Causal Mediation Analysis for Effect Heterogeneity
Observational Studies · 2026 · 12 (1) · 17-41
DOI: 10.1353/obs.2026.a985143 ↗Kurz gefasst
Fachartikel in der Zeitschrift Observational Studies, 2026. Verfasst von Jiaqing Zhang und 1 weiteren Autor:innen. Beteiligte Einrichtung: Adobe Inc. Der Volltext ist frei in PubMed Central verfügbar.
Abstract (Originalsprache)
Understanding the impact of an exposure on an outcome through a mediator is crucial in many fields of study. However, the effect often varies across subgroups. Addressing why and how an exposure gives rise to an outcome differently for subsets provides a deeper understanding of what works for whom and through which mediator(s). This paper explores causal mechanisms by focusing on explaining effect heterogeneity via causal mediation analysis. The counterfactual framework is extended to heterogeneous effect decomposition. We provide nonparametric definitions, identification assumptions, and analytical formulas for various direct and indirect effect heterogeneity measures. The decomposition of effect heterogeneity provides insightful causal implications regarding the question for whom and in what context the effect operates. The proposed methodology is demonstrated through an application of effect heterogeneity decomposition of neighborhood poverty's impact on mental health, as modified by sex, using data from the National Longitudinal Study of Adolescent to Adult Health (Add Health).
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Adobe Inc.
- Biostatistics Columbia University.
Zitieren
APA
Zhang, J., & Valeri, L. (2026). Causal Mediation Analysis for Effect Heterogeneity. Observational Studies, 12(1), 17-41. https://doi.org/10.1353/obs.2026.a985143
Vancouver
Zhang J, Valeri L. Causal Mediation Analysis for Effect Heterogeneity. Observational Studies. 2026;12(1):17-41. doi:10.1353/obs.2026.a985143
BibTeX
@article{Zhang2026,
author = {Zhang, Jiaqing and Valeri, Linda},
title = {Causal Mediation Analysis for Effect Heterogeneity},
journal = {Observational Studies},
year = {2026},
volume = {12},
number = {1},
pages = {17--41},
doi = {10.1353/obs.2026.a985143}
}
EUVIMED