Metric conjoint segmentation methods: A Monte Carlo comparison

The authors compare nine metric conjoint segmentation methods. Four methods concern two-stage procedures in which the estimation of conjoint models and the partitioning of the sample are performed separately; in five, the estimation and segmentation stages are integrated. The methods are compared conceptually and empirically in a Monte Carlo study. The empirical comparison pertains to measures that assess parameter recovery, goodness-of-fit, and predictive accuracy. Most of the integrated conjoint segmentation methods outperform the two-stage clustering procedures under the conditions specified, in which a latent class procedure performs best. However, differences in predictive accuracy were small. The effects of degrees of freedom for error and the number of respondents were considerably smaller than those of number of segments, error variance, and within-segment heterogeneity.