MODEL SPECIFICATION, MODAL AGGREGATION, AND MARKET SEGMENTATION IN MODE-CHOICE MODELS: SOME EMPIRICAL EVIDENCE

Abstract

Multinomial logit models that have four, five, and seven alternatives are described for work-trip mode choice. The most satisfactory overall specification is based on treating travel-time components as a single variable in additive and generic form and also on equating three rail alternative-specific constants (in the seven-alternative models) as a result of statistical tests. A simplistic method is used to aggregate rail and bus modes in the five-and four-alternative models, respectively. The resulting coefficients of corresponding variables among all the models are all consistent since they are statistically equal and numerically very close. Statistical tests show that at least alternative-specific constants, which account for over two-thirds of the total explanatory power of the models, are valued differently for the following markets: (a) one-versus two-car households, (b) commuters bound for the central business district versus others, and (c) low-and high-income households that also value service attributes unequally. Finally, coefficients estimated by means of observation of level-of-service attributes or by means of network models that estimate these attributes are also compared.