Warrants for Exclusive Left Turn Lanes at Unsignalized Intersections and Driveways

Accommodating left turns at unsignalized intersections is one of the most challenging problems in traffic engineering. Over the last forty years, a small number of studies developed guidelines for traffic engineers to help in deciding when a left turn lane is warranted for a given situation. Building on these previous attempts, this report describes the development of a refined decision support system (DSS) for assessing the likely benefits of left-turn lane installations as an aid to deciding whether a left-turn lane is warranted. The developed DSS is designed to predict these likely benefits based on several measures including delay savings, reductions in percent stops, increases in fuel efficiency, and reductions in emissions. The first step in developing the DSS was to use microscopic simulation to model several real-world unsignalized intersections with different geometric configurations and located in different area types. After carefully calibrating these models, several scenarios covering a wide range of operational conditions were simulated. The output from these simulation runs was then used to train a set of Multi-layer Perceptron Neural Networks (NNs), and to generalize the results from the models’ runs. The NNs were then incorporated into a DSS that can help an analyst quantify the impacts of a proposed new development as well as estimate the benefits of left-turn lane installations. The report also presents an investigation of the safety effects of exclusive left turn lane installation at unsignalized intersections. Crash prediction models were estimated using crash and volume data from intersections without left turn lanes by intersection category and crash category. Intersections throughout Connecticut were selected, representing six different types of intersections based on rural versus urban location, number of approach legs and number of through lanes on the main road. Negative binomial modeling was used with generalized estimation equations to account for the correlation among the crash counts for an intersection through the years. The expected number of crashes was predicted using the prediction models for intersections with left turn lanes assuming no left turn lanes were present. If the observed crash counts were significantly lower than the predicted, then there was evidence that the left turn lane created a safer condition. The results of comparing the observed and predicted crashes showed that the intersections were safer for same direction crashes when left turn lanes were installed, except for those on urban two lane roads, at which no safety effect is discernible.