Delta Region Transportation Development Program: Rural Safety Innovation Program Evaluation – Final Report

Chapter 3. Simpler Quantitative Evaluation

A simpler quantitative evaluation of the safety effectiveness of countermeasures was performed for one RSIP project. The specific countermeasure that was evaluated included:

• RSIP Project 33 – AHTD: Cable median barrier

A description of the RSIP project is provided below, followed by the results of the safety evaluation, including, descriptive statistics, analysis approach, and analysis results.

RSIP Project 33

Agency: Arkansas State Highway and Transportation Department (AHTD)

Focus of Evaluation: Cable Median Barrier

Project Background

AHTD received funding through the RSIP to extend an existing median cable barrier along I-55 from near Jericho to the Highway 63 Interchange near Lake David, approximately an 11-mi section of rural interstate. This section of highway was selected for improvement based upon a high frequency of crossover crashes. Prior to the RSIP project, AHTD had previously installed median cable barrier along an adjacent section of I-55 and along a section of I-40.

The roadway characteristics of the highway section where the cable median barrier was installed are as follows:

• Number of lanes: 4 (2 per direction)
• Lane width: 12 ft
• Paved inside shoulders: 4-6 ft
• Median type: grass median
• Average median width: 25 ft
• Speed limit: 70 mph

Shoulder rumble strips are present on both the inside and outside shoulders of this section of interstate.

The type of work involved in this project included:

• Earthwork (involved little change to the median cross slope, if any)
• Modifying drop inlets to provide adequate safety slopes
• Paving ditches
• Installing cable system (in the middle of the median)

Construction dates for installing the median cable barrier were from April 2009 through March 2010.

Because the cable median barrier installed as part of the RSIP project covered a limited number of miles and because AHTD had recently installed cable median barrier along two similar sections of rural interstates, it was decided to include in the analysis the safety experience from all three sections of rural interstates where AHTD had installed cable median barrier. The approximate locations of the cable median barrier and the installation dates are as follows:

• I-55, Section 11, log miles 12.62 to 23.87 (installed in 2009 through 2010 using RSIP
• funds)
• I-55, Section 11, log miles 8.75-12.62 (installed 2007)
• I-40 Sections 42 /43, log miles 204.5-213.0 (installed 2005)

Objective

The objective of this evaluation was to estimate the safety effectiveness of installing cable median barrier on rural interstates.

Descriptive Statistics

Three sections of rural interstates, covering approximately 23.6 miles, where cable median barrier was installed are included in the analysis.

Crash and traffic volume data were obtained for years 2004 through 2011 for the analysis. The number of before and after years varied among the three sites, and for two of the sites, only one year of data was available in one of the periods.

All crash types were considered in the analysis. Although cable median barrier would be expected to reduce cross median crashes and potentially increase SVROR crashes to the left, these crash types represented such a small percentage of the total crashes along such a short section of highway, that it was not reasonable to consider such crashes types in the analysis (e.g., 9 head-on crashes occurred in the before period and 10 in the after period, 5 sideswipe opposite direction crashes occurred in the before period and 10 in the after period, no distinction could be made between SVROR to the left vs. to the right). Therefore, the analysis focused on all crash types combined. Three crash severity levels were considered in the evaluation: total, FI, and FS crashes. Table 30 provides the years included in the before period and a summary of crash counts and crash rates [crashes/million vehicle miles traveled (MVMT)] during the before period for each of the treatment sites included in the analysis. Similarly, Table 31 provides the years included in the after period and a summary of crash counts and crash rates (crashes/MVMT) during the after period for each of the treatment sites included in the analysis.

Table 30. RSIP Project 33: Summary Statistics for the Before Period for Treatment Sites in Arkansas

Table 31. RSIP Project 33: Summary Statistics for the After Period for Treatment Sites in Arkansas

Analysis Approach

The analysis approach for this RSIP project differs from that for the three previous ones because the same sites were considered both before and after treatment installation, in other words, each site served as its own control. It also differs in that yearly crash counts were analyzed rather than crash counts summed over a given period, that is, site-years rather than sites were the basis for analysis. This approach thus required a special treatment of the data in the modeling approach: each "after site" was paired with its "before site" and the temporal correlation in crash frequencies at a given site from year to year was taken into account. The safety analysis of this treatment therefore consisted of a simple paired before/after comparison where each site served as its own control. A generalized linear model (GLM) with a negative binomial distribution and a log link was used to model the yearly crash rates (crashes/MVMT). A repeated measures correlation structure was included to account for the relationship in crash rates at a given site across years (temporal correlation). A compound symmetry covariance structure was used. General estimating equations (GEE) within PROC GENMOD in SAS were used to estimate the final regression parameters. For the three sites combined, a total of 20 site-years were considered for analysis.

The model for each crash type has the following form:

Figure 9. Equation 9 – General model to calculate predicted crashes.

where:

Analysis Results

The regression results are shown in Table 32 and include the following statistics for the three severity levels:

• The number of sites and site-years in Columns 2 and 3
• The estimates of a and b, their standard error, and associated p-value in Columns 5-7
• The percent change in crash rates due to treatment installation (i.e., before to after change): estimate and lower and upper 95-percent confidence limits in Columns 8-10
• The type 3 p-value associated with the treatment effect in Column 11

Table 32. RSIP Project 33: Regression Results and Treatment Effects by Crash Severity

Note: an empty cell indicates that the statistic is not applicable.

Interpretation of Results

The negative percent change in crash rate (Column 8) indicates that the crash rate decreased due to the installation of the cable median barrier for all three severity levels (conversely, a positive change would indicate an increase in crash rate). The 95-percent confidence limits of the percent change provide an assessment of whether the change, positive or negative, is statistically significant at the 95-percent confidence level: if the interval contains zero, then the change is not statistically significant (i.e., not different from zero) at the 95-percent confidence level; if the interval does not contain zero, then the change is statistically significant (i.e., different from zero) at the 95-percent confidence level.

The Type 3 p-values in the last column of Table 32 also provide an indication of whether cable median barriers have a significant effect on crash rates. These p-values correspond to the score statistics produced in the Type 3 GEE analysis and are generally more conservative than the p- values associated with the computation of the 95-percent confidence limits which are computed with the Wald statistic. Generally, these two p-values are in agreement with each other; however, when the two disagree, the Type 3 p-value should be the one on which to base conclusions.

For total crash rates, the Type 3 p-value (0.59) is close to the p-value associated with the confidence limits (0.54), both indicating that the percent change in total crash rates is not statically significantly different from zero. For FI and FS crash rates, however, the discrepancy between the two types of p-values is large; in both cases, the p-values associated with the confidence limits (0.0032 and 0.0012, respectively) indicate statistical significant at the 95- percent confidence level while the type 3 p-values (0.31 and 0.12) show no statistical significance at the 95-percent confidence level. Note, however, that the type 3 p-value (0.12) associated with the percent change in FS crash rate would indicate statistical significance at approximately 90-percent level.

In summary, the analysis showed the following:

• The installation of cable median barriers had a beneficial effect on total, FI and FS crash rates based on the negative percent changes in these crash rates: -13 percent for total, -53 percent for FI, and -49 percent for FS crash rates.
• There is insufficient evidence, however, to conclude that the treatment had a statistically significant effect on crash rates at the 95-percent confidence level.
• The installation of cable median barrier had a marginally significant effect in reducing FS crashes; the reduction is -49 percent and is statistically significant at the 88-percent confidence level.