Randomized policy trials are becoming an increasingly important part of social science research, particularly in economics. The claim is often made that we draw on methodologies developed for clinical trials in medicine, but there are some important differences, the most fundamental of which is that it is impossible to blind subjects as to their status when the treatment is a social program. This means that we cannot avoid placebo effects, but I argue that these are a legitimate part of the treatment effect of policies. 

Another important difference relates to the ways in which subjects are recruited into experiments. We may think that those willing to subject themselves to an experiment are less risk averse than the average person; this behavioral trait may not alter the effect of a drug, but is very likely to change the way that subjects respond to policy. Hence the treatment effect measured through an 'ethical' research design which obtains informed consent may not be the same as that which would be observed in the population as a whole. For this reason, many social science experiments do not inform treatment or control subjects that the implementation of the program is being done in a randomized fashion. This methodology reveals the correct treatment effect, but it has ethical problems and makes such experiments prone to collapse if treatment and control units communicate with each other during the course of the study.