Primary Submission Category: Causal Inference and SUTVA/Consistencies Violations
Estimating causal effects of interventions altering social connectivity patterns under network interference
Authors: Shinpei Nakamura Sakai, Laura Forastiere,
Presenting Author: Shinpei Nakamura Sakai*
Causal inference under network interference is an emerging topic as network data is ubiquitous across multiple disciplines. We say that the treatment effect ‘spills over’ to other units when one’s potential outcomes are affected by the treatment status of other units. Such a phenomenon is often due to social or physical interactions and depends on the social structure of the population. An intervention that alters social connectivity would alter the interference mechanism and consequently, the spillover effect. Current methods under interference estimate causal effects conditional on a known and fixed social connectivity graph. On the other hand, epidemic models have been used to predict the effect of hypothetical interventions altering social connectivity parameters to control the spread of infectious diseases. However, a formal and general definition of the causal effects of such interventions altering the social structure is lacking. We consider a stochastic network formation and propose causal estimands to estimate spillover effects with a modification of the network formation mechanism. These causal estimands are defined under interventions shifting the degree distribution or the network formation mechanism. We develop estimators for the counterfactual estimands under hypothetical interventions altering the network structure, and we investigate the finite sample bias and large-sample properties of these estimators.