Low-Degree Outcomes and Clustered Designs: A Combined Approach for Causal Inference Under Interference

Existing literature on estimating causal effects under network interference falls under two primary approaches: leveraging assumptions on the interference to inform the experimental design or leveraging assumptions on the potential outcomes to inform the choice of estimator. These strategies have traditionally been considered in isolation; our work aims to understand and quantify their synergy. We present a general pseudoinverse estimator for causal effects under low-degree outcome models that works for arbitrary experimental designs. Next, we analyze the bias and variance of this estimator both for arbitrary designs and more specifically in the case of cluster randomized designs. Throughout, we see that the combination of cluster randomized designs and low-degree outcome models leads to more favorable properties than either strategy ensures on its own.

Oct 18, 2023 9:30 AM

INFORMS Annual Meeting (2023)

Phoenix Convention Center

100 N 3rd St, Phoenix, AZ 85004