New Econometric Approaches to Models with Unobserved Heterogeneity

Princeton University, Princeton NJ

Investigators

Abstract

This project develops new methods for the statistical analysis of certain kinds of economic data. The first part of the research focuses on developing a new method for performing data analysis that requires less computational power than other competing methods. The result will speed data analysis for a class of problems in several different fields of economic analysis. The second part of the research considers a new method for using statistics to measure possible causal effects to measure treatment effects. The work will allow other researchers to improve empirical analysis in two different ways. Because computational costs are lower, researchers will be able to analyze larger data sets or consider larger and more realistic models. Second, the work will improve the methods empirical researchers use to test for causal effects. The PI plans to involve students in the research, and will make computer programs for the statistical methods available at no cost to the research community. The research has two components. The first project develops a new statistically and computationally efficient method of estimation by simulation. Methods such as the method of simulated moments and indirect inference are widely used. However, when a finite number of simulation draws are used, these estimators are consistent but in general are not efficient. The new method will render these estimators efficient with as little as one simulation draw per observation. The second project considers issues of inference in instrumental variable models with heterogeneous treatment effects. In these cases, researchers interpret two stage least squares results as estimating local average treatment effects. While this interpretation is popular, it poses significant challenges. In particular, in these kinds of models one can no longer use the J-test of overidentifying restrictions to test the validity of exclusion restrictions. The project will develop a novel methodology for diagnostics of the exclusion restriction, a method motivated by ideas of subgroup robustness analysis.