• Federal Highway Administration >
• Publications >
• Research Publications >
• 11024 >
• 002.Cfm >
• Safety Evaluation of the Safety Edge Treatment

Safety Evaluation of the Safety Edge Treatment

Chapter 2. Project Database

Evaluation of the safety edge treatment required data on roadway geometrics, traffic volumes, crashes, construction costs, and implementation projects for sites where the safety edge treatment was implemented and for other similar sites. This chapter describes the selection of sites and assembly of the project database.

2.1 Participating States and Site Selection

Three States agreed to implement the safety edge treatment and to participate in the study: Georgia, Indiana, and New York. Colorado also agreed to participate in the study, but no sites were resurfaced with the safety edge treatment in time for inclusion in the analysis. Sites for the study were selected with the assistance of the participating State highway agencies. However, the site selection approach varied for three types of study sites: sites that were resurfaced and treated with the safety edge (treatment sites); sites that were resurfaced but not treated with the safety edge (comparison sites); and sites that were similar to the treatment and comparison sites but were not resurfaced (reference sites).

Treatment sites were selected by the three participating States from among the sites considered for their normal resurfacing program for 2005. In Indiana and New York, the sites that received the safety edge treatment were selected by the State as representative resurfacing projects for which the safety edge treatment would be appropriate. In Georgia, the transportation department made a policy decision to include the safety edge treatment in all resurfacing projects let to contract in April 2005 or thereafter. The treatment sites for this evaluation were drawn from among the projects let to contract after that date.

Most of the sites selected by the State highway agencies were used in this evaluation. A few sites that were distinctly different from the remainder of the study sites were dropped from the evaluation. Based on a preliminary review of the available treated projects in Georgia, Indiana, and New York, the decision was made to focus the analysis on the following three types of roadway segments:

• Rural multilane roadways with paved shoulders with widths of 4 ft or less.
• Rural two-lane roadways with paved shoulders with widths of 4 ft or less.
• Rural two-lane roadways with no paved shoulders (i.e., unpaved shoulders only).

Comparison sites were selected from among projects that were resurfaced in 2005 but did not receive the safety edge treatment. In Georgia, the comparison sites were resurfacing projects that were let to contract prior to April 2005, the date on which the Georgia Department of Transportation began implementing the safety edge treatment in all resurfacing projects. The comparison sites were selected to include the same roadway types as the treatment sites. The comparison sites were located in the same highway districts as the treatment sites so they were in the same geographical area.

Reference sites in each participating State included sites that had not been resurfaced during the period before resurfacing of the treatment and comparison sites and were not expected to be resurfaced during the entire 3-year study period. The reference sites included the same roadway types as the treatment and comparison sites. The total length of reference sites selected in each State was at least the same length as the treated sites in the State and often larger. Reference sites were chosen from the same highway districts as the treatment sites so they were in the same geographical area. Input from district engineers was sought to ensure that the reference sites were similar to the treatment sites. No reference sites were selected in New York because the reference sites were needed only for the before-after EB evaluation and it appeared unlikely that an EB evaluation could be conducted for the limited set of treatment sites available in New York. The New York data were included in other evaluations without the need for reference sites.

Each resurfacing project was divided into smaller roadway segments as needed based on a review of site characteristics and traffic volumes to assure that each site was relatively homogenous with respect to lane width, shoulder type and width, and traffic volume. The project database included 415 sites: 261 in Georgia, 148 in Indiana, and 6 in New York. The individual sites ranged in length from 0.1 to 25.8 mi. The total length of all segments considered in the study was 685 mi in Georgia, 514 mi in Indiana, and 25 mi in New York. Table 1 summarizes the number of sites by State, roadway type, shoulder type, and site type.

Table 1. Summary of number and total length of sites.

T = Treatment sites resurfaced with safety edge.

C = Comparison sites resurfaced without safety edge.

R = Reference sites not resurfaced.

Table 1 shows that the project database included 90 treatment sites with a total length of 211 mi, with 57 treatment sites in Georgia, 30 treatment sites in Indiana, and 3 treatment sites in New York. The project database also includes 85 comparison sites with a total length of 240 mi and 240 reference sites with a total length of 774 mi.

2.2 Data Collection

A substantial amount of data was collected and assembled into a database for consideration in the analysis phase of the study. Data were collected for the period before resurfacing of the treatment and comparison sites and for 3 years after resurfacing. Information concerning data availability, data collection procedures, and contents is presented below for the following data types:

• Project locations and roadway characteristics.
• Crashes.
• Traffic volumes.
• Field measurements of pavement-edge drop-offs.

2.2.1 Project Locations and Roadway Characteristics

For each treatment, comparison, and reference site, the project database included the following data elements: location on the agency's highway system, project construction dates, and basic roadway characteristics. The basic roadway characteristics obtained included road type, lane width, and shoulder type and width. These data were obtained from State highway databases or published reports. All State data were verified and supplemented by field visits to the sites.

Analysis units for the study (i.e., study sites) were created by subdividing resurfacing projects into sections that were generally homogeneous with respect to roadway geometrics. The roadway characteristics used to define the site boundaries were monitored for changes other than resurfacing.

2.2.2 Crashes

The crash database for the study included all non-intersection crashes that occurred within the limits of each site during the study period. Crash data, provided by the participating agencies from their electronic crash record databases, contained sufficient summary information to identify the target crash types most likely to be affected by provision of the safety edge.

Where possible, it was desirable to limit the evaluation to specific target crash types that were most likely affected by the implementation of the safety edge. If the crash data for both the before and after periods included crash types that could not conceivably be affected by the safety edge treatment, then this "noise" could introduce unnecessary variability into the crash counts and mask the safety effect of the treatment. For example, the installation of the safety edge treatment is likely to have a greater effect on run-off-road crashes than on rear-end crashes. By limiting the analysis to include only run-off-road crashes, the likelihood of finding statistically significant effects may be improved. However, the more restrictive the crash type definition used, the smaller the crash counts available for analysis, making it more difficult to find statistically significant effects. Because of this tradeoff between the relevance of the target crash type to the treatment being evaluated and the number of crashes available for analysis, a range of target crash type definitions from more inclusive and less relevant to less inclusive and more relevant was considered.

The selection of the target crash types to be evaluated was guided by two recent studies of crashes related to pavement-edge drop-offs by Council and Hallmark et al.⁽¹⁾ These studies identified five scenarios (crash sequences) in which over-steering may result in a crash related to a pavement-edge drop-off. This report assumes that only these types of crashes and no others would be affected by provision of the safety edge.

The five types of crashes used to identify potential drop-off-related crashes are as follows:

• Head-on collision with an oncoming vehicle.
• Sideswipe collision with an oncoming vehicle.
• Run-off-road crash on the opposite side of the road.
• Overturning within the traveled way or on the opposite side of the road.
• Same-direction sideswipe collision on multilane roads.

Head-on crashes may involve a vehicle that crossed the centerline without first running off the road. Such head-on crashes were not classified as drop-off-related nor treated as target crashes.

The target crash types described represent potential drop-off-related crashes, defined as precisely as possible without obtaining and reviewing individual police crash forms. Past research by Council, which included a detailed analysis of hard-copy reports, indicated that a larger percentage of potential crashes were judged as probable or possible drop-off crashes when the officer had noted a shoulder defect. Therefore, if the agency's crash form had an item for "low shoulder" or "shoulder defect," then this item was used to identify potential drop-off-related crashes.

This methodology represents a narrow interpretation of drop-off-related crashes. Therefore, it was also recommended that crashes that showed evidence of a vehicle leaving the road and run-off-road crashes be included, such as the following:

• Run-off-road right, cross centerline/median, hit vehicle traveling in the opposite direction (head-on or sideswipe).
• Run-off-road right, sideswipe with vehicle in same direction (multilane roads).
• Run-off-road right, rollover (in road or on roadside).
• Run-off-road right, then run-off-road left.
• Single vehicle run-off-road right.

Selection of the crash types was based on descriptors in the crash database furnished by the participating States. The data fields used included sequence of events, location of first harmful event, type of collision, driver, and roadway contributing circumstances. The specific fields used to identify drop-off-related crashes in this study for each participating State are described in appendix A.

Crash severity levels considered in the evaluation are as follows:

• Fatal, injury, and property-damage-only (PDO) crashes (i.e., all crash severity levels combined).
• Fatal and injury crashes.
• PDO crashes.

The highest priority in assessment of the safety edge treatment is the evaluation of its effect on fatal and injury crashes because these categories include the most severe crashes among the target crash types of interest. Crashes of all severity levels (i.e., including PDO crashes) were considered because the larger crash sample size made it easier to detect statistically significant effects. It would have been more desirable to consider only PDO crashes that were severe enough for at least one vehicle to be towed from the crash scene since PDO tow-away crashes are more consistently reported than other PDO crashes. However, this exclusion was not applied because only one of the participating States (Indiana) identified tow-away crashes in its data.

Table 2 and table 3 summarize the crash data for total and fatal and injury crashes, respectively, including the breakdown of total, run-off-the-road, and drop-off-related crashes for each State, roadway type, shoulder type, and site type .

Indiana was able to provide only reference-point (i.e., milepost) information and latitude and longitude information for some of the crashes. Additionally, some of the reference-point information provided with the crashes indicated that the crashes occurred on side roads at intersections. Approximately 40 percent of the crashes had wrong or missing reference point or coordinate information but contained a verbal description of the crash. Extensive efforts to better locate these crashes were undertaken during the execution of the work plan.

Table 2. Summary of total non-intersection crash data for study sites.

¹ Does not include at-intersection or intersection-related crashes.

T = Treatment sites resurfaced with safety edge.

C = Comparison sites resurfaced without safety edge.

R = Reference sites not resurfaced.

Table 3. Summary of fatal and injury non-intersection crash data for study sites.

¹ Does not include at-intersection or intersection-related crashes.

T = Treatment sites resurfaced with safety edge.

C = Comparison sites resurfaced without safety edge.

R = Reference sites not resurfaced.

2.2.3 Traffic Volumes

Annual average daily traffic (AADT) volume data for all study locations were obtained through agency databases or published sources from each of the participating agencies, so no field traffic counts were required as part of the database development. When possible, separate AADT values for each year of the study period were obtained. When AADT values were not available for all years of the study period, values were interpolated or extrapolated for the missing years.

Table 4 summarizes the traffic volume data assembled for the project database. Ideally, the AADT ranges should be as similar as possible for the various site types within each State/road type/shoulder type combination. In particular, it was desirable for reference sites to cover the entire range of values of the treatment and comparison sites, as SPF performance outside the range of the reference sites is not optimum. It was also desirable that the comparison and reference sites have nearly identical ranges. The AADT ranges were found to be similar for most cases except for multilane highway sites with paved shoulders in Georgia. For these sites, the AADT ranges were higher for treatment sites than for comparison or reference sites. To a lesser extent, the same is true for two-lane highway sites with paved shoulders in Indiana.

Table 4. Summary of traffic volume data for study sites.

T = Treatment sites resurfaced with safety edge.

C = Comparison sites resurfaced without safety edge.

R = Reference sites not resurfaced.

2.2.4 Lane Width

Lane widths ranged from 9 to 13 ft across all sites and States, with the majority of lanes being 12-ft wide. The distribution of lane width is summarized in table 5 by State and site type. The variability in lane width was most evident for the unpaved shoulder type, so it was decided to include this variable in modeling efforts for these sites.

Table 5. Summary of lane widths for study sites.

T = Treatment sites resurfaced with safety edge.

C = Comparison sites resurfaced without safety edge.

R = Reference sites not resurfaced.

2.2.5 Field Drop-Off Measurements

Field visits were made to each treatment and comparison site to collect pavement-edge drop-off measurements and additional geometric design variables. Field measurements of pavement-edge drop-offs were made before resurfacing and during each of the 3 years after resurfacing. However, some of the project sites were resurfaced before field visits could be made, which prevented supplemental data collection before resurfacing at some sites. Drop-off height was measured 4 inches from the pavement edge for all sites. The types of data collected and the methodology for collecting these data are documented in appendix B.