CAREER: Robust Econometric Inference

Project Outcomes Statement

This project developed new methods for data-driven decisionmaking and the measurement of uncertainty in a range of economic contexts. Providing estimates to guide decision-making, for instance estimating the effects of different policies on outcomes important to policymakers and the public, is a central goal in much of social science research. In addition to providing estimates, it is also important to accurately quantify uncertainty, so that consumers of research know which research conclusions can be relied upon and which reflect our current best guess but are likely to be revised in the future. Unfortunately, in many important economic problems it is difficult to construct reliable measures of uncertainty, due to a range of factors including limited information in the data about our questions of interest, the risk of mean-reversion in estimated effects, and the fact that our models are almost always simplifications of reality, and so are wrong in important respects.

The papers written for this project provide new methods to make reliable, data-driven decisions and accurately quantify uncertainty in a range of important economic contexts. Examples include new methods to correct for mean-reversion in estimated effects and correct for bias in the published literature, methods to reliably quantify uncertainty in economic models where the data are not very informative, and novel approaches to account for the possibility that our economic models may be wrong. These methods were developed across 10 papers, all of which are published in peer-reviewed journals, and apply to a wide variety of economic questions, ranging from estimating the effectiveness of programs helping low-income families move to high-opportunity neighborhoods to estimating the extent to which publication bias distorts the published literature on the effect of minimum wage increases.

In addition to novel research, this grant also supported a pair of public lectures, recorded and posted publicly on the website of the National Bureau of Economics Research, on best practices for a set of problems where the data are not very informative about the quantities of interest (specifically, instrumental variables problems with weak instruments), and what researchers can do to address the resulting issues.