How many classes? Statistical modelling of a social network and a terrorist network, with a latent class model and Bayesian model comparisons
22 September 2015
Speaker: Professor Murray Aitkin, the University of Melbourne
This talk discusses the assessment of the number of classes in a social network, through the latent class model extension of the exponential random graph model. The assessment uses a new Bayesian method of model comparisons, based on the posterior distribution of the deviance for each of the competing models.
The approach is applied to a well-known social network in which the number of classes is known a priori, and to the Noordin Top terrorist network, analysed at length in the book by Everton (2012).
The performance of the model comparison method is illustrated with simulations from single population models and normal mixture models. This work is joint with Duy Vu and Brian Francis. Everton, S.F. (2012) Disrupting Dark Networks. Cambridge University Press.