A0736
Title: Robust high-dimensional inference for causal effects under unmeasured confounding and invalid IVs
Authors: Yunan Wu - The University of Texas at Dallas (United States) [presenting]
Lan Wang - University of Miami (United States)
Baolin Wu - University of California Irvine (United States)
Yixuan Ye - Yale University (United States)
Hongyu Zhao - Yale University (United States)
Abstract: A novel high dimensional robust estimation and inference procedure are considered for the causal effects in the presence of unmeasured confounding and invalid instruments based on observational data. Compared with the existing literature on causal inference using instrumental variables, the approach has several distinctive features. The prior knowledge is not assumed of a set of relevant instruments. The uncertainty of the availability of such a set is built into the inference procedure. In fact, the framework allows for the simultaneous violation of any of the three commonly imposed instrument validity conditions. The measured confounders are also allowed to be endogenous. The conditions for the identification of causal effects, estimation and inference procedures do not require the specification of an exposure model. In particular, the method allows for a nonlinear relationship among the exposure, the instruments and other variables. The proposed inference procedure allows for high-dimensional instruments and/or high-dimensional measured confounders. The new procedure exploits the sparsity of the observed data model to identify the causal effects with potentially invalid instruments or many weak instruments. The validity of the confidence intervals is established under relatively weak conditions without requiring prior knowledge of a subset of valid instruments.
