This project was created by myself and two co-authors exploring the non-parametric CATE estimation technique for MA750 Fall 2021 Boston University under Professor Ashis Gangopadhyay. My primary focus was on the application of CATE estimation to the effect of covid on voter turnout for different races.

Abstract:

As of the 21st century, there is a wealth of observational data in many industries and contexts. Often, researchers/scientists are interested in understanding the causal effect of some underlying mechanism of those data, but without paying for an expensive(and often impossible) randomized control trial. However, under certain modeling assumptions, the observational data can be used to estimate causal effects, or the average treatment effect (ATE), of interventions. Sometimes we wish to know the effect of an intervention in a subpopulation of observations, or the conditional average treatment effect (CATE). In the following paper, we explicate a few estimators of the CATE, and for the DR-Learner, derive theory regarding finite sample bounds on the error of estimation. We end with an observational study of the causal effect of COVID-19 deaths on voter turnout in different racialized populations.