Covariate Adjustment Cannot Hurt: Treatment Effect Estimation under Interference with Low-Order Outcome Interactions

Abstract: In randomized experiments, covariates are often used to reduce variance and improve the precision of treatment effect estimates. However, in many real world settings, interference between units, where one unit's treatment affects another's outcome, complicates causal inference. This raises a key question: how can covariates be effectively used in the presence of interference? Addressing this challenge is nontrivial, as direct covariate adjustment, such as through regression, can sometimes increase variance due to dependencies across units. In this paper, we study how to use covariate information to reduce the variance of treatment effect estimators under interference. We focus on the total treatment effect (TTE), defined as the difference in average outcomes when all units are treated versus when all are controlled. Our analysis is conducted under the neighborhood interference model and a low order interaction outcome model. Building on the SNIPE estimator from Cortez-Rodriguez et al. (2023), we propose a covariate adjusted SNIPE estimator and show that, under sparsity conditions on the interference network, the proposed estimator is asymptotically unbiased and has asymptotic variance no greater than that of the original SNIPE estimator. This parallels the classical result of Lin (2013) under the no interference assumption, where covariate adjustment does not worsen estimation precision. Importantly, our variance improvement result does not rely on strong assumptions about the covariates: the covariates may be arbitrarily dependent, affect outcomes across units, and depend on the interference network itself.