Evaluating the Mediating Processes of Educational Interventions in Clustered Settings

• In Event: Advanced Techniques in Multilevel Modeling and Treatment Effect Analysis

Abstract

Educational programs for individual subjects are often administered in clustered settings, complicating the evaluation of mediational processes using conventional parametric causal mediation analysis methods based on single-level models and standard error estimation techniques. This study suggests enhancing the estimation of natural direct and indirect effects through cluster bootstrapping and fixed effects modeling, offering robustness against cluster-level confounding and variation. A real data analysis of the early Head Start program's impact on vocabulary development illustrates how different approaches to addressing clustering may lead to different conclusions. A pilot simulation demonstrates that fixed effects modeling with cluster bootstrapping yields more accurate estimates than single-level approaches. This underscores the need to address clustered data structures for valid causal mediation analysis in educational research.

Authors

• Hanna Kim, University of California - Santa Cruz

• Jee-Seon Kim, University of Wisconsin - Madison