Exploring the spatial effects on the level of congestion caused by traffic accidents in urban road networks: A case study of Beijing

When traffic accidents occur on the urban link, the induced congestion propagates upstream and affects adjacent links. Since the congestion on an urban link is affected by its adjacent links, the spatial correlations of adjacent links should be considered when we model the congestion. Although there has been a proliferation of studies that investigate the determinants of the congestion caused by traffic accidents, the spatial correlations of adjacent links in traffic accident scenarios have not been well explored. To fill this gap, we explore the spatial effects on the level of congestion caused by traffic accidents in urban road networks. Both non-spatial and spatial statistical models are used to quantify the effects of causal factors that influence the level of congestion. We conduct numerical experiments using real data in Beijing and find that: (1) All the spatial models outperform the nonspatial model and the spatial Durbin model has the best performance in terms of the R2, AIC, BIC and likelihood ratio test; (2) All the spatial models show that the level of congestion on an urban link is significantly correlated with that of its adjacent links and the spatial lag coefficient is around 28%; (3) The spatial Durbin model shows that the level of congestion on an urban link is significantly correlated with the attributes of its adjacent links such as the link width; and (4) Both non-spatial model and spatial models show that the level of congestion on a link is mostly correlated with the occurrence time, the accident types and the traffic characteristics before the occurrence of the accident.