Finite Mixtures of ERGMs for Modeling Ensembles of Networks

Ensembles of networks arise in many scientific fields, but there are relatively few statistical tools for inferring their generative processes, particularly the presence both dyadic dependence and cross-graph heterogeneity. To address this gap, we propose characterizing network ensembles via finite mixtures exponential family random graph models (ERGMs), a class parametric that has been successful explicitly modeling complex stochastic processes govern structure edges network. Our proposed framework can also be used applications such as model-based clustering density estimation distributions. We develop joint approach to estimate number mixture components identify cluster-specific parameters simultaneously well obtain an identified model under Bayesian paradigm. Specifically, Metropolis-within-Gibbs algorithm perform inference, using strategy deliberate overfitting with sparse priors removes excess during MCMC. As true ERGM likelihood is generally intractable specifications terms, consider two tractable approximations (pseudolikelihood adjusted pseudolikelihood) facilitate efficient inference. run simulation studies compare performance these respect multiple metrics, showing conditions which useful. demonstrate utility ensemble political co-voting among U.S. Senators brain functional connectivity networks.