Estimating walking modal share: A novel approach based on spatial regression models and GIS

Abstract

Traditional 4-stage transport forecasting models allocate trips between modes based on their relative generalized cost (a weighted sum of travel time and cost). However, this approach takes little account of the relative convenience of different modes at the trip distribution stage, which is a particular limitation when forecasting public transport and pedestrian trips. For example, a zone with good pedestrian accessibility to nearby large employment centres would generate a lot more short distance walking trips than another zone located far from employment and with poor pedestrian infrastructure. This paper explores the potential of spatial regression techniques and GIS in understanding and forecasting the proportion of walking trips as a function of the characteristics of the local road network, car ownership, average age and population and jobs within a zone's catchment area. The results presented include plots of spatial correlation for proportion of walking trips, and plots of error terms for simple linear regression models and the Spatial Durbin Model, using Census data for the city of Leeds, UK. This work provides an unprecedented insight into the effect of spatial patterns on walking and it is shown that spatial regression models can produce improved modal split estimates at the trip generation stage by taking into account detailed spatial characteristics of surrounding zones. The results from this work are relevant to accessibility, land use and transport planning policies as well as to the wider application of GIS and spatial regression techniques in transport modelling.