Urban Transportation Demand: Dealing with the Curse of Dimensionality



Urban Transportation Demand: Dealing with the Curse of Dimensionality

Authors

VIAUROUX C, GREMAQ, France

Description

Transportation demand analysis is of capital importance to generate efficient transport policies. Transport projects like investments in infrastructure, changes in operating and pricing policies cannot be adopted without a prior analysis of what the consu

Abstract

Transportation demand analysis is of capital importance to generate efficient transport policies. Transport projects like investments in infrastructure, changes in operating and pricing policies cannot be adopted without a prior analysis of what the consumer behaviour will be. In particular, it is useful to forecast the response of users to changes brought about by these measures. This is a hard task since the decision maker cannot observe variables which determine the traveller's choice in terms of mode of transport or destination of the trip for example and is only able to attribute choice probabilities to individuals. The objective of the paper will be to try to improve the probability the analyst attributes to the traveller. In this aim, we use a microeconomic approach to derive a travel demand model which is disaggregate at the individual level. Travel demand is described with discrete variables, in particular by hisher choice in terms of mode of transport and destination of hisker trip. Moreover, this choice will be called an alternative .The selection of the different alternatives is based on the principle of utility maximisation or (dis)utility minimisation, which represents preferences of the individual. Nevertheless, this (dis)utility is not known with certainty and contains a random component. The principle of utility maximisation applied to a set of alternatives leads to the construction of individual choice probabilities. In most analysis of the literature, agents have a very restrictive choice set. For technical reasons, studies consider a small number of alternatives of choice, which decreases the interests of results obtained. Or, when they treat lots of alternatives of choice, they do not take into account individual heterogeneity. The individual is assumed to belong to a group of individuals having the same characteristics. But, working in a centre area can not be tackled in the same way by an individual living in the centre of the town than by somebody living in Montpellier outskirts. The decision will depend for example on the number of vehicles at disposal in the household. Then, we can distinguish two main issues: how to treat individual heterogeneity and analyse a choice when alternatives are numerous. In the literature, the simplest model derived from discrete choice analysis is the Standard MultiNomial Logit model. It is characterised by its linear-in-parameters and linear in variables utility function. It has been extensively used for its closed form mathematical structure and its computational simplicity. However, when applied to large choice sets, it does not treat individual heterogeneity. In this paper, we show that a similar approach can be adopted by using a simple standard household activity survey. We formulate a variant of the MultiNomial Logit model which we will denote by the N.L.M.N.L model (Non Linear Multinomial Logit Model) with a specification of the utility function that accounts for population heterogeneity. In addition, this specification is parsimonious in the number of parameters to be estimated allowing for the introduction of a large number of alternatives of choice. It remains tractable and is estimated using simple maximum likelihood. The functional form is then tested against the S.M.N.L model and results are compared. Market shares and their elasticities are computed and underline the flexibility of our functional form. The paper is organised as followed In Section 2, we recall the S.M.N.L model. In section 3,we present our theoretical economic N.L.M.N.L model and show how it leads to a flexible substitution pattern. Section 4 makes a short analysis of the data and describes the estimation procedure. In section 5, we comment the determinants of the joint choice as well as market shares and their elasticities. We compare these results to those of the S.M.N.L. Section 6 concludes and gives some extensions of our research.

Publisher

Association for European Transport