Synthesizing Psychometrics and Causal Inference: Applications of Latent Variable Models to Treatment Heterogeneity, Psychological Networks, and Learning Transfer

Abstract

Evaluating the effectiveness of educational interventions presents two distinct challenges: measurement of outcome variables through psychometrics and estimation of program impact through causal inference techniques. Education research has been unnecessarily constrained by the separation of these two domains, and their synthesis is a promising area of methodological research that can inform ongoing substantive debates in the field. I contribute to this synthesis through three dissertation studies. In Study 1, I demonstrate how the common approach to estimating treatment heterogeneity on test score outcomes using interaction effects is susceptible to bias if treatment effects are correlated with item easiness. I show how analysis of item-level treatment effects can eliminate the bias. In Study 2, I explore network psychometrics, in which psychological traits are considered complex systems rather than unidimensional continua. While common in other fields, network psychometric models are rare in education research, in part due to computational constraints. I demonstrate how to leverage item response theory modeling approaches to make inferences about causal effects on network structures when direct estimation is not possible and apply the proposed approach to randomized controlled trials in education and related fields. In Study 3, I extend an analysis of a content literacy intervention to examine potential mechanisms of interdisciplinary learning transfer from reading to math using latent mediation analysis. I conclude by summarizing the implications of these studies for education research.