Predictive Testing of Consumer Choice Models Not Subject to Independence of Irrelevant Alternatives

Abstract

The probabilistic modeling of the relationship between objective or perceived characteristics of multiattribute alternatives and consumer choice is receiving increased attention in marketing and other disciplines. Many marketing applications use the Luce choice axiom, the LOGIT model, and the independent PROBIT model. All these formulations involve the “independence of irrelevant alternatives” assumption, which is not realistic in many consumer behavior contexts. The author suggests new product introduction situations in which the consequences of the assumption are not intuitively appealing. Models recently developed by transportation researchers and one extension developed by the author, all of which are not constrained by the independence restriction and may be applicable in modeling consumer choice, are described. In an empirical application, cross-sectional data on consumers’ perceptions of transportation modes on several characteristics and their choices of modes in the San Francisco Bay area are used to compare models with and without the independence property on diagnostic and predictive criteria.