Prediction of vehicle kilometres travelled: a multilevel modelling approach

This paper builds on previous regression-based approaches that endeavour to account for spatial effects underlying differences in vehicle kilometres of travel (VKT) by private vehicle. A multilevel modelling (MLM) approach is developed with the intent of isolating the variability in VKT attributable to various levels of geographic aggregation. The approach is applied to the prediction of VKT for the Sydney Statistical Division using information from a major household travel survey, supplemented with measures of accessibility, density, and land-use designed to capture different spatial influences. MLM null models show that around half the variation in VKT is attributable to effects at the higher spatial unit while the development of full models (i.e., with all the independent variables included) reduces the unexplained variance in VKT substantially. Diagnostics of model fit showed the MLMs offered small improvements over current OLS methods.