A1085
Title: Causal inference under interference with dependent outcomes due to treatment and outcome spillover
Authors: Subhankar Bhadra - Pennsylvania State University (United States) [presenting]
Michael Schweinberger - Pennsylvania State University (United States)
Vishesh Karwa - Temple University (United States)
Abstract: Causal inference is considered under interference, with dependence among outcomes arising from treatment and outcome spillover. Two statistical contributions are made. First, the direct and indirect causal effects are characterized as a function of model parameters. Second, consistency results are established along with rates of convergence for the direct and indirect causal effects. Both are the first such results when outcomes are dependent on treatment and outcome spillover. In addition, the intervention network is allowed to be random, and the probability law that governs the intervention network is inferred, providing insight into the network-generating mechanism and helping quantify the uncertainty about the network-generating mechanism. Two approaches are developed to estimate the direct and indirect causal effects: a low-rank approximation and a minorization-maximization algorithm.
