Enhancing algorithmic base for discrete choice modelling: Fid-Bau Portal

Enhancing algorithmic base for discrete choice modelling

Roh, Hyuk-Jae / Khan, Ata M.

Abstract Frequently, discrete choice modelling process is constrained due to lack of information on alternative estimation methods and the inability of commercial software to enable comparative efficiency tests. Further, the underlying numerical operational procedures used in the software are not displayed to the user. This paper reports research on the comparative efficiency of four selected algorithms that are combined with the Maximum Likelihood method in computationally logical way for estimating the parameters of Multinomial Logit travel mode choice model. The algorithms were formulated and numerical experiments were designed for the comparison of their estimation performance in an access to airport mode choice case study. Also, the important factors of econometric model estimation, namely step size, convergence criterion, and initial guessing of the starting points are studied. The process used here can be applied to mixed logit or other discrete choice model formulations. The findings of this research are expected to enhance discrete choice modelling in transportation planning and will be useful to researchers who wish to develop computer code encompassing estimation algorithms of their choice.