Left-turning maneuvers are considered to be the highest risk movements at intersections and two-thirds of the crashes associated with left-turns are reported at signalized intersections. Left-turning vehicles typically encounter conflicts from opposing through traffic. To separate conflicting movements, transportation agencies use a protected-only phase at signalized intersections where each movement is allowed to move alone. However, this could create delays and thus the concept of a protected-permissive phase has been introduced to balance safety and delays. However, the permissive part of this phasing scheme retains the safety concerns and could increase the possibility of conflicts resulting in crashes. This research developed a model that can predict the number of crashes for protected-permissive left-turn phasing, based on traffic volumes and calculated conflicts. A total of 103 intersections with permissive-protected left-turn phasing in Kentucky were simulated and their left-turn related conflicts were obtained from post processing vehicle trajectories through the Surrogate Safety Assessment Model (SSAM). Factors that could affect crash propensity were identified through the Principal Component Analysis in Negative Binomial Regression. Nomographs were developed from the models which can be used by traffic engineers in left-turn phasing decisions with enhanced safety considerations.

The identification of roadway sections with a higher than expected number of crashes is usually based on long term crash frequency data. In situations where historical crash data are limited or not available, surrogate safety measures, based on characteristics such as road geometries, traffic volume, and speed variation are often considered. Most of existing crash prediction models relate safety to speed variation at a specific point on the roadway. However, such point-specific explanatory variables do not capture the effect of speed consistency along the roadway. This study developed several measures based on the speed profiles along road segments to estimate the crash frequency on urban streets. To collect speed profile data, second-by-second speed data were obtained from more than 460 GPS-equipped vehicles participating in the Commute Atlanta Study over the 2004 calendar year. A series of speed data filters have been developed to identify likely free-flow speed data. The quantified relationships between surrogate measures and crash frequency are developed using regression tree and generalized linear modeling (GLM) approaches. The results indicate that safety characteristics of roadways are likely a function of the roadway classification. Two crash prediction models with different set of explanatory variables were developed for higher and lower classification roadways. The findings support the potential use of the profile-based measures to evaluate the safety of road network as the deployment of GPS-equipped vehicles become more prevalent.