Understanding temporal correlation in severe traffic conflicts: evidence from cycle-level at signalized intersections

This study aims to develop traffic conflict models that accommodate temporal correlation across observations while considering the characteristics of overdispersion and the preponderance of zeros in severe conflict data. The Poisson Lognormal (PLN) model, PLN-Lindley model, PLN-Lindley linear time (PLN-Lindley-LT) model, and PLN-Lindley autoregressive time (PLN-Lindley-AT) model, were proposed. The PLN-Lindley model addresses overdispersion, the preponderance of zeros in severe conflict counts. The PLN-Lindley-LT and PLN-Lindley-AT models incorporate the temporal correlation between cycles. Severe traffic conflicts and shock wave characteristics were collected from six signalized intersections in British Columbia, Canada. Results showed that the PLN-Lindley model better fitted the data than the PLN model, while PLN-Lindley-LT and PLN-Lindley-AT highlighted significant temporal correlations, with the latter performing best. Key findings indicated fewer severe conflicts with lighter traffic volume and shock wave area but more with lower platoon ratio. These insights can guide real-time safety interventions such as safety warnings at intersections.