Bayesian spatial modeling for speeding likelihood using floating car trajectories

Speeding likelihood is usually used to measure drivers' propensity of committing speeding. Albeit some studies have analyzed speeding likelihood, most of them are inadequate in considering spatial effects when analyzing speeding behaviors on urban road networks. This study aims to fill this knowledge gap by modeling speeding likelihood with spatial models and then evaluate the influence of contributing factors. The percent of speeding observations (PSO) is adopted to represent the speeding likelihood. The speeding behaviors and PSO of each floating car (i.e., taxi) are extracted from the GPS trajectories in Chengdu, China. PSO is modeled by several Bayesian beta general linear models with spatial effects, namely the beta model, beta logit-normal model, beta intrinsic conditional autoregressive (ICAR) model, beta Besag-York-Mollié (BYM) model, and beta BYM2 model. Results show that the beta BYM2 model performs better than other models in terms of data-fitting. According to the estimates from the beta BYM2, spatial correlation is the main reason for the model variability. The roads with more lanes and roads linked by elevated roads are found to increase the speeding likelihood, while higher speed limits, intersection density, traffic congestion, and roadside parking are associated with lower speeding likelihood. These findings provide valuable insights for designing effective anti-speeding countermeasures on urban road networks.