ERGM parameter estimation of very large directed networks: implementation, example, and application to the geography of knowledge spillovers

Speaker:
Alex Stivala, University of Swinburne, Australia

Abstract:
The recently published Equilibrium Expectation (EE) algorithm for exponential random graph model (ERGM) parameter estimation has allowed such models to be estimated for networks far larger than previously possible. Here we demonstrate the extension of this algorithm to directed networks, with an implementation that overcomes some technical problems limiting the sizes of networks that could be practically estimated. We apply this method to estimate ERGM parameters for an online social network with approximately 1.6 million nodes, and a patent citation network with approximately 3.8 million nodes. The latter model allows us to test the geographic knowledge spillover hypothesis (that knowledge spillovers are geographically localized) using patent citation data, without having to treat the patent citation network as exogenous.

Biography:
Dr Stivala is a research fellow in the Centre for Transformative Innovation at the University of Swinburne. He is a member of the MelNet Social Network Research Group and an international fellow at the SoNAR-C Social Network Analysis Research Center in the Institute of Computational Science at Università della Svizzera italiana (Lugano, Switzerland). Dr Stivala's research is largely in the area statistical and computational methods for social network analysis, especially relating to exponential random graph models (ERGMs).

Host: Prof. Ernst Wit