A0745
Title: Estimation strategies for treatment and spillover effects under network interference
Authors: Yichen Qin - University of Cincinnati (United States) [presenting]
Abstract: A novel approach is introduced for treatment and spillover effects estimation in observational studies on social networks with arbitrary interference. The direct covariate balancing estimator is proposed, which is robust for modeling misspecification and avoids the extreme weights to gain finite sample efficiency. To the best of knowledge, this is the first attempt to adopt direct covariate balancing strategies in causal effect estimation under interference. The balancing estimator is further improved with the regrouping strategy to accommodate the limited sample sizes and vertex heterogeneity. Balancing the individual covariates as well as the network embeddings is also advocated to safeguard the complexity of the data-generating process. Both theoretical and numerical justifications are established. Through the analysis of a real social experiment, the proposed method reveals the heterogeneity of conditional treatment effects, which sheds some light on the complexity of networked experiments.
