It's Not That I Don't Care, I Just Don't Care Very Much: Confounding Between Attribute Non-attendance and Taste Heterogeneity



It's Not That I Don't Care, I Just Don't Care Very Much: Confounding Between Attribute Non-attendance and Taste Heterogeneity

Authors

S Hess, ITS, University of Leeds, UK; A Stathopoulos, University of Trieste, IT; D Campbell, V O'Neill, Queen's University Belfast, UK; S Caussade, LAN Chilean Airlines, CL

Description

In this paper we put forth a framework that allows us to discern between cases where attributes are ignored and ones where extreme sensitivity can explain preferences.

Abstract

Standard practice in choice modelling assumes that all information presented in a choice task is attended to. In the context of exploring different failures of normative expected utility theory a large body of work has posed the question of whether some attributes are in fact ignored in the choice process. Numerous empirical models have been proposed and non-attendance appear to has become the most frequent application of non-logit behaviour in the choice experiment literature. Recently, the main emphasis has been on the use of latent class approaches. In this approach latent classes correspond to different combinations of attendance, imposing a zero coefficient for ignored attribute-combinations. Following such parameter restrictions, unobserved class probabilities are estimated across the population. Overall the findings of this body of work point towards a significant portion of people ignoring attributes, including payment-variables, leading to important implications for retrieved welfare measures and behavioural reactions to changes in relevant/irrelevant attributes. Importantly, a common shortcoming of the overviewed work is the failure to account for the possibility of taste heterogeneity within the classes reflecting non-attendance. Indeed, recent evidence suggests that in many cases the attribute merely has a low importance rather than being ignored. This implies that we run a risk of generating significant bias if we mistake low sensitivity for zero-utility and allow coefficients to drop out of LC utility function as is currently done. Moreover current work does not explore the links between the retrieved or 'apparent' ignoring behaviour and stated information concerning attribute importance, attribute attitudes and cognitive styles. In this paper we put forth a framework that allows us to discern between cases where attributes are ignored and ones where extreme sensitivity can explain preferences. This is achieved by incorporating a mixing distribution in the latent class structure. We obtain significant improvements in fit, more reasonable implied willingness to pay measures, and reductions in the implied rates of attribute non-attendance.

Publisher

Association for European Transport