Abstract:•This paper is an exploratory study of the relationship of geometric design features, land use, traffic volume, and travel speeds to safety.•It uses a comparatively new method of measuring speeds by extracting GPS data from taxis operating on Shanghai’s urban network.•Bayesian inference using four different models, Poisson-lognormal (PLN), PLN with Maximum Likelihood priors, hierarchical PLN (HPLN), and HPLN with Maximum Likelihood priors was used to estimate crash frequencies.•Results showed the HPLN-ML models had the best goodness-of-fit and efficiency, and models with ML priors yielded estimates with the lowest standard errors.•Higher average speeds were associated with higher crash frequencies during peak periods, but not during off-peak periods.

Safety modeling of urban arterials in Shanghai, China