Skip to content

Abstract Search

Primary Submission Category: Causal Inference in Networks

Bayesian inference for the estimation of causal effects under network interference

Authors: Seungha Um, Samrachana Adhikari,

Presenting Author: Seungha Um*

Causal inference in the presence of interference is challenging in observational studies on social network since the effect of treatment on a unit spills over connected units. The fact that the spillover effect is commonly confounded with latent homophily and depends on social influence, which varies by unit, makes the estimation more challenging. To disentangle the effect of the individual treatment and neighborhood treatment, we employ the estimation strategy based on generalized propensity score. Also, the homophilous attributes are estimated by relying on latent space positions and social influence is evaluated based on pairwise distances among each pair of nodes. Within Bayesian framework, the uncertainty in propensity score is quantified while avoiding model feedback and imputation of missing potential outcome is implemented. We design a simulation study to assess the performance of our proposed method and examine the spillover effect varying characteristics of network topology such as size, density and community structure. The causal effects of exposure to belief about teaching mathematics are examined on teacher advice-seeking network.