Safety Evaluation of Edge-Line Rumble Stripes on Rural Two-Lane Horizontal Curves

Chapter 7. Before–After Evaluation Results

This chapter presents the results of the before–after evaluation, including aggregate analysis for both Kentucky and Ohio and disaggregate analysis of the Ohio data. Disaggregate analysis of the Kentucky data was not conducted because the sample size was too small.

Aggregate Analysis

Table 22 provides the estimates of expected crashes in the after period without treatment, the observed crashes in the after period, and the estimated CMF and its SE for all crash types considered in Kentucky. Table 23 presents the results for Ohio.

*Statistically significant results at the 95-percent confidence level.

*Statistically significant results at the 95-percent confidence level.

The results for Kentucky indicated statistically significant reductions for all crash types except nighttime ROR crashes at the 95-percent confidence level. Nighttime crashes had the smallest CMF (which translates to the greatest reduction) with a value of 0.63. Total, injury, and ROR crashes had CMFs of 0.75, 0.64, and 0.74, respectively. The CMF for nighttime ROR crashes was 0.75 and was consistent with the same CMF from Ohio; however, it was significant only at the 80-percent level, suggesting that sample size was the reason for the lack of statistical significance at the 95-percent confidence level. The CMFs were smaller than—but consistent with—those found in the most comprehensive and reliable study of SRSs to date.⁽¹⁰⁾ Based on a before–after EB analysis, the project team found that milled SRSs had a crash reduction of 16 percent (SE = 8) for all SVROR crashes and 36 percent (SE = 10) for SVROR injury crashes.⁽¹⁰⁾ However, the analysis results for SRSs in Torbic et al. considered segments with both horizontal tangents and curves; therefore, a direct comparison of results cannot be made.⁽¹⁰⁾ It is also important to remember that all crash types considered in this research excluded intersection-related and animal crashes.

The results for Ohio indicated statistically significant reductions for all crash types. Nighttime ROR crashes had the smallest CMF with a value of 0.71. Total, injury, ROR, and nighttime crashes had CMFs of 0.79, 0.79, 0.78, and 0.75, respectively. As with the Kentucky results, the CMFs were smaller than but consistent with those found in Torbic et al.⁽¹⁰⁾ The resulting Ohio installation CMFs reflected the installation of ELRSs on horizontal curves as well as the impact of the statewide signing program.

A subset of both treatment and reference curves received sign upgrades, including chevrons, curve ahead signs, and speed advisory signs, all of which target crash types (i.e., nighttime and ROR) similar to those targeted by ELRSs but through a different mechanism (i.e., rumble strips target distracted or drowsy drivers through a haptic alert). The initial set of reference sites accounted for the impact of the signing upgrades. Additional analyses of the reference sites indicated a spillover effect of the horizontal curve signing program on curves that did not receive treatments as well as shorter tangents; therefore, longer tangent segments were used to determine the expected trend in the after period had no treatment (i.e., the signing program or ELRS installation) occurred. Owing to the spillover effects of the signing program, further analyses involving curves that specifically received new or additional signs were not fruitful (i.e., the resulting CMFs could not separate the effects of the signing program from those resulting from ELRS installation).

Disaggregate Analysis of Ohio Data

The disaggregate analysis sought to identify those conditions under which the treatment was most effective. Because ROR, nighttime, and nighttime ROR crashes were the focus of this treatment, the project team focused on these crash types for the disaggregate analysis. In addition, disaggregate results are presented for total crashes and fatal and injury crashes. The data sample for Kentucky was too small to perform disaggregate analyses; therefore, disaggregate analyses focused only on Ohio data.

Several variables were identified as being of interest and available for both States, including degree of curve, posted speed limit, paved shoulder width, lane width, AADT, and before-period expected crash frequency. Disaggregate results are provided by AADT in table 24 and before-period expected crash frequency in table 25. The number of crashes in the after period is presented for each CMF to indicate the sample size available. Several of the estimated CMFs rely on small samples, especially for nighttime crashes and nighttime ROR crashes.

*Statistically significant results at the 95-percent confidence level.

*Statistically significant results at the 95-percent confidence level.

As shown in table 24, smaller CMFs (i.e., larger safety benefits) were found for all crash types for sites with an AADT of 4,000 or more vehicles per day; however, the 95-percent confidence intervals overlap for each crash type. At AADTs lower than 4,000 vehicles per day, for example, an ROR crash CMF of 0.82 was estimated versus a CMF estimate of 0.64 for AADTs of 4,000 vehicles per day or greater. A similar difference was found for all other crash types. The 4,000 vehicles per day AADT cutoff is consistent with previous research by Patel et al. and Lyon et al.^(19,30)

For the before-period expected crash frequency, as shown in table 25, the project team found larger safety benefits for all crash types for higher before-period expected crash frequency. The 95-percent confidence intervals did not overlap for total crashes and ROR crashes. Owing to the differences in the frequencies of different crash types, the before-period expected crash frequency cutoff varied for each crash type. For example, an ROR crash CMF of 1.13 was estimated for horizontal curves with an ROR before-period expected crash frequency of less than 0.20 crashes/yr. This can be compared with a CMF of 0.66 for horizontal curves with 0.20 or more before-period expected crashes/year. Note that the CMF of 1.13 for an ROR before-period expected crash rate less than 0.20 is not statistically significant. Similar results were found for all other crash types.

Caution should be used in interpreting and applying these disaggregate CMF results because of correlation among variables and because they were not robust enough to develop crash modification functions. A crash modification function is a formula used to compute the CMF for a specific site as a function of its site-specific characteristics. For example, crash modification functions would allow the estimation of CMFs for different levels of AADT and before-period crash frequency. In addition, the disaggregate analysis results used the EB analysis data, which include the effects of the statewide horizontal curve signing program. However, the disaggregate analysis CMF results can be used to inform the process of prioritizing treatment sites for ELRSs. For example, sites with a high proportion of ROR crashes and high AADTs could have high priority for receiving this treatment because those are the sites likely to benefit the most.