Methodology

74 intersections covering four categories of VEWF intersection warning systems were evaluated.

1. Category 1 – Overhead Signs and Flashers at the Intersection on Major, Loop on Minor (24 sites)
2. Category 2 – Overhead Signs and Flashers at the Intersection on Minor, Loop on Major (19 sites)
3. Category 3 – Post Mounted Signs and Flashers in Advance of Intersection on Major, Loop on Minor (23 sites)
4. Category 4 – Locations with Combination of Category 1 through Category 3 (8 intersections)

Treatment sites were located in urban and rural areas with mainline approach speed limits ranging from 35 mph to 55 mph, although the majority of sites were rural, isolated, high speed facilities. The intersection annual average daily traffic (AADT) ranged from approximately 3,000 to 30,000 vehicles entering per day. The type of mainline facilities varied with the intersection geometry including 56 two-lane at two-lane intersections, 11 four-lane divided at two-lane intersections, and 7 multi-lane undivided at two-lane intersections.

A before and after crash analysis was performed at each intersection utilizing the Traffic Engineering Accident Analysis (TEAAS) software that accessed the North Carolina Traffic Records Database. The time periods analyzed for each location varied depending on when the treatment was installed. In most cases, the ending dates for the analyses were determined by the available crash data at the time the crash analysis was completed, which was through October 31, 2011. Some of the analysis periods were over ten years in duration, and the average before period was approximately five years.

Empirical Bayes before and after techniques were utilized to account for selection bias and to overcome the threat of regression to the mean, along with other potential deficiencies in a naïve before and after analysis. The Empirical Bayes approach requires the use of reference sites as well as before period data from the treatment site to estimate the expected safety of the treatment site had no improvements been made. Approximately five reference sites per treatment site were chosen.

Findings

• For all 67 treatment locations, the results of the Empirical Bayes analysis with the traffic volume adjustment yielded
  • 3 percent (+/- 5 percent) reduction in target crashes
  • 7 percent (+/- 4 percent) reduction in total crashes
  • 6 percent (+/- 6 percent) reduction in injury crashes
  • 16 percent (+/- 16 percent) reduction in severe injury crashes
• There was a noticeably greater reduction in total crashes for Categories 3 and 4 (19.3 percent, 25.1 percent) than Categories 1 and 2 (-9.1 percent, 3.5 percent). Category 1 actually showed an increase in total crashes.
• Crash analysis results for four-lane divided at two-lane intersections (11 of the intersections studied) showed no apparent reduction in crashes. Overall, the locations with this intersection geometry did not experience a reduction in total, target, injury or severe injury crashes. The use of VEWF systems at four-lane divided at two-lane intersections may be more a “band-aid” treatment that does not address the root cause of crashes in this situation and a geometric change (i.e. directional crossover, offsetting minor road legs, or crossover closure) may be a more appropriate solution to address crash patterns likely related to gap-acceptance.