Deep Learning for Propensity Score Estimation

• In Event: Sharp Statistics in Educational Research

Abstract

Machine Learning (ML) in the age of big-data in education is still in its infancy but has already proved to be a useful analytic tool, especially with supervised ML methods such as decision trees, and neural networks. As educational researchers continue to use propensity score methods to estimate causal effects from observational data, new approaches from ML should be evaluated to see how they perform against traditional methods. In this simulation study, we applied logistic regression and Deep Neural Networks (DNN) to estimate propensity scores to evaluate whether DNN could outperform the standard logistics regression. Results showed that DNN and Logistic regression performed equally well across conditions and sample sizes.

Authors

• Alberto Guzman-Alvarez, University of Pittsburgh

• Paul Wesley Scott, University of Pittsburgh