Reducing sign violation for VTTS distributions through recognition of an individual's attribute processing strategy

Abstract

A number of authors have recently argued that the selection of the distributional assumptions in mixed logit models, used in deriving distributions of valuation of travel time savings (VTTS) to capture taste heterogeneity, has a significant impact on the empirical evidence, including the incidence of sign change across the distribution. The stream of research by Hensher and his colleagues on accounting for the attribute processing strategy (APS) in stated choice studies, the main data source of VTTS, suggests that the existence of intuitively implausible signs for a subset of the sampled population is likely due, to some extent, to the manner in which the information in the stated choice experiment is actually input into the estimation of the choice model. In this paper we show evidence of what happens when we take into account a specific attribute processing strategy where respondents indicate that they ignored one or more attributes in making a choice. Accounting for the APS significantly reduces the incidence of intuitively implausible VTTS, even with unconstrained distributions. Allowance for APS in real world applications will require information on the incidence of alternative attribute processing rules in constructing VTTS estimates. (a)