New Econometric Methods for Estimation and Inferences in Nonlinear Econometric Models

Princeton University, Princeton NJ

Investigators

Abstract

This project develops new tools and insights in three topics that will be useful to empirical microeconomists and social scientists. The first concerns the sample selection problem in which data are not representative of the population of interest. Not dealing with this problem can lead to incorrect conclusions. The investigator provides a better understanding of this problem and develops methods to avoid strong assumptions that are typically imposed by traditional approaches. The second topic is related to panel data that contain a large number of individuals or firms with more than one-time period. The investigator examines more general versions of existing nonlinear models for panel data by asking whether it is possible to exactly learn the object of interest even with infinite amounts of data. If the answer is negative, this project further constructs bounds for the parameters of interest. The third topic is concerned with how researchers calculate the statistical uncertainty associated with the analysis of a data set. The investigator develops tools that can be used when existing tools are computationally infeasible, as opposed to tools that are better from a theoretical point of view. The tools developed in this research can be used in many areas of economics and social sciences. To facilitate this, the investigator also produces computer programs for other researchers to utilize in their own work. This research develops new tools and insights that will be useful to empirical microeconomists. The project has three parts. The first topic is sample selection models. These models have a long history in economics with applications in many areas. To estimate sample selection models, previous studies often assumed exclusion restrictions, in that some variables influence the selection into the sample, but do not have an effect on the outcome of interest. This project provides a better understanding of this issue and develops methods to alleviate it. The second part of the project investigates panel data. The investigator focuses on models where the variable of interest is binary and investigates the extent to which it is possible to estimate the parameters of such models. Finally, this research develops tools for researchers to more easily make statistical inference. Specifically, it develops a simpler version of the so-called "bootstrap", which can be computationally more convenient for complicated models than alternative procedures. 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.