Primary Submission Category: Causal Inference in Networks
Causal inference with uncertain interference structure
Authors: Daniel Nevo, Bar Weinstein,
Presenting Author: Daniel Nevo*
As an alternative to the no-interference assumption, an interference structure is often represented using a network. Ubiquitously, the network structure is assumed to be known and correctly specified. However, correctly encoding the interference structure in a network can be challenging. For example, edges might be measured with error or censored, network structure can change over time, and contamination between clusters might be present. Using the exposure-function framework, we quantify the bias of commonly used estimators when the network interference structure is misspecified.
To overcome the problem of network misspecification, we propose two solutions. First, we propose a novel estimator utilizing multiple networks simultaneously, which is unbiased if one of the networks correctly represents the interference structure. As an alternative, we propose a sensitivity analysis framework that quantifies the impact of a postulated interference structure misspecification on the causal estimate as a function of parameters governing a misspecification mechanism.
We illustrate the bias arising from incorrectly specified network and study the bias-variance tradeoff entailed in our proposed misspecification-robust estimator. We demonstrate the utility of our methods in two real examples involving two different interference structures: a social network field experiment and a cluster-randomized trial.