HNCI: High-Dimensional Network Causal Inference

We propose a new method of high-dimensional network causal inference (HNCI) that provides both valid confidence intervals for the average direct treatment effect on the treated (ADET) and valid confidence sets for the neighborhood size affecting the interference effect. We adopt the model framework from Belloni et al. (2022), which has key advantages such as nonparametric modeling of interference effects and allowing certain types of heterogeneity in node interference neighborhood sizes. Utilizing the nested matching property of the network interference effect, we reformulate the original nonparametric model into a linear regression model where the regression coefficients, corresponding to the underlying true interference function values of nodes, exhibit a latent homogeneous structure. This formulation enables us to leverage existing literature on homogeneity pursuit to conduct valid statistical inferences with theoretical guarantees. The resulting confidence intervals for the ADET are formally justified through asymptotic normality with estimable variances. By employing the repro samples approach, we further provide the confidence set for the interference of neighborhood size with theoretical guarantees. The practical utility of the newly suggested methods is demonstrated through simulations and real data examples. This is a joint work with Rundong Ding, Wenqin Du and Yingying Fan.