Modelling Park-and-ride in the PRISM Model for the West Midlands Region

Modelling Park-and-ride in the PRISM Model for the West Midlands Region


J Fox, RAND Europe, UK


This paper describes the development of park-and-ride models, their integration within the overall PRISM forecasting model, and results of model predictions for park-and-ride facilities.


RAND Europe and Mott MacDonald have recently developed a model system to analyse the distributional impacts of policies and investments across the West Midlands Region on the local population. The design of this model system was driven by stakeholder consultation, and a key policy issue identified during the course of this consultation process was the need to represent park-and-ride use in the model system. In response to this policy requirement, an explicit representation of park-and-ride access to train and metro was included in the model system in order that demand for existing and planned park-and-ride facilities could be accurately forecast.

Some data on park-and-ride was available from a Household Interview (HI) carried out in the West Midlands during 2001, which collected information from around 12,000 households and has formed the principal data source for the development of the PRISM model system. However as train (and metro) represent a small proportion of total travel in the West Midlands, it also proved necessary to use specific train and metro surveys in order to obtain sufficient data for both the analysis and modelling effort. The specific train surveys used were standard platform surveys undertaken by CENTRO, the local public transport operator, for their own research. Consequently there was no need for additional data collection.

The modelling followed a three stage process:

§ Analysis of park-and-ride data;
§ Development of stand-alone park-and-ride models;
§ Incorporation of the park-and-ride models in the overall mode and destination choice structure.

These three stages are discussed in brief below.

The analysis of train and metro park-and-ride data revealed a number of important findings which impacted upon the modelling effort. The first finding was that both car-driver access, often termed ?park-and-ride?, and car-passenger access, often termed ?kiss-and-ride?, are important. It is important to separate out park-and-ride car access in order to forecast demand for parking at park-and-ride facilities. Following this analysis three access modes were represented in the models: car-driver, car-passenger and other (walking, cycling and access by bus). Another important outcome from this analysis was that access mode shares vary significantly by model purpose and therefore purpose-specific models for access mode choice were required.
The stand-alone models, developed from the train survey data collected by CENTRO, represented both the choice of access mode, and for car-access the choice of station, using hierarchical tree-logit models. The models comprised three sets of variables:

§ Level-of-service parameters, which for car access reflect the attractiveness of different station alternatives, for example the choice between a long car access journey and a short journey by train, and a short car access journey and a long journey by train.
§ Socio-economic parameters, representing differences in access mode preferences across the population identified in the analysis discussed above.
§ Size variables to represent the attractiveness of different park-and-ride sites, for example by the number of parking spaces.

A key outcome from the stand-alone models was that station choice for car access to train is more elastic than access mode choice, i.e. travellers are more likely to switch between park-and-ride stations than change their access mode.

The final stage in the modelling process was to combine the stand-alone models of access mode and station choice with the disaggregate models of mode and destination choice estimated from the HI data. Hierarchical tree-logit models were used to account for the different elasticities of the various choice decisions represented in the models, specifically:

§ the relative elasticity of main mode and destination choice;
§ the relative elasticity of main mode and access mode choice for train and metro;
§ the relative elasticity of access mode and station choice for car access to train and metro.

Furthermore the combined models of access mode, main mode and destination choice enabled the joint estimation of public transport level-of service parameters for in-vehicle time, access time, wait time and so on. Due to the large volume of additional public transport data incorporated in the models, the joint parameters were estimated with greater statistical significance, and had more acceptable values, than the parameters estimated from the mode-destination choice data alone.

The problems of joint estimation have been overcome, providing a methodology for determining the relative elasticities of each choice decision using commonly available data sources. The model has now been implemented and validated in the PRISM model system. In implementation the car access legs of train and metro trips are loaded onto the car network together with car main-mode trips, ensuring that park-and-ride journeys contribute properly to congestion.

The models are now being used to forecast demand for park-and-ride facilities across the West Midlands. The paper will present future forecasts of demand, both for existing facilities, and for stations along a new metro corridor, indicating the extent to which park-and-ride can be effective in reducing car traffic.


Association for European Transport