Alternative models of preference heterogeneity for elicited choice probabilities

Discrete choice experiments (DCEs) often present concise choice scenarios involving product attributes that may appear incomplete to respondents. To allow respondents to express uncertainty arising from this incompleteness, DCEs may ask them to state probabilities with which they expect to make specific choices. The workhorse method for analyzing the elicited probabilities involves semi-parametric estimation of population average preferences. Despite flexible distributional assumptions, this method presents challenges in estimating unobserved heterogeneity in individual preferences. We introduce a fractional response model, specifically a beta regression model for panel data, that incorporates a random parameter logit specification of preference heterogeneity. The model enables researchers to uncover preference heterogeneity under comparable parametric assumptions as adopted in conventional choice analysis, and can accommodate scale heterogeneity that make the semi-parametric method inconsistent. Using a DCE on alternative fuel vehicles, we illustrate the complementary roles of the parametric and semi-parametric approaches. We also undertake a separate analysis in which respondents are randomized to either a DCE employing a conventional choice elicitation format or a parallel DCE employing the probability elicitation format.