Experimental Estimation of Heterogeneous Treatment Effects Related to Self-Selection

Social scientists widely regard the random-assignment experiment as the gold standard for making causal inferences about the world. We argue that it can be improved. For situations in which self-selection and heterogeneity of treatment effects exist, an alternative experimental design that retains random assignment to treatment or control and supplements it with some self-selection of condition offers a clear advantage. It reveals the average treatment effect while also allowing estimation of the distinct effects of the treatment on those apt and inapt to experience the treatment outside the experimental context.