Using geographically weighted regression to forecast rail demand in the sydney region

This paper is centred on using Geographically Weighted Regression (GWR) to investigate the spatial variability in the factors which influence rail demand patterns in the Sydney region of New South Wales, and to explicitly incorporate these variations in the demand forecasting process. It explores whether allowing spatial parameter variability to be included in demand models enhances their explanatory power and improves their forecasting capability. The paper firstly reviews the different methodologies in use for estimating and forecasting demand, both in the Sydney region and elsewhere around the world. Against this background the methodology of GWR is explained, identifying the key differences in both inputs and outputs of a GWR based model. The paper then describes the development of 'conventional' global regression models of rail demand in the Sydney region, and reports on the recalibration of the most successful model using GWR. A large number of explanatory variables are tested in the models, including catchment population and employment, household size, income and age profile, car ownership levels, train frequency, bus interchange potential, bicycle storage provision and distance to the city centre. The results from the global and GWR models are then compared, with the significance of the spatial variation in the GWR models tested using an AICc-based criterion. This comparison informs a discussion of the extent to which modelling spatial parameter variation contributes to a better understanding of rail demand and its forecasting. The discussion concludes by identifying the relative merits of the developed GWR models as compared to other models for forecasting rail travel.