Causal inference often neglects the presence of interference. This takes place when treatment exposure of one unit also affects other units connected through physical, social or economic interactions in a network structure. Extensive work has been done to assess the role played by spillover effects in policy evaluations, but most of the literature focuses on randomized experiments under cluster interference.
This paper presents a methodology to draw causal inference in a non-experimental setting subject to network interference. Specifically, we develop a generalized propensity score-based estimator to estimate both direct and spillover effects of a continuous treatment.
Spillover effects are defined by the exposure to the network treatment, that is, a summary of the treatment received by connected units.
Our estimator also allows to consider asymmetric network connections characterized by heterogeneous intensities. To showcase this methodology, we investigate whether and how spillover effects shape the optimal level of policy interventions in agricultural markets. Our results show that, in this context, neglecting interference may underestimates the degree of policy effectiveness.
Download paper (PDF)