Title:

Modelling large, messy, and sampled network data – problems and prospects for principled analysis of real social networks

Speaker:

Johan Koskinen - Lecturer in Social Statistics, Cathie Marsh Institute for Social Research, University of Manchester

Talk Abstract:

The social network analysis (SNA) paradigm has proved a powerful and intuitive explanatory perspective that has recently gained new currency under the more general moniker of network science. SNA draws on graph theory and social entities are conceived of as nodes in a graph connected by lines (or edges), representing how people, organisations, countries, etc. are relationally tied. Having a network perspective results in compelling pictures of the social web; insightful summary measures capturing both positions of individuals and properties of the network; as well as the possibility of statistical modelling of how nodes are connected. Here we consider the latter - statistical modelling of the network. In particular we are considering modelling tie-variables using exponential random graph models (ERGM). ERGMs have their origin in statistical mechanics and are intimately related to Markov random fields and the classic Ising and Potts models. For social networks, in contrast to, say, particle spins on a lattice, the modelling often turns out to be much more complicated but also giving rise to richer models. Here we discuss some issues associated with modelling social networks based on empirical data. Among the challenges of this domain are specifying the boundary of the network and accounting for the often partial nature of data. Additionally, seeing as social networks cannot typically be collected automatically (through for example scraping on-line sources), we have to rely on observational data that is often laden with error. Among the opportunities that real social network data offers is that you may explore the full complexity of people’s relations. People are not only tied to other people but also affiliated with organisations, places, and events. We present these different aspects in the context of a number of illustrative datasets where information is collected and collated from different sources.

Speaker Bio:

Johan Koskinen joined the Department of Social Statistics at the University of Manchester in 2011 having previously worked at the Universities of Stockholm, Melbourne and Oxford. Dr Koskinen has contributed extensively to methodological development in social network analysis to enabled innovative applications by several disciplines. He is one of the co-authors of the RSiena statistical network analysis package for longitudinal network analysis and a contributor to the MPnet software package, one of the most commonly used statistical software packages for network analysis. His methodological contributions are often developed in collaboration over substantive research projects with applied researchers and he is active in disseminating best practices through frequent workshops. He has also co-written two books on social network research methods aimed at practitioners. One of them a book on exponential random graph models (Cambridge University Press) that was awarded the 2016 Harrison White Book Award by the American Sociological Association. His current research concentrates on extending current statistical methodology for modelling social interaction to social networks of multiple types of nodes using data collated and collected from different sources

Live Stream:

bluejeans.com/8474912527

Related Info