Primary Submission Category: Heterogeneous Treatment Effects
A causal mediation framework for examining treatment effect heterogeneity in longitudinal studies
Authors: Hanna Kim, Jee-Seon Kim,
Presenting Author: Hanna Kim*
Heterogeneity in treatment effects have been investigated as a means to obtain contextual information on how a treatment works. In longitudinal studies where treatments are provided for multiple time periods and individuals may participate with different patterns over time, variability in such treatment participation patterns can be an important factor contributing to treatment effect variability. In this study, we propose to conceptualize the effects of participating in the national Head Start program with various patterns from age three to four on children’s cognitive development as causal mediation estimands. Considering that Head Start attendance of the Head Start Impact Study (HSIS) participants was only randomized in the first year and not in the second year, causal effects such as the benefit of attending Head Start one year earlier at age three in addition to attending it at age four need to be defined as corresponding causal estimands, which is in this case the controlled direct effect of Head Start at age three given Head Start at age four. Directed acyclic graphs (DAGs) are presented to describe confounders and identification assumptions specific to each research question. Estimation methods are applied to the HSIS data to illustrate that the causal mediation framework can naturally address longitudinal treatment participation patterns as a novel source of treatment effect heterogeneity and provide substantive insights into refining well-known social interventions.