Quantitative models for social networks?

Quantitative models for social networks were proposed by Frank and Strauss (1986) and by Strauss and Ikeda (1990) in the Journal of the American Statistical Association. The models were formulated in terms of a Markov random field assumption that, at first sight, might seem an attractive starting point. The subsequent decade has seen a flurry of activity in developing these models, fitting them to available data via maximum pseudolikelihood estimation and making substantive interpretations of the corresponding parameter values. Much of this work, in which S. Wasserman and P. Pattison have been key players, has been incorrect or of doubtful value. This talk will review the modeling assumptions and statistical methods that have been used and develop some more rigorous alternatives in a modern computing environment. Some of the many remaining issues may also be relevant more widely in quantitative social science.