Enhancing Rail Passenger Demand Models to examine Station Choice and Access to the Rail Network

INTRODUCTION AND OBJECTIVES Much analysis of rail travel demand in Great Britain has been undertaken using time-series direct demand models, for example Jones and Nichols (1983), and Owen and Phillips (1987). In these models, changes in demand over time are explained as a function of independent variables that change incrementally over the same time period. However, such an incremental approach is of no use for forecasting the demand from new stations, or for other new rail services. Furthermore, this approach does not handle competition between different stations, nor the impact of access on either rail demand or rail elasticities. There is, therefore, a need for cross-sectional models which can forecast demand for journeys from new stations, or in response to population changes, changes in station accessibility or radical service quality changes. Previous examples include Tyler and Hassard (1973), Holt and White (1981), Shilton (1982), Jones and White (1994), and Wardman (1996). These authors were unable, for obvious reasons, to take advantage of the new opportunities for developing such models which have been presented by the increased availability of machine-readable Geographical Information Systems (GIS) data on populations and road networks. Such data can be combined with data on rail passenger flows and revenues and on rail service quality, these latter data being those already used to develop time-series models. Arising on the growth of computing power, a further opportunity is now presented for potentially more sophisticated cross-sectional models, which may not be amenable to linear regression, to be calibrated using non-linear regression. Some initial attempts to build more sophisticated models have already been reported (Lythgoe and Wardman, 2002; 2004). The objective of this paper is to generalise the station choice model from the earlier work and to show how various limitations have been overcome. There is an emphasis on replacing the MNL station choice form by a particular cross-nested logit form, with different dissimilarity parameters between given station i and each of its competing stations. Introducing such a cross-nested logit form enables the proportion of new journeys from station i abstracted from its competitors to be dependant, inter alia, on the proximity of station i to each of those competitors. In particular, what we propose here is an improvement in that the previous model could only be applied to a subset of origin stations, namely Parkway stations. Also, when fitting the data, a specification error that had been previously identified is remedied by introducing a population elasticity. The origin station choice model described in this paper builds on the original Parkway station model (Lythgoe and Wardman, 2002; 2004) and can predict the demand for inter-urban rail journeys of over 40km between pairs of stations in Great Britain. It is based upon 10,324 observed demand levels from 329 existing stations to 334 destination stations. The aim is that it should be more straightforward to apply than existing techniques for forecasting demand from new or greatly revised stations and services, and that it should provide consistent results.