The Importance of Attribute Interactions in Conjoint Choice Design and Modeling

Within the context of choice experimental designs, most authors have proposed designs for the multinomial logit model under the assumption that only the main effects matter. Very little attention has been paid to designs for the attribute interaction models. In this paper, we present Bayesian D-optimal interaction-effects designs for the multinomial logit models and compare their predictive performances with those of main-effects designs. We show that in situations where a researcher is not sure whether or not the attribute interaction effects are present, incorporating interaction effects into both design stage and model estimation stage is most robust against misspecification of the underlying model for making precise predictions.