A novel approach to detect attribute by covariate interactions in discrete choice models

This paper introduces a novel and simple method to identify attribute by covariate interactions in discrete choice models. This is important because incorporating such interactions in choice models can be an effective way to account for systematic taste variation or “observable preference heterogeneity” across individuals. Using simulated data sets to mimic a well-known phenomenon of selective attention to design attributes, we tested our proposed approach in a banking service context. Our proposed approach was successful in detecting the attribute by covariate interactions implied by the data generation process and outperformed a model with all covariate interactions. The proposed method contributes to the choice modelling literature by providing one of the “tricks of trade” to model observed preference heterogeneity. The simplicity of this approach has advantages for both academics and practitioners in marketing, transportation, healthcare and other fields that use choice modelling.