Latent class mixture models for analyzing rating responses

Abstract

Latent class methodology has been used extensively in market research. In this approach, segment membership and parameter estimates for each derived segment are estimated simultaneously. A popular approach for fitting latent class models to rating responses is to assume mixtures of multivariate conditional normal distributions. An alternative approach is to assume a Proportional Odds model. These two approaches are compared empirically in a Monte Carlo study, assessing segment membership and parameter recovery, goodness of fit and predictive accuracy.