this paper, we use simulation methods to quantify those power costs and to explore remediation costs in terms of increased sample sizes.

Authors

• Akiva Liberman, Urban Institute

• Michael Jacob Lebron, University of Maryland

• Robert Brame, University of Maryland