In-Service Performance of Traffic Barriers (2003)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

7CHAPTER 2 LITERATURE REVIEW INTRODUCTION While agreement is almost universal on the importance of in-service evaluation for roadside safety appurtenances, the process that should be used has never been formalized. One goal of this project was to recommend an in-service evaluation process that could be used to evaluate the performance of a roadside safety appurtenance. This process was developed largely by examining prior in-service evaluations and identi- fying features and techniques that either did or did not result in a meaningful examination of the appurtenance’s performance. “In-service evaluation” here implies that actual sites were visited and examined shortly after a collision occurred. Stud- ies retroactively examining accident and maintenance records are referred to as “historical studies” in this report. In other cases, data collection was performed concurrently with the study period by police or maintenance personnel, but no site visits occurred. Without examining the site of a collision, the data can be difficult to interpret. Such studies are referred to as “collision studies” to differentiate them from “in-service evaluations,” which involved site visits. A literature review was performed to identify prior in-service evaluation efforts. Materials were identified in three ways: 1. A TRIS search (10); 2. A review of FHWA reports, NCHRP reports, TRB papers, and other roadside safety literature; and 3. A survey of roadside safety professionals. From these sources, 57 reports were obtained that described some type of roadside feature evaluation. The survey was sent to approximately 240 people in state DOTs, FHWA division offices, and FHWA regional offices; roadside hardware man- ufacturers; and other roadside safety professionals. Ninety- five people responded to the survey, representing a total of 45 states. Table 2 summarizes the survey results. Nineteen of the 45 states responding had performed some type of in-service evaluation in the past. Only 18 of the states had some type of roadside hardware inventory, a few of which were outdated according to the survey. Table 3 summarizes the data sources used by respondents to perform roadside hardware evaluations in their states: police reports, hardware inventory, maintenance reports, or on-site investigations. Most of the respondents named police or main- tenance reports as data sources with 21 and 20 responses, respectively. On-site investigations were used by 16 of the respondents and inventory reports by 5 of the respondents. Table 4 presents a summary of the roadside devices iden- tified by the respondents. Devices studied included various types of end treatments, guardrails, median barriers, and impact attenuators. HISTORICAL STUDIES A historical study in this review indicates a study that retroactively examines collision and/or maintenance records. The data collection is performed some time after the colli- sions occurred, and there is no investigation of the site. Kentucky, 1975—Bridges (11) In 1975, Agent reported on an investigation of collisions associated with highway bridges and overpasses in the state of Kentucky. This investigation was performed to satisfy the FHWA’s request that data be collected as part of routine bridge inspection. The objective of the study was to “identify those principal features of bridges and appurtenances which may be related to collision frequency and severity and to provide some further insights into highway safety.” Most of the study data were collected from police reports of collisions that occurred between 1972 and 1973. Non-fatal collision data were collected for 1972 only. Each type of device was assigned a severity index (SI) based on the sever- ity of the injuries associated with the collision. The smallest value for the SI was 1.00 for property damage collisions. The highest value possible was 9.50 for a device that resulted in fatality or severe injury in all collision cases. If more than one fatality or severe injury occurred within a single event, this was noted but did not affect the SI. Collisions with bridge piers, gaps between bridge open- ings, entrance posts and wing walls, bridge railing or curb, bridge railing and guardrail, and approach guardrail were investigated, where each type of object was categorized as a different device. In addition, impacts with another vehicle on the bridge and those due to bridge geometry were examined. Although the specific results of the report are beyond the scope of this synopsis, it appears that the most common cause

of collisions associated with bridges and overpasses was icy or inclement weather conditions. The researchers also inves- tigated whether there was a correlation between warning and flashing “ice on bridge” signs and a reduction in collisions. In general, investigation results showed a reduction in the number of collisions at high-incident locations. The data presented in this report are based on a large sam- ple base and consequently have statistical merit. The size of the database was sufficient for an investigation of the statis- tical significance of some of the factors, such as icy condi- tions, but such analyses were not performed. Correlating the type of device struck with the severity indices could have been investigated but was not. Iowa, 1979—Cable Guardrail (12 ) In 1979, Schneider reported the results of a study of cable guardrail collisions in the state of Iowa. The purpose of the study was to determine the performance of the light post cable guardrail within the state using collision statistics. Data collection for the evaluation consisted of an exami- nation of existing collision data sources such as the state maintenance property damage reports, the Accident Location and Analysis System (ALAS), and the police reports of motor vehicle accidents. Two years of collision data were used in the study. Once the data were collected, the performance of the guardrail was categorized according to whether the guardrail redirected the vehicle, kept the vehicle from entering the haz- ardous area, and if the barrier was economical to construct and maintain. Sixty maintenance reports were examined from the 2-year study period. Of these, 31 were matched with a corresponding police crash report from the ALAS database. Results showed that the average property damage and collision severity were lower for the cable guardrail than for all guardrail collisions in the state during the study period. Approximately 32 percent of the vehicles impacting the cable barrier penetrated it, and one fatality was recorded. Average costs for the installation, maintenance, and repair of the guardrail were also computed. 8 Although the author states that time was limited for the completion of the study, a more thorough evaluation would have been obtained had data been collected concurrently with the collisions themselves. This would have allowed on- site investigations of the collisions and periodic examina- tions to determine unreported incidences, which could have changed report results and increased the number of docu- mented collisions from which the conclusions were drawn. New York, 1983—Turned-down Guardrail End Treatment (13) In 1983, Fortuniewicz, Bryden, Hahn, and Phillips described crash testing and a subsequent collision and site study per- formed on a turned-down guardrail end treatment for heavy- post, blocked-out W-beam median barriers. The purpose of the evaluation was to determine the field collision performance and maintenance requirements of the end treatment. Most of the data were collected retroactively by examining computerized collision files and maintenance records. The remainder of the data were collected during site examinations to find indications of unreported impacts. These unreported events were compared to reported collisions occurring at or near the end treatments involving guardrail or median barriers. Sites were classified as having sustained either minor or severe damage. Minor damage consisted of small dents, scratches, and paint marks. Severe damage included bent or misaligned posts and rail sections. When the collision study and field investigations were per- formed in 1981, 62 turned-down end treatments for heavy- post median barriers existed in the state of New York. Police crash reports were difficult to obtain and only one was found in the files. A vehicle rollover occurred but no injuries were reported. All sites were examined once during the field inves- tigation. Sixty-one impacts were documented; 53 were clas- sified as minor and eight as severe. The authors noted that it was difficult to differentiate between multiple- and single- event impacts at any given location. Additionally, no main- tenance activities had occurred at these sites since the first installations in the mid 1970s. This study provided useful information for the implemen- tation of the ramped end treatment for heavy-post guardrail in the state of New York. The criteria by which the impact with the turned-down treatment were evaluated included misaligned posts. Poor installation or soil settlement could account for misaligned or non-vertical posts. The study illus- trates the difficulty of obtaining an adequate number of cases with a small installed inventory and a relatively short data collection period. Iowa, 1989—Bridges (14) In 1989, Schwall reported on a study of collisions occur- ring at or near bridges in the state of Iowa. The purpose of TABLE 2 Inventory and evaluation data Yes Responses No. of States Evaluations? 25 19 Inventories? 20 18 TABLE 3 Data sources for evaluations Yes Responses Percent of Total Evaluations Police 21 84% Maintenance 20 80% On-site 16 64% Inventory 5 20%

the study was to examine the collision data at bridges on Iowa’s primary system and the cost-effectiveness of installing approach guardrail. This information was used to justify a pro- gram to upgrade approach guardrail at primary road bridges in Iowa. The majority of the data came from a previous study’s report, “Iowa Fixed Object Accident Analysis,” by Dominic Vi-Minh Hoang of the Iowa Division of the FHWA. Police crash reports were collected for all fatal collisions at bridges and culverts in Iowa since 1981. Sixty-one of the 90 fatal bridge collisions involved impacts with unprotected bridge ends. The study also identified the percentage or length of approach guardrail that was below current standards for W-beam approach guardrail. Installation of approach guardrail was determined to cost on average $7,000. The average maintenance cost of these 9 systems was also determined. In this assessment, only the cost to repair the guardrail after an impact was considered. To determine the feasibility of upgrading unprotected and substandard guardrail on bridge ends, a benefit-cost (B/C) analysis was performed. The societal cost for each type of injury was assigned based on the Iowa value loss index. The average cost of a collision for an unprotected bridge end was then computed. It was then assumed that the use of approach guardrail would reduce the cost of an collision (based solely on severity) by 24 percent. Average maintenance costs were then added to the cost of the guardrail based on the average daily traffic (ADT) of the road where each bridge was located, and a B/C analysis was performed. An alternative was consid- ered feasible if the B/C ratio was 1.2 or higher. Based on the analysis, installing approach guardrail on bridges with an ADT of 1,700 vehicles per with a 610 mm offset would result in a TABLE 4 Roadside devices studied Devices Studied Report No Report Planned Total End Treatments BCT 6 1 0 7 ET-2000 2 2 2 6 Bluntend/Turndown 4 1 0 5 SENTRE 2 1 0 3 TREND 2 0 0 2 VAT 1 0 0 1 CAT 8 1 2 11 MELT 1 0 1 2 CO 3F 1 0 0 1 SRT 0 0 3 3 Brakemaster 4 0 0 4 Guardrail and Median Barrier W-beam GR 4 2 2 8 Cable GR 5 1 0 6 Box-Beam GR 3 1 0 4 WY-Box Beam 0 0 1 1 Mod. Thrie-Beam 2 0 0 2 Quick-Change 1 0 1 2 SERB 2 1 0 3 IBC Mark VII 2 0 0 2 Impact Attenuators Frangible Tube 1 0 0 1 CIAS 1 0 0 1 NCIAS 1 0 0 1 Tire-Sand Bar. 1 0 0 1 GREAT 1 0 1 2 CTMA 1 0 0 1 Triton Water 1 0 1 2 Modular CC 1 0 1 2 Hi-Dro Cell 1 0 0 1 Hex-Foam 1 0 0 1 Tor-Shok 1 0 0 1 LMA 0 0 1 1 REACT 0 0 2 2 Quadguard 0 0 1 1 ADIEM II 0 2 2 4

B/C ratio of 1.2. All bridges on roads with an ADT of at least 2,700 vehicles per day had a B/C ratio of 1.2 for approach guardrail with a 3-m offset. This study provides useful information for the upgrade of approach guardrail associated with bridges in the state of Iowa. To this end, the report did fulfill its purpose of exam- ining collisions involving bridges and justifying a program to upgrade approach guardrail on primary road bridges. How- ever, the methods of data collection and analysis could have been improved: (1) Mention was made of the reduction in the number and severity of collisions involving bridge ends, but it appears these factors were not incorporated into the final analysis. (2) Costs for routine maintenance such as replacing rotted posts, corroded guardrail, vandalized devices, etc. were neglected. Unfortunately, these repairs can be relatively costly and should have been included. The study was focused solely on fatal collisions. While fatal collisions have the highest soci- etal cost, including other injury collisions may have changed the results. New York, 1989—Cable Median Guardrail (15 ) In 1989, Tyrell and Bryden described a study of the cable median barrier in the state of New York. The purpose of the evaluation was to determine the field performance of cable guardrail used as a median barrier. Data collection for this report relied solely on police- reported collisions involving the 15 sites under investigation. Photologs were examined to verify that the collisions involved one of the cable barriers. Collisions were then classified according to the most severe injury in the event, the occur- rence of a secondary collision, and how the vehicle interacted with the barrier (i.e., contained, penetrated, snagged, etc.). No attempt was made to determine the number of unreported collisions. It is important to note that only passenger vehicles are allowed on the parkway. According to the collision data, 99 collisions occurred with the barriers during the 3-year evaluation period. Personal injury was reported in 24 incidents. In six cases, only a single vehicle was involved and no personal injury was recorded. There were, however, four cases where the impacting vehi- cle was not contained by the barrier. Two of these were attributed to the height of the barrier. The guardrail in the study was constructed before the standard height of cable barriers was lowered. In the other two cases, the vehicle impacted a tree. It is uncertain in the latter cases whether the vehicles penetrated the barrier or if the barrier deflected enough to allow the vehicles to strike the trees. Performance data and the costs associated with the use of the system deter- mined that the cable median barrier’s performance was sat- isfactory in this study. This study presents useful information on the implemen- tation of the cable barrier system when used as a median bar- rier. However, the exclusion of certain information from the study lowered the quality of the evaluation’s findings: (1) As 10 stated above, no attempt was made to determine the number of unreported collisions. This lowered the number of suc- cessful re-directions. (2) No attempt was made to determine possible problems with the installation or use of the systems. Interviews with maintenance personnel could have helped. California, 1991—Median Barriers (16) In 1991, Seamons and Smith of the California Department of Transportation (CalTrans) published a study of past and current median barrier practices in California. The report contained a collision study relating to the assumptions made when identifying and recommending a site for median bar- rier placement. The purpose of this study was to verify the accepted values for the increase of the collision rate after the placement of a median barrier at a particular location. Collision data were acquired from California’s Traffic Acci- dent Surveillance and Analysis System (TASAS), which is “the computerized record system of traffic accidents and high- way features for the (California) state highway system (16).” TASAS was used to identify sites where median barriers were the primary item on a contract and had been completed 5 to 6 years prior to the study to allow before-and-after study. Twenty-four freeway sites and five non-freeway sites were used in the collision study. Installation of a median barrier greatly reduces the fre- quency of cross-median collisions but increases the total fre- quency of collisions. From previous studies cited within the report, the expected increase in the collision rate after place- ment of a median barrier was 20 to 30 percent. The results of the new study, however, indicated an increase of 10 to 20 per- cent on freeways and 50 percent on non-freeways was more likely. As a result of the collision analysis, it was recom- mended that the new percentages be used to determine the cost-effectiveness for the placement of median barriers within the state of California. Texas, 1991—Turned-down Guardrail End Treatment (17) In 1991, Griffin reported on the performance of turned- down guardrail terminals in the state of Texas. In 1990, a FHWA memorandum was issued stating turned-down termi- nals were to be replaced on high-volume and high-speed roadways and were not to be used in new installations. Since the turned-down end terminal was effectively the only end treatment used in Texas at the time, the Texas State Depart- ment of Highways and Public Transportation asked the Texas Transportation Institute to investigate the number of collisions involving guardrails, the severity of these colli- sions, the number of vehicles overturning on turned-down treatments each year, the type of highway and traffic vol- umes where the collisions occur, and other variables con- tributing to these collisions. The objective of the study was

to examine the frequency of vehicle overturn and accidental death or injury associated with turned-down end treatments in order to assess the cost-effectiveness of replacing them with different end treatments. Data for this study came from the Texas traffic accident database and police crash reports. In 1989, there were 190,512 police-reported collisions on state-maintained highways in Texas, including 4,047 that reportedly involved guardrails. From these 4,047 collisions, all collisions that resulted in fatalities were extracted. Next, every fourth non-fatal colli- sion (based on accident number) was extracted, resulting in a 25 percent sample of the non-fatal data set. The author reviewed the fatal and non-fatal collision reports and coded them based on two supplemental variables: the point of impact on the guardrail (turned-down end, not end, unknown, not guardrail) and whether the impacting vehicle rolled over. Of the 987 non-fatal collisions, 152 involved turned-down ends, 604 occurred at other known parts of the rail, 115 occurred at undetermined parts of the guardrail, and 116 collisions were actually non-guardrail collisions (e.g., they were mis-coded). Of the 100 fatal collisions, 32 occurred at a turned-down end, 46 occurred at other known points of the rail, nine occurred at undetermined points on the rail, and 13 were non-guardrail collisions. The resulting data were ana- lyzed using standard crash report variables. While the analysis of the data in this study is very thor- ough, the author is careful to mention possible uncertainties in the data collection method: (1) Determination of guardrail/ non-guardrail collisions and other classifications was based largely on the narratives contained within the police reports. The author pointed out that the narratives in many cases were vague, so collision classification was subject to misinterpre- tation. (2) Data collection in this study was retroactive in nature. Report findings were based on interpretations of the reporting officers at the site as recorded in the police report. (3) Unreported collisions were ignored. The author hypothe- sized this would create a lopsided view of the end-hit to not- end-hit ratio because of the presumed higher rate of collision severity associated with end terminals. North Carolina, 1993—Across-Median Collisions (18) In 1993, Lynch, Crowe, and Rosendahl prepared a report on a collision study involving across-median collisions in the state of North Carolina. The objectives of the study were to • Use collision histories to identify interstate locations that had an unusually high number of across-median collisions; • Determine what safety improvements could be made; • Develop a priority listing of the high incident locations; and • Develop a model that will help identify potentially dan- gerous locations on North Carolina Interstates based on 11 relevant variables such as median width, traffic volume, and other geometric and operational characteristics. Data for the study were collected from police reports on file for the period of April 1, 1988, through October 31, 1991. During this period, the North Carolina reporting form did not indicate median involvement. Collisions involving medians were, therefore, identified as follows: 3,121 collisions listing “run off the road left,” “head on,” “sideswipe opposite direction,” “hit fixed object,” or “hit other object” as the first harmful event were extracted from all the collisions during the study period. These collisions were then more thoroughly examined. Construction zone collisions, reports of vehicles entering the median and recovering with- out incident, and non-reportable collisions (no personal injury and less than $500 property damage) were eliminated from the data. This resulted in 2,922 collisions that were eligible for the study. Reports for the 2,922 collisions were copied and sent to the NC DOT Area Traffic Engineers. Each collision and loca- tion was investigated. Roadway characteristics and conditions for each location were collected using standard data collec- tion forms developed by the researchers. The resulting data were entered into a computer database. The researchers then screened the data and extracted cases where a vehicle crossed a median and encroached on oncoming traffic. The final num- ber of collisions used in the study was 751. Seventy-one of the 751 across-median collisions resulted in at least one fatality. This represents 32.2 percent of all fatal- ities occurring on North Carolina interstates. Twenty-four high-collision locations were identified for potential installa- tion of median barriers. A B/C analysis was used to rank the sites. The study was very concise and well organized. The use of computer forms to standardize the data collected was an excellent addition to the study. The use of area engineers to determine site characteristics illustrates the importance of site visits for in-service evaluations. Without the site data, it would have been difficult to make useful conclusions about potentially high-accident locations. Oklahoma, 1993—Guardrail End Treatments (19) In 1993, Gattis, Varghese, and Toothaker published a paper in Transportation Research Record 1419 describing an evalu- ation performed on exposed, turned-down, and flared-end treatments in the state of Oklahoma. The purpose of the 3-year study was to determine whether there were differences among the three devices based on vehicle vaulting, vehicle overturn- ing, and accidental death or injury. Data for the study were collected from police collision records on interstate, U.S., and state highways; accident reports; and video logs of the collision sites. In the case of the video logs, it was assumed that the collision site had not changed since the time of the collision (e.g., repairs were

performed on a replacement in-kind basis). Collisions were classified as approach or trailing end and by whether the col- lision was a presumed end hit, presumed-but-questionable end hit, or not an end hit based on the description in the police report. For the purposes of the study, only presumed and presumed-but-questionable end hits occurring at the approach end of the guardrail were used to determine the general per- formance of the end treatments. Because of the low number of collisions involving flared-end treatments, these collisions were eliminated from the data set. After the collisions were classified, the significance in the difference between the different end treatments was deter- mined based on the Games-Howell multiple comparison sta- tistic. The statistical data indicated there was no significant difference between the exposed and turned-down ends in terms of the proportion of severe (A+K) injury collisions to the total number of collisions, or in terms of injuries from rollover/vaulting of the vehicles for the two devices. Other observations were also noted within the report. Approximately 53 percent of the collisions occurred on the 10 percent of the roadway system with higher ADTs. The severity associated with rollover/vault collisions was signif- icantly higher than non-rollover/vault incidents. This article provides valuable insight into the correlation of collision severity with the performance of the vehicle dur- ing a collision. It also shows that a majority of collisions can occur on a limited segment of high-volume roads and com- pares the vaulting/rolling incidences between the exposed and turned-down end treatments. However, changes in data collection and comparison could have greatly enhanced the quality of the study. Spearing incidences should have also been considered due to the inclusion of the exposed end treat- ments in the study. This was the original reason for the devel- opment of the turned-down end and therefore should not be absent from the study. Data should have been collected in a way to allow on-site investigations of the sites. As stated in the report, the video logs did not prove to be useful and in some instances lowered the quality of the data collected. It was often unclear from the video log whether a blunt end, break- away cable terminal (BCT), or trailing anchor was involved. The conclusions, therefore, cannot be accepted with much confidence since there was so much uncertainty about the types of devices struck. South Carolina, 1994—Brakemaster and Crash Cushion Attenuating Terminal (20 ) In 1994, the South Carolina DOT prepared a report describ- ing an evaluation of the Brakemaster and crash cushion atten- uating terminal (CAT) used in the state. The purpose of this study was to examine the effectiveness of using these sys- tems to protect median bridge piers. This study was more of a pre-evaluation of the CAT and Brakemaster systems since no actual collisions involving the CAT or Brakemaster were documented. Instead, collisions 12 involving median guardrails were analyzed over a 2-year period. Fatalities, incapacitating injuries and property dam- age collisions were assigned societal costs based on the 1992 National Safety Council Formula (21). These costs were compared with the average installation costs of the CAT and Brakemaster during the same time period with the presump- tion that the injuries would have been prevented if these devices had been used at the site. In addition to the comparison described above, a review of other in-service evaluations involving the CAT and Brake- master was performed in order to evaluate how the CAT and Brakemaster were used in other states. COLLISION STUDIES In a collision study, police and maintenance records are collected concurrently with the study period. This type of evaluation does not include investigation of collision sites. Connecticut, 1977—Tire-Sand Inertial Barrier System (22) In 1977, Button described an in-service evaluation of the tire-sand inertial barrier system in the state of Connecticut. Two sites were used for the 2-year evaluation to obtain data concerning the performance and cost of the system. It is unclear how the data for this study were obtained, but photographs of the systems were taken prior to and after impacts to the device. Also, photographs of one of the vehicles were taken after a collision, indicating that perhaps both the police and DOT were involved with the study. It is assumed that the cost data were obtained from maintenance records. Four minor and two major hits were recorded on the sys- tem during the 2-year examination period. No injuries or fatalities were recorded for the six impacts. One impact caused a substantial amount of debris to be scattered on the roadway. Cost comparisons of the system showed the initial cost of the system to be six times lower than comparable systems. The evaluation showed the average cost per collision to be the highest of the three systems used for comparison (e.g., $195 more than the Fitch Barrel system). Repair estimates were based on one impact with the system, but not all impacts would be as severe so the average repair cost may have been overstated. Cost data should have been collected for the minor and unreported collisions to the system for a fairer comparison. A B/C analysis could have been performed to provide a more precise comparison among the barrier system alternatives. New York, 1977—Guardrail, Sign and Luminaire Supports, Impact Attenuators (23) In 1977, Carlson, Allison, and Bryden reported on a colli- sion study in the state of New York involving lightweight-

post guardrail, slip-base sign posts, frangible luminaire sup- ports, and impact attenuation devices such as sand-filled plastic barriers, water-filled cell sandwich units, water-filled cluster units, and empty steel drums. The purpose of the eval- uation was to document field performance of the lightweight- post barriers at New York’s newer mounting height and to investigate the field performance of the slip-base sign posts, frangible luminaire supports, and impact attenuation devices. Data collection for the study varied by the type of device being investigated. For the light-post barriers, data were col- lected over a 4-year period between 1971 and 1975 for all bar- riers constructed between 1969 and 1971 on state roads. It was assumed that these barriers would have the newer New York standard height of 661.5 mm to the center of the rail. Data were also collected for light-post barriers on the New York State Thruway during a 6-month period in 1973. The report did not say why the data collected from the Thruway only lasted 6 months or why this area was included in the study. At the beginning of the study, 47 selected breakaway sign supports and frangible-base luminaire supports were moni- tored. Few impacts occurred at these 47 sites. Therefore, the frangible-base luminaire supports and breakaway sign sup- ports along the New York State Thruway were added to the observation. The sites along the Thruway were then moni- tored for the 6-month period during which the guardrails were under observation. In addition to the Thruway, a 20.12-km section of I-90 containing 392 frangible-base luminaire sup- ports was also added to the data collection. All impact attenuation devices on state roads were moni- tored during 1971–75. It is unknown how many impact atten- uation devices were initially under observation, but by the end of the study, 70 devices were being monitored. Most of the data were collected using three separate forms developed by the researchers. One form was created for guardrail, luminaire, and sign support collisions occurring on state roads; one form for these same devices on the New York State Thruway; and the third form collected collision information for impact attenuation devices. All the latter were located either on state highways, in New York City, or under the jurisdiction of the Port Authority of New York and New Jersey. All these agencies used the same form for describing collisions involving impact attenuation devices. The two forms involving state-maintained roadways were provided to the NY-DOT Highway Maintenance Subdivi- sion. All information except repair data was provided by the foreman assigned to the site being repaired for the guardrail, sign, and luminaire support collisions occurring on state roads. The repair data, (e.g., amount of device reused, cost data, etc.) for these devices were provided by the Resident Maintenance Engineer. For collisions involving impact attenuation devices occurring on state roads, the Resident Maintenance Engineer filled out the entire form. When an impact attenuation device collision occurred in New York City or in the areas maintained by the New York and New Jersey Port Authorities, the same form was used but filled in by different personnel within these 13 agencies. In all cases involving guardrail, frangible-base lumi- naire supports, slip-base sign supports, and impact attenua- tion devices, an attempt was made to obtain a police accident report for the crash being investigated. This information, once collected, was then forwarded to the NY-DOT Engi- neering and Research Development Bureau. The form for collisions occurring on the New York State Thruway was handled by the Thruway Authority’s Traffic and Safety Engineer. A form was given to the state police officer reporting the collision, who filled in information about the type of barrier hit, the actions of the vehicle, etc., attached a copy of the police accident report, and forwarded this infor- mation to the Traffic and Safety Engineer. A form was also given to the maintenance supervisors in charge of the repair of the damaged devices. The maintenance supervisors filled in the same areas of the form as the police officers and also provided information on the repair materials needed for the device. The Traffic and Safety Engineer provided repair cost information, which was obtained from the Bureau of Account- ing. After all of the information had been compiled, the data were forwarded to the NY-DOT Engineering and Research Bureau. Collisions involving the luminaire supports on I-90 were obtained from the utility company. This information was then used to obtain police accident reports for these impacts, if avail- able. Cost information for these collisions was not recorded due to the nature of the utility company’s records. The performance criteria for the devices were based on the severity of injury experienced in the impact, the vehicle reac- tion, and the maintenance requirements needed to repair the device. It is important to note that unreported collisions were not included as a part of this study. Severity of injury was clas- sified as none, minor, severe, or fatal. Minor injuries included cuts, scrapes, and non-specific complaints of neck or back pain. Injuries were only classified as severe if hospital admis- sion was required. Vehicle reaction described whether the device performed as intended. Impact attenuators should gently decelerate a vehicle to a stop or redirect the vehicle. Sign and luminaire supports should easily release without excessive decelera- tion. Barriers were evaluated based on whether penetration had occurred. Post-impact vehicle trajectories were also noted for all the devices. Maintenance requirements included collision damage repairs and routine maintenance costs. Collision damage repairs included costs for equipment, materials, and labor. Routine maintenance was recorded such as winterizing the attenuators. Maintenance for the sign and luminaire supports was considered negligible and was not included in the study. A total of 392 collisions occurred involving roadside and median barriers. Minor and severe injuries were combined for each of the barrier types. The chi-square test was used to access differences among the barrier types at the 95 percent confidence level, both in terms of severity of injury and cost of maintenance and repair. Specifically, the box beam guardrail

experienced lower injury severity but higher maintenance cost than the W-beam guardrail. For the barriers, only the differ- ence between the W-beam and box beam median barriers was statistically significant. In general, the low number of serious injuries and the complete absence of fatalities indi- cated good performance for the barriers. Penetration rates were also examined for the roadside and median barriers. Although the reporting forms developed by the researchers contained three methods of penetration (over, under, or through), the collisions were classified only by whether or not penetration had occurred. Except for cable guiderail, penetration rates were relatively low. However, there were few mid-section hits for the cable guiderail. Of the 29 collisions involving all types of end sections, most impacts were unreported, indicating good performance of the barriers. Collision damage and repair costs were examined for the barrier impacts. The length of rail damaged tended to decrease as the stiffness of the rail increased. Therefore, cable guide- rail had the highest average length of barrier damaged in an impact. However, the difference in repair costs among the different rails was very minor. Ten hits were recorded involving slip-base sign supports. Three impacts resulted in injuries with two minor and one serious. The base appeared to have performed properly in all 10 collisions. Due to the high incidence of secondary colli- sions associated with this device, pole repair data were com- bined with repair costs of other roadside features. Therefore, no cost data were available for the sign supports. Seventy-eight collisions were recorded involving frangible- base luminaire supports. Fifty-nine of these were investi- gated by the police and 19 incidents were unreported. A total of 15 incidents resulted in injury with 12 minor and 3 severe injuries. Eleven of the 15 injury collisions involved secondary collisions. The base also failed to release in a few instances. However, this was attributed to the low-impact speeds of the vehicles in these cases. Few repair data were available for the same reasons as the slip-base sign posts. In those instances where repair data were available, the cost to reuse the same pole averaged $362 while the total cost to install a new one averaged $715. The utility company made the repairs on I-90 so cost data for this area were unavailable. A total of 393 collisions involving impact attenuators were recorded during the evaluation period (1971–76). The 35 sand barrel installations were impacted 242 times, the 21 Hi-Dro cell sandwich installations were impacted 63 times, 84 impacts occurred at the 11 water-filled cluster installations, and four impacts were recorded at the three steel barrel installations. Four of the impacts with sand barrels resulted in injuries (three minor and one severe). Four impacts with the Hi-Dro sandwich units resulted in injury (three severe and one fatal). The fatality in this case was attributed to an older Hi-Dro design and poor installation practices at that location. Fifteen impacts with the water-filled clusters resulted in injury (14 minor and one severe). No injuries were recorded at the empty steel drum sites. 14 Maintenance costs were also considered for the attenua- tors. Attenuators repair costs ranged from $18 to $2,718. In general, the Hi-Dro cell sandwich units were the least expen- sive to repair, while the sand barrel systems were the most expensive to maintain. The study provides extensive information about the devices under observation. The following modifications to the study would have enabled the information to be presented in a more concise manner: • Narrowing the data collection to one area would have reduced the coordination required to obtain data from the different agencies. The inclusion of the I-90 data effectively eliminated repair data for all the luminaire supports and increased the number of agencies involved (i.e., the utility company). • A uniform data collection period is needed for the devices being studied. The inclusion of the Thruway may have improved the data if the data collection period had been the same as the rest of the areas. The short data collection period of this area did not take into account variances in weather, change in vehicle fleet, etc. • An explanation of why the Thruway was incorporated into the study would have been helpful. It is unclear why this area was included in the study. The short collection period of this area made the data obtained irrelevant. Kentucky, 1984—Breakaway Cable Terminal and Median BCT (24) In 1984, Pigman, Agent, and Creasey reported on the in- service performance evaluation of the breakaway cable termi- nals (BCTs) and median breakaway cable terminals (MBCTs) in Kentucky. Kentucky began using the BCT and MBCT in 1974. When the study was performed in 1983, there were 3,633 BCTs installed on Kentucky roadways with an average installation cost of $515. In addition to the BCTs, there were 573 MBCT installations with an average installation cost of $627. The purpose of the in-service evaluation was to deter- mine if the BCT and MBCT were performing as expected in real-world collisions. The number and location of BCT and MBCT collisions in 1980–1982 were identified using Kentucky’s crash-reporting system. An inventory of BCT and MBCT installations was apparently developed using the accident database. This inven- tory was used to select routes that had a history of BCT and MBCT collisions. The study team then contacted maintenance engineers in the districts containing the selected routes (Ken- tucky DOT Districts 5, 6, and 7) and established a notifica- tion procedure. Whenever the maintenance engineer became aware of a collision involving a BCT or MBCT, the study team was contacted. The report is unclear whether the data collection was performed by the study team or whether the data were collected by the maintenance staff and forwarded to the study team. The study team also obtained information

from the district operations managers on prior (i.e., before 1984) BCT and MBCT collisions. This involved searching the maintenance files for police reports, repair reports, and photographs of damaged installations. The study group also collected some data serendipitously. When they observed damaged BCT or MBCT installations while driving around the state, the research team took pho- tographs and searched the police crash reports for a case that matched the location. In some instances they were able to match the damaged BCT or MBCT to a police report, while in other cases no record of a collision at that location could be found. In general, it appears most of the data were col- lected “after the fact” by searching through DOT mainte- nance records. Data collected were of three types: police crash reports, maintenance reports, and photographs. The police reports seemed to provide the most detailed information. It is unclear what specific information was contained within the mainte- nance reports although the authors reported repair costs rang- ing from $730 to $920. Interestingly, there were many cases where barrier repair cost exceeded the average initial cost of the barrier installation. This suggests that the repair cost may include guardrail repairs outside the terminal area. Photo- graphic data were also limited. Generally, no photographs of collisions were found in maintenance files and repairs were often made before a study team member could photograph the damaged installations. Fifty BCT and 19 MBCT collisions were identified using this procedure. Of the 69 total collisions, police crash reports were obtained in 50 cases, repair forms in 33 cases, and pho- tographs in 33 cases. A complete set of data (i.e., police report, maintenance report, and photographs) was found for only six of the 69 collisions. The purpose of the study was to assess performance of the BCT and MBCT barriers. To do this, the study team con- structed criteria for defining proper performance. Proper per- formance was defined by whether the first breakaway post failed and no spearing of the vehicle occurred. Using these criteria, the performance was judged to be “proper” in 75 per- cent of the BCT cases and in 50 percent of the MBCT cases. Seventy-six percent of the BCT cases and 75 percent of the MBCT cases resulted in a serious or fatal occupant injury. It is interesting that in many cases where the barrier perfor- mance was classified as “proper,” the collision was classified as “severe.” While the Kentucky study helps provide some much- needed information about BCT and MBCT collisions, the sampling technique and type of data collected limit the information that can be extracted from the data. For example • The data sampling method was not rigorous. It is not clear how thoroughly or systematically maintenance records were searched. • Cases were collected sporadically from different years, some as early as 1977 and some as late as 1983. If the 15 study team came across a BCT or MBCT collision they included it regardless of when or where the collision occurred. Adding serendipitous information polluted the statistical validity of the data set. • The inventory was based on collision data, so there was no real assessment of how many devices were actually covered by the study, and no systematic method to check the accuracy of the “collision-based” inventory. • No attempt was made to assess whether the BCT or MBCT was installed or located consistent with the Ken- tucky State Standards. Several cases were noted where no BCT flare or an incorrect offset occurred and it seems likely that many other subtle installation deficiencies were present but not recorded. Without knowing which devices were correctly installed and which were not, it is difficult to compare the performance of the BCT specified in the state’s standards. Since the data sampling technique was not systematic, the percentages of “properly” and “improperly” performing devices and the percentages of “severe” and “un-severe” col- lisions were not reliable. North Carolina, 1988—Self-Restoring Barrier Rail (25) In 1988, Strong prepared a report describing an in-service evaluation of the self-restoring barrier rail (SERB) in the state of North Carolina. The project was initiated as a part of FHWA’s Demonstration Project Number 64, “Traffic Bar- rier Systems.” The purpose of the project was to evaluate the SERB under real-world conditions. It is unclear how the data were collected for the project, but it is assumed the collision data relied solely on police and maintenance-reported incidents. Maintenance personnel were interviewed to determine how the device was installed and to estimate the expenses required to repair the device during its 2-year evaluation period. Field investigations were made of the site before and after the installation of the SERB. Collision data were collected from the period of January 1977 to June 1988. During the period prior to the SERB’s installation, eight reported colli- sions occurred at this location. During the study, three colli- sions were reported and numerous unreported collisions were noted from the field investigation performed at the close of the project. Repair costs from the previous device were noted for com- parison with the SERB’s repair costs. However, upon inter- viewing the maintenance engineer, it was found that minimal repairs were needed to the SERB. The field inspection later revealed an impact which damaged the SERB to the point of repair, possibly from an unreported incident. Strong’s report on the evaluation of the SERB in the state of North Carolina was relevant and informative. This study, cou- pled with other studies involving the device, gives valuable

insight into the utilization of the SERB in real-life applications. However, this study alone could have been more comprehen- sive with a few inclusions and changes to the evaluation: • The only field evaluation, which was performed at the end of the evaluation period, revealed evidence of numer- ous unreported incidents. Periodic site investigations could have produced a more accurate picture of the total number of impacts with the device during the study. • More SERBs could have been installed and evaluated at different sites to increase the number of impacts and more accurately determine the SERB’s performance under in-service conditions. • No indication was given of the condition of the occu- pants of the vehicle in the impacts with the SERB. Types and severity of injuries incurred can be indicators of the performance of the device being evaluated. The inclusion of these suggestions could have altered the data in the study and given broader insight into the field per- formance of the self-restoring barrier rail. Colorado, 1989—IBC Mark VII Barrier System (26) In 1989, Woodham prepared a report describing an in- service evaluation performed on the IBC Mark VII barrier system in the state of Colorado. The purpose of this evalua- tion was to determine the performance of the barrier under field conditions and evaluate the costs associated with the use of the barrier. Data for this study were collected from police crash reports and maintenance records. Sample police reports were included in the final report, along with a breakdown of the costs incurred from the use of the barrier during the 3-year evaluation period. Nine known collisions occurred with the barrier during this time period, two of which required repairs to the barrier. Costs to repair the barrier in these two instances proved to be relatively high compared to the New Jersey concrete safety shape. In addition, the initial costs to install the barrier were twice that of the pre-cast concrete barriers. North Carolina, 1989—GREAT (27 ) In 1989, Stanley and Strong reported on an evaluation of the in-service performance of the GREAT system in North Carolina. The purpose of the study was to determine the performance of the device in field situations and the cost- effectiveness of the device from an initial cost, maintenance, and repair standpoint. The performance of the GREAT sys- tem was then to be compared to other devices used within the state. 16 Collision data for the study came mainly from police crash reports. The maintenance personnel were interviewed to deter- mine the number of unreported incidents and the costs associ- ated with installation, maintenance, and repair of the devices. The results of these interviews were summarized on a one- page report sheet. Nineteen collisions occurred with GREATs during the October 1984 to early 1989 evaluation period. Although the GREAT system comprised 80 percent of the installed devices used in the state and in the study, no reported impacts occurred with the device. Instead, most impacts occurred with one of the devices intended to be compared to the GREAT system. One unreported impact did occur with the GREAT system, and it performed satisfactorily. Due to this unexpected result, the report then focused on the performance of the device most often struck (15 hits in all), the Hi-Dro Sandwich system. Arizona, 1990—TREND and SENTRE (28) In 1990, Lattin reported the results of an in-service evalu- ation of the TRansition END treatment (TREND) and the Safety barrier ENd TREatment (SENTRE) in Arizona. When the evaluation began, only two terminal designs were speci- fied by the Arizona DOT (AZ-DOT). These were the stan- dard BCT and Arizona’s standard attenuator assembly. At the time of their installation, the SENTRE and TREND sys- tems were classified as experimental under the old FHWA classification system. The purpose of the project was to deter- mine the in-service performance of these two systems. Only one site was used for each of the two systems. At both sites, attenuators were installed on both the upstream and the downstream ends of a bridge. Therefore, there were four installations of each barrier for a total of eight. Prior to the installation of the systems, each site was inves- tigated for previous collision history using information from the AZ-DOT Traffic Studies Branch. Collision totals were obtained from the period of 1973-88. For the TREND site, there were three collisions during this period. For the SENTRE installation site, 55 collisions were reported. When Lattin’s report was prepared, there had been only four collisions at the sites. These collisions all involved heavy con- struction vehicles from surrounding projects and the author did not feel these events were typical for what the attenuators would normally experience. No conclusions could be made about the in-service performance of either system in more typ- ical situations. Installation costs were totaled and compared with the AZ- DOT’s estimates for other systems such as the BCT and the standard AZ-DOT attenuator. A BCT (including guardrail) at one site was estimated at $4,019. The cost for an AZ-DOT attenuator was estimated at $3,558. These costs were signif- icantly lower than the installation costs of the SENTRE and TREND at $7,421 and $8,600 respectively. However, it was reported that the performance of the standard barriers was not

equivalent to the SENTRE and TREND systems, making direct cost comparisons questionable. It was suggested that a system of weighting factors be assigned to each design objec- tive to differentiate the importance of each in the cost com- parison and the overall evaluation of the systems. This evaluation does a good job of including site-specific concerns in cost comparison. Sites with a higher incidence of impacts should have been chosen for an in-service evalua- tion. In this case, the TREND site had experienced only three collisions in a 15-year period. It was highly probable that no impacts would occur with this system during the 2-year eval- uation period; therefore, no concrete conclusions could be made about this system based on this study. Site selection in this study seems to have negatively affected the ability to collect data. For the purposes of an in- service evaluation, a site with a high likelihood of impacts should be selected. More impacts would therefore most likely occur and reasonable estimates could be made regarding the performance and maintenance costs of the systems. Iowa, 1990—IBC Mark VII Median Barrier (29 ) In 1990, Marks described an in-service evaluation of the International Barrier Corporation’s Mark VII median barrier (IBC Mark VII) in Iowa. The purpose of the evaluation was to determine the cost and performance of the barrier in the field and compare them with those of the New Jersey con- crete safety shape. Data collection methods were not documented in the report other than the mention of an annual field inspection. It is assumed from the contents of the report, however, that maintenance personnel were interviewed during the course of the study. Data from only one site were used for the study. During the 4-year evaluation of the barrier, no severe impacts occurred. Upon performing the yearly inspection of the barrier, evidence of four minor impacts where the barrier was scraped or dented were found. Due to the minor nature of the impacts encountered during the evaluation period, no maintenance was necessary during the study. The relatively high initial cost of the barrier was noted, however. This study presents important information pertaining to the installation of the IBC Mark VII and its performance with minor impacts. Due to the limited number of impacts expe- rienced and the relatively long evaluation period performed in this case, there is evidence that the study methodology could have been improved: (1) No indication of a preliminary study for site selection was given in the report. A more suit- able site, with a higher impact probability, should have been used if available. (2) If economically possible, more sites with high-impact probabilities could have also been used to help increase the chances of more severe impacts with the barrier. These types of impacts would provide more insight into the performance of the barrier and give some indication of the repair costs associated with the use of the IBC Mark VII. 17 Alaska, 1991—Side Slopes (30) In 1991, Botha described a study performed to determine cost-effective side-slope safety countermeasures for Alaska’s highways. The purpose of the study was to determine the relationship between slope steepness, embankment height, and other factors relevant to side-slope design and collision severity. Data collection utilized existing police-reported collision data. Initially, collisions were examined for the years 1985 and 1986 and restricted to roads in northern Alaska. Later, more funds became available and the data collection period and area were expanded. This new data included the initial region along with data from the Parks Highway in central Alaska for the years 1984 to 1987. The objective of the collision search was to identify colli- sions in which a side slope had been involved. This task used the Alaska DOT computerized police crash database. Ini- tially, very broad categories were used to narrow the number of collisions to 1,077 records. Records were then extracted based on whether a guardrail was present, the vehicle hit a fixed object, damage occurred prior to the vehicle leaving the roadway, the collision occurred at an intersection, and other vehicles were involved. After this process, 538 sites remained for further investigation. Each of these sites was investigated to determine the height and slope of the side slope involved in the collision. During the course of this process, some sites were eliminated due to the presence of guardrail and other objects. A total of 330 sites was left for analysis. The data were used to perform a site cost analysis to deter- mine if a correlation existed between the cost of a collision and the side slope. Collision cost was determined by injury severity. These costs were analyzed using an extensive pro- cedure described in the literature, and a correlation coeffi- cient was determined to relate side slope to cost of collision (i.e., collision severity). A correlation coefficient of 0.33 was obtained from the data. However, in some instances, a nega- tive correlation coefficient was determined, indicating that the cost of the collision decreased as the slope associated with it increased. These results were attributed to the high cost associated with fatalities (700 times greater than prop- erty damage only [PDO] collisions). Despite this, the corre- lation between side slope and collision cost or severity was determined to be relatively low. Further studies were recom- mended to develop a more reliable relationship between side slope and collision severity. Indiana, 1992—Vehicle Attenuating Terminal and Crash-Cushion Attenuating Terminal (31) In 1992, Gendron prepared a report on the in-service eval- uation of the vehicle attenuating terminal (VAT) and the crash-cushion attenuating terminal (CAT) in the state of Indi- ana. In 1988, it had been decided that the VAT was to be used and evaluated in the state. To create a reliable sample base, a

large number of VATs were installed within Indiana. Later, when the CATs were installed within the state, they were added to the study. The purpose of the study was to collect data on the ease and cost of installation, maintenance, and repair and collision performance. Data collection for the study was obtained primarily from on-site investigators, although it is assumed some communi- cation with local law enforcement agencies occurred because information typically provided on police crash reports was included in the data. In addition to the reports by the site investigators, photographs were also taken. During the 3-year study period, 21 impacts were investi- gated and reported. Some of the collision summaries in the report were incomplete because of removal of evidence prior to the investigator’s arrival on the scene. The report includes many field problems experienced with the installation of the CAT and VAT such as splicing difficul- ties between the 10-gauge and 12-gauge guardrail sections. The report also mentions possible problems with the design that were associated with the one fatal collision in the study. In addition, collision summaries included detailed installation discrepancies experienced in the impacted devices. New York, 1992—Light-Post Barriers (32 ) In 1992, Hiss and Bryden reported on a collision study of light-post barriers in the state of New York. The purpose of this investigation was to perform an evaluation of these devices and to relate collision severity to mounting height of the barriers. Data for this study were collected over a 1-year period from July 1, 1982, to June 30, 1983. Accident reports were the primary data source. On-site investigations were used to determine highway and barrier characteristics at the impact sites. Unfortunately, these measurements were typically made 1 to 2 years after the impact and repair of the barrier so it is questionable whether the measurements reflected the state of the system at the time of the collision. A total of 1,726 colli- sions involving light-post barriers were identified during the evaluation period. Six types of barriers were examined in the study. Cable, W-beam, and box beam guardrails were investigated along with cable, W-beam, and box beam median barriers. These devices were rated according to occupant injuries, post-impact vehicle trajectory, and secondary collisions. These factors were then subdivided based on barrier height. The box beam guardrail experienced 623 collisions and the median barrier experienced 308 impacts that were inves- tigated in the evaluation. Collision severity did not vary sig- nificantly over the range of the barrier heights (i.e., 533 to 838 mm). Containment of the vehicles for the box beam sys- tem was high with a containment percentage exceeding 90 percent in nearly all the height categories. This device also performed similarly for a number of secondary events with a 4.3 percent overturn rate and a 16.7 percent fixed-object hit 18 rate. Similar observations were made for the box beam median barrier. Weak-post W-beam barriers were also evaluated based on 306 guardrail and 46 median barrier collisions occurring dur- ing the study period. Injury rates increased when the height of the barrier was below 762 mm. Redirection rates were high for barrier heights above 584.2 mm. The chance of a secondary event occurring was higher when the barrier height was below 685.8 mm. The relatively small sample size for the weak-post W-beam median barriers prevented the researchers from prop- erly evaluating the performance of this device. Based on the 427 cable guardrail and 16 median barrier collisions, it was determined that injury rates were relatively unaffected by barrier heights over 588 mm. Vehicle trajec- tory and the incidences of secondary collisions were lowest in the 588 to 710.5 mm height range. However, when the bar- rier height exceeded 710.5 mm, an increase of adverse vehi- cle trajectories was noted. The relatively small sample size for the cable median barriers prevented the researchers from evaluating the performance of this device. A recommenda- tion was made to set the standard center of weak-post barrier rail heights to 588 mm based on the results of this evaluation. Colorado, 1993—Brakemaster, CAT, and 10-gauge Guardrail (33) In 1993, Outcalt reported on an in-service evaluation of the Brakemaster, crash cushion attenuating terminal (CAT), and 10-gauge guardrail systems. The objectives of the study were to evaluate the devices under field conditions and to determine the installation, maintenance, and repair costs asso- ciated with the use of the devices. Data collection for the Brakemaster and CAT systems involved communication with maintenance personnel. Also, because of the descriptive nature of the collisions mentioned in the report, it is assumed that police reports were used in the study. For the 10-gauge guardrail evaluation, only data from the maintenance forces were used to determine the guardrail performance. During the course of the evaluation, three impacts were recorded involving the Brakemaster systems. One injury and no fatalities resulted from the impacts, and one vehicle was able to drive away after colliding with the terminal. In addi- tion to these collisions, one instance was documented where a Brakemaster was damaged from an impact with a boulder. Only one site was used for the evaluation of the CAT. No collisions were documented during the 2-year evaluation period. Two types of 10-gauge guardrail, galvanized and corro- sion resistant railing, were examined in the evaluation. Cor- rosion resistant rail is used in Colorado for its aesthetic appearance. The main objective of the 10-gauge rail evalua- tion was to determine if the use of the thicker railing would be advantageous at sites that frequently need repair. Based on the interviews with maintenance personnel, the railing is as

easy to work with as 12-gauge rails and does not require as much maintenance as the thinner 12-gauge railing. Michigan, 1994—Breakaway Cable Terminal (34) In 1994, Morena and Schroeder reported the results of an in-service evaluation on breakaway cable terminals (BCTs) in the state of Michigan. Since the late 1970s, the BCT had been the standard guardrail end treatment in Michigan. When the study began, there were over 14,000 BCTs in the state. At the time of the BCT’s acceptance as a standard in Michi- gan (i.e., 1970s), the BCT’s poor performance with the 820 kg test vehicle was considered moot since a more suitable end treatment was not available. By the early 1980s, how- ever, more expensive end-treatment alternatives had been developed that provided better performance. The purpose of this study was to examine the in-service performance of BCTs in Michigan and to provide a basis for a subsequent cost-versus-performance comparison. Data were collected during calendar years 1984–86 and 1988–90. The first data collection period relied initially on information obtained from maintenance records. The study group then developed a one-page reporting form for main- tenance personnel to fill out each time a BCT was repaired. The forms were collected and cross-referenced with match- ing police crash records. During the matching phase of this process, additional collisions involving BCTs were discov- ered in the police crash reports and added to the data. Most of the data came from police crash records. The main- tenance reports appear to have been used primarily in the first data collection period to gather information regarding instal- lation procedures and costs. Installation details were not checked in the second data set, and it is uncertain whether any maintenance records were used as data sources in the second phase of the study. It is also difficult to ascertain what infor- mation was available from the accident reports and mainte- nance forms because no sample forms were included in the final project report. No site visits were made during the course of this study. Fifty collisions were observed in the 1984–86 period and 83 were observed in the 1988–90 period. Of the 50 collisions in the 1984-86 data set, 24 were matched to police accident reports. A majority of these collisions (42) happened on free- ways while the remaining occurred on non-freeways zoned for 55 mph. All 83 impact events in the 1988–90 data occurred on two segments of interstate highways. These two segments were upgraded to BCTs in the mid-and late-1980s so the guardrail terminals better reflected then-current MI-DOT policy. Once a collision was identified, five steps were required to collect the data: 1. The end treatment struck was first verified from MDOT’s inventory. 2. Verification of an end hit was made based upon the description in the accident report. 19 3. Collisions where a secondary hit was involved were eliminated. 4. Car size was determined along with vehicle impact point, spin prior to, and roll after the impact. 5. The degree of injury and seat-belt usage were deter- mined. A higher percentage of drivers were injured (60 percent) in the 1984–86 data than in the 1988–90 data (35 percent). Changes in the vehicle fleet, seat-belt usage, number of non- freeway collisions, and improved BCT design and installa- tion were identified as possible reasons for the improved performance observed in the 1988–90 data. Examination of 1984–1990 general collision data showed similar trends dur- ing this time period. The reduction in driver injury and fatality percentages between the two data sets may have been due to passage of a mandatory seat-belt law in 1985, and improved BCT design and construction techniques. The authors used these factors to justify elimination of the first data set; therefore, the study’s findings and conclusions were based primarily on belted drivers impacting BCTs between 1988 and 1990 (i.e., the second data set). The first data set was used, however, to explore the advantage of the flared BCT over an earlier tan- gent design. Based on the data, occupants impacting a tan- gent BCT design were two to three times more likely to be injured than those impacting a flared design. Further examination showed the percentage of occupants involved in a collision and reported wearing seat-belts was much higher (nearly 3 to 2) than known percentages in the overall vehicle fleet. Fines for not wearing safety belts may have caused occupants to report having seat-belts engaged during a collision when, in fact, they were not. The study group thus concluded that the belt-use data may have been contaminated. The study team compared driver injury to BCT impact, type of vehicle, point of impact, right/left side placement of the guardrail, and spin/roll of the vehicle. The following con- clusions were made based on the results of these comparisons: 1. Vehicles impacting a BCT were more likely to result in passenger injury than vehicles impacting a guardrail (8
5), 2. BCT performance improves as vehicle weight increases, especially for vehicles greater than 1,020 kg, 3. Left-side impacts with BCTs are more likely to result in driver injury than impacts involving other regions of the vehicle, 4. Vehicles spinning prior to impacting a BCT did not have a higher risk of driver injury, 5. Ten collisions were observed where the vehicle rolled over after an impact, 6. 67 percent of drivers purportedly wearing safety belts still received moderate to severe injuries, and

7. Impacts involving BCTs located on the right side of the road resulted in more moderate or severe injuries (27 percent) than those located on the left side (7 percent). It was hypothesized that better grading in the median areas resulted in more consistent application of flared BCTs than on the shoulder. New Hampshire, 1994—Modified Eccentric Loader Terminal (35 ) In 1994, McDevitt summarized an evaluation of the Mod- ified Eccentric Loader Terminal (MELT) in the state of New Hampshire. The MELT was used to replace the outdated turned-down ends used prior to this study. The degree to which the various agencies were involved in the report was not specified. However, it is known that on-site photographs were taken and police accident reports were collected. Main- tenance personnel were also used in the study, although it is not known how. The period of the data collection was not specified in the report, although the report includes police crash reports from 1991 to 1993, a 3-year time span. During this time, approxi- mately 25 collisions occurred involving the MELT. No fatal- ities or major injuries were reported in any of the docu- mented collisions. In addition, no spearing, vaulting, or other negative vehicle behavior was reported. Most of the MELT units which were struck during this period went unreported because the impacts were minor. Consequently, valid cost information for the repair of the ter- minals was unavailable. However, it was reported that the cost of a new MELT was $1,200. This was considerably less than the other terminals such as the ET-2000, SENTRE, and the Brakemaster under consideration by New Hampshire for replacement of the turned-down ends. Problems occurred with the device when snow plows hit the terminal, knocking the rail off the shelf clip. To remedy this problem, more shelf clips were added to the device. More careful construction of the device was also recommended. IN-SERVICE EVALUATIONS In this report, an “in-service evaluation” includes exami- nation of the collision sites shortly after a collision occurs. These studies also usually use police crash records and main- tenance records as data sources. Connecticut, 1977—Frangible-Tube Bridge Barrier System (36 ) In 1977, Lane reported on an evaluation of the Frangible- Tube Bridge Barrier System used in the state of Connecticut. This system was based on an idea originating from NASA and applied to the area of roadside safety. The purpose of the 20 2.5-year evaluation was to observe the construction and monitor the performance of the barrier system. Data were collected for a 3-year period prior to the sys- tem’s installation (i.e., September 1971 to September 1974) to provide the basis for a before-after study. These data were obtained from the state’s police department and the traffic engineering services section within the state. Although not directly stated, it is assumed no effort was made to determine the number of unreported collisions during this period. Data were collected from multiple sources. Reported colli- sions were obtained from police data. Field inspections were performed at regular intervals to help determine the number of unreported collisions. The effects of winter maintenance, the change in skid numbers for the period, and initial installa- tion and repair costs for the system were also noted. Data for the evaluation of the frangible tube system were collected over a period of 2.5 years from July 1974 to Feb- ruary 1977. All data obtained for the study were acquired from one site. During this period, nine collisions involving ten vehicles were documented. Two of the vehicles were driven away, two minor injuries were reported, and a drop in the total number of reported collisions for this site was noted, even though the AADT increased during this period. New York, 1977—Guardrail Systems (4) In 1977, Van Zweden and Bryden reported on an in- service evaluation of the light and heavy-post guardrail sys- tems used in the state of New York. The heavy post barriers consisted mainly of W-beam guardrail while the light-post barriers included cable, W-beam, and box beam systems. The purpose of the study was to determine the performance of the then newer light-post and older heavy-post barriers based on actual collision experience in New York. The sec- ondary objective of the evaluation was to expose problems associated with the field use of the light-post barriers. Data were collected over a 2-year period for each of the barrier systems but was not concurrent. Collisions on the entire state highway system, which involved all barrier types, were evaluated during the period of 1967–69. Data on median barrier collisions on the New York State Thruway were also collected during the same period. Collision data along the Taconic State Parkway were collected for a period between September 1968 and December 1970. Multiple data sources were used for this study. Maintenance foremen completed a “barrier accident reporting form” for all barrier collisions occurring on state highways and the Taconic State Parkway during the study periods, which included basic data such as barrier type, length of rail damage, and rail pen- etration. They also provided information on repair costs. Police officers investigating collisions on the Taconic State Parkway were asked to fill out a one-page supplemental form in addition to the usual collision report.

Occupant injury was used as the primary means of char- acterizing the barrier performance. Collisions were classified based on the most severe injury occurring within the event. Initially, injuries were classified into the five categories of fatal, hospitalization required, minor, none, and unknown. As the study progressed, however, the researchers noted that many injuries were being classified as unknown, which sug- gested more minor injuries. The classification was revised into three categories where “minor,” “none,” and “unknown” were combined into “other.” Vehicle characteristics and behavior were also used to evaluate the performance of a particular system. Vehicles were classified by whether they penetrated the barrier, where the impact occurred on the system (within 15.2 m of end or in the middle), and the weight of the vehicle. The weight classi- fication was intended to differentiate passenger car impacts from truck and service vehicle impacts. The barrier type was also recorded. In reviewing standards for the heavy-post barriers, the researchers discovered that a minimum of 22 different combinations of rails, posts, and blockouts were used. Therefore, the heavy-post barriers were classified only by the rail type used and the placement of the barrier (i.e., normal or median). Most of these types of heavy-post barriers are now considered obsolete. Three types of data were collected to determine the main- tenance performance of the barriers: the number of posts reset and replaced, rail length re-erected or replaced, and the length of barrier damaged per collision. Additionally, repair costs for some collisions were also recorded to determine average repair costs for each barrier type. Additional factors affecting barrier performance like the impact angle of the vehicle, braking prior to impact, vehicle type, and roadway geometry were also considered. Due to the difficulty in coding these items, however, they were excluded from the data collection. During the 2-year collection period, 4,213 collisions involving guardrails were reported, including 717 light-post crashes and 3,496 heavy-post impacts. Comparisons were made between the various barrier types and differences between barrier types were analyzed using the Chi-square contingency test. Specific results are beyond the scope of this synopsis; however, general results are as follows: • Fatality and serious injury rates were lower for light- post barriers, but rates for median barriers of both types were nearly identical. • End-section impacts resulted in higher injury rates than mid-section impacts. • In general, light-post barriers were penetrated less fre- quently than the heavy-post barriers. There was also a positive correlation between penetration and the sever- ity of a collision. • Field investigations made during the course of the study showed that the low mounting height of the light-post barriers could have contributed to vehicle penetration. 21 The New York study was very thorough and well organized, especially considering its early date, 1977. A large sample and statistical analysis helped provide a degree of confidence in the results. A few more minor revisions to the methods of the data collection could have improved the evaluation: • Data for the different barrier types should have been collected concurrently. This would not have been as important if the types of barriers being examined were thoroughly mixed within the study areas. However, since the Taconic Parkway study involved primarily the box-beam median barrier, comparisons with other sys- tems may not be valid due to unconsidered factors that change over time. • Comparisons between the systems may also not be valid due to differences in installation details and location. The light-post barriers were generally installed on newer roadways while the heavy-post systems were normally installed on older highways and involved a variety of obsolete designs. Care should be taken to ensure valid comparisons are made between different types of barri- ers installed at different locations. Connecticut, 1980—Connecticut Crash Cushion (37) In 1980, Carney and Larson reported on an in-service inves- tigation of the Connecticut Crash Cushion truck-mounted attenuator in the state of Connecticut. This portable device is mounted on the rear of vehicles protecting minor maintenance activities. Prior to 1977, Hydrocell units served this purpose, but they were heavy and did not perform adequately in field tests. The purpose of the evaluation was to investigate the ease of installation, removal, and replacement; the cost of con- struction and repair; and the level of acceptance by the main- tenance personnel. At the onset of the 3-year study, three devices were in use, increasing to eight by the end of the evaluation. All existing devices were observed for impacts during the course of the evaluation. Data for the study were mainly collected from mainte- nance personnel already on site when the impacts occurred. Data were also gathered from accident reports filed by the maintenance personnel after the impacts. Photographs and schematics of the collisions were included to help illustrate device performance. Three impacts occurred during the evaluation period. Infor- mation about the first impact was not available, so descriptions and conclusions were based on the two later collisions. No major injuries and no damage to the maintenance vehicles were reported for these impacts. The repair cost estimate for the unit in the third impact, based on 1980 data, was $1,323. The original cost of the device in 1977 was $2,000.

Indiana, 1980—Breakaway Cable Terminal (38 ) In 1979, the state of Indiana was concerned about the effec- tiveness of the breakaway cable terminal (BCT) because of a large number of serious collisions involving BCTs. An in- service evaluation was performed to determine if the BCT was performing as expected. The study had three objectives: 1. Determine if the barriers were being installed properly, 2. Observe the field performance of the terminal, and 3. Identify possible design and installation procedure changes that might improve BCT performance. Data were collected primarily by highway maintenance personnel. Once a collision location was identified, a field inspection was performed at each site, the damaged BCT was photographed, police crash reports were obtained, and police photographs of the damaged vehicles were collected. All per- tinent information from these sources was then condensed into a one-page report form developed for the study. Wit- nesses and occupants of the vehicles were not interviewed. This procedure initially provided 21 impacts for examina- tion over a 1-year period. Two collisions were eliminated because they involved tractor-trailer impacts for which the BCT was not designed. Nine of the remaining 19 collisions were eliminated because the collision date was not known. Without the date, it was difficult to obtain the accident report. Photographs of 10 collisions were identified for further study. Six of the 10 collisions (60 percent) involved fatalities. Five of these fatal collisions involved the guardrail penetrating into the vehicle’s passenger compartment. Based on these data, the author proposed changes to the design of the BCT including the diaphragmed end. New Jersey, 1980—Breakaway Cable Terminal (39) In 1980, Baker reported on a breakaway cable terminal (BCT) in-service evaluation in the state of New Jersey. When the BCTs were first installed there in 1976, the FHWA requested that an in-service examination of the device be per- formed. The New Jersey DOT performed a 2-year study of the BCTs in the state to collect field performance data for the BCT and compare these impacts with experimental crash tests. Fifty BCT sites were initially selected for study. Organi- zational difficulties with this number of sites necessitated reducing the number of sites to four. The small number of collisions observed at these sites prompted the researchers to ask NJ-DOT maintenance personnel to keep track of all BCT sites in one of New Jersey’s maintenance districts. Thirteen major collisions and six minor collisions were examined. Types of data included on-site investigation of all collision sites, accident reports involving impacts on or near a BCT, and photographs of the damaged BCT and the collision site. These data were condensed to a one-page collision summary form developed specifically for this study. 22 While some cases exhibited performance similar to crash tests, many of the impacts involved spearing, rebounding, displaced footings, and ramping of the vehicle. The displaced footings were attributed to poor installation practice, and the rebound phenomenon was thought to be due to the corrective actions of the driver. The remaining factors, however, were attributed to site problems or poor BCT design. The study provided useful information on the field per- formance of the BCT including: (1) a correlation between unflared BCTs and spearing and (2) a correlation between BCTs placed near curbs and vaulting of the vehicle. The sam- pling method used seems to have resulted in a bias toward more severe crashes. Different maintenance garages may have interpreted the study objectives differently resulting in an ad hoc sample. There were also many cases where incorrectly installed BCTs were observed but the data was not segre- gated by proper or improper installation. Colorado, 1988—Type 3F End Treatment, Self-Restoring Barrier, and Modified Thrie Beam Guardrail (40) In 1988, a report was prepared describing an in-service evaluation performed on three devices in the state of Col- orado. These devices were the Colorado Type 3F end treat- ment, the self-restoring barrier (SERB), and the modified thrie beam guardrail. The purpose of this project was to deter- mine the performance of the systems under operating condi- tions as well as noting the damage repair costs due to field impacts. Data for this project were collected from September 1983 to January 1988. Data sources included police reports and, in the case of the SERB and Modified Thrie Beam devices, site visits. One SERB, three Modified Thrie Beam installations, and over 200 Colorado Type 3F end-treatment sites were analyzed. Collisions were documented for all the devices in the study. Four impacts occurred with the SERB. No fatalities were recorded, and no repair was necessary for three of the impacts. Six impacts were reported at the Modified Thrie beam sites. Again, no fatalities were recorded for these col- lisions and the system required only minimal repair in most cases. However, there were two collisions involving the SERB and the Modified Thrie Beam guardrail where the vehicles penetrated the railing. Both incidences involved heavy vehicles for which the devices are not designed (two military transports and a semi tractor-trailer). These incidents involved occupants being ejected from the vehicles after pen- etration. Six impacts were documented with the Colorado Type 3F end treatments during this time. One incident involved skeletal injury to a person who was not wearing a seat-belt, and minor injuries were recorded in the rest of the cases. Maintenance for the SERB and the Modified Thrie Beam system was, in most cases, relatively minor. However, the cost to repair the Colorado Type 3F end treatment nearly

matched the initial cost of installation. Modifications were suggested for all three devices to reduce maintenance. Connecticut, 1988—Connecticut Impact-Attenuation System (41) In 1988, Lohrey prepared a report describing an in-service evaluation of the Connecticut impact-attenuation system (CIAS). The systems were first installed in Connecticut in 1984 under the joint cooperation of the Connecticut DOT (Conn-DOT) and the FHWA. The purpose of the 3-year study was to determine the field performance of the CIAS and examine the maintenance requirements and other prob- lems associated with field use. Although the author never explicitly states where the infor- mation for this study was obtained, police reports and main- tenance personnel were apparently used as data sources. Also, judging from the included pictures of the impact events, site visits were probably performed. This study did not quantify collision severity except to state that no major injuries were sustained to the vehicle occupants in any of the impact events. Barring this information, the report seemed to focus more on the maintenance requirements and minor problems with the design of the system. Site visits involved measuring of system displacements and stating which cylinders needed replacement. The researchers also noted problems with the cover of the system, which is intended to keep debris and snow out of the crash cylinders. The design of the cover was not adequate to withstand the wind loading from passing heavy trucks on the freeway. The fastening system was redesigned and tested in the field. The study also examined the maintenance requirements of the system. If the cylinders needed to be reshaped, they needed to be transported to a garage and reshaped by main- tenance personnel. The study did examine on-site refurbish- ment possibilities, but these proved to be too time-consuming and inadequate for reassembly. Indiana, 1988—SENTRE (42 ) In 1988, Noureldin prepared a report describing an in- service evaluation of a SENTRE end treatment in the state of Indiana, including the methods used in the evaluation. The report also includes data collection forms from the study. The objectives of the study were to evaluate the field performance of the SENTRE system and analyze the cost effectiveness of the barrier. The 2-year study involved four agencies. The Indiana Division of Design, maintenance personnel, the Divi- sion of Research, and the contractor all inspected the sites after a collision and had input into the project. Post-collision investigation was to include an overall per- formance of the barrier during the collision, the reusability of its components, the estimated cost of repair, photographs of the damage, and police reports relating to the collision. 23 An initial cost of the SENTRE (i.e., $5,747) was deter- mined based on the average cost of the four systems involved in the study. This cost was much more than the possible alter- natives such as the BCT (i.e., $1,200/unit) and the conven- tional buried end (i.e., $479/unit). During the course of the evaluation, only one impact was observed and investigated. A police report, ambulance report, barrier report form, and cost estimate report were all obtained after the collision and used to draw the conclusions made in the study. From these reports, it was estimated that the ter- minal would cost $1,194 to repair or approximately 20 per- cent of the initial cost and, more important, the vehicle was not speared and did not overturn during the collision. North Carolina, 1990—SENTRE (43) In 1990, Stanley prepared a report describing a field eval- uation of the SENTRE guardrail anchor system in the state of North Carolina. Six units were installed and observed in the study. The purpose of the project was to evaluate the vehic- ular impact performance of the SENTRE and determine the costs associated with repair. The units were installed during the fall of 1987, with the final two devices being completed in December of that year. Data were then collected over a 2-year period. Data collection involved biannual site investigations and relied primarily on police crash reports. Information was also obtained from maintenance personnel regarding device and cost history, as well as interviews with police officers. The average installation cost of the devices was $5,000. No reported collisions occurred during the course of the study. The researchers stated the sites did not have a high AADT and that many impacts were not expected. Also, it was observed from the site investigations that no hit-and-run or unreported collisions had occurred. Minor vandalism of the devices was observed, however. Kentucky, 1991—BCT, MBCT, CAT, Brakemaster, and Turned-down End (44, 45, 46) Pigman, Agent, and Creasey prepared several reports between 1991 and 1993 describing an in-service evaluation performed on the breakaway cable terminal (BCT), Ken- tucky’s version of the median breakaway cable terminal (MBCT), the crash cushion attenuating terminal (CAT), the Brakemaster system, and the Type 7 weakened turned-down end treatment in the state of Kentucky. The purpose of this study was to determine whether these devices were perform- ing as designed in the field. Data for the evaluation were obtained from police crash reports, maintenance personnel, on-site visits, and observa- tions while traveling through the state. The data were then sorted according to the performance of the device. A device was considered “proper” if it performed as designed when

impacted. Data involving impact severity, including vehicle and property damage and injury severity, were not used as criteria to judge the performance of the device. A previous study was supplemented with an additional 2 years of data for the final analysis. A total of 5,706 sites were used for the BCT evaluation. From these sites, 232 impacts were found. Since some colli- sions lacked data, BCT performance was rated for only 158 of the 232 recorded collisions. These 158 collisions were then categorized by their geometry as straight, simple curve, and parabolic flare. From this investigation, it was deter- mined that the curvature, and especially the parabolic flare, improved the performance rating of the BCT. At the time of the report there were 848 sites using the MBCT and 66 impacts involving the MBCT. The performance of the MBCT was determined in only 33 of the cases because some colli- sions lacked important information. Performance was judged to be proper in 64 percent of the 33 impacts. Three collisions involving the MBCT resulted in fatalities. Spearing of the vehicle occurred in two of the cases and the third resulted in a vehicle rollover. The MBCT was recommended to be redesigned or eliminated from Kentucky standards, follow- ing evaluation of the data. The criteria used to determine “proper” CAT performance was based on the intended design of the device. This included the release of the breakaway posts during an impact event. Factors such as guardrail damage, vehicle damage, and injury severity incurred in a collision were not used as criteria to determine the performance of the device. During the 4-year evaluation period, 34 collisions involving the CAT were doc- umented. Accident reports were available for 23 of the inci- dents, and repair forms obtained from maintenance forces identified the remaining 11 impacts. The performance of the CAT was judged to be “proper” in 28 of the 34 impacts. Improper performance of the CAT was attributed to improper rail height and the raised medians in which the devices were installed. Based on these data, researchers recommended fur- ther use of the device within the state. Twenty Brakemaster sites existed in Kentucky during the course of this evaluation. From these, two collisions were recorded during the evaluation. One collision was judged to be severe although the device performed “properly” in this case. Due to the small number of collisions, further evalua- tion was recommended before the performance of this device could be completely assessed. A total of 3,781 installations of the Type 7 weakened turned-down end treatment existed in Kentucky during the period of this study. At these, 67 collisions were documented during the evaluation. Due to the lack of information for some of the impacts, the performance of the device was determined in only 61 of the cases. Based on this data, performance was judged to be “proper” in 51 of the 61 collisions. All the improper collisions involved vehicle rollover. Two fatalities were recorded, both involving rollover of a small automobile and ejection of the occupant from the vehicle. 24 While the study illustrates the hazardous nature of rollover collisions, an evaluation of performance should also involve recording the collision severity. A collision that resulted in serious injuries should not be classified as performing properly simply because the vehicle did not roll over or penetrate the barrier. Collision severity should have been used in defining the performance criteria. The ultimate goal of roadside hard- ware is to reduce fatalities and serious injuries. Proper deter- mination is not possible when this type of data is excluded from an evaluation. North Carolina, 1993—TREND and CAT (47) In 1993, Stanley described an in-service evaluation of the TREND and CAT devices in the state of North Carolina. The purpose of this study was to determine the impact perfor- mance of the devices, and to identify any problems associ- ated with the installation and maintenance of the hardware. As of the writing of the final report, six CAT and two TREND devices had been installed and were under observa- tion by the study group. The period of observation for the devices was 3 years, although four CAT terminals had been installed after the initiation of the collision data collection. Data collection for the study involved gathering police crash reports, interviews with maintenance personnel for informa- tion on upkeep, and periodic site visits to help identify unre- ported collisions. The final report indicated no impacts had occurred with any of the eight devices in the study area. North Carolina, 1993—Quick-Change Movable Concrete Median Barrier (48) In 1993, Stanley produced an interim report describing an ongoing in-service evaluation of the quick-change movable concrete median barrier (MCB) in the state of North Car- olina. The purpose of this study was to determine the ease of use of the device in the field, its vehicle redirection charac- teristics, and problems associated with its use. Data for this study included collisions occurring at the construction sites during the use of the MCB and information provided by the maintenance staff. Due to the nature of the device, this study had continuous on-site monitoring. This provided researchers with daily inspections of the device for incidental impacts occurring during the off-hours of the con- struction and maintenance crews. At the time of the second interim report, three construction sites had been used to evaluate the device, with four more sites proposed and one other approved for future use of the device. As of the second interim report, a total of 77 recorded collisions had occurred at the construction sites. However, not all these collisions involved the MCB; a few of the impacts involved either the Vehicle Mounted Impact Atten- uator (VMIA) or both the VMIA and the MCB. Although

the exact number of collisions involving the MCB is not given in the report, it is stated that over three out of four of the 77 collisions involved the MCB system. Of the reported incidents involving the MCB, only one collision resulted in an incapacitating injury; the rest of the reported injuries were minor. Problems associated with the use of the device were also noted. In four of the collisions, the MCB may have con- tributed to trapping water on the roadway causing the vehi- cles to hydroplane and lose control. Chipping and spalling of the corners of the barrier sections was also observed. Connecticut, 1994—Narrow Connecticut Impact-Attenuation System (49 ) In 1994, Lohrey described an in-service evaluation of the narrow Connecticut impact attenuation system (NCIAS). The NCIAS was installed at five high-hazard locations within the state of Connecticut in 1991 and evaluated for a period of 3 years. The purpose of this study was to evaluate the in-field performance of the NCIAS by examining occupant injury and comparing repair costs of the system to similar devices used on Connecticut roadways. Data collection for this study involved three major sources: official police crash reports, site visits, and state property damage reports. The police crash reports, when available, along with the on-site measurements and photographs, pro- vided the researchers with information necessary to evaluate the barrier’s safety performance. The state property damage reports were used to determine costs of repair of the barrier, including parts, labor, and equipment. In addition to these data, damage reports for similar devices were obtained for a cost-effectiveness comparison to other attenuators. The five evaluation sites that were used in the study were chosen based on four major factors: size of the hazard to be shielded, available space for the cushion, need for an upgraded safety device, and impact frequency. Although more collisions were expected based on histori- cal data, only one reported collision occurred at the sites dur- ing the course of the 3-year period. The researchers sug- gested the increased visibility of the system may have played a role. However, this may have also been attributed to an increase in safety performance associated with the new bar- rier. In the only reported collision, the occupant was treated and released for a head injury the same day as the collision. Although the number of reported collisions was very low, cost estimates for the repair of the system could be made in the other impact events from the state property damage reports. However, five of the six hit-and-run collisions were not sig- nificant enough to require cost reports. Therefore, the repair cost estimates for the NCIAS in this study are based on two impact events. 25 Virginia, 1994—Quick-Change Movable Concrete Barrier System (50) Cottrell reported in 1994 on an in-service evaluation of the quick-change movable concrete barrier system in Virginia. The barrier system was used on two concurrent projects dur- ing the period of February 1991 to April 1993. The main goal of the evaluation was to develop guidelines for use and eval- uate the system’s effectiveness in the field. More specifically, its performance during setting, resetting, and during a colli- sion were of concern, as well as the costs, advantages, and disadvantages of the system. Data were collected from maintenance personnel and police crash reports. In addition, photographs of the incidents were taken. The conditions of the road, time of day of the col- lision, number of vehicles involved, as well as injury data, were collected. The data collection took place between May 1991 and August 1992. Forty-two collisions occurred involving the barrier system during the evaluation period. No collisions resulted in a fatal- ity and 33 percent of the impacts resulted in an injury. In one case involving a tractor-trailer, the vehicle broke through the barrier. In general, most incidents were brush hits and resulted in little or no damage to the barrier system. The initial costs of the system included the cost for the vehicle to move the barriers and the actual barriers them- selves. The cost of purchasing the vehicle was $246,000 in project one and $275,000 in project two. The cost per meter of the barrier was $312 and $241 for project one and project two, respectively. In-place costs were also determined. These included minor maintenance to the placement vehicle and the system. Estimates per meter were $13 for project one and $14 for project two. The total cost for the systems was $2.1 mil- lion for project one and $1.4 million for project two. Pennsylvania, 1995—Brakemaster and Crash Cushion Attenuating Terminal (51) Snyder describes a 1995 in-service evaluation of the crash cushion attenuating terminal (CAT) and the Brakemaster sys- tems in the state of Pennsylvania. The purpose of the 32-month evaluation was to explore the feasibility of using the CAT and the Brakemaster Systems as an alternative to the GREAT system for concrete barrier end terminals in narrow median width (1200 mm) applications. Four different sites were used for the evaluation of the Brakemaster and CAT. The contractor chose which barrier was to be installed at each site. Between the four projects, a total of 22 CAT and two Brakemaster devices were evaluated for costs and collision experience. The sites were generally inspected after a collision, how- ever, one collision that occurred after the end of the official evaluation period did not include a site visit. Maintenance

forces were interviewed to determine installation and mainte- nance costs for the devices. The average installation costs for the CAT and Brakemaster systems were $8,570 and $7,700, respectively. Attempts were made to obtain police reports cor- responding to each collision. Only one collision involving a Brakemaster system occurred during the study period. No police crash report was filed for the collision. The site was visited the day after the collision and photographs were taken of the damaged device. The device performed as designed, but the panels designed to flare outward upon impact protruded into the roadway travel lane. When personnel from the manufacturer were later ques- tioned, they recommended a minimum median width of three meters for placement of the Brakemaster. The cost to repair the device was $4,305.83. One collision involving a CAT system occurred shortly after the end of the study period and was also described in the report. A police crash report was obtained from a Pennsyl- vania Engineering District office. No photographs of the scene or site visits were made by the researchers for this incident. The sole occupant (driver) sustained minor injuries that did not require medical attention. Extraction of the wooden posts from the galvanized sleeves was very difficult. The cost to repair the device was $3,257.27. Although both devices performed as designed and were less expensive to install and repair than the GREAT system, the CAT and Brakemaster systems were not recommended as alternatives to the GREAT system for narrow median applications. This was because the Brakemaster’s flaring pan- els protruded into the traveled way, and the CAT was classi- fied as a “gating” device. Based on NCHRP Report 350, “gat- ing” devices should have a recovery area 6 m behind the rail and 23 m along the length of the rail and are not appropriate in narrow median applications. This report provides valuable information for the use of the Brakemaster and CAT systems in narrow median applica- tions. The study also reasserts the need to carefully consider all site characteristics when selecting an impact attenuation device. Not nearly enough collisions were observed during the course of the study to make accurate determinations for the typical repair costs or device performance. Perhaps the selection of more collision-prone sites could have increased the number of impacts encountered. Two Brakemaster units are not sufficient to make a reasonable determination of the costs and performance of the device. More of these devices should have been installed. Ohio, 1996—ET-2000 Guardrail Terminal (52) In 1996, the Ohio DOT reported on an in-service perfor- mance evaluation of the ET-2000. The ET-2000 is a propri- etary guardrail terminal that has been permitted by Ohio DOT since 1992. This device was used where grading require- 26 ments limit the use of the MELT. Between February 1992 and December 1995, 6,421 ET-2000s were installed in Ohio at an average cost of $2,205. Also during this period, 214 ET-2000s were rebuilt with an average cost of reconstruction of $1,640. The purpose of this in-service evaluation was to examine the safety performance of the ET-2000. Information on the average installation costs of the new guardrail terminals was obtained from the Ohio DOT records following the study. During the 40-month data collection period, 306 collisions involving ET-2000s were reported and investigated. The data were collected by DOT district staff who visited the collision site prior to repairs, took photo- graphs, and filled out a one-page collision summary. The ET-2000 terminal performed very well according to the study. Only 39 of the 306 collisions (12.8 percent) involved occupant injuries. Of these 39 collisions, only five injuries (1.6 percent) were moderate or serious. Most of the collisions that resulted in an injury involved impact speeds of 88 km per hour or more (82 percent). The average cost to repair the ET-2000 was relatively high at $1,640, or 74 percent of the original installation cost. The data collection was one of the most systematic and complete in-service performance evaluations found in the literature. Site visits were performed by maintenance per- sonnel, guaranteeing evaluation prior to the terminal’s repair. A one-page reporting form was developed for maintenance personnel, which probably simplified the data collection and limited the data collected to only those areas which were part of the study. Certain data sources and data types excluded from the study may have provided better insight into the perfor- mance of the ET-2000. It was unclear what was included in the on-site estimate of the repair costs for the ET-2000. Some of the estimates possibly included labor costs, while others did not. This would have an impact on the average estimate for repair of the terminal. The installation cost for the ET-2000 was compared with the average installation cost of the MELT. The cost of installing the MELT, how- ever, did not include the cost of providing proper grading around the terminal. The difference between the installation costs for the ET-2000 and the MELT is probably less than what is reported. Texas, 1996—ET-2000 Guardrail Terminal (53, 54) A presentation and report in 1996 described an in-service performance evaluation of the ET 2000 guardrail end treat- ment in the state of Texas. The objective of this study was to determine the field performance of the ET-2000 and to refine the design to improve its safety and ease of installation. Data collection for this study consisted of interviews with district design personnel, area engineers, project inspectors,

maintenance and warehouse personnel, examination of police accident reports, the witnessing of an installation of a GET by a contractor, and the review of several of the impact sites. When the data had been collected, a task force consisting of the original designers, the manufacturer, researchers, and FHWA was formed to examine the performance, installa- tion characteristics, and repair issues in order to improve the device. This group suggested many changes that were implemented in later designs of the system. Among these design changes were the removal of legs and rubber bumpers which were deemed unnecessary, a change in the universal posts to simplify construction, changing the attachment bracket from a one-hole to three-hole system to allow for minor changes in post placement, and the addition of delin- eators for visibility. Although it is unclear when the collision data collection ceased, the results of the Texas study are based on a period in 1993 and 1994. During this period, 37 collisions involving the ET-2000 were investigated. Of these 37 collisions, 92 percent resulted in no injuries or only minor injuries to the occupants. Three crashes resulted in incapacitating injuries to at least one of the occupants of the impacting vehicle. These collisions included a side impact, an unrestrained occu- pant in the bed of a pickup truck, and a possible misreported injury. No fatalities were recorded. Because of the limited number of collisions recorded, no attempt was made to ana- lyze the data statistically. In addition to collision severity data, undesirable perfor- mance characteristics of the device were also identified. In some cases, the extruder head was pushed into a traveled lane of traffic and remained there, blocking the roadway. Concern was also expressed that the extruded rail might encroach on lanes of traffic if installed in medians less than 7.6 m wide. Perhaps the most important observation was that of a rail becoming sheared prior to being extruded in the GET, which resulted in the rail spearing the vehicle. PROCEDURAL STUDIES A procedural study is an evaluation or discussion of data collection procedures. Connecticut, 1975—Photographic Surveillance (55, 56) In 1973 and 1975, Bowers prepared two reports summa- rizing experience with photographic surveillance of roadside devices in the state of Connecticut. The purpose of the reports was to describe the problems encountered with the installa- tion and use of the system. A Rich Hy-Dro Cell sandwich unit and a Fitch Inertial Sand Barrier were observed at sites with a relatively high fre- quency of impact. Ease of camera and surveillance system 27 mounting was considered in selecting sites. One system was located in an overhead sign bridge and the other was located on a wooden pole at an I-95 off-ramp. The surveillance systems consisted of high-speed water- proof cameras encased in wooden housings and mounted strategically at the site. The detection system consisted of 50-mm diameter by 50-mm deep magnetic sensors mounted in the concrete gore area upstream of the attenuation device. These sensors, when tripped, sent a signal to the camera. To minimize the time lapse between the signal and the first exposure, the camera motor ran continuously. An estimated 38 impacts and false alarms could be recorded before film replacement was necessary. These systems were plagued with many problems. The author noted at least two instances per system where camera motor gears had to be replaced due to continual use. There was a high instance of false alarms. In the initial debugging stage, site counters recorded 100 false alarms per day at one site and an average of 300 per day at the other site. Many explanations were given for the high number of false alarms, including rush hour traffic, gore area crossovers, near misses, and thrill-seeking motorists. To remedy both problems, the systems were shut off during peak traffic hours. The systems were also disengaged during nighttime hours due to lighting constraints. The rate of false alarms, however, still exceeded 40 per day and cameras still required daily film replacement. The systems were prematurely removed and the project ended because of high false alarm rate and infrequency of impacts at either site. These reports illustrate the many potential problems that must be considered when considering photographic surveil- lance techniques. The author suggested ways to improve sim- ilar studies in the future, including: • Site selection should consider collision experience and potential false alarm rates, • Still frame cameras should (as of 1975) have their motors running continuously to reduce lag time between system activation and film exposure, and • Using continuously erasing video tape systems should be considered to prevent running out of film and there- fore eliminate the false alarm concern. The author also suggested that the system shut down after an impact had occurred. NHTSA, 1982—Longitudinal Barrier Special Study (57, 58) The National Accident Sampling System (NASS) was established to aid the National Highway Traffic Safety Admin- istration (NHTSA) in the reduction of the number of fatalities, injuries, and economic losses resulting from motor vehicle crashes on the nation’s highway. The Longitudinal Barrier Special Study (LBSS) was one of three studies initiated to

provide more in-depth information on particular types of col- lisions. The LBSS addressed collisions involving longitudi- nal barriers, both guardrails and median barriers, but not bridge rails. The LBSS data were collected using a stratification sam- pling based on area type (rural, urban), highway type (free- way, non-freeway), vehicle size (small # 1,134 kg, large >1,134 kg), and longitudinal barrier types (G1, GR1, G2, G3, G4(1W), G4(2W), G4(1S), G4(2S), G9, W-beam, Blocked- out W-beam, Concrete safety shape, MB1, MB2, MB3, MB4W, MB4S, MB5, MB7, MB8, MB9, and MB10). A sample size of 82 was requested in each stratum. Priority for sampling was given to all collisions involving end treatment type BCT and guardrail/bridge rail transition. Data collectors visited police agencies weekly, biweekly, or monthly and identified collisions that met the study crite- ria. All collisions in the Special Study had at least one impact involving a guardrail or median barrier, and the collision was reported by police at the scene with all involved vehicles and drivers present. Note that this excluded unreported collisions. All vehicle types were included with the exception of motor- cycles when their involvement in the collision was hitting the barriers. If motorcycles were involved in collisions where the other vehicle hit the barrier, the crash was included. Data collectors then investigated selected collisions in fur- ther detail. Documentation of barrier damage including pho- tography and field measurements, impact and trajectory data, and vehicle damage data sufficient to obtain the Collision Deformation Classification (CDC) had to be present to include the collision in the LBSS. Extensive photographs were taken at the scene with the scale identified. The basic requested photographs included: general photographs along the path of the vehicle starting from 3 m behind the tire marks, with mul- tiple “path pictures” if the path was over 15 m long; photo- graphs of the point of impact and vehicle rest positions; photographs of the road and terrain in the direction of travel beginning 305 m upstream with photographs at 60-m incre- ments and 30-m increments after the vehicle left the road- way; at least two general photographs of each impact taken in the direction of travel at different distances; at least two gen- eral photographs of each impact taken opposite the direction of travel at different distances; one or more photographs along the paths between impacts; a general photograph of each road- side structure/object struck; and at least one close-up photo- graph displaying the damage at each roadside structure/object struck. Data collection was organized into the following four cat- egories: header information, location identification, impact sequence, and longitudinal barrier information. The header category identified the Primary Sampling Unit (PSU), the investigator that completed the form, and general informa- tion required for data processing. The location identification category identified the location of the collision. The impact sequence category identified the impact sequence in the col- lision for each impact by the object contacted and its lateral offset. The longitudinal barrier category described the char- 28 acteristics of the struck barrier, the roadside cross-section, the extent of damage to the barrier, the vehicle dynamics and trajectory during impact, and the barrier performance. The data collection requirements, coding instructions, and field procedures used in the study were detailed in the Coding/Editing and Field Procedures Manual. The manual was intended for use by PSU investigators for data collection and Zone Center (ZC) personnel in their review process. The manual provided an introductory page for each category that identified the name of the category, the applicable variable numbers, a description of the data category, and references used in formulating the definitions and coding instructions for the variables. For each variable or group of variables, the variable number, variable name, format, beginning column, element value (range and individual codes or responses), the source, any remarks (level of data collection, descriptions and definitions, coding instructions, and illustrations), field procedures, and related variables were provided. The manual also contained a section that identified editing and consis- tency checks to aid PSU investigators and ZC personnel when they reviewed the special study forms. Erinle et al. discussed an analysis of the LBSS data in a 1994 report (59). Some of the results included the following: • Weak-post barriers were less associated with driver injury than strong-post and fixed barriers. • Drivers were more often injured when their vehicles returned to or crossed the roadway than when their vehicles remained on top of, penetrated, or overrode the barriers. • The subsequent impact that most often resulted in dri- ver injury was rollover. Rollover was also associated with higher serious injury (A+K) rates. • Serious driver injuries were more often associated with blunt and turndown ends than with the length-of-need section of guardrails. • In many cases of barrier failure, the impact conditions were “unusual,” i.e., they differed significantly from the crash test impact conditions. DISCUSSION AND CONCLUSIONS In-service evaluations provide a means to test assumptions and conclusions resulting from crash testing roadside appur- tenances. Typically, these devices are designed and tested based on worst-case scenario crash test conditions. Since worst-case conditions represent a small percentage of actual crash occurrences, it could be concluded that most in-service collisions will simply give better performance results than crash testing. During most crash tests, however, these devices are installed and tested under ideal conditions and do not take into account variations in weather, unusually high corrosion due to the use of de-icing salts, driver behavior due to the installation of the device, and other factors not under the con- trol of the test facility. In short, the real world presents sub-

stantially more variables than can be completely accounted for under laboratory conditions, making in-service evalua- tions a necessary and vital tool for the development of road- side devices. The rationale for observing real-world performance of a roadside device is well founded. Historical studies, consid- ered a predecessor to in-service evaluations, are documented as far back as the early 1970s. These studies use primarily police crash reports to examine the performance of existing devices. For example, an early Iowa study focused on colli- sions with light post cable guardrail systems retroactively using state data sources (3). Although historical studies are valuable in evaluating existing devices, states often require data on the performance and cost-effectiveness of newer devices. DOTs have con- ducted various types of collision studies since at least the late 1970s. A collision study usually involves performing data collection as collisions occur, rather than searching records retroactively. Maintenance personnel as well as police offi- cers generally are involved in the data collection, and costs associated with the installation and maintenance of the devices are recorded to some degree. New York DOT, for example, conducted several studies on the performance of guardrails and other roadside appurtenances that relied heav- ily on maintenance personnel for data collection (4, 23). Based on such collision studies, a number of improved methodologies have been developed and implemented to improve data collection and analysis. In-service performance evaluations represent a more exhaustive study of collisions. This method of data collection is performed concurrently with the time period for which the data are being collected. This simultaneous data collection enables individuals involved to make personal visits to the collision sites and collect all the necessary information for the performance evaluation of the devices. Unfortunately, many times the individual responsible for data collection lacks proper training, which may lead to incomplete data. To remedy this, an in-service evaluation will typically require the development of a comprehensive, stan- dardized reporting form to be used by those collecting infor- mation from site visits. These forms can then be distributed as supplementary documentation to the individual responsible for filing accident reports, usually the police officer at the scene. A few studies have examined methods of data collection and how they could be improved. One Connecticut study documented an attempt to obtain data from cameras mounted along selected highways (55, 56). Although this early study failed to achieve its objectives, it was valuable in stressing the importance of accurate data collection from on-site obser- vation. More recently, the “Coding/Editing and Field Proce- dures Manual” for the NASS Longitudinal Barrier Special Study outlined methods for collecting, recording, and veri- fying data for use in in-service evaluations (57). Collision characterization and evaluation are typically governed by the number of devices available for observation. 29 When only a few devices or installations are investigated, field evaluations are relatively easy to perform. However, as noted in many of the reviewed reports, this may lead to insuf- ficient sampling of collisions. Conversely, when more devices are examined in an evaluation, more collisions are typically experienced, but the ability of the study team to perform on- site investigations is impaired. In addition, in-service evalu- ations often target devices under development, which do not already exist within the geographical area of study. This gen- erally has a crippling effect on the number of sites available for study. New devices, however, typically exhibit high instal- lation costs compared to those already existing within an area. In general, if a device must be installed prior to its eval- uation, not many are constructed. Therefore, in most of the evaluations reviewed, the number of devices studied was not sufficient for researchers to make valid determinations regard- ing the performance of the device. Uncertainties associated with new devices and their high installation costs typically contributed to the sparse number of devices available for study. This, in turn, has limited the number of observation sites. For example, many reports noted that bid estimates to install new devices varied greatly and tended to be substantially higher in comparison to the actual cost of installation. In many cases, contractors who lacked experience installing new devices would leave a mon- etary buffer for possible unforeseen installation costs. When only a small number of devices can be constructed, sites should be selected which are appropriate for the device and facilitate collision experience with the appurtenance. In con- trast in the literature review, a few studies occurred where some sites did not experience a single hit over a 2-year period. In some other studies, while the sites selected were appropri- ate for the function of the hardware, often they were not cho- sen to encourage collision experience with the devices. In the most successful studies, site selection involved research- ing the potential sites and determining which ones had the highest collision probability. These sites were then typically chosen as construction and observation sites. In this manner, all but one study reported reasonable collision experience given the number of devices under observation. As devices become more widely used and contractors more experienced in installing them, costs will naturally decrease, and opportu- nities for in-service evaluation will increase. The lack of collision experience with devices under obser- vation limits the validity of any conclusions drawn from a particular study. Ideally, the quantity of data collected should be sufficient for statistical analyses, but cost constraints limit the number of observation sites and therefore collision expe- rience. With limited collision experience, many of the reports were primarily anecdotal in nature with perhaps an average cost of installation or percent injured by class noted. Statisti- cal evaluation was mostly confined to the larger historical and collision studies, although a few in-service evaluations used statistical methods to analyze their data.

Despite the absence of statistical analyses in most of the in-service evaluations, valuable information can still be obtained. The inclusion of a collision report form filled out by the police or the investigators can improve the quality of the data collected by targeting only those data that are rele- vant to the study. Reports that were based solely on police accident reports and accident databases were not as flexible as studies that included the use of some sort of data collec- tion form. Data collected in this manner were also available from a centralized data repository. In many instances, how- ever, if necessary data were not available from the accident database, the source police crash report had to be obtained, which frequently did not produce the desired information. On several occasions, the police crash report itself could not be found, thus reducing the data pool even further. Mainte- nance information for these events was often even more dif- ficult to obtain. Simultaneous data collection did have some drawbacks, however. Cases were noted where the researchers experienced difficulties in obtaining information from the data collec- tors. Good communication among all individuals and agencies involved is, therefore, critical. This type of data collection also requires that data be obtained simultaneously with the evalua- tion period. That is, the study period must be the present. Data must be collected for a period of time before any con- clusions can be drawn about the performance of a device. Many other problems were experienced during the course of the in-service evaluations. Often, especially under cir- cumstances associated with “new” devices, the hardware was not installed properly as determined by the researchers or 30 other DOT personnel. These cases reinforced the importance of inspection of the devices to ensure correct installation. Perhaps the most difficult problem resulted from unreported incidents. Although occupant injury and vehicle and device damage were typically minor in these instances, these types of collisions are still important because the device apparently performed correctly and had no serious consequences. Rou- tine inspections of the devices were the primary method used to document these events. Unless these inspections were per- formed regularly, however, many unreported collisions could occur between inspections, making it difficult to determine the actual number of unreported collisions. Even when these events were documented, details of the collisions were impos- sible to determine. If unreported collisions were documented in a report, injuries were usually classified as PDO and the devices were generally judged to have performed “properly.” Fifty-seven reports on collision studies and in-service evaluations were reviewed in the literature survey, most typ- ically performed by state DOTs. Two particularly good stud- ies are an evaluation of the MCB system in Virginia and an evaluation of the ET-2000 in Ohio (50, 52). In-service evaluations have been recognized as an impor- tant part of the development of roadside devices (5)(2). Full- scale crash testing can only provide information about the performance of a device under ideal installation conditions and prescribed impact conditions. The basis of evaluation is of utmost importance to a collision study or in-service eval- uation. Factors such as difficulty in installation, poor appli- cation, and environmental factors can decrease the perfor- mance of a device and should therefore be included.