Disentangling Groupwise Heterogeneity in the Mediating Relationships of Yearly Renewable Education Programs

Hanna Kim, Jee-Seon Kim

DOI: 10.64028/poae705266

Keywords: causal mediation analysis, effect heterogeneity, potential outcomes framework, treatment-mediator interaction, regression-based estimation

Abstract:

Although recent advances in causal mediation analysis have clarified the necessary conditions for identifying direct and indirect effects using intuitive closed-form estimation, these effects are often assumed to remain constant between subpopulations. In the context of yearly renewable education programs, this assumption implies that the effects of initial program attendance - both its long-term effects and the mediated effects through subsequent program attendance - are deemed homogeneous, despite the presence of distinct subgroups who may benefit differently from such programs.

We introduce a set of regression-based causal mediation estimators that account for such groupwise heterogeneity not only in the main effects of the exposure and mediator on the mediator and the outcome, but also in the exposure-mediator interaction within the outcome model. A real data analysis illustrates the importance of examining groupwise heterogeneity in causal mediation analysis, highlighting markedly different effects of the early and regular Head Start programs on English receptive vocabulary skills among Dual Language Learners and primarily English-speaking children. We conclude by discussing how extensions of causal mediation analysis methods can advance the study of heterogeneity in developmental processes and outline directions for future research.

Submission Number: 23