Assessing the Treatment Effect Heterogeneity with a Latent Variable
The average treatment effect (ATE) is popularly used to assess the treatment effect. However, the ATE implicitly assumes a homogenous treatment effect even amongst individuals with different characteristics. In this paper, we mainly focus on assessing the treatment effect heterogeneity, which has important implications in designing the optimal individual treatment regimens and in policy making. The treatment benefit rate (TBR) and treatment harm rate (THR) have been defined to characterize the magnitude of heterogeneity for binary outcomes. When the outcomes are continuous, we extend the definitions of the TBR and THR to compare the difference between potential outcomes with a pre-specified level c. Unlike the ATE, these rates involve the joint distribution of the potential outcomes and can not be identified without further assumptions even in randomized clinical trials. In this article, we assume the potential outcomes are independent conditional on the observed covariates and an unmeasured latent variable. Under this assumption, we prove the identification of the TBR and THR in non-separable (generalized) linear models for both continuous and binary outcomes. We then propose estimators and derive their asymptotic distributions. In the simulation studies, we implement our proposed methods to assess the performance of our estimators and carry out a sensitive analysis for different underlying distribution for the latent variable. Finally, we illustrate the proposed methods in two randomized controlled trials.