Safety Evaluation of Intersection Conflict Warning Systems

Chapter 7. Before–After Evaluation Results

Aggregate Analysis

Table 18 through table 21 provide the estimates of expected crashes in the after period without installation, the observed crashes in the after period, and the estimated CMF and its standard error for all crash types considered. Results are provided separately for each State as well as for all States combined. Results are provided separately for individual States because the application practices varied for all three. Minnesota used a variety of post-mounted signs, including static signs with flashers, blank-out signs with flashers, visual displays, and signs with LED arrows indicating the direction of conflicting vehicles. Missouri used post-mounted static signs with flashers exclusively. North Carolina used static signs that were a mix of post mounted and overhead where the overhead signs were installed at the intersection. The results were combined to further draw inferences on the overall effect of ICWSs.

The results for Minnesota in table 18 were inconsistent across crash types and by number of through lanes on the major route. Statistically significant reductions at the 95-percent confidence level were found for fatal and injury crashes for two-lane at two-lane intersections only. However, the sample sizes were quite small, so readers should use caution when attempting to draw meaningful conclusions from the results.

Statistically significant results at the 95-percent confidence level are indicated in boldface.

The results for Missouri in table 19 indicate reductions for nearly all crash types for both two‑lane at two-lane intersections and four-lane at two-lane intersections. No statistically significant results were found for two-lane at two-lane intersections because of small sample sizes; however, significant reductions were found for four-lane at two-lane intersections. Reductions were found for total crashes, fatal and injury crashes, and nighttime crashes that were significant at the 95‑percent confidence level.

Statistically significant results at the 95 percent confidence level are indicated in boldface.

The results for North Carolina in table 20 indicate statistically significant reductions at the 95‑percent confidence level for all crash types except nighttime crashes for two-lane at two-lane intersections. A statistically significant reduction at the 95-percent confidence level was found for nighttime crashes for four-lane at two-lane intersections; however, insignificant decreases were found for total crashes, fatal and injury crashes, and right-angle crashes. These findings are explored further in the Disaggregate Analysis section of this chapter.

Statistically significant results at the 95-percent confidence level are indicated in boldface.

The combined results in table 21 indicate reductions for all crash types analyzed for both two‑lane at two-lane and four-lane at two-lane intersections. The reductions were statistically significant at the 95-percent confidence level for all crash types except nighttime crashes for two-lane at two-lane intersections. The reductions were statistically significant at the 95-percent confidence level for all crash types except for rear-end crashes for four-lane at two‑lane intersections.

Statistically significant results at the 95-percent confidence level are indicated in boldface.

For two-lane at two-lane intersections, the crash type with the smallest CMF (which translates to the greatest reduction) was rear-end with a CMF of 0.425, which was statistically significant at the 95-percent confidence level. Total, fatal and injury, and right-angle crashes had estimated CMFs of 0.733, 0.701, and 0.803, respectively, which were also statistically significant at the 95‑percent confidence level. Nighttime crashes had an estimated CMF of 0.898, which was not statistically significant at the 95-percent confidence level. It is important to consider the sample size used to develop each CMF when interpreting the results. For example, the sample sizes used to develop CMFs for rear-end and nighttime crashes were relatively low, resulting in larger standard errors and confidence intervals compared with the CMFs for total, fatal and injury, and right-angle crashes.

For four-lane at two-lane intersections, the crash type with the smallest CMF (which was statistically significant at the 95-percent confidence level) was nighttime crashes, with a CMF of 0.612. Total, fatal and injury, and right-angle crashes had estimated CMFs of 0.827, 0.802, and 0.850, respectively, which were also statistically significant at the 95-percent confidence level. Rear-end crashes had an estimated CMF of 0.973, which was not statistically significant at the 95-percent confidence level.

As discussed in the literature review, the most comprehensive study to date of ICWS applications was conducted by Simpson and Troy using data from North Carolina.⁽¹⁾ This reportincludes recommended CMFs for two-lane at two-lane intersection but does not provide recommended CMFs for four-lane at two-lane intersections because the small sample size precluded a rigorous analysis. Simpson and Troy recommended a CMF of 0.897 for total crashes and 0.878 for injury crashes at two-lane at two-lane intersections.⁽¹⁾ Greater crash benefits were indicated in the present study, which were attributed to the following characteristics of the present study:

• Included only four-legged intersections.
  
  
• Limited the number of study years to no more than 5 years before and 5years after installation.
  
  
• Used SPFs to account for changes in traffic volumes.
  
  
• Used annual multipliers to account for trends at reference sites.
  
  
• Used a multistate database.

Disaggregate Analysis

The disaggregate analysis sought to identify those conditions under which the ICWS strategy was most effective. Because total, fatal and injury, and right-angle crashes were the focus of this strategy, these crash types were the focus of the disaggregate analysis. Several variables were identified as being of interest and available for all three States, including installation category, message, presence of turn lanes, presence of lighting, presence of additional countermeasures, major and minor route AADT, major and minor route posted speed limit, and expected crash frequency in the before period.

For installation category, the categories developed by NCDOT were expanded for use in this study. Categories for further analysis were as follows:

• Category 1—Overhead signs and flashers at the intersection on major; loop on minor.
  
  
• Category 2—Overhead signs and flashers at the intersection on minor; loop on major.
  
  
• Category 3a—Post-mounted signs and flashers in advance of the intersection on major; loop on minor.
  
  
• Category 3b—Post-mounted signs and flashers at the intersection on minor; loop on major.
  
  
• Category 4—Locations with a combination of category 1 through category 3.

For two-lane at two-lane intersections, all categories were considered in the disaggregate analysis. For four-lane at two-lane intersections, categories 3a and 3b were included in the disaggregate analysis. Category 1 and category 2 systems were found only in North Carolina, and these systems were installed at the intersection on both the major and minor road. Category3a signs were found only in Missouri and North Carolina and were installed in advance of the intersection. Category 3b systems were found only in Minnesota and Missouri and were installed at the intersection.

Table 22 provides the disaggregate results by category for two-lane at two-lane intersections and four-lane at two-lane intersections. The number of intersections is indicated for each installation category. For each crash type, the estimated CMF, standard error (in parentheses), and sample size in terms of observed crashes in the after period is provided. It is important to consider the sample size used to develop the CMFs when applying the CMFs.

For two-lane at two-lane intersections, results indicate statistically significant reductions at the 95-percent confidence level for all crash types for category 1, 3a, and 4 systems. Considering the standard errors of the CMFs, it was difficult to draw a conclusion about the relative effectiveness of categories 1, 3a, and 4; with the exception of the CMFs for right-angle crashes, the results were not statistically different at the 95-percent confidence level. The majority of the category 4 sites consisted of a combination of categories 1 and 2 or a combination of categories 3a and 3b.

For four-lane at two-lane intersections, the results indicate statistically significant reductions at the 95-percent confidence level for all crash types for category 3a and for total crashes only for category 3b systems. The CMFs for categories 3a and 3b were not significantly different for any crash type.

Statistically significant results at the 95-percent confidence level are indicated in boldface.
In each cell containing results is the estimated CMF, standard error (in parentheses), and the sample size in terms of observed crashes in the after period.
N/A = Not applicable.

It was not appropriate to compare the effectiveness of overhead versus post-mounted applications on the major route from the study results because the placement of treatment differed for the two groups. Post-mounted ICWSs were installed in advance of the intersection, whereas all overhead signs were installed at the intersection. Ideally, to address the difference between post-mounted and overhead signs, the placement should be taken into consideration. The MUTCD states that warning signs should be placed to provide an adequate PRT. This suggests that the findings in table 22 may have been influenced by system placement, which could not be addressed in this research.⁽²⁾

Table 23 presents the disaggregate results for intersections by sign message. For two-lane at two‑lane intersections, the ICWS strategy appeared to be slightly more effective when the message specifically stated “WHEN FLASHING,” compared with signs that did not have the message. Considering the standard errors of the CMFs, there was no statistical difference between the two conditions. All systems in Minnesota located at two-lane at two-lane intersections had a “WHEN FLASHING” message and were therefore considered in this category. There was no apparent difference by message for four-lane at two-lane intersections.

Statistically significant results at the 95-percent confidence level are indicated in boldface.

Table 24 presents the disaggregate results by the presence of intersection lighting. There was no apparent difference by lighting presence for two-lane at two-lane intersections. For four-lane at two-lane intersections, the strategy appeared to be more effective at sites with intersection lighting. The difference was statistically significant at the 95-percent confidence level for fatal and injury crashes.

Statistically significant results at the 95-percent confidence level are indicated in boldface.

Table 25 presents the disaggregate results by expected crash frequency in the before period. There was no apparent difference by expected crash frequency for two-lane at two-lane intersections. For four-lane at two-lane intersections, the ICWS strategy was more effective when the expected crash frequency was higher in the before period. This is logical because the strategy was often used at intersections with unusually high crashes or issues related to limited sight distance. For total crashes, there did not appear to be a benefit if the expected crash frequency was less than or equal to three crashes per year before installation; however, there was a significant reduction for sites with more than three expected crashes per year in the before period. The results for right-angle crashes were significantly different from each other for sites with less than or equal to 2.5 expected crashes per year versus sites with more than 2.5 expected crashes per year before installation. There did not appear to be a benefit if the expected fatal and injury crash frequency was less than or equal to two crashes per year before installation; however, there was a significant reduction for sites with more than two expected fatal and injury crashes per year in the before period.

Statistically significant results at the 95-percent confidence level are indicated in boldface.