Random Regret Minimization and Random Utility Maximization in the Presence of Preference Heterogeneity: An Empirical Contrast

Abstract

Random regret minimization (RRM) interpretations of discrete choices are growing in popularity as a complementary modeling paradigm to random utility maximization (RUM). While behaviorally very appealing in the sense of accommodating the regret of not choosing the best alternative, studies to date suggest that the differences in willingness to pay estimates, choice elasticities, and choice probabilities compared to RUM are small. However, the evidence is largely based on a simple multinomial logit (MNL) form of the RRM model. This paper revisits this behavioral contrast and moves beyond the multinomial logit model to incorporate random parameters, revealing the presence of preference heterogeneity. The important contribution of this paper is to see if the extension of RRM-MNL to RRM-mixed logit in passenger mode choice widens the behavioral differences between RUM and RRM. The current paper has identified a statistically richer improvement in fit of mixed logit compared to multinomial logit under RRM (and RUM) but found small differences overall between the empirical outputs of RUM and RRM, with no basis of an improved model fit between these two nonnested model forms. The inclusion of both model forms should continue to inform the likely range of behavioral outputs during investigation of a broader range of process heuristics designed to capture real world behavioral response.