Deep Learning for Investigating Causal Effects with High-Dimensional Data: Analytic Tools and Applications to Educational Interventions

Alberto Guzman-Alvarez

Quantitative education research has the potential to be revolutionized by recent developments in machine learning. Realizing these possibilities, however, requires scholars to bridge the worlds of educational research and computer sciences. Through my dissertation, I aim to merge advances in deep learning and causal inference to enable researchers to assess program impacts using quasi-experimental methods with high-dimensional data. First, I will develop a new analytical procedure incorporating advances in Deep Learning, specifically Deep Neural Networks, to estimate propensity scores with procedures that flexibly accommodate both high-dimensional data and complex relationships between treatment selection and observable characteristics. Preliminary results suggest that these methods outperform most traditional modeling strategies, particularly when complexities (e.g., non-linearities, interactions) are present in the selection model. Second, I will incorporate this approach into a causal mediation framework that employs a propensity score-based weighting strategy to allow researchers to test potential mechanisms underlying treatment effects. In addition to the methodological contributions, my dissertation also will make substantive contributions to the applied literature. I will employ these methods to evaluate a large-scale college access intervention that offered high school students critical supports to ease their transition into college during the COVID-19 pandemic. In addition to academic papers, I will produce an open-source R package to allow applied researchers to conveniently implement my method.

About Alberto Guzman-Alvarez

Alberto Guzman-Alvarez is a Ph.D. student in the Department of Educational Foundations, Organization, and Policy at the University of Pittsburgh, School of Education. Proudly raised by Mexican immigrant parents, Alberto approaches his research through his experience as a first-generation student, believing that data can be a tool for social justice. His research focuses on applying and developing quantitative methods for evaluating the effectiveness of education policies and interventions. In particular, he is interested in college access issues affecting first-generation, historically marginalized students. He is broadly interested in educational data science, computational social science, and causal inference. In his dissertation, he merges his methodological and policy interests by developing new methods for evaluating quasi-experimental interventions with high-dimensional data and applying these methods to a large-scale college access intervention.He has published both methodological and policy-focused articles in journals such as Educational Evaluation and Policy Analysis and the Journal of Research on Educational Effectiveness. In his recent policy manuscript, he and his co-author analyzed the administrative cost to colleges and universities of the FAFSA verification mandate. This work has received national attention and has been cited in a congressional committee hearing on college access.He is the K. LeRoy Irvis Pre-Doctoral Fellowship recipient and was recognized as an Equity and Inclusion Fellow by the Association for Public Policy Analysis and Management (APPAM). Prior to his doctoral studies, Alberto earned his B.S. in Psychology from the University of California, Davis, and a Masters in Research Methodology from the University of Pittsburgh.