A Multinomial Probit Conjoint Preference Simulator Incorporating Varying Importance Weights

Abstract

Many probabilistic conjoint preference simulators, such as those based on either a multinomial logit (MNL) share allocation or a version of the Bradley-Terry-Luce rule, are known to suffer from the problem of Independence from Irrelevant Alternatives (IIA). Johnson (1992) has developed a simulator based on the Multinomial Probit (MNP) distribution that provides a way to simulate preference shares not subject to IIA. In this paper we extend this conjoint simulator by applying individual-level importance weights to an alternative similarities matrix and discuss how these weights can be estimated and used in the simulator. We then illustrate the use of the simulator in a hypothetical situation and compare results with other simulators including the original Johnson MNP preference simulator.