2022 Volume 14 Pages 2093-2112
This study aims to employ the negative binomial model and empirical Bayes model to estimate the accident risk for vulnerable road users at the road intersections in Toyohashi, Japan. A comprehensive dataset including 831 pedestrians and 3874 bicycle accidents at 16,283 intersections over a 10-year period (previous five years and next five years) from 2009–2018 is utilized. The obtained results indicate that accident estimation with the empirical Bayes model is highly correlated (r = 0.882) with the previous accidents and moderately correlated (r = 0.445) with the future accidents, which are considerably better than those assessed by the negative binomial model (r = 0.269 and r = 0.266), for bicycle as well pedestrian accidents. Moreover, the empirical Bayes model reveals the effect of “intersection angle deviation” alone on each intersection by exposing the high-risk intersections, which is insignificant in the negative binomial model.
