A comparison of collision-based and conflict-based safety evaluation of left-turn bay extension

There is a need to validate the results of conflict-based before-after (BA) evaluations in terms of their collision reduction estimation. This study compares the results of a collision-based and a conflict-based BA safety evaluation of left-turn bay extension. Both collision and conflict data were collected and used in the evaluation. The collision-based BA evaluation was conducted by developing a Poisson-Lognormal intervention (PLNI) model which was estimated by the full Bayesian approach. The conflict-based BA analysis was performed by an extreme value theory (EVT) model developed with the generalized extreme value (GEV) distribution. The comparison showed consistent results from the two approaches. The results showed a reduction of 56.2% in actual collisions from the PLNI model and a reduction of 63.9% in estimated collisions from the EVT model. These results provide support for the validity and reliability of using traffic conflicts in BA studies.