Heterogeneity of Interactions and Assessment of Treatment Effects: An Approach by Spatial Dependency Effects

In the specific context of the two stage model of causal effect evaluation, this article deals with the analysis of the validity of the assumptions assuming no interdependency. More precisely we question the CIA and the SUTVA assumptions. We use a spatial analytical framework to characterize the interactions between individuals. We consider the heterogeneity of neighborhoods according to individuals and their neighbors, whether or not they are involved in the public policy. The evaluation model combines two models. A probit model with a spatially autocorrelated errors makes possible to take account for unobservable factors potentially affecting individuals' decisions to take part in the policy. A spatial autoregressive model on the outcome allows us to integrate the heterogeneity of neighborhoods and the potential selection bias spatially adjusted. We use the ZRR (revitalization of rural areas) policy as an empirical application. We show that if the average treatment effects of the policy on the creation of business establishments are significant and positive with models without interactions, these effects are not significant anymore when we consider spatial dependencies.