Choice of Mode for Long Distance Travel: Current SP-based Models from Three European Countries



Choice of Mode for Long Distance Travel: Current SP-based Models from Three European Countries

Authors

K W Axhausen, ETHZ-IVT, CH; M de Lapparent, INRETS-DEST, FR; A Frei, ETHZ-IVT, CH

Description

We use a panel mixed Logit approach to analyse long distance travel mode choice between 4 alternatives using SP data. The modelling allows for non linear and interaction effects, random coefficients, error components, and individual random effects.

Abstract

KITE stands for "a Knowledge base for Intermodal Transport in Europe". It is a FP6 research project that provides relevant information about passenger intermodal behaviour and that allows developing and evaluating transport policies and measures to favour it. It integrates and it disseminates current existing and future information and data. We refer the reader to http://www.kite-project.eu/ for more detailed information. Part of the project develops a strategic approach for the collection of long distance and intermodal survey data. In the present article, we use data from an associated SP survey that was conducted at the end of year 2008 and the beginning of year 2009 in three countries: Switzerland, Czech Republic, and Portugal. The sample size is 900 individuals. It distributes between the three countries as 40% for Czech Republic, 40% for Switzerland, and 20% for Portugal. Each decision maker is faced to a sequence of 4 choice experiments, thereby giving a total number of 3600 observations. It focuses on the choice of a main mode of transport for long distance travel. Four modes of transport are considered: car, air, train, bus. Each is described with at most five attributes: in-vehicle time, access time, cost, departure time of the day, and number of transfers. Even though it does not consider explicitly combination of modes of transport, it sheds light on potential demand for intermodal travel as air, train, and bus are modes for which travellers access and egress by generally using other modes of transport.
We present the results of an econometric analysis of choice behaviours that we postulate to be consistent with random utility maximization. It is estimated and compared several parametric formulations of a panel mixed Logit model. From a general point of view, the specification of the strict part of the indirect utility function is flexible enough to account simultaneously for nonlinearities and interaction effects. Its random part is designed to allow existence of continuous unobserved taste heterogeneity, error components, individual random effects, and idiosyncratic errors. Fit statistics are presented for all the estimated models but only the detailed results of the most informative one according to some statistical criteria are reported and discussed.
Nonlinearities are modelled by means of either Box-Cox transformation and/or multiplicative terms that incorporate the travel attributes. Interaction effects pass either by defining the parameters of the model as functions of the characteristics of the decision maker or by entering directly the utility function through the multiplicative terms. Individual random effects model framing effects that may exist due to the repetitive nature of the SP experiments. Error components shed light on likely unobserved substitution patterns (e.g. substitution between collective modes of transport and/or between ground modes of transport) and unobserved taste heterogeneity models random variation of the parameters that weigh the independent variables (i.e. each decision maker has specific tastes). Note that distributional assumptions will be set to keep consistent with economic demand theory (e.g. negativity of price and time coefficients).
As a result, the values of travel time savings are defined as functions of travel attributes and socioeconomic characteristics. Not only are they key figures that will be used in the KITE project for development of an aggregate cost-benefit analysis tool to assess intermodal transport projects but they also give important information on disaggregate behavioural responses to modification of transport supply, particularly as it regards the incentives to relieve the use of a car for long distance travel. To that extent, it is formally derived, computed, and then discussed, a series of values on the basis of our work assumptions.
The outline of the article is as follows. After an introductory section, data are presented in a second section. The econometric specification is then developed in a third section. The results are discussed in a fourth section. In a last section, it is drawn conclusions and it is proposed further extensions of the proposed approach.

Publisher

Association for European Transport