Mostly Harmless Statistical Decision Theory

Cornell University, Ithaca NY

Investigators

Abstract

This award will support research to develop innovative methods for policy evaluation. The wide adoption of experimental and observational data to evaluate policy changes has revolutionized policy making in the twentieth century. However, policy makers face several complications when making data-driven policy choices. For example, there may be concerns about whether the results from the experiments might be generalizable or the experiments may feature imperfect compliance. These problems imply that the available data only provide partial information about the welfare associated with a new policy. This award will support research that uses the framework of Statistical Decision Theory to develop new decision rules that can be used to make policy choices in these complicated, but realistic, situations. The methods to be developed are easy to solve mathematically and allow researchers to make the most out of the available data when making policy decisions. The new methods will be useful in several areas of economics, econometrics, biostatistics, and other social sciences. The researchers will write a statistical code to implement the new methods. Besides improving economic science, the results of the research supported by this award will improve policy design and thus improve economic performance and the wellbeing of US citizens. This award supports research that uses three projects to develop new methods of estimation and inference by connecting Statistical Decision Theory to economics and econometrics. The first project studies a binary policy choice problem in which payoff relevant parameters are partially identified. A key research output will be a novel decision rule that makes explicit use of an estimator of the identified set for the payoff relevant parameters. The rule will be justified by a new finite-sample optimality theory that refines the minimax regret principle. The second project concerns a policy maker uncertain about the true policy effect but has access to data generated by an RCT with imperfect compliance. The research will give a decision theoretic interpretation of the usual “IV” estimators, irrespective of whether it admits a causal interpretation. The researchers will apply these results to study data-driven policy making with experimental and observational data. The results of this research will improve methods of data-driven policy decision-making. Besides improving economic science, the results from the research supported by this award will improve policy design and thus improve economic performance and the wellbeing of US citizens. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.