The Augmented Latent Class Model: Incorporating Additional Heterogeneity in the Latent Class Model for Panel Data

Abstract

Marketing researchers have actively sought to account for the heterogeneity in households' preferences and responses to marketing-mix elements (e.g., price, promotion) when analyzing scanner-panel data. This is because not doing so biases response coefficient estimates that can potentially alter the conclusions drawn from such analyses. In this article, the authors present an extension to the latent class model (LCM) that accounts for additional heterogeneity by allowing households' preferences to be a continuous mixture of segment-level preferences. The authors show how their model generalizes the LCM by the mere addition of S + 1 parameters, where S refers to the number of segments. Because a limitation of the traditional LCM is that it does not adequately account for heterogeneity in responses across households, the authors' model mitigates this criticism leveled at the LCM. They then compare their model with the traditional LCM and the equivalent continuous mixture (random-effects) models in terms of model selection criteria and predictive ability. The results indicate that the authors' model outperforms both the LCM and the equivalent random-effects (logit) model on these criteria.