Primary Submission Category: Interference and Consistency Violations
Dyad-level Weighting Estimators for Causal Effects on Changes in Network Ties under Interference
Authors: Qixiang Xu, Laura Forastiere,
Presenting Author: Qixiang Xu*
Public health interventions, designed to improve health outcomes, may also influence participants’ social network structures. For example, in health education programs, participants who receive the intervention may become more popular and attract new social connections, while ties between untreated individuals may be weakened. We develop a novel causal inference framework to measure the effect of interventions on network change, where the treatments and outcomes are defined at the dyadic-level. We allow for dependence in the network formation between dyads by assuming the potential presence of interference from a pre-specified set of units, defined for each dyad and called interference set. Under this interference assumption, we define the direct effects of the treatment status of a dyad of units on the formation or dissolution of a directed tie between them as well as the spillover effects from the treatment of the whole interference set or a subset of it, such as common friends between two units, by changing or fixing the dyadic treatment while fixing or changing the distribution of the treatment in the interference set or in the subset of interest, respectively. We develop new Horvitz-Thomson and Hajek estimators for these network-based causal estimands and derive their asymptotic properties under dyadic-level data. We then apply our estimators to a two-stage randomized trial of a health education program in Honduras.