Nutzenermittlung in wahlbasierter Conjoint-Analyse: Ein Vergleich von Latent-Class- und hierarchischem Bayes-Verfahren

Summary

Two different concepts of disentangling noise from systematic deviations in Choice-Based Conjoint evaluations are compared: The Latent Class Technique and the Hierarchical Bayes procedure. In addition, a probabilistic interpretation of LC estimates is presented as an interims model. Conceptual differences between these models are discussed and hypotheses on resulting differences in estimates are derived. These are tested in a large-scale empirical study. The relative performance is evaluated in two distinct application areas: segment/level and individual/level estimates. The expected patterns are confirmed only partly by empirical evidence. It is shown that the structure of the underlying heterogeneity concept influences the achievable outcomes. Contrary to expectations it is shown that the segment-level estimates are highly stable across methods. While individual Hierarchical Bayes estimates are often of questionable quality, they are to be preferred against the Latent Class estimates, because they detect outliers reasonably well and provide more flexibility in the data evaluation.