Abstract

This study assessed the accuracy of mixed logit (MXL), latent class logit (LCL), and latent class mixed logit (LCML) models in identifying preference-based segments in a world with different levels of preference heterogeneity. Over the past few decades, LCL has become one of the most popular segmentation techniques for capturing this type of heterogeneity. By contrast, MXL has rarely been used to identify segments, whereas LCML has recently received increasing attention. While the existing literature provides valuable insights, there remains an opportunity to compare the predictive accuracy of LCL segments with these other discrete choice models that account for consumer heterogeneous preferences. To address this gap, this study conducted a Monte Carlo simulation. In addition to the classes predicted by LCL and LCML, this study uses individual-specific posterior (ISP) distributions of coefficients for segmentation. The results show that LCL outperforms the other two models only in a world with low preference heterogeneity (i.e., two segments). A novel finding is that segmentation based on the ISP distributions for the LCL or LCML models is superior to traditionally predicted classes when consumers exhibit considerable preference heterogeneity. These results challenge the common market segmentation practice that relies on classes and probabilities predicted by the LCL and suggest that model selection is highly dependent on the underlying preference heterogeneity. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025.