Assignment procedure biases in randomized policy experiments

Randomized controlled trials (RCT) have become a dominant empirical tool in applied economics. The internal validity of RCTs crucially depends on the (implicit) assumption that the procedure assigning subjects to treatment and control groups has no effect on behavior. We show theoretically that this assumption is violated when people are motivated by belief-dependent preferences and care about the intentions of others. The choice of assignment procedure influences subjects’ behavior and, consequently, the RCTs’ findings. Strikingly, even a credible and explicit randomization procedure does not guarantee an unbiased prediction of the impact of a general introduction of the policy.