Policy Evaluation in presence of interferences: a spatial multilevel DID approach

During the last decades SUTVA has represented the "gold standard" for the identification and evaluation of causal effects. However, the presence of interferences in causal analysis requires a substantial review of the SUTVA hypothesis. This paper proposes a framework for causal inference in presence of spatial interactions within a new spatial hierarchical Differencein-Differences model (SH-DID). The novel approach decomposes the ATE (Average Treatment Effect), allowing the identification of direct (ADTE) and indirect treatment effects. In addition, our approach permits the identification of different indirect causal impact both on treated (AITET) and on controls (AITENT). The performance of the SH-DID are evaluated by a Montecarlo Simulation. The results confirm how omitting the presence of interferences produces biased parameters of direct and indirect effects, even though the estimates of the ATE in the traditional model are correct. Conversely, the SH-DID provides unbiased estimates of both total, direct and indirect effects. In addition, this model is the more efficient compared both to the traditional and a Spatial modified Difference-in-Differences estimator.