Preference Heterogeneity and Attribute Non-attendance: Can They Be Separated?
D Campbell, S Hess, ITS University of Leeds, UK
In this paper we compare the potential of discrete and continuous mixture distributions in accommodating and distinguishing preference heterogeneity and attribute non-attendance.
Within the context of discrete choice experiments, when offered a choice between a bundle of goods, respondents are expected to choose the good that delivers them the highest utility. In making this choice, it is assumed the respondent evaluates each and every attribute and makes trade-offs between them. This gives rise to the continuity axiom, which infers unlimited substitutability amongst attributes, and is synonymous with passive bounded rationality, where respondents process all available information. There is growing empirical evidence, however, highlighting that there is process heterogeneity in the way that respondents evaluate bundles of attributes. Importantly, in many situations it is found that respondents do not comply with this continuity assumption. In such cases, respondents adopt an attribute processing strategy, whereby they behave in a rationally adaptive manner. One specific strategy adopted by respondents is their decision to pay attention to only a subset of the attributes, ignoring all other differences between the attributes (referred to as attribute non-attendance).
The estimation of discrete choice models with random coefficients has all but become standard practice since it helps uncover the unobserved heterogeneity. However, given the growing literature on attribute non-attendance it may be unclear if the uncovered preference heterogeneity is influenced by non-attendance. In the same way, where attempts have been made to accommodate attribute non-attendance, it may be difficult to ascertain if it is preference heterogeneity or attribute non-attendance that is retrieved.
In this paper we compare the potential of discrete and continuous mixture distributions in accommodating and distinguishing preference heterogeneity and attribute non-attendance. To test our methodology we use four simulated stated preference datasets and a real dataset. Our results show that attribute non-attendance exaggerates the degree of preference heterogeneity and, similarly, preference heterogeneity leads to an overestimation of the extent of attribute non-attendance. We demonstrate the need to discriminate between preference heterogeneity and attribute non-attendance, resulting in significant improvements in model fit and more accurate estimates.
Association for European Transport