Treatment Effects with Dyadic Data in the Presence of Spatial Dependence

In this paper, we propose a framework to estimate treatment effects with dyadic data in the presence of spatial dependence. In this setting, we build three latent-variable models combined with spatial autoregression for the corresponding data generating process and define individual and average direct and indirect treatment effects. We construct IV-2SLS and quasi-maximum likelihood (QML) estimators for their estimation accordingly. The estimators are consistent and asymptotically follow a properly-centered normal distribution. We also design specification tests for modeling choices. We apply the framework to estimate the treatment effects of COVID-19 policies on the US airline market.