🤖 AI Summary
This paper addresses the failure of causal identification in longitudinal panel data due to spatiotemporal interference—where an individual’s outcome is affected by others’ past treatment assignments. We propose a design-based causal inference framework that, under minimal assumptions (unknown interference structure and sequential ignorability), formally defines and identifies separable direct effects and spatiotemporal spillover effects for the first time. We demonstrate that conventional fixed-effects and difference-in-differences (DID) estimators suffer from systematic bias under interference. To overcome this, we construct a new estimator with consistency and asymptotic normality. Theoretical analysis, Monte Carlo simulations, and replications of two canonical empirical studies validate our approach: it substantially reduces estimation bias in spillover effects and effectively corrects the failure of standard panel methods under complex interference patterns.
📝 Abstract
Many social events and policy interventions generate treatment effects that persistently spill over into neighboring areas, resulting in a phenomenon statisticians refer to as"interference"both in time and space. In this paper, I put forward a design-based framework to identify and estimate these spillover effects in panel data with a spatial dimension, when temporal and spatial interference intertwine in intricate ways that are unknown to researchers. The framework defines estimands that enable researchers to measure the influence of each type of interference, and I propose estimators that are consistent and asymptotically normal under the assumption of sequential ignorability and mild regularity conditions. I show that fixed effects models in panel data analysis, such as the difference-in-differences (DID) estimator, can lead to significant biases in such scenarios. I test the method's performance on both simulated datasets and the replication of two empirical studies.