Informative Identification-Robust Inference for Subvectors and User-Chosen Parameters

University Of Colorado At Boulder, Boulder CO

Investigators

Abstract

Economists and policy makers lack rich data sets and statistical methods to draw valid conclusions given complex human behavior and the statistical models used to study them. This award funds research that will develop new econometrics and statistical methods that allow researchers to draw valid conclusions from data in many settings. This research will yield new theoretical insights for economists and statisticians as well as develop new methods for applied research in economics, statistics, other social sciences, biostatistics, epidemiology, and medicine. In addition to theoretical advances, the research will also produce statistical software that will implement the new methods, thus allowing applied researchers to use these new methods in their research. Results from this research will provide new and improved methods that increases efficient decision making in social, behavioral, and medical sciences. The research results will improve efficiency, increase productivity, economics growth, and wellbeing of citizens not only in the US but also around the globe. This award supports research that focuses on two leading classes of identification-robust inference problems in econometrics: inference robust to weak and partial identification. The current econometrics literature has only developed identification-robust subvector inference methods that are either specific to a small number of models or produce tests and confidence sets that are uninformative and computationally costly for applied researchers. This research will develop a generally applicable reparameterization procedure that allows informative identification-robust subvector inference by allowing researchers to plug estimates of nuisance parameters into test statistics, rather than projecting over them. Additionally, current econometrics literature on partial identification-robust inference lacks tools for inference on treatments or policies chosen by estimating the identified set of best-performing treatments or policies. The proposed research will develop new methods of producing valid confidence intervals for how well one can expect chosen treatments or policies to perform in these settings. The results of this research will provide improved and innovative methods for decision making in social, behavioral, and medical sciences. The research results will improve efficiency in decision making and improve the wellbeing of citizens not only in the US but also around the globe. 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.