Abstract
Ethical concerns aside, there is nothing inherently wrong with using randomized control trials for intellectual inquiry in development economics. A fundamental problem arises, however, in claiming that results from experimental and quasi-experimental methods are more credible than other sources of evidence for policy. Specifically, there is a contradiction between rejecting econometric assumptions required for identifying causal relationships using non-experimental data, and accepting assumptions required for extrapolating experimental results for policy. I explain this tension and its implications, then discuss recent efforts -- including the use of replication and machine learning methods -- to circumvent it. Such attempts remain inadequate, and assertions in the 2019 Nobel Award are therefore either premature or misplaced. Use of pluralistic approaches negates these sharp contradictions, but requires abandoning any special status for experimental methods.