Primary Submission Category: Difference in Differences
Identifying and estimating causal effects using difference-in-differences under network dependency and interference
Authors: Michael Jetsupphasuk, Didong Li, Michael Hudgens,
Presenting Author: Michael Jetsupphasuk*
Differences-in-differences (DiD) is a causal inference framework for observational panel data that allows for unmeasured confounding but assumes parallel outcome trajectories among treatment groups under the (possible) counterfactual of receiving a specific treatment. We study DiD under network dependency and interference, where outcomes may be correlated and treatments assigned to a unit may affect outcomes in neighboring units. The proposed methods accommodate general interference provided there exists a known exposure mapping that summarizes treatments in interfering units. This framework includes the bipartite setting where treatment and outcome units are different. The main estimand of interest generalizes recently proposed estimands and is a time-varying analogue of the average treatment effect among the treated where potential outcomes may depend on multi-valued or continuous exposure histories. We identify the causal estimand under parallel trends and propose outcome regression, inverse probability weighted, and doubly robust estimators. The methods are evaluated in simulations and applied to study the effects of adopting emission control technologies in coal power plants on county-level mortality due to cardiovascular disease.