Bayesian hierarchical spatial count modeling of taxi speeding events based on GPS trajectory data

Speeding behavior, especially serious speeding, is more common in taxi driver than other driving population due to their high exposure under traffic environment, which increases the risk of being involved in crashes. In order to prevent the taxi and other road users from speed-related crash, previous studies have revealed contributors of demographic and driving operation affecting taxi speeding frequency. However, researches regarding road factors, and spatial effect are typically rare. For this sake, the current study explores the contributions of 10 types of road characteristics and two kinds of spatial effects (spatial correlation and spatial heterogeneity) on taxi total speeding and serious speeding frequency. Taxi GPS trajectory data in a Chinese metropolis were used to identify speeding event. The study then established four kinds of Bayesian hierarchical count models base on Poisson and negative binominal distribution to estimate the contributor impacts, respectively. Results show that Bayesian hierarchical spatial Poisson log-linear model is optimum for fitting both total and serious speeding frequency. For the analysis, it is found that drivers are more likely to commit speeding on long multilane road with median strip, and road with non-motorized vehicle lane, bus-only lane and viaduct or road tunnel. Roads with low speed limit, and work zone are associated with increasing speeding as well. In terms of serious speeding, bus-only lane is not a contributor, while road speed camera number and one-way organization are significantly positive to the speeding frequency. Furthermore, it reveals that two spatial effects significantly increase the occurrence of speeding events; the impact of spatial heterogeneity is more critical.