Title: Analysing populations of networks with mixtures of generalized linear mixed models
Authors: Mirko Signorelli - Leiden University, Mathematical Institute (Netherlands) [presenting]
Abstract: Until recently, collecting data on populations of networks (PoN) was cumbersome and rare. However, the increasing availability of automatic monitoring devices and the growing scientific interest in networks make such data more widely available. From sociological experiments involving cognitive social structures to fMRI scans revealing large scale brain networks of groups of patients, there is growing awareness that we urgently need statistical methods to analyse and summarize PoN. We will show how mixtures of generalized linear mixed models can be employed to model PoN in a thrifty but interpretable way. This model-based clustering approach to PoN allows identifying subpopulations of networks that share certain topological properties (degree distribution, community structure, effect of covariates on the presence of an edge, etc.) of interest. We will discuss how the proposed model can be estimated by combining adaptive gaussian quadratures with the EM algorithm and assess its classification performance using simulated data. We will conclude by illustrating an example application of the proposed method to a PoN representing how employees perceive advice relationships within a small business, paying particular attention to model specification and the interpretation of the outputs of our model.