Patient enrollment in medical trials: Selection bias in a Randomized Experiment

Abstract. Self-selection can bias estimates of treatment effects from ran-domized experiments if one is interested not merely in the effect of treatment on the treated, but in extrapolating results to the general population. This paper employs the Roy model to study this problem in the context of medical trials. The main insight is that, as the probability of receiving active treat-ment rises, patients who are less optimistic about new treatment will begin to enroll and estimates of treatment effects will fall. This, in turn, implies that selection bias is positive. These findings are confirmed with data from trials of ulcer medications