Primary Submission Category: Interference and Consistency Violations
A Two-Stage Experiment Design for Causal Inference under Interference
Authors: Mayleen Cortez-Rodriguez, Christina Yu, Matthew Eichhorn,
Presenting Author: Mayleen Cortez-Rodriguez*
Network interference is becoming increasingly relevant in our interconnected world. While most approaches to causal inference under interference rely on knowledge of the underlying network to make headway, recent work uses low-order potential outcomes models and a staggered rollout experimental design to obtain unbiased causal effect estimators without requiring network information. However, the required polynomial extrapolation can lead to prohibitively high variance. To address this, we propose a two-stage experiment that selects a sub-population in the first stage and restricts treatment rollout to this sub-population in the second stage. We prove theoretical guarantees for the bias and variance of a polynomial interpolation-style estimator under this design, showing improved performance even without network knowledge. For settings where the researcher may have access to some network knowledge, we also explore the role of clustering in the first stage. Bias increases with the number of edges cut in the clustering of the interference network, but variance depends on qualities of the clustering that relate to homophily and covariate balance. There is a tension between clustering objectives that minimize the number of cut edges versus those that maximize covariate balance across clusters, highlighting an interesting direction for future work.