The mixed logit model has become the state-of-the-art tool for the estimation of different willingness-to-pay indicators. This has resulted in a significant amount of research into the choice of distribution in such models. An issue that has often been overlooked is the possible correlation between the distributions. This paper will apply the two approaches termed modeling in preference space and willingness-to-pay space to estimate indicators for two stated preference data sets. Furthermore, the paper allows for more general correlation structures and investigate the effect of explanatory variables. The results show that models allowing for correlation outperform standard models in both preference space and willingness-to-pay space. The results also show that the inclusion of correlation can have an impact on the evaluation of willingness-to-pay indicators. The main conclusion of the paper is that the choice of correlation structure is as important as the choice of marginal distributions in mixed logit models.
Abstract