Causal Inference for Stochastic Networks by Duncan Clark, Atalan Tech

Causal Inference for Stochastic Networks by Duncan Clark, Atalan Tech, Monday November 6, 1:10 – 1:50pm, North Science Building 017, Wachenheim

This talk focuses on a method to derive causal inference in the setting of a social network. For example, the effect of having 1 more connection to another node on a nodal outcome of interest. Claiming causal inferences in this setting necessitates careful consideration of the often complex dependency between outcomes for actors. Of particular importance are treatment spillover or outcome interference effects. I consider a model for causality when the underlying network is endogenous; where the ties between actors and the actor covariates are statistically dependent. I develop a joint model for the relational and covariate generating process that avoids restrictive separability and fixed network assumptions, as these rarely hold in realistic social settings. While the framework can be used with general models, I develop the highly expressive class of Exponential-family Random Network models (ERNM) of which Markov Random Fields (MRF) and Exponential-family Random Graph models (ERGM) are special cases. I will present potential outcome based inference within a Bayesian framework, and propose a modification to the exchange algorithm to allow for sampling from ERNM posteriors. Finally, demonstrating the value of the framework in a case-study of smoking in the context of adolescent friendship networks