A Bayesian generalized rank ordered logit model

Abstract

Using rank-ordered logit regression, researchers typically analyze consumer preference data collected with Best-Worst Scaling (BWS) surveys. We propose a generalized rank-ordered logit (GROL) model that allows flexibility in modeling preference heterogeneity. The GROL and mixed rank-ordered logit model (MROL) accommodate preference heterogeneity. However, the GROL also allows one to model heterogeneity as a function of demographic or environmental variables. A Monte Carlo experiment compares the estimates of accuracy and precision of the proposed GROL estimation with the MROL specification. Simulation results suggest that the GROL model performs comparatively well when the GROL or the MROL is the true data-generating process (dgp). Coefficient and willingness-to-pay estimates of the GROL are more precise and accurate compared to the MROL when the MROL is the true dgp. We surmise that the increased precision of the GROL estimator arises from the added flexibility for modeling different sources of heterogeneity. An empirical application analyzes a BWS survey on consumer preferences for single-use eating-ware (SUEW) products made from biobased materials. Findings suggest that consumers value most product degradability and using non-plastic materials to fabricate SUEW. Consumers also valued the rapidity of product degradability and using non-plastic materials to make SUEW plates. Respondent attentiveness also affected willingness-to-pay (WTP) estimates across attributes. Results suggest attentive respondents were about $3.00 more WTP for biodegradable SUEW than inattentive respondents.