B0656
Title: A statistical platform for discrete and dependent attribute and network data generalizing GLMs
Authors: Cornelius Fritz - Pennsylvania State University (United States) [presenting]
Michael Schweinberger - Pennsylvania State University (United States)
David Hunter - Pennsylvania State University (United States)
Abstract: The world of the twenty-first century is interconnected and interdependent, as demonstrated by recent events that started as local problems and turned into global crises (e.g., pandemics, political and military conflicts, economic and financial crises). More often than not, such events are unique and cannot be replicated. To learn from dependent events involving attributes and networks, a statistical platform is introduced for discrete and dependent attributes and connections. The proposed framework is (a) flexible, in the sense that it can capture a wide range of attribute-attribute, attribute-connection, and connection-connection dependencies; (b) interpretable, in that it builds on the proven statistical platform of generalized linear models; and (c) scalable, in that it allows large populations to be more heterogeneous than small populations. Scalable composite likelihood $M$-estimators are introduced and are placed on firm statistical ground, by providing theoretical guarantees based on a single observation of discrete and dependent attributes and connections. Simulation results and an application to political discourse on Twitter are presented.
