A Comparison of Two-Stage Segmentation Methods for Choice-Based Conjoint Data: A Simulation Study

Due to the increasing interest in market segmentation in modern marketing research, several methods for dealing with consumer heterogeneity and for revealing market segments have been described in the literature.In this study, the authors compare eight two-stage segmentation methods that aim to uncover consumer segments by classifying subject-specific indicator values. Four different indicators are used as a segmentation basis.The forces, which are subject-aggregated gradient values of the likelihood function, and the dfbetas, an outlier detection measure, are two indicators that express a subject’s effect on the estimation of the aggregate partworths in the conditional logit model. Although the conditional logit model is generally estimated at the aggregate level, this research obtains individual-level partworth estimates for segmentation purposes. The respondents’ raw choices are the final indicator values. The authors classify the indicators by means of cluster analysis and latent class models. The goal of the study is to compare the segmentation performance of the methods with respect to their success rate, membership recovery and segment mean parameter recovery. With regard to the individual-level estimates, the authors obtain poor segmentation results both with cluster and latent class analysis. The cluster methods based on the forces, the dfbetas and the choices yield good and similar results. Classification of the forces and the dfbetas deteriorates with the use of latent class analysis, whereas latent class modeling of the choices outperforms its cluster counterpart.