CAREER: Sensitivity Analysis in Econometrics

Duke University, Durham NC

Investigators

Abstract

Economists use data and theory to perform two key tasks for improving national welfare: they evaluate the performance of existing public policies and predict the outcomes of newly proposed policies. These evaluations and predictions often rely on behavioral assumptions which are difficult or impossible to verify with data. Researchers use many approaches to assess the importance of these assumptions, but many of these approaches have serious problems which are not widely understood. This CAREER research project will develop new and better tools to allow researchers to test whether underlying assumptions are correct and therefore improve the accuracy of predictions and policy evaluations. The research consists of three projects that focus on measuring inappropriate assumptions, how to deal with models that are known to be false, and building a set of toolkits to test whether research results significantly change when the underlying assumptions change. The research project will create a series of publicly available videos and software to ensure that the new methods are widely disseminated, understood, and used by practitioners. The results of this research project will improve economic and other policy decision making as it improves the quality of prediction and policy evaluation. In so doing, it significantly contributes to economic growth and living standards of US citizens. This CAREER research project advances the literature on sensitivity analysis as it builds a set of tools used to assess the robustness of empirical research and policy evaluation. The research project has three components designed to improve on current methods. In the first project, the PI will study ways for researchers to measure and calibrate violations of assumptions. Specifically, this project will address which form of assumption violations should researchers be worried about, and how researchers can tell when these violations are large. The second project will study how to deal with models which are known to be false because they usually fail overidentification test, a common occurrence in structural models. The third project will develop general tools for doing sensitivity analyses in a wide variety of models. Together, these projects will help make the practice of sensitivity analysis more robust, thus improving the quality of empirical analysis throughout economics and the social sciences more broadly. The results of this research project will decision making, hence improve the living standards of Americans as well as establish the US as the global leader evaluation models. 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.