Modelling mobility: the use of simulation to choose between near-equivalent models

The application of the log-linear model to frequency counts has rekindled interest in the traditional social mobility table. This recent literature is to a degree divided over several issues, including how to develop an acceptable model of mobility flows and how to choose between near-equivalent models which generate almost identical fitted counts but have different substantive implications. In reviewing some of these issues, we draw upon Goldthorpe's development of Hauser's “levels” model and present a general mobility model which, we argue on theoretical grounds, should be broadly applicable to modern industrial societies. We develop the model using Australian data, extend it to British and American mobility data, and use Monte Carlo methods to choose between alternative models. In developing our substantive interpretations of different models, we illustrate models that underfit, overfit, and adequately fit the observed mobility flows.