Abstract:

A longitudinal traffic accident data set was constructed based on the collected in 4 years severe crush(involving light injury and more severe) and related traffic data on 87 4legged signalized intersections in Beijing city. The data were fitted with the basic negative binominal model(NBM) and the generalized estimation equation(GEE) based NBM with 3 different working correlation structures. The goodnessoffits of the data samples by the 4 regression models were compared by the cumulative residual tests, and the results show that the GEE NBM with autoregressive correlation structure has the best goodnessoffit. The model was used to estimate the effect of each significant factor on severe crash occurence. The estimations were compared with those by the basic NBM, and the effect of the temporal correlation on the model estimation was revealed.

Key words: engineering of communication and transportation safety, accident analysis, generalized estimating equations, signalized intersection, accident frequency, negative binomial regression