Beyond Averages: Portraying Treatment Effect Variation

Jee-Seon Kim, Graham Buhrman, Xiangyi Liao

DOI: 10.64028/xvhr927961

Keywords: Causal Inference, Heterogeneous Treatment Effects, Causal Machine Learning, Conditional Average Treatment Effect (CATE), Random Slope Models, Latent Class Analysis, Unobserved Heterogeneity

Abstract: This paper addresses distinct forms of treatment effect variation that commonly arise in social and behavioral science research. Differences in the nature of relationships among variables or in the contexts where treatments are implemented can lead to both quantitatively and qualitatively different patterns of treatment effect heterogeneity. Such variation may involve interactions between treatments and covariates, conditional average treatment effects defined by observed characteristics, random treatment coefficients across clusters, or differences in treatment effects across unobserved subpopulations. By highlighting distinctive features of treatment effect variation, this paper emphasizes the importance of addressing heterogeneity as an integral part of study design and research objectives, rather than treating it as a secondary or post hoc concern. This paper concludes by emphasizing the need for a structured and conceptually grounded framework to better identify, interpret, and apply heterogeneous treatment effects.

Submission Number: 20