Run-Off-Road Collisions

Objectives

Strategies

15.1 A—Keep vehicles from encroaching on the roadside

15.1 A1—Install shoulder rumble strips (T)

15.1 A2—Install edgeline "profile marking," edgeline rumble strips or modified shoulder rumble strips on section with narrow or no paved shoulders (E)

15.1 A3—Install midlane rumble strips (E)

15.1 A4—Provide enhanced shoulder or in-lane delineation and marking for sharp curves (P/T/E)

15.1 A5—Provide improved highway geometry for horizontal curves (P)

15.1 A6—Provide enhanced pavement markings (T)

15.1 A7—Provide skid-resistant pavement surfaces

15.1 A8—Apply shoulder treatments

• Eliminate shoulder drop-offs (E)

• Widen and/or pave shoulders (P)

15.1 B—Minimize the likelihood of crashing into an object or overturning if the vehicle travels beyond the edge of the shoulder

15.1 B1—Design safer slopes and ditches to prevent rollovers (see "Improving Roadsides," page V-36) (P)

15.1 B2—Remove/relocate objects in hazardous locations (see "Improving Roadsides," page V-36) (P)

15.1 B3—Delineate trees or utility poles with retroreflective tape (E)

15.1 C—Reduce the severity of the crash

15.1 C1—Improve design of roadside hardware (e.g., bridge rails) (see "Improving Roadsides," page V-36) (T)

15.1 C2—Improve design and application of barrier and attenuation systems (see "Improving Roadsides," page V-36) (T)

Note: The following page explains (T), (E), and (P) demarcations.

Technical Attributes

Target

Drivers of errant vehicles, using sound and sensation to directly alert the individual of encroachment or pending encroachment.

Expected Effectiveness

On freeways, shoulder rumble strips have proven to be a very effective way to warn drivers that they are leaving or are about to leave the road. According to FHWA, several studies have estimated that rumble strips can reduce the rate of ROR crashes by 20 to 50 percent. Further statistics regarding effectiveness for specific programs are documented below. However, these crash reduction statistics apply to freeways.

While this strategy is currently implemented on nonfreeways by a number of jurisdictions, there is little information on the safety effectiveness of shoulder rumble strips on these roads. Further evaluation is clearly needed. Care should be taken in extrapolating freeway application experience to the two-lane highway system. On one hand, the rumble strips could be less effective since freeway design provides the errant driver with a wider clear zone in which to recover after hitting the strip. On many two-lane roads, the clear zone—often just a shoulder—is much more limited. In such cases, the driver has little opportunity to recover even when given a warning. However, rumble strips could be more effective on two-lane roads for basically the same reason: since two-lane roads have much less clear zone and much more hazardous roadsides (less breakaway objects, more severe sideslopes, objects closer to roadway), a higher proportion of excursions from the travel lane may become crashes. Moreover, the quality of the roadway alignment is generally worse on two-lane versus freeway facilities, and hence the need for such warning to keep drivers on the road is greater. Similarly, most freeways commonly include full 12-foot lanes, while there are many high-speed two-lane rural highways with lane widths as narrow as 10 feet. Thus, if the shoulder rumble strips are effective, they could prevent more crashes per excursion. While it is not possible to determine which set of assumptions is correct, shoulder rumble strips should produce measurable benefits somewhat consistent with those demonstrated in studies for freeways. In the absence of such information, the following studies provide effectiveness estimates for shoulder rumble strips on freeways and expressways.

The New York State Thruway Authority (NYSTA) installed continuous milled-in shoulder rumble strips on all four shoulders of 485 roadway miles of thruway between 1992 and 1993. In its before/after evaluation, NYSTA used accident data provided by the state police assigned specifically to the toll road system. One year of before data (1991) and 1 year of after data (1997) were used for the study (Exhibit V-4). Only single-vehicle ROR crashes with certain "causes" were selected for the study because "it was believed that these specific run-off-road crashes were indicative of those that could be mitigated by the use of continuous shoulder rumble strips and correcting the driver's behavior" (Perrillo, 1998). These causes included use of alcohol or drugs, driver inattention or inexperience, fatigue, illness, passenger distraction, and glare. Exhibit V-4 shows the reduction of crashes observed from 1991 to 1997.

In a companion study by the New York DOT of 300 miles of additional nonthruway mileage, the reduction of ROR crashes, resulting from driver inattention, fatigue, and drowsiness, is reported to be 65 percent with the installation of milled-in shoulder rumble strips (New York State DOT, 1998). The initial study also developed benefit-cost ratios for the rumble-strip installation program. The cost of installation was $3,995 per roadway mile for continuous rumble strips on all four paved shoulders. Hence, the total cost of installation for 485 roadway miles was more than $1.9 million. Using the cost of highway crashes as defined by the FHWA and assuming a yearly accident savings as summarized in Exhibit V-4, the total accident savings per year is $58.9 million. Assuming that the shoulder rumble strips have a maintenance-free lifespan of 6 years and that the yearly accident savings is as calculated by comparing 1991 data and 1997 data, the benefit-cost ratio equaled 186. Such a high benefit-cost ratio indicated shoulder rumble strips to be an extremely beneficial treatment.

In a recent study, the FHWA used data extracted from the Highway Safety Information System (HSIS) to study continuous rolled-in shoulder rumble strips installed on 284 miles of rural and urban freeway in Illinois and 122 miles in California. Where possible, the author used two different before/after methodologies, one involving "yoked" or paired comparison sites and one involving a nonpaired comparison group. In contrast with the more restricted group of accident types in the New York Thruway study, all single-vehicle ROR crashes were studied here. The Illinois data indicated an 18.3-percent reduction in single-vehicle ROR crashes on all freeways combined and a 13-percent reduction in single-vehicle ROR injury crashes. Both reductions were statistically significant. Comparable reductions on Illinois rural freeways were 21.1 percent for single-vehicle ROR crashes and 7.3 percent for injury crashes. California data for the combined urban and rural freeways indicated a 7.3-percent reduction in single-vehicle ROR crashes, but the finding was not statistically significant.

It is difficult to specify a crash reduction factor for shoulder rumblestrips on rural two-lane roads. There have been no effectiveness studies on such roads, and the effect could be logically hypothesized to be either less than or greater than on freeways. There are also differences in the estimated effects on freeways, with crash decreases ranging from 7 percent of total single-vehicle crashes to 90 percent of single-vehicle crashes related to driver inattention or fatigue. Part of this wide range is the result of differing crash types being studied (i.e., the more selective the crash type, as in the New York studies, the greater the effect will be). Part may also stem from effectiveness differences between milled-in rumble strips (in the New York studies) and rolled-in strips (in the FHWA study). However, no study has been identified that specifically addresses this potential difference in effectiveness. A "best guess" at this time might be a 20- to 30-percent reduction in single-vehicle ROR crashes on rural freeways, with less effect on urban freeways. For the reasons cited above, it is difficult to define even a "best guess" for two-lane rural roads. With no specific study on these roads, one might assume a similar effect to that seen on rural freeways—a 20- to 30-percent reduction in single-vehicle ROR crashes.

Keys to Success

If the use of shoulder rumble strips on freeways continues to be as effective as studies indicate, states should readily adopt them on these roads. The key to increased installation on two-lane and other nonfreeway roads would appear to be further proof of effectiveness on these roads and resolution of incompatibility issues such as bicycle use, noise, etc. (See "Potential Difficulties" below.) The use of prototype studies is suggested to establish the validity of extending this strategy to nonfreeway facilities. It will also be important to identify appropriate road sections—sites where ROR crashes are a problem and continuous shoulder rumble strips can be installed.

Potential Difficulties

Incompatibilities may exist between shoulder rumble strips and bicycle use. Since the transportation community encourages increased bicycle use, this may become a serious issue. In a recent Draft Technical Advisory on Roadway Shoulder Rumble Strips, FHWA has noted its full support of AASHTO's position, as stated in the 1999 AASHTO Guide for the Development of Bicycle Facilities, that

Rumble strips or raised pavement markers . . . are not recommended where shoulders are used by bicyclists unless there is a minimum clear path of 0.3 m (1 foot) from the rumble strip to the traveled way, 1.2 m (4 feet) from the rumble strip to the outside edge of paved shoulder, or 1.5 m (5 feet) to adjacent guardrail, curb or other obstacle. (Draft Technical Advisory on Roadway Shoulder Rumble Strips)

In that same advisory, the FHWA describes current state efforts to develop bicycle-friendly rumble strip programs and stresses the need for states to regularly sweep shoulders to remove debris where rumble strips and bicycles coincide in order to allow the bicyclists to use the outer rather than inner part of the paved shoulder.

It is also noted that the Association of Pedestrian and Bicycle Professionals (APBP) has commented on these guidelines (see http://www.apbp.org). Key suggestions for locations with bike traffic include only using rumble strips on two-lane roads where there is a significant, demonstrated crash problem (rather than a systemwide approach), minimizing the depth of the cut to 3/8 inch, preferably retaining 8 feet of clear paved shoulder outside the rumble strip, installing the strip at or under the edgeline rather than leaving the 1-foot "no man's land" between the edgeline and rumble strip, using 12-inch-wide strips with gaps, and no installation of rumble strips where there will be 4 feet or less of clear paved shoulder after installation without "overwhelming justification" and without warning signs to bicyclists.

In its early use of rumble strips, Pennsylvania would only use raised (edgeline) rumble strips where there was at least 4 feet of paved shoulder in order to accommodate bicycle use. The state required a minimum of 4 feet of paved shoulder for shoulder rumble strips and preferred 6 to 8 feet. Because of these concerns, Pennsylvania has developed a design to make shoulder rumble strips "bicycle-tolerable." Working for the Pennsylvania DOT, the Pennsylvania Transportation Institute researched alternative designs to alert motorists without being disruptive to bicyclists. The resulting design, which is used on shoulders at least 6 feet wide, is a 3/8-inch-deep cut that is 5 inches wide with a 7-inch space between cuts. The rumble strips begin 6 inches off the edge of the pavement. The Transportation Institute also recommended a similar pattern, except with a 6-inch space between cuts for lower-speed roads. Research in Pennsylvania continues on an appropriate design for roadways with narrower shoulders (2 to 4 feet). (See Appendix 1 for detailed drawings.) Due to similar concerns, California DOT (Caltrans) tested the vibration, noise, and subjective comfort levels of 11 different rumble strip configurations using passenger cars, trucks, volunteer bicyclists and State Highway Patrol motorcyclists. Based upon a combination of results from the different tests, Caltrans adopted new standard rolled-in and milled-in rumble-strip designs for routes with bicycle usage. Where the shoulder is less than 5 feet wide, the policy allows for the use of raised/inverted profile thermoplastic traffic strips as the edgeline. See Exhibit V-5.

Note that a similar raised edgeline design was modified in Great Britain due to bicycle and motorcycle concerns. The raised ribs in the final design are approximately 1/4 inch high. Details can be found at http://www.roads.dft.gov.uk/roadnetwork/ditm/tal/signs/02_95/index.htm. Of course, discouraging bicycle use on roadways prone to ROR crashes may be the appropriate thing to do (or providing safer, separated bicycle facilities within the same general corridor). To the extent that shoulder rumble strips would be used in a site-specific versus systemwide basis, this apparent conflict may be manageable. At least one state noted that motorcyclists may not be able to recover as well from riding along a rumble strip as from a normal paved shoulder. However, testing by Caltrans involving a very small sample of four state highway patrol motorcyclists indicated that the motorcyclists had no problems traversing any of the designs tested.

Other potential pitfalls include complications with snow removal, shoulder maintenance requirements, and noise. With respect to adverse weather, ice and snow can collect in rumble strips. When the trapped water freezes, icy conditions may occur. However, the drainage designed for shoulders, as well as the speed, turbulence, and vibrations from passing vehicles, tends to knock the ice from the rumble strips. Continuous shoulder rumble strips also have proven to be an asset to truck drivers during inclement weather. The shoulder rumble strips aid in determining the edge of the roadway when low visibility makes it difficult to see painted roadway edges and markings. (Note, however, that North Carolina has found the raised/inverted profile edgelines do not tolerate snowplowing.)

With respect to maintenance, Pennsylvania has not noted any additional maintenance required for the rumble strips installed on interstates with shoulders in good condition. Neither Massachusetts nor New York has noted any degradation over the past 3 years. Indeed, in some user states, rumble strips have been shown to help snowplows find the edge of the travel lanes. While some states have expressed a concern that the installation of rumble strips might lead to pavement deterioration, the FHWA "Rumble Strip Community of Practice" Web page indicates that this does not occur with proper installation. Finally, with respect to degradation, Kansas is changing its rumble strip policy, which allowed rolled-in strips, to one requiring milled-in strips. This change is due to Kansas's observation that rolled-in strips have a tendency to "heal over" and reduce effectiveness over time.

There have been reports of noise complaints where shoulder rumble strips have been installed. New installations should acknowledge this concern and make provisions where necessary. Implementing a program of rumble strips systemwide should consider local sensitivities to maintain support for such a program.

Finally, there is not a crash-proven rumble strip design for two-lane roads without paved shoulders or with very narrow paved shoulders (e.g., 2 feet wide). This is a significant problem for some state agencies and many county and local agencies where most or all two-lane roads do not have paved shoulders. It is possible that the effectiveness of shoulder rumble strips may well be lessened from freeway experience, by poor or narrow shoulders that exist on many two-lane highways, so that even an "alerted" motorist might not be able to safely recover. However, given the numbers of such miles in the United States, there is clearly a need to test some potential designs. (See sections below concerning possible experimental strategies.)

Appropriate Measures and Data

Process measures of program effectiveness would include the number of miles of road or the number of hazardous locations where rumble strips are installed.

Impact measures include the number of ROR crashes reduced at these locations and the changes in total crashes. If possible, the impact measure should include potential "crash migration" (i.e., crashes occurring on downstream sections where rumble strips have not been applied, but where drowsy drivers may still be on the road) effects on adjacent roadways.

The advent of low-cost vehicle-sensing and recording devices might allow for the use of a surrogate measure based upon the number of encroachments onto the shoulder over a specific section of road (e.g., a curve). In addition to process and crash data, the agency should also collect information on acceptance by the public and by bicyclists and on any adverse noise problems for adjacent properties.

Associated Needs

There have been a few reports of people who mistook the sounds produced by the rumble strips as car trouble. A public information or education campaign, as well as standard installation, should eliminate such misinterpretations. However, current moves to standardized use on freeways may provide the most effective public training.

Organizational and Institutional Attributes

Organizational, Institutional, and Policy Issues

First, if the agency does not have a design policy for rumble strips that can be retrofitted to shoulders, one may need to be developed. Additionally, a policy regarding the types of two-lane road sections where placement is acceptable may be necessary. While many states have established specific design and placement policies for shoulder rumble strips on freeways and other access-controlled facilities, specific criteria for two-lane or other nonfreeway roads were much more limited. For example, Minnesota policy states that "Rumble strips can also be placed on the shoulders of two-lane roads at the discretion of the District." Since 1991, the Kansas DOT has had a policy requiring shoulder rumble strips to be included on all reconstruction or new construction projects with a full width (8- to 10-foot) shoulder. Such strips were also required if full-width shoulders were being overlaid with a minimum of 1 inch of asphalt. This policy primarily pertains to freeways and expressways since few two-lane rural roads have full-width shoulders. However, Kansas is installing the rumble strips on its "Super-Two" sections—sections with 12-foot lanes and full-width shoulders. Finally, rolled-in strips on asphalt pavements tend to deform over time, thus reducing the size of the cuts and lessening their effectiveness. Due to these problems with "healing" rolled-in strips, Kansas is now considering a revision of this policy, which would mandate milled-in rumble strips. Review of freeway-related policies from Connecticut, New Hampshire, New Jersey, Massachusetts, Maine, and Minnesota indicate that factors to be considered in such policies include bicycle accommodation/routes, minimum shoulder width where allowable, offset from edgeline, placement on or near bridge decks, use at intersections, speed limits, and other factors. Second, while this strategy is implemented by the state DOT, there is a clear need for the inclusion of bicycle transportation offices or groups to be involved early in the planning process for treatment of nonfreeways.

Issues Affecting Implementation Time

Shoulder rumble strip programs can be implemented quickly, certainly within a year of an agency deciding to proceed. They can be implemented as components of both new construction and rehabilitation projects.

Costs Involved

Due to increased installation and technological advances, the cost of continuous shoulder rumble strips has decreased over the years. For instance, in 1990, the New York DOT reported paying $6.18 per linear meter compared with $0.49 per linear meter in 1998. Specific cost of installation on the New York Thruway was reported to be $3,995 per roadway mile for rumble strips on all four shoulders. The cost includes milling in the rumble strips, sweeping and discarding excess asphalt, and maintaining and protecting traffic. The Pennsylvania DOT reports an average cost of $0.25 per foot or $2,640 per mile for the installation of milled-in rumble strips on the shoulders on both sides of two-lane roads. Incremental costs would be even less for rumble strips being implemented concurrently with reconstruction or resurfacing of a highway.

Training and Other Personnel Needs

There appear to be no special personnel needs for implementing this strategy. Either agency personnel or contractors could do the installation. The need for training will depend on whether the agency has been using retrofitted rumble strips on freeways or other roadways. If not, either agency personnel or contractor personnel will need to be trained in proper installation techniques.

Legislative Needs

There do not appear to be any special legislative needs.

Other Key Attributes

One benefit of shoulder rumble strips is that, unlike other safety measures whose effectiveness may decrease over time as their "novelty" wears off, rumble strips primarily affect only drowsy or other inattentive drivers. Concern has been expressed that if fatigue-related crashes are prevented on one section of roadway, the problem may be transferred to another section. While the FHWA attempted to examine this issue, no data have been found to support or dispel the theory. Such a possibility may be reduced by public education urging fatigued drivers (particularly those who ride over the rumble strips and recover their vehicle from running off the road) to stop and rest before continuing.

Technical Attributes

Target

Drivers of vehicles entering potentially hazardous curves.

Expected Effectiveness

At least limited evaluations of all three types of devices have been conducted. Based on these studies, well-placed shoulder delineators are a proven crash-reducing strategy, at least for roads with average or higher designs. The on-pavement treatments aimed at warning the driver or providing an increased sense of hazard have been evaluated in terms of speed reduction, but not crash reduction. The positive findings with respect to speed reductions would place these treatments in the "tried" category.

In a very well-designed early study of post-mounted delineators on rural two-lane curves, Foody and Taylor (1966) found them to reduce ROR crashes by 15 percent. In a more recent nonaccident study, the "curve following behavior" of drivers was studied before and after rural, two-lane curves were treated with different combinations of chevron signs, post-mounted delineation, and raised pavement markings. Vehicle speed and the placement of the vehicle in the lane were measured at 46 sites in Georgia and 5 in New Mexico. The results for nighttime hours show that vehicles moved away from the centerline when chevrons were used (i.e., closer to the edgeline) and even farther away when raised pavement markers were used. When post-mounted delineators were used, vehicle placement on right curves shifted toward the centerline (Zador et al., 1987).

Contrasting findings for raised reflector posts were found in a Swedish study by Kallberg (1993). (Note that this study was not restricted to posts on curves.) The author concluded that "reflector posts on narrow, curvy, and hilly roads can significantly increase driving speeds and accidents in darkness." Specifically, reflector posts increased accidents on roads with relatively low geometric standards and 50-mph posted speed limits. Although the specific effects of reflector posts on the lateral position remain unclear, it is clear that the shift in lateral position (if there is a significant shift) is toward the edge of the road. This before and after study with control sites was conducted on roadway segments in Finland. The counterintuitive findings are supported by the human factors concept of selective visual degradation. This theory explains that reflector posts do not improve the driver's ability to detect potential hazards but do improve the driver's ability for orientation tasks. This may reduce the frequency of ROR collisions, but it also may increase speeds and therefore increase the severity of those ROR crashes that do occur. With respect to warning messages placed on the pavement, the Insurance Institute for Highway Safety (IIHS) conducted a study for a single, very sharp curve (~90°) on a suburban two-lane secondary road in Northern Virginia with a posted speed limit of 35 mph (Retting and Farmer, 1998). The pavement marking consisted of the word "SLOW" in 8-foot-high white letters, a white 8-foot-high left curve arrow, and an 18-inch-wide white line perpendicular to the road at the beginning and end of the text/symbol. Results were based on before/after changes in mean speed, 90th-percentile speed, and percentage of vehicles exceeding 35 mph, 40 mph, and 45 mph, as compared with similar data from a nearby comparison curve that was not treated. The pavement marking was associated with a decrease in vehicle speed of 6 percent overall and 7 percent during daytime and late night periods.

The same pavement marking was used in a 1999 study at six sites in Pennsylvania (Retting, 1999). A before/after study of effects on vehicle speeds showed that these pavement markings had little effect on the average speed and the 85th-percentile speed. However, the 95th-percentile speed was reduced significantly. This year, the marking will be implemented at 200 sites statewide, and IIHS will again evaluate the effect.

Evaluations of markings on the pavement to slow drivers by heightening "apparent danger" have been conducted for a number of years both in the United States and internationally. In a 1979 study for the Ohio DOT, the effects of yellow-bar pavement markings installed perpendicular to the direction of travel were studied. There were "reported reductions in traffic speeds, most notably high speeds" resulting from the pavement markings installed prior to curves (Retting and Farmer, 1998). In a somewhat limited 1980 before/after study of one particularly hazardous curve on a rural two-lane road in Meade County, Kentucky, the treatment involved transverse lines of reflective tape in an ever-tightening pattern designed to slow a vehicle from 55 mph to 35 mph before entering the curve. The pattern consisted of 30 stripes with a total pattern length of 810 feet, designed to give the illusion of acceleration unless the driver slowed down. Daytime mean speeds decreased from 41.3 mph to 33.9 mph immediately after installation. The mean speed increased slightly to 34.9 mph 6 months after installation. Nighttime mean speeds decreased from 40.5 mph to 35.1 mph immediately after installation and increased to 39.1 mph 6 months later. Average crashes per year decreased from 7.7 in the preceding 6 years to three crashes the year after installation. An estimated benefit-cost ratio of 45.9 was calculated, and the authors concluded that the treatment was more effective than signs alone and should be used at other curves where excessive speed is an accident factor (Agent, 1980). In a more recent study of "optical speed bars" at approaches to workzones, Meyer (2001) examined the issue of whether the decrease in speed from transverse striping was due to the perceptual effects of "increasing speed" with a pattern of stripes with gradually decreasing spacing or simply from the "warning" given. The author examined changes in speed as a free-flowing vehicle passed through three adjacent patterns when entering the work zone. The first pattern of transverse stripes were equally spaced, the second had ever-decreasing spacing, and the third were much wider stripes further apart. The data indicated speed reductions of 2 to 3 mph in the mean and 85th-percentile speeds for each of the patterns. The authors concluded that there was both "perceptual" and "warning" effects present. They drew no conclusions concerning which pattern was more effective. No crash data were analyzed. (The effects could differ between curves and work zones due to the driver's judgment of "hazard" related to each.)

In the earlier noted 1999 study of hazardous curve sites in Pennsylvania, transverse striping giving the illusion of acceleration was studied at several sites (Retting, 1999). Unlike the pavement arrow described above, the before/after study showed that these pavement markings had little effect on the average, 85th-percentile, or 95th-percentile speeds.

Other pavement markings designed to increase the "apparent danger" of the curvature have also been evaluated, but not for rural, two-lane curve situations. In a 1998 study of three urban exit ramps in Virginia and one ramp in New York, an experimental pavement-marking scheme was investigated. The treatment narrowed the apparent lane width of the entry to the ramp curve and the ramp curve itself by using a gradual inward taper of existing edgeline or exit gore pavement markings. Studies of vehicle speeds at three of the four ramps indicated that the proportion of passenger vehicles exceeding the posted speed limit by more than 10 mph decreased 20 to 30 percent while speeds at the control site and upstream site remained the same or increased. Similar or slightly larger decreases in the percentage of large trucks exceeding the posted advisory speed by more than 5 mph were also found at the three sites where the equipment differentiated trucks from other vehicles (Retting et al., 2000).

Finally, in the Netherlands and other European countries, an experimental use of edgelines has been tried on curves on narrow, low-volume roads where no edgeline was used in the past (Steyvers and Waard, 1997). Both a solid edgeline and a dashed edgeline caused vehicles to move away from the roadway edge when compared with a completely unmarked curve and with a curve with only a centerline. Driving speeds were slightly higher with the edgelines than with no lines, but slightly lower than when a centerline only was present. No crash analysis was conducted. While experimentation with such markings deserves further testing for these low-volume roads, current Manual of Uniform Traffic Control Devices (MUTCD) guidelines should be taken into consideration. In summary, there are few studies of the accident-related effects of these innovative treatments. Based upon the only crash studies available, post-mounted delineators might be expected to reduce ROR accidents on curves by approximately 15 percent. There is some question concerning the cost-effectiveness of continuous use of such devices on narrow, hilly, curvy roads with lower design standards. While warning symbols on the pavement prior to the curve, pavement markings "narrowing" the lane, and some transverse markings have been shown to reduce either mean speed or 95th-percentile speeds, there are no sound accident-based studies available. Thus, there continues to be a need for well-designed before/after pilot evaluations of crash experience, particularly for the pavement arrow and transverse striping treatments. The ongoing work in Pennsylvania should provide data on the arrow treatment.

Keys to Success

The development of design standards, based upon sound evaluation studies of these innovative markings, will be important. The ability of interested states to have access to evaluations in other states will be important to achieve acceptance.

Potential Difficulties

If these treatments are targeted to curves with actual or expected safety problems, there appear to be few potential difficulties. The Pennsylvania study of the initial transverse-bar sites noted some motorists driving on the shoulder to avoid the lines. This could be a problem with unpaved shoulders (but it is less likely to occur without paved shoulders) and if the vehicle makes a sudden avoidance maneuver without reducing speed (which, again, may not be likely to occur). Pennsylvania also noted that some drivers (presumably commuters) would drive across the centerline or onto the shoulders to avoid transverse rumble strips. Further observations of traffic behavior at treatment sites are needed to determine whether these are true problems. An attribute of these special treatments is their uniqueness and hence high level of notice by drivers. Overuse of these treatments could lead to them losing this uniqueness and ultimate effectiveness. A final possible difficulty could include maintaining the pavement markings over time, given that they are being crossed by all traffic.

Appropriate Measures and Data

In the evaluation of these delineation programs, process measures would include the number of hazardous curves treated.

Impact measures involve comparison of crash frequencies or rates (with the study appropriately designed) for the period before and after modifications. A useful surrogate measure is the change in speed for vehicles entering selected curves. The advent of low-cost vehicle-sensing and recording devices might also allow for the use of a surrogate measure based upon the number of encroachments onto the shoulder over a specific section of road (e.g., a curve). Sufficient data/information will be needed to target these treatments to the correct location. The expert system software noted in "Personnel and Other Training Needs" below will help in this effort.

Associated Needs Services

The transverse strips and the pavement arrow are new treatments, and a relatively modest public information effort may be helpful in garnering support for the effort. If evidence is found that a significant proportion of motorists do drive on the shoulder to avoid the transverse lines (see "Potential Difficulties" above) and if this is found to be a safety problem, then a more significant public education effort will be needed for this treatment.

Organizational and Institutional Attributes

Organizational, Institutional, and Policy Issues

These strategies will be implemented by state and local roadway agencies, and it does not appear that extra coordination with other agencies or groups is needed. If these treatments prove effective and are accepted by states for implementation, both specific design policies and placement policies will be needed. There are two different approaches in selecting delineators for a curve—local practice/policy and the MUTCD. Some of the "newer" pavement markings may have to be approved for use by FHWA as an experimental marking and then eventually adopted as an acceptable device for the MUTCD. However, until a standard is adopted, engineers should consider the effects of implementation inconsistencies on violating driver expectancies.

Jennings and Demetsky (1983) investigated the three post-mounted delineator systems used in Virginia (chevron, special striped delineator [on post], and reflector on a post) for their effectiveness in controlling ROR crashes and to recommend a standard policy regarding use of the system. The resulting simplified policy states that for moderate curves (less than 7 degrees) where delineation is necessary, standard delineation should be used as recommended in the MUTCD. If the curve is greater than 7 degrees, chevrons give better delineation information and the spacing should be 2 to 3 times MUTCD recommendation. Other policy-related advice on delineation selection and placement can be found in expert system software developed by Zwahlen and Schnell (1995). (See "Training and Other Personnel Needs" below.)

Issues Affecting Implementation Time

Since these devices are relatively inexpensive and standard, they could be implemented very quickly.

Costs Involved

The cost of the arrow pavement marker is about $2,000 per site (both directions) according to Pennsylvania's experience. Cost figures are not available for the other treatments. However, many states already use chevrons and other delineators in certain locations and may have cost figures of their own.

Training and Other Personnel Needs

There appear to be no special personnel needs for implementing this strategy. Either agency personnel or contractors would do the installation.

Since there are various low-cost devices available to the engineer, there is need for some guidance on treatment design and placement. Zwahlen and Schnell (1995) developed a PC-based expert system software package that helps the designer choose an appropriate treatment and place the devices for maximum effect. This expert system considers devices such as flexible post delineators, object markers, and various size chevrons.

Legislative Needs

None identified.

Other Key Attributes

None identified.

Information on Agencies or Organizations Currently Implementing this Strategy. As noted in the Effectiveness section, various states (e.g., Ohio, Virginia, Pennsylvania, Kentucky, and New York) have implemented limited installations of delineation and warning systems on curves. As documented in Appendix 1, the most recent of these is Pennsylvania, which is implementing and testing an innovative "pavement arrow" treatment on curve approaches.

Technical Attributes

Target

While the treatment will target hazardous or potentially hazardous curves, the ultimate target is a vehicle that runs off the roadway on these curves.

Expected Effectiveness

Research by Zegeer et al. concerning this proven strategy provides estimates of the effect of curve flattening for various degrees of curve on two-lane rural roads (assuming that the central angle remains constant, and therefore the less-sharp treated curve will be longer and will "replace" some tangent in the initial layout). While more detailed estimates based upon type of curve (isolated versus nonisolated) and central angle (10 to 50 degrees) can be found in the full report, Exhibit V-10 indicates ranges of estimated percent reduction in total crashes for such treatments. For example, flattening a 30-degree curve to 10 degrees is predicted to reduce total crashes on the section by 61 to 67 percent. As noted in a recent review of this study and others, in work related to development of accident modification factors (AMFs) for use with FHWA's Interactive Highway Safety Design Model (Harwood et al., 2000), the estimates provided by this cross-sectional modeling effort would be expected to be less accurate than results from well-conducted before/after studies of actual curve flattening efforts. However, in the absence of such before/after studies in the literature, these results were accepted by the AMF expert panel.

As noted below, curve flattening along two-lane roads may be combined with other safety strategies, including lane and shoulder widening, to provide an additional safety benefit. Indeed, in the process of realigning a curve, the agency would simultaneously provide a new roadside, which itself could provide a positive contribution to safety. Exhibit V-10 summarizes the reductions possible. For instance, assume a 20-foot roadway (with two 10-foot lanes) is to be widened to 22 feet of paved surface with 8-foot gravel shoulders. Exhibit V-11 indicates that these improvements would reduce curve accidents by 5 percent (due to lane widening of 1 foot per side) and 24 percent (due to widening unpaved shoulders by 8 feet per side). Note that the 5-percent and 24-percent accident reduction values cannot merely be added numerically.

In summary, improving the geometry of horizontal curves can lead to significant crash reductions. These reductions change with the amount of curve flattening or widening, as shown in Exhibits V-10 and V-11. It is noted that these reductions are related to percentages of total crashes, rather than just to ROR crashes. While these treatments clearly affect ROR crashes, specific percentages for this subset are not presented in the study. The authors have noted that since curve flattening and widening affects almost all crash types, percent reduction in total crashes were considered to be the most appropriate measure.

Keys to Success

Since this is a relatively expensive treatment, one of the keys to success would appear to be targeting higher-hazard curves. Since ROR crashes increase with degree of curve, the targeting could be based primarily on prior crash history, curve degree, ADT, and speed limit.

Potential Difficulties

As noted above, the estimated effects of this treatment may be inflated due to the fact that they are not based on before/after studies. If the implementing agency "expects" effects this large for a given site or project and after-treatment experience is lower, the agency might curtail similar future efforts. However, given the size of the predicted effects, even if the true effects are much lower (e.g., half as high), this will still remain one of the most effective treatments for ROR crashes on curves.

Appropriate Measures and Data

In estimates of program implementation effectiveness, appropriate process measures would include the number or proportion of "hazardous" curves that are flattened (perhaps categorized by the change in curvature).

The impact measure would be the number of total crashes reduced in the roadway section replaced by the new design.

Targeting will require data on crash frequencies, degree of curve, length of curve, speed limit, and ADT. The factor most likely missing from computerized state files is the degree of curve.

Associated Needs

None identified. This is a standard treatment requiring no additional public information (except as part of any required environmental study).

Organizational and Institutional Attributes

Organizational, Institutional, and Policy Issues

This strategy will be implemented by the state DOT or local roadway agency, and it does not appear that coordination with other agencies will be needed. (The exception would be coordination with environmental agencies if new right-of-way were required.) Since curve flattening is a standard treatment, it would appear that new policy efforts are not required.

However, a slightly different "institutional safety philosophy" may be needed here in comparison with other strategies in this guide. Given the higher cost of this treatment (but coupled with the higher potential payoff), the agency must be prepared to implement more than just low-cost improvements.

Issues Affecting Implementation Time

Given that the treatment will require some form of design and reconstruction and will usually require purchase of additional right-of-way (and thus involve the environmental process), this treatment period will be relatively long.

Costs Involved

Costs will depend on the amount of reconstruction necessary and on whether additional right-of-way is required. In general, this is one of the higher-cost strategies recommended. It is also one of the most beneficial.

Training and Other Personnel Needs

There appear to be no special personnel or training needs for implementing this strategy, given that it involves "standard" reconstruction efforts.

Legislative Needs

None identified.

Other Key Attributes

Since curve flattening would require significant reconstruction, it would be very easy to combine this treatment with lane widening and shoulder improvement treatments noted elsewhere. In addition, it should provide some benefit for bicyclists using the shoulders since it reduces the number of vehicles that leave their lane.

Technical Attributes

Target

Drivers of vehicles who might leave the roadway because of inability to see the edge of the pavement in the roadway section ahead.

Expected Effectiveness

Enhanced lane markings are an appropriate treatment if it is assumed that drivers leave the roadway because they cannot see the pavement edge in the downstream roadway sections. While some driver guidance is needed in such cases, the question is: How much should be added without changing the roadway geometry or the roadside design? Since some evaluations have raised questions about the overall effect of enhanced markings and RPMs, these features are considered a "tried" strategy at this time.

For example, past research (Pendleton, 1996) and research being conducted by Bellomo-McGee, Inc., for NCHRP indicate a lack of significant effect or even a possible increase in crashes on some locations. This could be because drivers tend to drive faster when presented with a clearer delineation of the lane edge. Note, however, that evaluations of such treatments reflect studies of projects involving delineation that was implemented in conjunction with resurfacing. What is not clear is whether speeds increase because of simultaneous resurfacing and remarking or because improved markings were added without alignment or shoulder treatments.

A review of earlier studies on wider edgelines in NCHRP Report 440 noted that, in general, the effectiveness of 8-inch edgelines to reduce ROR crashes is "questionable." The study recommends that they be used only on roads with 12-foot lanes, unpaved shoulders and ADT between 2,000 and 5,000 vehicles per day. In contrast, a 1988 study by the New York DOT indicated that sections of curving two-lane rural roads with new 8-inch edgelines resulted in higher crash reductions than similar sections with new 4-inch edgelines. The study indicated greater safety effects for total crashes (a 10-percent decrease for the wider edgelines versus a 5-percent increase for standard edgelines); for injury crashes (15-percent decrease versus 10-percent decrease, respectively); and for fixed-object crashes (33-percent decrease versus 17-percent decrease, respectively). The study appears to have controlled for the regression to the mean bias by choosing both sets of experimental and control sites from a listing of high-crash locations. It is not clear whether the choice was made randomly.

Effectiveness studies of RPMs have been conducted by states in before/after analyses of treatments at high-hazard locations. (It should be noted that accurately evaluating a treatment at a high-crash location is difficult because of the "regression to the mean" phenomenon. Whether the following studies controlled for such potential biases is unknown.) In southern New Jersey, RPMs have been used on two separate routes, both two-lane rural highways totaling 53.5 miles. The total project cost was $122,730 (1985 dollars). Using data from 2 years before and 1 year after, there was a statistically significant reduction in various types of nighttime accidents including total, injury, head-on, fixed object, overturn, and between intersection accidents. The calculated benefit-cost ratio was 19.89 (State of New Jersey, 1986).

In northern New Jersey, RPMs were installed on six routes (over 126 miles), generally rural two-lane roads. The total project cost was $314,242 (1985 dollars). Again, using data from 2 years before and 1 year after, there was a statistically significant reduction in various nighttime crashes including total, injury, property damage, overturn, head-on, fixed object, and between intersection crashes. The calculated benefit-cost ratio was 15.45 (State of New Jersey, 1986).

For projects with fewer than 800 markers, state forces (not independent contractors, as above) do the installation. For six different route sections totaling 47.8 miles, the construction cost was $151,493. Analysis results show a statistically significant reduction in accidents in every nighttime accident category (total, fatal, injury, property damage, head-on, fixed object, wet surface, and between intersections). The benefit-cost ratio was 25.51.

In Ohio, marker studies were conducted at 184 locations that had high accident rates prior to 1977, including horizontal curves, narrow bridges, stop approaches, and interchanges. Over 3,200 accidents at marker locations were analyzed 1 year before and 1 year after. The results show a 9.2-percent reduction in accidents and a 14.9-percent decrease in injuries. Markers were determined to be effective in all types of driving conditions, including nighttime (5.3-percent reduction) and adverse weather conditions (5.5-percent reduction in crashes at the same time precipitation increased by 10.6 percent). The study concluded that "a dollar spent on raised reflective highway markers in Ohio has returned $6.50 in savings due to accident reduction." As of 1981, nearly 700,000 RPMs were installed in Ohio (The Ohio Underwriter, 1981).

In a 1997 report, the New York State DOT concluded from prior evaluations that raised snowplowable pavement markers (RSPMs) can reduce "guidance-related accidents" (fixed-object collisions, ROR, and encroachment) by approximately 19 percent if selectively applied at locations having high percentages of such crashes. (It is not clear from the report whether the regression-to-the-mean bias has been accounted for.) Based upon an evaluation of 1992 data and a review of studies from other states, the DOT further concluded that RSPMs should not be applied systemwide, since they are somewhat costly and would have no effect or a possible negative effect on crashes at such nonspecific locations.

In summary, the effectiveness of RPMs as a general "systemwide" treatment appears questionable. The effectiveness of RPMs at high-hazard sites may also be less clear than first thought. This is not to say they should not be tried. Their relatively low cost argues for experimentation. However, at this point, it is not possible to specify a crash reduction factor for these devices. Clearly, well-designed before/after studies of effectiveness at such sites are needed—studies that account for the "regression-to-the-mean" bias. Thus, although this treatment may be effective in reducing crashes, careful targeting, monitoring, and evaluation are needed.

Similarly, the effectiveness of wider edgelines is also difficult to specify based upon past studies. While the NCHRP Report 440 review found wider edgelines "questionable" in general, the New York State DOT study indicated that implementation on high-crash sites on two-lane roads might result in a 10- to 15-percent decrease in ROR crashes.

Keys to Success

Based upon the effectiveness studies, the key to success is the targeted application of this treatment to sites where more guidance is needed for the driver, but where vehicle speeds will not be increased to unsafe levels.

Potential Difficulties

A potential difficulty with RPMs is the damage to the reflector or possible dislodging of the reflector during snow plowing. However, these concerns have lessened due to the creation of plowable RPMs. Another potential pitfall is nontargeted or erroneously targeted application of the devices on high-speed two-lane roads. This could result in adverse safety effects, which might negatively affect opinions about the treatment and therefore keep it from being implemented where needed.

Appropriate Measures and Data

In agency evaluations of implementation effectiveness, process measures would include the number of hazardous curves treated and the type of treatment applied.

Impact measures would involve before/after changes in crash frequencies or rates (with the study appropriately designed) and changes in speed from before to after treatment.

It would also appear that data are needed to better target the treatment, targeting to sites where additional visual guidance is needed, but where speeds are less likely to be increased. This is a difficult task. It may be aided by use of video logs and conduct of safety audit types of studies.

Associated Needs

No new public information efforts appear to be needed since this is a publicly accepted treatment on other roads. (Efforts to train the public to use them correctly—i.e., not to increase speed—are not expected to be effective).

Organizational and Institutional Attributes

Organizational, Institutional, and Policy Issues

This strategy could be implemented by the state DOT or a local roads agency, and it does not appear that additional cooperative efforts with other agencies are needed. The only exception might be if the enhanced delineation led to increased speeds. In this case, targeted speed enforcement could be needed.

After effectiveness is established and targeting methods are developed, a design and placement policy is needed to facilitate implementation, along with AASHTO support and guidance.

Issues Affecting Implementation Time

Since these devices are relatively inexpensive and are standard devices, they could be implemented in a very short timeframe.

Costs Involved

An old cost figure states that Ohio's average cost is $14.71 per unit for 35,000 units. A 1997 New York DOT report indicates that an RSPM (which is more expensive than a standard RPM) costs approximately $25­30 to install and $6­8 each 3 years for reflector replacement. Installation was found to increase the cost of delineation from approximately $2,000 to $5,300 per mile. However, states have most likely developed their own cost estimates, since these treatments are being widely used.

Training and Other Personnel Needs

There appear to be no special personnel or training needs for implementing this strategy. The installation would be done by either agency personnel or contractors and indeed is already being done in most state agencies.

Legislative Needs

None identified.

Other Key Attributes

None identified.

Technical Attributes

Target

Treatment will target locations where skidding is determined to be a problem, in wet or dry conditions. The ultimate target, however, is a vehicle involved in a crash due to skidding, usually on wet pavement. With respect to ROR or head-on crashes, the target vehicle is one that runs (skids) off the road due to insufficient skid resistance or becomes involved in a head-on crash either by skidding into the opposing lane or by crossing into the opposing lane after an over-correction from an initial ROR maneuver caused by insufficient skid resistance.

Expected Effectiveness

There are many different specific countermeasures that may be implemented to improve skid resistance. This may include changes to the pavement aggregates, adding overlays, or adding texture to the pavement surface. The effectiveness of the countermeasure not only depends on that measure selected, but also will vary with respect to location, traffic volume, rainfall propensity, road geometry, temperature, pavement structure, etc.

The New York State DOT has implemented a program that identifies sites statewide that have a low skid resistance and treats them with overlays or microsurfacing as part of the maintenance program. A site is eligible for treatment if its 2-year wet accident proportion is 50 percent higher than the average wet accident proportion for roads in the same county. Between 1995 and 1997, 36 sites were treated on Long Island, resulting in a reduction of more than 800 annually recurring wet road accidents. These results and others within the state support earlier findings that treatment of wet road accident locations result in reductions of 50 percent for wet road accidents and 20 percent for total accidents. While the reductions in ROR or head-on crashes cannot be extracted from the data at this time, it appears that reductions in these types would be at least the same as for total crashes.

While these results could be subject to some regression-to-the-mean bias, the New York staff has found that untreated sites continue to stay on the listing until treated in many cases—an indication that these reductions are clearly not totally due to regression. The New York State DOT is planning a more refined data analysis to account for possible biases in these effectiveness estimates. Based on the current knowledge, this identification/treatment strategy would be classified as "proven."

Keys to Success

Monitoring the skid resistance of pavement requires incremental checks of pavement conditions. Evaluation must identify ruts and the occurrence of polishing. Recent research (Galal et al., 1999) has suggested that the surface should be restored between 5 and 10 years in order to retain surface friction, but the life span is affected by site characteristics such as traffic volume.

In addition, spot- or section-related skid accident reduction programs will be clearly most successful if targeted well. The New York State DOT program noted above provides a methodology for such targeting. In addition, in a 1980 Technical Advisory, the FHWA provided a detailed description of a "Skid Accident Reduction Program," including not only details of various treatments, but also the use of crashes and rainfall data in targeting the treatments.

Potential Difficulties

Skid resistance changes over time. This requires a dynamic program and strong commitment. As noted in the preceding section, it also requires good "targeting." When selecting sites for skid resistance programs, it is important to somehow control for the amount of wet-pavement exposure. This will help decrease the identification of sites that have a high wet-accident proportion or that rate simply because of high wet-weather exposure with no real pavement-friction problems. Unfortunately, it is difficult or impossible for an agency to develop good wet-pavement crash rates per vehicle mile for all roadway sections due to the lack of good wet-weather exposure data for all sites. Such data would require both good rainfall data for all potential sites and good measures of traffic volume during wet and dry weather. In its Skid Accident Reduction Program (SKARP), the New York State DOT uses a surrogate for such detailed data. The DOT compares the proportion of wet-weather crashes at each site with the proportion for similar roads in the same county. The assumption here is that rainfall (and thus wet-pavement exposure) would be similar across a county, a reasonable assumption.

Appropriate Measures and Data

Data are needed on traffic crashes by roadway condition. In addition, measures of traffic exposure that identify and reflect both dry and wet periods are needed. Finally, measurements of road friction and pavement water retention should be documented both before and after implementation of a strategy.

Associated Needs

None required. Relatively unnoticed by the public.

Organizational and Institutional Attributes

Organizational, Institutional, and Policy Issues

Implement by state DOT; no coordination required. Policy may be needed in order to determine the most appropriate pavement aggregate statewide and at special locations. Additionally, guidelines may be needed to highlight when pavement groove cuts should be considered. These countermeasures may also require cooperation within an agency, especially if these types of safety treatments are to be tied to routine maintenance.

Issues Affecting Implementation Time

Depends upon the treatment. Grooving can be done quickly, but overlays require more time. Nevertheless, all strategies being suggested should have short implementation periods.

Costs Involved

Highly variable depending upon the specific treatment. The New York State DOT estimates that its resurfacing/microsurfacing projects are approximately 0.5 miles long, with an average treatment cost of approximately $20,000 per lane mile (1995 dollars).

Training and Other Personnel Needs

No special personnel needs for implementing this strategy. Either agency personnel or contractors could do installation.

Legislative Needs

None identified.

Other Key Attributes

None identified.

Technical Attributes

Target

The targets of this package of strategies are vehicles that stray from their lanes onto the shoulder area. The ultimate targets are the drivers of these vehicles, who are being provided with an opportunity for a safe recovery.

Expected Effectiveness

Even though there have been numerous studies of both shoulder widening and paving and limited studies of pavement edgedrop elimination, there is still some uncertainty about the true effect of such treatments. A recent unpublished literature review by Hauer (2000) demonstrated this uncertainty, noting some studies of shoulder widening/paving that indicated effects as large as 30- to 40-percent reductions and other studies that indicated no effect or even a possible increase in crashes for certain ADT levels. (If true, such an increase could be attributed to increased speeds resulting from shoulder improvements without changes in curvature or other factors.) The major shortcoming in the large body of research is that most findings are not based on well-conducted before/after studies where shoulders have actually been improved in the field. Instead, most are "cross-sectional" studies, in which different segments of roads with different shoulder characteristics are used in statistical models that estimate the effect of a change in width by changes in model output. However, based on the best available research, shoulder widening and paving would be considered "proven" strategies. Even though their safety benefits would appear to be "obvious," strategies related to edgedrop elimination would have to be considered "experimental," since no research into effectiveness is available.

With respect to shoulder widening and paving, in a recent FHWA effort related to determining AMFs for use with the Interactive Highway Safety Design Model (Harwood et al., 2000), a panel of experts attempted to develop a best estimate of shoulder treatment effectiveness based on a review of a number of research studies. Their estimate of effectiveness of shoulder widening on two-lane rural roads is shown in Exhibit V-17. Here, the base shoulder is a 6-foot-wide paved shoulder, and the AMFs shown for different ADTs are relative to this base shoulder. For example, a roadway with 500 vehicles per day and a 2-foot shoulder would be expected to have 30 percent more "related crashes" than the same road with a 6-foot shoulder (i.e., an AMF of 1.3). In like fashion, a two-lane rural road with 2,000 vehicles per day and an 8-foot shoulder would be expected to have 13 percent fewer related crashes than the same road with a 6-foot shoulder (i.e., an AMF of 0.87). Note that these reductions are not for total crashes, but for "related crashes," which include single-vehicle ROR, multivehicle opposite-direction (i.e., head-ons and opposing sideswipes), and multivehicle same-direction sideswipe crashes. To obtain the percentage reductions in total crashes, these AMFs would be multiplied by the percentage of total crashes they represent (typically, 35 percent for two-lane rural highways).

In the same study, the panel also defined AMFs for turf, composite, and stabilized gravel shoulders relative to the paved shoulder of the same width. As shown in Exhibit V-18, these effects change with shoulder type and shoulder width. For example, for an 8-foot width, turf shoulders are expected to experience 11 percent more "related crashes."

Much less is known about the effectiveness of edgedrop treatments, since it is difficult to specifically define the percentage of ROR or head-on crashes, which is the result of "overcorrection" by vehicles that run off the road first. Whatever that percentage, Humphreys and Parham (1994) concluded that a 45-degree-angle asphalt fillet at the lane edge would virtually eliminate this type of crash, even in cases where the shoulder is unpaved and suffers subsequent erosion damage.

Keys to Success

As with other ROR treatments, keys to success will include treatment targeting, such that funds are used as efficiently as possible. Targeting them to higher-speed roads with high ROR crash frequencies and rates could enhance all three strategies. Implementation of the edgedrop treatments will be enhanced by the identification of "champion" states that have implemented edgedrop treatments as a standard part of their repaving efforts and have found the treatments to be both low cost and effective. If an edge-fillet program is to be implemented, an additional key to success will be the development of an inclusive pavement specification and the necessary equipment modifications.

Potential Difficulties

While not evaluated extensively, it appears that the edge fillet or other edgedrop treatments would not have significant potential difficulties unless the use of this treatment resulted in less maintenance of unpaved shoulders.

However, if wider paved shoulders are added to high-speed roads with poor alignment and hazardous roadsides, they possibly could lead to an increase in vehicle speeds and total crash frequency and severity. Thus, careful targeting and monitoring is needed.

Appropriate Measures and Data

In the evaluation of strategy implementation effectiveness, process measures would include the number of road miles or number of hazardous locations where these shoulder treatments are installed, as well as the type of installation. Impact measures will include the number and rate of ROR (and head-on) crashes reduced at these locations. However, due to possible adverse effects, changes in total crashes also should be studied.

Data on ROR crashes would be needed to target the shoulder widening/paving treatment. If the state decided to use only the pavement edge treatment at selected locations (rather than as a standard add-on to resurfacing activities), criteria would need to be developed to define those critical locations, and data (e.g., crash or edgedrop inventory) would be needed to identify the locations. In addition, as noted above, since the edge-fillet treatment has not been evaluated, if a state were to implement the wedge, it is critical that the necessary treatment location, crash, and roadway inventory data on possible confounding factors be collected.

Associated Needs

Since these are somewhat "standard" treatments, there does not appear to be a critical need for public information or education efforts.

Organizational, Institutional, and Policy Issues

This strategy can be implemented by the state DOT or local roadway agency, and it would appear that there is no need for cooperative efforts with other agencies. Since these are "standard" treatments in general, no significant policy action appears needed other than a possible design policy for the pavement edge fillet.

Issues Affecting Implementation Time

Unless shoulder widening requires additional right-of-way, these treatments can be implemented in a relatively short timeframe. While all three would involve retrofits to existing pavements, it seems that the most opportune time to implement them would be in conjunction with repaving efforts.

Costs Involved

Shoulder widening costs would depend on whether new right-of-way is required and whether extensive roadside moderation is needed. Shoulder pavement costs should be similar to lane pavement costs and depend on how much shoulder stabilization is required.

Humphreys and Parham (1994) note that the cost of adding a pavement edge fillet when resurfacing a roadway is very low--perhaps 1 to 2 percent of the typical resurfacing cost.

Training and Other Personnel Needs

There would appear to be no special personnel needs for implementing these strategies, since they are similar to other paving/construction activities. The only new training needed would be for paving forces (whether state or contract) that would place the pavement edge fillet.

Legislative Needs

None identified.

Other Key Attributes

None identified.

Technical Attributes

Target

The targets for roadside improvement treatments are vehicles that leave the roadway, including those that return to the roadway out of control due to poor roadside design. However, the primary focus would be vehicles that strike objects on the roadside or overturn.

Expected Effectiveness

Three strategy areas have been covered for roadside improvements--rollover reduction due to flattening sideslopes, single-vehicle crash reduction due to flattening and widening sideslopes, and single-vehicle crash reduction and crash-severity reduction related to improvements in roadside hardware (e.g., replacing older hardware designs with newer designs). Historically, most roadside design and roadside hardware design improvements have been based on crash testing and recently on computer simulation. Due to the difficulty in collecting the necessary roadside inventory data to conduct a well-designed, accident-based study, few such studies exist. Thus, crash-based estimates of effectiveness are limited. Based upon these limited studies, the first two sets of strategies relating to sideslope and clear zone improvements are considered "proven" strategies. The replacement of older (approved) hardware with newer designs would be considered "tried" but not "proven" at this point due to the lack of crash-related evaluations. The following narrative describes some of the more important studies.

Zegeer, et al. (1987) examined the effects of sideslope on both rollover crashes and total single-vehicle crashes. They used field-measured crash, sideslope, cross section, and traffic data from approximately 1,800 miles on rural two-lane roads in three states. The rollover data were limited, making analysis of individual slope categories difficult. However, the authors found that rollover rates were significantly higher on slopes of 1:4 or steeper as compared with slopes of 1:5 or flatter. That is, in terms of rollover crashes, the 1:4 slopes were similar to the steeper 1:3 and 1:2 slopes. While the Roadway Design Guide would indicate the need for guardrail protection for 1:2 slopes, it would not for 1:3 slopes. The latter are generally considered to be "traversable ­ nonrecoverable," indicating that the vehicle would be expected to either stop on such a slope or continue to the bottom of the slope without overturning.

Based upon the same study (and a much larger sample size), it is concluded that single-vehicle ROR crashes (which include, but are not limited to, rollovers) can be significantly reduced by flattening existing sideslopes to 1:4 or flatter. As shown in Exhibit V-24, the estimated reduction in single-vehicle ROR crashes on two-lane rural roads ranges up to approximately 27 percent (i.e., for flattening a 1:2 slope to 1:7 or flatter). Because ROR crashes are a major component of total crashes on two-lane rural roads, and because flatter and safer sideslopes can decrease some head-on and sideswipe crashes due to safer recoveries, the corresponding decrease in total crashes for this example is an estimated 15 percent. These estimates are made under the assumption that the clear zone width stays the same and that the resulting sideslope is relatively free of rigid objects.

The Washington State DOT funded a study by Allaire et al. (1996) to determine whether past sideslope flattening projects had reduced ROR collision frequencies and severities. Unlike other studies, the authors were able to conduct a before/after study of the effects of slope flattening based upon a detailed review of 60 3R projects implemented in 1986­1991. Each of these 60 projects called for sideslope flattening in at least some portion of the project. The authors were not able to develop benefit estimates for specific degrees of flattening (e.g., flattening a 1:3 slope to 1:6) due to insufficient data on the precise "before" conditions. However, they were able to examine the before-to-after reductions in crashes by severity level for the treated sections and to compare these changes with a series of "control" changes. These comparisons included comparisons of actual ROR collision rate per mile (by severity level) in the after period with predicted after rates corrected for "other improvements" such as object removal and clear zone widening; predicted after rates based on the experience of the entire 3R project length, much of which did not include slope flattening; and predicted after rates based on changes in the statewide rate for similar roads during the same time period. In almost all cases, a statistically significant benefit of slope flattening was found. The percent reduction in ROR collision rates varied by comparison and by injury severity class from approximately 3 to 50 percent. Based upon examination of the tables, the estimated "median" reduction in ROR crash rate is approximately 25 to 45 percent.

Zegeer et al. (1987) also estimated the effects of clear zone widening on two-lane rural roads. If the existing recovery area measured from the edgeline is less than 10 to 15 feet, Exhibit V-25 presents the expected percentage reduction in "related crashes" (i.e., ROR, head-on, and sideswipe) due to clear zone widening by a given amount. For example, widening by 10 feet is predicted to result in a 25-percent reduction in these crashes.

With respect to removing roadside hardware from the clear zone or relocating it farther from the travel way, a study by Zegeer et al. (1990) developed the effectiveness estimates shown in Exhibit V-26 for two-lane rural roads. As can be seen, for example, moving culvert headwalls from 5 to 15 feet from the roadway would result in an expected 40-percent reduction in culvert headwall collisions on two-lane rural roads. Placing guardrails an additional 5 feet from the roadway would be expected to reduce the corresponding guardrail accidents by 53 percent. These estimates are based upon the assumptions that removal of a specific object leaves a wider clear zone and that other potentially hazardous objects do not remain at the same distance from the roadway. For example, if the culverts are at the edge of a row of large trees, then it is likely that culvert crashes will only be replaced by additional tree crashes.

The third strategy noted above involves the upgrading of existing roadside hardware. In a recent study, Ray (2000) examined the possible effects of upgrading guardrail terminals (e.g., BCT and MELT designs) to a newer design (the ET-2000) that does pass upgraded crash test standards. The author examined both past accident-based studies of the older designs in five states and recent data on the older and newer designs in three states. He used data from both police-reported and nonreported (maintenance) cases where available. He concluded that while the samples were small and the results varied greatly across the studies, he could detect no statistically significant difference in injury severity among the three designs for properly installed terminals. The author stressed the need for proper installation, since there is evidence that the BCT device was not installed properly (i.e., improper flare and offset from the travel lane) in a significant number of cases. And some data in earlier studies indicate that these improper installations are more hazardous (Morena and Schroeder, 1994; Agent and Pigman, 1991). Thus, one might expect an improvement from upgrading improperly installed BCT devices, for example.

Keys to Success

Keys to success would include accurate targeting; appropriate levels of funding (since most of these strategies can be relatively high cost); and a cooperative program among all agency divisions that can affect the roadside crash problem when a high-crash site is identified (e.g., traffic or safety engineering); during construction or reconstruction (e.g., roadway design division); or during normal maintenance operations (e.g., roadway maintenance forces). Appropriate targeting and analysis will require the development and regular use of methods for identifying sites with ROR crash problems related to the roadside.

Potential Difficulties

As noted above, simple clearing and grading to create small additions to the clear zone (e.g., 5 to 10 feet) on steeper sideslopes (e.g., 1:3 or 1:4) under high-speed conditions may not prove to be effective, since the vehicles entering the roadside may continue, due to the effect of the slope, to either overturn or traverse the clear zone and strike objects at its far edge. The other potential pitfall could be public reaction to tree cutting without appropriate public education, as well as coordination with environmental and other public groups.

Appropriate Measures and Data

The most appropriate measures will depend upon the strategy implemented, but would include process measures such as miles of roadside treated, number and type of hazardous objects removed or relocated, and number and type of older devices upgraded.

Impact measures must include both crash frequency or rate and crash severity, since some strategies will be successful even if only severity is affected.

Since targeting and site analysis appear to be keys to success, data needs would include accident, roadway and roadside inventory, and traffic data of sufficient accuracy and detail. The data most often missing are for the roadside inventory.

Associated Needs

As noted above, since tree clearing is a major component of some of these strategies, and since the public can view such tree removal negatively, there is a need for public information before tree-related strategies are employed. See Volume 3 of this report for more information.

Organizational and Institutional Attributes

Organizational, Institutional, and Policy Issues

While the primary agency would be the highway agency responsible for the right-of-way, certain strategies would clearly require participation of other agencies and public and private groups (e.g., strategies involving tree removal or utility pole relocation or removal). (See note on the need for a cooperative, multiagency program under "Keys to Success" above.) Since cooperation among various governmental and private groups is necessary if tree clearing is anticipated, see the more detailed discussion in Volume 3 of this report.

An organizational safety philosophy is needed that includes willingness to implement more than just low-cost improvements to optimize results. Many of these strategies are higher-cost strategies, but offer higher potential payoff.

Issues Affecting Implementation Time

The timeframe required will depend on the strategy chosen. It could be relatively short for treatments such as replacing older hardware at a specific location, but much longer if applied to an entire corridor or route system or if the treatment involved new right-of-way acquisition.

Costs Involved

Costs of these strategies can vary widely depending on the strategy chosen. Factors include whether new right-of-way is required or whether the treatment can be implemented as part of other rehabilitation or original construction efforts.

Training and Other Personnel Needs

Since most of these strategies are being implemented by many state highway agencies as part of construction or rehabilitation, there would appear to be no special personnel or training needs for implementing these strategies.

Legislative Needs

None identified.

Other Key Attributes

None identified.