Latent Space Models for Dynamic Bipartite Social Networks

• Activity Number: 552 - Recent Developments in Latent Variable Network Models: 15 Years After the Work of 'Latent Space Approaches for Social Network Analysis'
• Type: Invited
• Date/Time: Wednesday, August 2, 2017 : 2:00 PM to 3:50 PM
• Sponsor: Social Statistics Section
• Author(s): Adrian E. Raftery*
• Companies: University of Washington
• Keywords: Latent position model ; Dynamic social network ; Company board ; Board interlock ; 2008 financial crisis ; Markov process

Abstract

We consider the problem of analyzing bipartite networks, consisting for example of organizations and their members, that are evolving in time and are observed at discrete time intervals. We develop a new statistical model for this situation by embedding the positions of both members and organizations in a Euclidean latent space and allowing their positions to change over time. This typically does not account for short-term inertia, however, and we represent this by modeling the process of links changing conditional on the latent space as a Markov process. We use this model to analyze the temporal bipartite network of the leading Irish companies and their directors from 2003 to 2013, encompassing the end of the Celtic Tiger boom and the ensuing financial crisis in 2008. We focus on the evolution of company interlocks, whereby a company director simultaneously sits on two or more boards. A useful aspect of our modelling approach is that it accommodates the development of metrics which can be used to assess the evolution of interlockingness over time. This is joint work with Nial Friel, Riccardo Rastelli and Jason Wyse.