Safety Evaluation of Centerline Plus Shoulder Rumble Strips

Chapter 7. Before-After Evaluation Results

Aggregate Analysis

Table 16 through table 19 provide the estimates of expected crashes in the after period without treatment, 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 all States combined.

The results for Kentucky in table 16 indicate reductions for all crash types that are statistically significant at the 95-percent confidence level. All treatment sites in Kentucky had SRS or edgeline rumble stripes prior to treatment, so the results indicate that CLRS further reduce run-off-road crashes.

Table 16. Results for Kentucky.

Bold indicates CMF estimates that are statistically significant at the 95-percent confidence level.
CMF = Crash modification factor.
EB = Empirical Bayes.

The results for Missouri in table 17 also indicate reductions for all crash types that are statistically significant at the 95-percent confidence level. Prior to treatment, rumble strips were not present. It is logical that the CMF for total crashes is smaller than that for Kentucky, for which the CMF pertains to the addition of CLRS on roadways that previously had SRS or stripes. That the CMF for run-off-road crashes in Missouri is larger than in Kentucky, where SRS previously existed, is not intuitive, but such comparisons for specific crash types can be influenced by how crash types are defined in different jurisdictions and the extent of the overrepresentation of specific crash types prior to treatment.

Table 17. Results for Missouri.

Bold indicates CMF estimates that are statistically significant at the 95-percent confidence level.
CMF = Crash modification factor.
EB = Empirical Bayes.

The results for Pennsylvania, where rumble strips were not present prior to treatment, are shown in table 18. These results indicate reductions in total, run-off-road, and sideswipe-opposite-direction crashes, and increases in injury and head-on crashes. However, none of these results are statistically significant. Nevertheless, these results are still of interest because they could be used to increase the significance of a combined CMF based on the results of all three States. More discussion of the Pennsylvania results is provided in the next section on disaggregate analysis, while the combined results for the three States are presented next.

Table 18. Results for Pennsylvania.

¹None of the CMF estimates are statistically significant at the 95-percent confidence level.
CMF = Crash modification factor.
EB = Empirical Bayes.

The combined results in table 19 indicate reductions for all crash types analyzed that are statistically significant at the 95-percent confidence level. It should be noted that combining the results of the three States produces a more robust CMF because the standard error of the combined CMF estimate (relative to the CMF) is smaller than that from any State or from the combination of the two States with the most significant results (Kentucky and Missouri). The crash type with the smallest CMF (which translates to the greatest reduction) is head-on with a CMF of 0.632. Run-off-road and sideswipe-opposite-direction crashes have estimated CMFs of 0.742 and 0.767, respectively. For all crash types combined, CMFs of 0.800 for all severities and 0.771 for FI were estimated. For run-off-road, head-on, and sideswipe-opposite-direction crashes combined (i.e., lane departure crashes), the estimated CMF is 0.733. It is important to remember that all crash types considered exclude intersection-related and animal crashes.

As discussed in the literature review, the most comprehensive and reliable study to date of both SRS and CLRS is published in NCHRP Report 641. This report does not include recommended findings for the combination of SRS and CLRS but does recommend CMFs for these treatments separately. A comparison of the results for the combined treatment with the recommended CMFs is encouraging.

In NCHRP Report 641, for SRS, a CMF of 0.85 is recommended for SVROR crashes. The results for other crash types were not statistically significant and so were not recommended. These results included a 6-percent reduction in total crashes and an 8-percent increase in FI crashes. In comparison with the new results, it appears that the effect of combining CLRS and SRS further reduces run-off-road crashes with a CMF of 0.742 for dual application versus 0.85 for SRS alone.

In NCHRP Report 641, CMFs of 0.91 for total crashes, 0.88 for FI crashes, and 0.70 for head-on plus sideswipe-opposite-direction crashes were recommended for CLRS. The new results, which estimated CMFs of 0.800 for all severities and 0.771 for FI for dual application, indicate that SRS further reduce these crashes. However, the CMF of 0.70 for head-on plus sideswipe-opposite-direction crashes suggests that dual application does not further reduce crashes of this type, which is intuitive.

Table 19. Results for combined States.

Bold indicates CMF estimates that are statistically significant at the 95-percent confidence level.
CMF = Crash modification factor.
EB = Empirical Bayes.
HO = Head-on.
ROR = Run-off-road.
S-OD = Sideswipe-opposite-direction.

Disaggregate Analysis

While the combined results for all States provide results that meet expectations and are statistically significant, the results are not consistent amongst the three States. The results for Kentucky and Missouri show statistically significant crash reductions for all crash types, with the exception of sideswipe-opposite-direction crashes in Kentucky, for which the CMF of 0.891 is not statistically significant.

For Pennsylvania, the results are much different. The CMFs estimated for Pennsylvania are all very close to 1.0, ranging from 0.920 to 1.021-none of which are statistically significant. These results are initially surprising. They differ from the findings for Kentucky and Missouri, and results in NCHRP Report 641 for two-lane roads in Pennsylvania indicated large crash reductions for some crash types. For SRS, the NCHRP Report 641 reports CMFs of 0.76 for total crashes and 0.56 for SVROR crashes, which were both statistically significant at the 95-percent confidence level. For CLRS, a non-significant CMF of 0.74 was estimated for head-on plus sideswipe-opposite-direction crashes. Anecdotally, this may be explained by the fact that Pennsylvania has been installing rumble strips on two-lane roads for many years with a goal of blanket coverage of their two-lane rural road system. Given this fact, it is likely that most higher-crash locations have already been prioritized and treated and that the sites that were evaluated in the current study did not have a high target crash issue and so logically will not have exhibited a large safety benefit compared with those that did and were evaluated for NCHRP Report 641. PennDOT indicated that sites are selected to prioritize high-volume locations and those with a high run-off-road or head-on crash frequency. A comparison of the summary statistics for the Pennsylvania data in NCHRP Report 641 and the current study support this hypothesis. Table 20 shows the crash rates per mile-year before treatment and the proportion of total crashes for both the current and previous study. For run-off-road crashes, the crash rate of 0.87 in NCHRP Report 641 is much higher in than the rate of 0.16 for data used in the current study. Similarly the NCHRP Report 641 crash rate of 0.31 for head-on plus sideswipe-opposite-direction is much higher than the rate of 0.08 (0.05 for head-on and 0.03 for sideswipe-opposite-direction) for data used in the current study.

Table 20. Comparison of Pennsylvania crash rates.

The different results in the present study for Pennsylvania compared with Missouri and Kentucky illustrate how the extent of the target crash problem at a location will affect the crash reduction benefits that can be expected. The before period crash rates in table 10 show that run-off-road and sideswipe-opposite-direction crash rates were higher in Missouri and Kentucky than in Pennsylvania, and the head-on crash rate in Kentucky was higher than in Pennsylvania.

The disaggregate analysis sought to identify those conditions under which the treatment is most effective. Since run-off-road, head-on, and sideswipe-opposite-direction crashes are the focus of this treatment, these crash types are the focus of the disaggregate analysis. Several variables were identified as being of interest and available for all three States, including speed limit, shoulder width, lane width, AADT, and the expected crash frequency per mile prior to treatment.

The analysis found no clear trend between the CMF and values for posted speed, lane width, or shoulder width.

For AADT, as shown in table 21, larger percentage crash reductions were found for run-off-road crashes for higher AADTs with some stability reached at an AADT of approximately 3,200. At AADTs above 3,200, the estimated CMF does not change significantly. At AADTs lower than 3,200, a run-off-road crash CMF of 0.851 is estimated versus 0.702 for AADTs at 3,200 or greater. For head-on+sideswipe-opposite-direction crashes, the stability in the CMF is reached at an AADT of approximately 9,200, and the trend is reversed with a CMF of 0.679 at AADTs under 9,200 and 0.817 for AADTs over 9,200. A possible explanation for a larger CMF value for head-on+sideswipe-opposite-direction crashes is that at higher AADTs, there are fewer passing opportunities, and not all head-on or sideswipe-opposite-direction crashes are due to vehicles drifting out of their lane.

For the expected crash frequency per mile-year without treatment as shown in table 21, larger percentage crash reductions were found for run-off-road crashes for higher crash frequencies with some stability reached at a crash rate of approximately 0.500/mi-year. At rates lower than 0.500, a run-off-road crash CMF of 0.840 is estimated versus 0.621 for rates at 0.500 or greater. For head-on+sideswipe-opposite-direction crashes, the stability in the CMF is reached at approximately a rate of 0.065, and the trend is reversed with a CMF of 0.608 at rates under 0.065 and 0.715 for rates over 0.065. Since expected crashes increase with volume as seen in the SPFs developed, the trend of a larger CMF at higher crash rate for head-on+sideswipe-opposite-direction crashes would be expected, given the results for AADT.

Caution should be used in interpreting and applying these disaggregate results because they are not robust enough to develop CMFunctions. A CMFunction is an equation that would allow the estimation of CMFs for different levels of AADT and expected crash frequency. However, they may be used in prioritizing treatment sites. For example, sites with a high proportion of run-off-road crashes and high AADTs will have higher priority than sites with high AADTs and a high proportion of head-on+sideswipe-opposite-direction crashes.

Table 21. Results disaggregated by ranges of AADT and expected crash frequency.

AADT = Average annual daily traffic.
CMF = Crash modification factor.