Model-Assisted Inference for Covariate-Specific Treatment Effects with High-dimensional Data

Covariate-specific treatment effects (CSTEs) represent heterogeneous across subpopulations defined by certain selected covariates. In this article, we consider marginal structural models where CSTEs are linearly represented using a set of basis functions the We develop new approach in high-dimensional settings to obtain not only doubly robust point estimators CSTEs, but also model-assisted confidence intervals, which valid when propensity score model is correctly specified an outcome regression may be misspecified. With linear and discrete covariates, both intervals for CSTEs. contrast, from existing methods specified. establish asymptotic properties proposed associated intervals. present simulation studies empirical applications demonstrate advantages method compared with competing ones.