A well-designed and operated traffic signal can increase the capacity of its intersection, reduce the prevalence of certain types of crashes, especially right-angle and left-turn crashes, and provide crossing opportunities for users facing heavy, opposing traffic streams. One important decision in designing a signal is the appropriate treatment given to left-turns (LT), and traditionally the choice has been between protected left-turns, where the left-turning drivers are given right-of-way and opposing traffic required to stop, and permitted left-turns, where left-turning drivers are required to yield to opposing traffic but may make the turn if an adequate gap appears. Protective turn phases can reduce the occurrence of left-turn crashes and allow the turns to be made when opposing traffic is heavy, but giving protective phasing when left-turn demand is light reduces the overall capacity of the intersection and, when opposing traffic is light, protective-only phasing can increase the delay experienced by left-turning drivers. Protective-permissive left-turn treatments (PPLT), where a protective interval is followed or preceded by a permissive interval, attempts to capture some of the benefits of protective LT phasing while avoiding some of the costs. A perennial difficulty however has been providing drivers with indications that clearly distinguish protected from permitted turns, and in 2003 a report commissioned by the National Cooperative Highway Research Program recommended that the flashing yellow arrow (FYA) indication be used to indicate permitted left-turns One potential advantage of four or five-section signal heads with FYA is that protective, permissive, or PPLT treatments can be varied throughout the day as traffic conditions might warrant. However, while tools exist for predicting the operational effects of within-day variation in traffic conditions, predicting how the risk of LT crashes might vary is more problematic. A review of existing literature indicated that while models do exist for predicting annual totals of LT crashes, predicting how risk varied within a day is still an open question. For this project, in order to accommodate data availability and data quality issues, it was decided to employ a matched case-control study design rather than the traditional cross-sectional design. The cases consisted of 436 left-turn crash events occurring at signalized intersections operated by MnDOT, identified in part from data provided by the Highway Safety Information System and in part using the MNCMAT crash mapping tool. For each case, five hourly periods for the same intersection approach and on the same day as that of the case were then randomly chosen. For both the case and control hours the left-turn volume, the opposing volume, and the opposing left-turn volume were estimated by developing statistical models for adjusting available turning movement counts to the appropriate days and hours. These hourly volumes were then used as independent variables in logistic regression models which used the traffic volumes to discriminate times when crashes occurred from times when crashes were absent. Because a matched case-control design does not allow one to analyse the effect of features that are common to both the cases and controls, such as speed limits or geometric features, the crash-occurring intersections were classified according to opposing speed limit, the type of LT protection, and whether or not the intersection’s geometrics indicated that a left-turning driver’s sight distance could be obstructed by an opposing left-turning vehicle. Separate statistical models were then estimated for each intersection type. For three of the intersection types, the available sample sizes were sufficient to reliably identify how left-turn crash risk varies as opposing and left-turn traffic volumes vary. The resulting statistical models were then incorporated in a spreadsheet tool which asks a user to enter available hourly turning movement counts for an intersection approach, along with several measurements describing the intersection’s geometry. The tool then determines if a potential for sight distance obstruction exists and estimates the hourly left-turn and opposing traffic volumes for those hours not included in the turning movement counts. Finally, using the statistical model appropriate for the intersection approach’s type, the tool computes how the relative risk for a left-turn crash varies as the hourly traffic volumes vary throughout the 24 hours of the day. (Author/publisher)
Abstract