Analysing speeding behaviour: A multilevel modelling approach

Abstract

This paper examines the variability in speeding for 147 motorists over a five-week period using data collected from Global Positioning System (GPS) technology. A multilevel modelling approach is employed to decompose speeding behaviour into four major levels of variation, namely: inter-individual variation, temporal variation, trip-level variation, and segment level variation. Initially, we estimate a null model (i.e., excludes the explanatory variables) to assess the variations at each level. Results suggest that the driver is more of a factor in speeding as the speed limit increases but that the majority of variation in speeding goes unexplained. This is followed by progressively including explanatory variables (e.g., age, gender, vehicle type, trips purpose etc) at each of the four levels to assess how much more of the variation in speeding can be explained. Results suggest that the reduction in unexplained variance in speeding varies markedly by speed zone, indicating the disproportionately different impacts of explanatory factors.