When estimating causal effects, researchers would like to control for certain confounding factors, but these variables may not be perfectly observed. A widely adopted approach is to use proxy variables in place of the unobserved ideal controls. However, this approach generally suffers from measurement error bias. In this paper, I develop a new identification strategy that addresses this issue. In particular, I use proxy variables for the unobserved confounding factors to construct a random variable conditional on which treatment variables become exogenous. The key idea is that under appropriate conditions, there exists a one-to-one mapping between the distribution of unobserved confounding factors and the distribution of proxies. Thus, "matching" on the conditional distribution of the proxy variables ensures that matched units have the same conditional distribution of the confounding factors. To satisfy an overlap/support condition, I use an additional variable, termed excluded variable, which satisfies certain exclusion restrictions and relevance conditions. I illustrate wide applicability of my method through examples.
About the Speaker:
Kenichi Nagasawa is an Assistant Professor of Economics at the University of Warwick. His main research interest lies in theoretical econometrics. He has worked on bootstrap-based inference procedures for estimators with non-standard asymptotics and identification of causal effects in selection models in economics. He obtained a Ph.D. in Economics from the University of Michigan in 2019.
Zoom Meeting :https://us02web.zoom.us/j/88054427850
Meeting ID：880 5442 7850
Your participation is warmly welcomed!