Exploiting network information to disentangle spillover effects in a field experiment on teens' museum attendance.

Nudging youths to visit historical and artistic heritage is a key goal pursued by cultural organizations. The field experiment we analyze is a clustered encouragement design (CED) conducted in Florence (Italy) and devised to assess how appropriate incentives assigned to high-school classes may induce teens to visit museums in their free time. In CEDs, where the focus is on causal effects for individuals, interference between units is generally unavoidable. The presence of noncompliance and spillover effects makes causal inference particularly challenging. We propose to deal with these complications by creatively blending the principal stratification framework and causal mediation methods, and exploiting information on interpersonal networks. We formally define principal natural direct and indirect effects and principal controlled direct and indirect effects, and use them to disentangle spillovers from other causal channels. The key insights are that overall principal causal effects for sub-populations of units defined by the compliance behavior combine encouragement, treatment and spillovers effects. In this situation, a synthesis of the network information may be used as a possible mediator, such that the part of the effect that is channeled by it can be attributed to spillovers. A Bayesian approach is used for inference, invoking latent ignorability assumptions on the mediator conditional on principal stratum membership.