Prediction Models for Configural Frequency Analysis

This article takes the conversation about prediction models within Configural Frequency Analysis (CFA) a step further. Our starting point is the link between these prediction models and logit models, a correspondence that von Eye and Bogat laid out in a companion piece appearing in the same issue. From that foundation we work out a broader set of CFA prediction models than has been described before.

The models we arrive at are not restricted to simple one-to-one setups. They can accommodate several predictors at once, and they can handle more than one criterion variable. We walk through what each configuration looks like in practice.

For every model under discussion, we lay out the design matrix explicitly and write down the logit model that goes with it. Pairing the two makes the structure of each prediction model easier to see and ties the CFA formulation back to the familiar logit framework.

Throughout, we rely on worked data examples rather than abstract description alone, using real figures to show how the models behave and how the results should be read.