Speaker: Rohit Bhattacharya, Assistant Professor, Williams College, USA - Rohit Bhattacharya – Computer Science
Abstract: A recurring question in network studies is whether two connected units resemble each other because one influenced the other (contagion) or because they were alike due to unmeasured background conditions (latent confounding, of which homophily is the canonical case). These are famously hard to separate from a single observed network. Under the assumption that these mechanisms do not co-occur, we use graphical models of interference to derive coding likelihood-ratio tests that distinguish contagion from confounding layer by layer---for baseline covariates, treatments, and outcomes---and propose estimators of network effects that are consistent once the mechanisms are known or correctly inferred. The latter extends auto-g-computation to handle unmeasured confounding between units. Our tests rely on an unexpected connection: to recover standard asymptotics from dependent data, we use units that are pairwise at least six degrees of separation apart---exactly the bound made famous by the small-world principle. This turns a piece of social-network folklore into a concrete statistical resource, albeit a scarce one: the more tightly connected the world, the fewer such units exist. We examine when this is affordable across several online social networks.
This will be a free hybrid seminar. To register to attend remotely, please click here: https://cam-ac-uk.zoom.us/meeting/register/T9b4ci7NTKaGOvP1TK5fFw