After a brief lull, the debate over randomistas seems to have revived itself. Martin Ravallion’s interesting short article, ‘Should Randomistas Rule?’, is a sharp indictment of the tendency to allow the method to overcome the questions we need to ask of development interventions. I highly recommend the article for a thorough read, and shall leave you with some highlights: (1) Randomizing the selection of projects and locations for randomized control trials (RCT) is unlikely; (2) variations in ‘take-up’ in treatment populations a serious problem and challenge claims of internal validity; (3) spill-over effects and ‘corruption’ of control populations is also a serious challenge to the robustness of the study; and (4) the problems with external validity, a criticism that has been done to death when arguing against RCTs
However, the point I found most striking is probably one of the most obvious ones (that my dumb brain probably knew but never articulated): An intervention tested using an RCT ‘mixes low-impact people with high-impact people’; and in complete contrast, an actual intervention will tend to have higher representation from the high impact type of people because it will be non-random and scaled up. This then completely alters the nature of the intervention itself and its potential for change through the positive externalities it creates.