A Two-Level Random Intercept Logit Model for Predicting Pedestrian-Vehicle Crash

The issue relating to pedestrian with motor vehicles crashes has received more attention in recent years. In this paper the two-level random intercept model for pedestrian crashes severity prediction that accounts for the interdependent characteristics was introduced. A thought is proposed in this paper by using Bayesian probability inference to calculate the parameters rather than the maximum likelihood method, setting parameter distribution for each parameter, using Monte Carlo Markov (MCMC) algorithm to generate model parameter distributions. To demonstrate this approach, pedestrian crashes data from Colorado is divided into spatial categories: urban road intersection, urban road section, and driveway access. The result shows that the prediction accuracy of Bayesian probability inference is better than the traditional ones of urban road intersection and urban road section. The results prove that causes of pedestrian death or injury crashes at intersections are: pedestrian age, pedestrian direction, vehicle type, and driver speed. Significant factors causing pedestrian death or injury on urban road sections including: road condition, lighting condition, belt use, and driver speed. The analysis results reflect the objective feasibility of the model and the method for model parameter estimation optimization. It is expected that the two-level intercept model and Bayesian probabilistic inference model optimization method can help more researchers, transportation officials, and urban city community planners to make more efficient treatments to proactively improve pedestrian safety.