Statistical Hypothesis Testing and Private Information∗

This paper provides a rationalization for using classical one-sided hypothesis tests for approving innovations, such as new drugs or treatment programs. I consider statistical testing as a strategy of the regulator in a game against informed proponents who can profit from the regulator’s acceptance decision even for ineffective innovations. Lower test size deters proponents from attempting costly clinical trials to gain approval of innovations they know ex ante to be ineffective. I show that test size equal to the ratio of the proponents’ cost of collecting data to their benefit from approval of a null innovation achieves minimax optimality for a regulator who prefers not to place any prior on the quality of potential innovations. It is also the limit of decision rules for Bayesian regulators as the assumed share of potential proponents with bad innovations converges to one.