ACIC 2022 Poster Presentation

Poster Session 1 on May 24th 5:00pm-7:30pm PT

Berkeley, CA

Abstract

For the first time, I present empirical results to the motivating example of the instrumented common confounding (ICC) approach: Returns to education. Specifically, we estimate the causal effect of obtaining a BA degree on household net worth at 35. This typical econometric identification problem features an obvious common confounder, ability or cognitive skill. Yet, endogeneity also stems from education being (at least in part) a result of optimal choice by economic agents with heterogeneous costs and benefits. Many previous solutions have been criticised for their endogenous or weak instruments. ICC is a useful alternative in the sense that it explicitly accounts for the common confounder. Hence, otherwise clearly endogenous instruments can be used. Our instrument, pre-college test scores, is strongly significant for obtaining a BA degree. It is also strongly related with ability, which is perfectly fine in ICC. We adopt the ICC identification approach to identify a locally linear model conditional on background covariates. Estimation of the locally linear parameters is conducted with generalised random forests.

Date
24.05.2022 17:00 — 19:30
Location
University of California, Berkeley
2495 Bancroft Way, Berkeley, CA 94704
Christian Tien
Christian Tien
PhD Student in Economics (3rd yr)

My research interests include causal inference, specifically identification, and machine learning.