Critiquing and Learning Model Structure

Abstract

This chapter discussed tools for checking and improving the structure of Bayesian networks. We are interested in fit indices that highlight particular kinds of model misfit that we know can appear in assessment data and distort our uses of the model. The initial sections introduce some fit indices and looks at the technique of posterior predictive model checking (PPMC). Subsequent sections address graphical methods for assessing model fit, differential task functioning, and model comparison. Key ideas from the model selection literature and the pitfalls in attempting to learn "causality" from data are reviewed.