A Semi-parametric Estimation of Personalized Dose-response Function Using Instrumental Variables

In the application of instrumental variable analysis that conducts causal inference in the presence of unmeasured confounding, invalid instrumental variables and weak instrumental variables often exist which complicate the analysis. In this paper, we propose a model-free dimension reduction procedure to select the invalid instrumental variables and refine them into lower-dimensional linear combinations. The procedure also combines the weak instrumental variables into a few stronger instrumental variables that best condense their information. We then introduce the personalized dose-response function that …

analysis application causal causal inference confounding free function inference paper parametric personalized refine them variables