The value of consideration data in a discrete choice experiment

• Published in: Journal of choice modelling Vol. 45; p. 100374
• Main Authors: Assele, Samson Yaekob, Meulders, Michel, Vandebroek, Martina
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.12.2022
• Subjects: Consideration set; Heterogeneity; Mixed logit model; Screening rule; Stated preference; Two-stage choice models; Heterogeneity; Mixed logit model; Screening rule; Consideration set; Two-stage choice models; Stated preference
• ISSN: 1755-5345; 1755-5345
• DOI: 10.1016/j.jocm.2022.100374

In stated preference surveys, data regarding the considered alternatives is sometimes collected prior to the preferred alternative. When the chosen alternative is not in the stated consideration set, the consideration data is inconsistent with the choice data. Several modeling approaches have been used in such situations. Some researchers ignore the consideration data and assume all alternatives are considered. Others only use the consistent choice data and delete the inconsistent observations. The most intricate methods use a latent consideration set formation approach in modeling the choice process. We extend the latent consideration set formation model to incorporate the stated consideration data but allow for inconsistencies in consideration and choice data, and allow for individual-level heterogeneity in the consideration and the choice process. We compare the recovery of the mean population preference parameters of our model with the existing approaches through simulation. The results show that if there is a similar effect of the attributes in both the consideration phase and the choice phase, the mixed logit model is not outperformed by the two-stage models. In contrast, when there is a sufficiently different effect of attributes in the consideration and the choice phase, two-stage models can recover the mean population preference parameters better than the mixed logit model. Furthermore, we can conclude that having stated consideration data barely improves the recovery of the mean preference parameters compared to a latent consideration set choice model that only uses choice data. Finally, we illustrate the models using empirical data about preferences for mobile phones. •We model the consideration and the choice data dealing with their inconsistencies.•Modeling screening is important if consideration and choice parameters are different.•A mixed logit model performs well if consideration and choice parameters are similar.•Consideration data barely improves the recovery of the mean preference parameters.