Primary Submission Category: Causal Inference in Networks
Data-adaptive exposure thresholds for the Horvitz-Thompson estimator of the Average Treatment Effect in experiments with network interference
Authors: Vydhourie Thiyageswaran, Jennifer Brennan,
Presenting Author: Vydhourie Thiyageswaran*
Randomized controlled trials on network data often suffer from interference, a SUTVA vi-
olation in which a unit’s treatment assignment affects the outcomes of its neighbors. A popular method
to reduce the bias, caused by interference, in estimating the Average Treatment Effect (ATE) is to apply
the Horvitz-Thompson estimator of the ATE with an exposure mapping: a function that identifies which
units in a given randomization are not subject to interference. For example, an exposure mapping may
specify that any unit with at least X% of its neighbors having its same treatment status does not experience
interference. In this work we propose a data-adaptive method to select this “X%” threshold, which greatly
affects the mean squared error of the Horvitz-Thompson estimator but is often difficult to elicit from domain
experts. Our method estimates the bias and variance of the Horvitz-Thompson estimator under different
thresholds using a linear dose-response model of the potential outcomes. We present simulations illustrating
that our method improves upon non-adaptive choices of the threshold for cycle graphs (and their 2k-degree
extensions). Furthermore, we demonstrate that our method is robust to deviations from the linear potential
outcomes model.