Primary Submission Category: Causal Inference in Networks
Model-Based Inference and Experimental Design for Interference Using Partial Network Data
Authors: Steven Wilkins-Reeves, Tyler McCormick, Arun Chandrasekhar, Shane Lubold,
Presenting Author: Steven Wilkins-Reeves*
The stable unit treatment value assumption states that the outcome of an individual is not affected by the treatment statuses of the individual’s neighbors. In many common scenarios, ranging from economics to epidemiology, this assumption is not met. For instance, an individual’s likelihood of being infected once given a vaccine likely depends on whether their close contacts received the vaccine. In many empirically relevant situations, full network data (required to adjust for these spillover effects) is too costly or logistically infeasible to collect. Partially or indirectly observed network data (e.g., subsamples, Aggregated Relational Data (ARD), egocentric sampling, or respondent-driven sampling) reduce the logistical and financial burden of collecting network data, but the statistical properties of treatment effect adjustments from these design strategies were, until now, largely unknown. In this paper, we present a framework for the estimation and inference of treatment effect adjustments using partial network data. Further, we demonstrate how to use partial network data to inform randomization in experimental settings to reduce the variance of the treatment effect estimate. In addition to our theoretical results, we evaluate this approach using simulated experiments on observed graphs, as well as an application to information diffusion.