Menu-based Choice Models for Customization: On the Recoverability of Reservation Prices, Model Fit, and Predictive Validity

An increasing number of companies engage in the customization of products to reap the benefits of increased sales and the potential to charge
higher prices. Online configurators, which allow customers to assemble “customized” versions of a product, play an important role in
customization. In particular, they help companies learn about user preferences and generate reliable forecasts. Therefore, menu-based choice
experiments have gained increased attention in recent years. Despite their high relevance, little is known about the properties of the different
modeling approaches under varying study conditions. We compare four prominent modeling approaches in an extensive simulation study that
systematically varies respondent heterogeneity, choice menu complexity, the available sample size, the number of individual tasks, and the
underlying behavioral model. We evaluate the models for reservation price recoverability, model fit, and predictive validity. The findings show that
in most cases, one obtains better results for simple and straightforward representations of respondent behavior (e.g., separate multinomial logit
models for different functional areas) than for the more sophisticated modeling approaches (e.g., probit-based approaches).