A0496
Title: A regression framework for studying relationships among attributes under network interference
Authors: Cornelius Fritz - Pennsylvania State University (United States) [presenting]
Michael Schweinberger - Pennsylvania State University (United States)
David Hunter - Pennsylvania State University (United States)
Subhankar Bhadra - Pennsylvania State University (United States)
Abstract: When network data are collected, the structure of networks is often of secondary interest compared to the question of how networks affect individual or collective outcomes. The well-established class of models known as generalized linear models (GLMs), which includes linear and logistic regression, assumes that the response of a given unit depends on predictors measured on that unit but is unaffected by predictors and responses of other units. A statistical framework is introduced that captures complex and realistic dependencies among attributes and connections while retaining the virtues of GLMs. The framework helps study relationships among attributes under network interference and is applicable to binary, count-valued, and real-valued attributes. The framework is demonstrated to be amenable to scalable statistical computing based on convex optimization of pseudo-likelihoods using minorization-maximization methods. Theoretical guarantees are established based on a single observation of dependent attributes and connections, and simulation results are presented along with an application to hate speech on the social media platform X, along with the theoretical guarantees.
