Nonlinear Algebra and Statistics (NLASTATS) Seminar by Pavel Krivitsky: Exponential-Family Random Graph Models for Multilayer Networks

Speaker: Pavel Krivitsky, Senior Lecturer in Statistics, School of Mathematics and Statistics, University of New South Wales

Title: Exponential-Family Random Graph Models for Multilayer Networks

Abstract: Networks with multiple layers of relationships are of increasing interdisciplinary interest. Such networks arise in a host of contexts where more than one type of relation may be observed among a common set of actors or vertices. The ability to model dependence processes giving rise to such systems within the exponential-family random graph framework has previously been limited to dependence arising from just two layers. To address this limitation, we introduce an extension to estimate a joint exponential random graph model over all separate measurement types which retains the (possibly correlated) layered nature of the data while facilitating estimation of dependence effects for arbitrary numbers of relations. Specifically, we extend the Conway-Maxwell-Binomial distribution for the sum of edges while simultaneously modelling joint dependence in multi-layer networks arising from cross-layer graph features. Model terms include analogues of familiar ERGM effects for arbitrary numbers of layers in the network and employ a novel "layer logic" in their specification. We present a number of empirical examples from multilayer datasets.

 

Nonlinear Algebra and Statistics Seminar