Competing methods for representing random taste heterogeneity in discrete choice models. Working paper, Danish Transport Research Institute

This paper reports the findings of a systematic study using Monte Carlo experiments aimed at comparing the performance of various methods in retrieving random taste heterogeneity in a discrete choice context. Specif-ically, the analysis compares the performance of the four most commonly used continuous distribution functions to that of two more advanced ap-proaches discussed in this paper. The first of these two approaches improves on the flexibility of a base distribution by adding in Legendre polynomials. The second approach uses a discrete mixture of multiple continuous distri-butions, where, in the present study, the base distributions are all Normal. The analysis shows very constant performance for the two approaches across a host of different scenarios, with varying levels of gains in performance when compared to the more basic distributions. Finally, and rather worryingly, the analysis shows that, across the various case studies, the poorest overall performance is obtained by the commonly used Normal distribution, which is in fact even outperformed by the Uniform distribution