Split failures occur when the green phase of a traffic signal fails to adequately serve the demand on a specific approach. As an important performance measure within automated traffic signal performance measures (ATSPMs), split failures provide valuable insights into both mobility and safety at signalized intersections and across the broader traffic system. At signalized intersections, split failures are associated with increased queuing, delays, and emissions from idling vehicles. From a safety perspective, split failures may contribute to risky driver behaviors, such as red-light running, which can increase crash risk. Understanding the factors influencing split failures is crucial for effective traffic signal management. This study investigates the factors contributing to split failures using high-resolution event-based video data from Miovision 360 cameras at 16 signalized intersections across Pima County, Arizona, combined with probe-vehicle speed data from INRIX. Several factors, including temporal characteristics, signal timing parameters, traffic flow characteristics, and geometric characteristics, were analyzed to assess their impact on split failure occurrences. To account for the zero-inflated and overdispersed nature of split failure data, negative binomial (NB), zero-inflated NB (ZINB), and hurdle NB (HNB) models were applied. Results indicate that the ZINB model provided the best fit, likely due to its ability to handle excess zeros in the split failure data. Key findings show that the frequency of split failures increases on major roads, on Fridays, at higher volumes, and at intersections with longer crossing distances. Conversely, the frequency of split failures decreases during morning peak hours, on Mondays, with higher operating speeds and posted speed limits, longer green ratios and yellow times, and a greater number of lanes. These insights offer practitioners valuable guidance for optimizing signal operations to enhance both traffic flow and safety.
Geffen et al. (Mon,) studied this question.