Nonparametric doubly robust estimation of causal effect on networks in observational studies

Abstract

Interconnection of nodes takes great challenge to the estimation of causal effect in the network. In this study, we develop a nonparametric doubly robust (NDR) estimator to identify the causal effect in the presence of general interference on network observational data. The estimator combines the strengths of doubly robust mapping and nonparametric regression. Thus, it is consistent when either the treatment or the outcome model is properly specified and is free of parametric assumptions. The asymptotic properties of the proposed estimator are also proved. We demonstrate the robustness and effectiveness of NDR by simulation studies and apply this method to investigate the impact of installation of SnCR on ambient ozone concentration of 473 power plants in America.