Roselinde Kessels, Assistant Professor, University of Antwerp and Maastricht University Robert Mee, William and Sara Clark Professor of Business Analytics, University of Tennessee
Past discrete choice experiments provide clear evidence of primacy and recency effects in the presentation order of the profiles within a choice set, with the first or last profiles in a choice set being selected more often than the other profiles. Existing Bayesian choice design algorithms do not accommodate profile order effects within choice sets. This can produce severely biased part-worth estimates, as we illustrate using a product packaging choice experiment performed for P&G in Mexico. A common practice is to randomize the order of profiles within choice sets for each respondent. While randomizing profile orders for each subject ensures near balance on average across all subjects, the randomizations for many individual subjects can be quite unbalanced with respect to profile order; hence, any tendency to prefer the first or last profiles may result in bias for those subjects. As a consequence, this bias may produce heterogeneity in hierarchical Bayesian estimates for subjects, even when the subjects have identical true preferences. As a design solution, we propose position balanced Bayesian optimal designs that are constrained to achieve sufficient order balance. For the analysis, we recommend including a profile order covariate to account for any order preference in the responses.
... View more
This poster provides a practical guide using JMP on the design and analysis of a discrete choice experiment (DCE) run in collaboration with researchers from the Department of Transport and Regional Economics of the University of Antwerp to study the housing preferences of students in higher education in Antwerp (Belgium). In the DCE, 1047 students evaluated 22 choice sets involving different housing accommodations that range from collective housing to individual studios. Each choice set contains two profile accommodations that are described in terms of communal living facilities and housing type, location, size, price and whether they are furnished or not. An important attribute was price to be able to estimate the students’ willingness to pay for the accommodations. For each choice set, the students had to indicate their preferred housing accommodation. The statistical analysis of the students’ choices revealed that students prefer a large furnished studio without communal living facilities, which is close to the university and inexpensive. Trading-off these ideal housing settings at the market equilibrium, students are willing to pay a substantial amount in order to shift from a traditional room with communal living facilities to a studio with no communal living facilities.
... View more