Recommended Guidelines for the Selection of Test Levels 2 Through 5 Bridge Railings (2021)

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.

82 DEVELOPMENT OF BRIDGE RAIL SELECTION GUIDELINES The available analysis methods and software products presented in the literature review assess the probability of a roadside feature being struck, the severity of the crash if it has occurred and the resulting crash costs through conditional probabilities then perform a benefit-cost analysis of roadside design alternatives to determine the most cost- effective design. RSAPv3 has the added capability of tabulating crash costs such that the lifetime risk of a specified crash severity can be calculated independent of the direct costs (i.e., construction, maintenance and repair costs). The risk analysis and cost/benefit analysis capabilities of RSAPv3 were employed in the development of the selection guidelines. RSAPv3 uses this conditional probability model [Ray12]: E(CC)N,M = ADT ∙ LN ∙ P(Encr) ∙ P(Cr|Encr) ∙ P(Sev|Cr) ∙ E(CCs|Sevs) where: E(CC)N,M = Expected annual crash cost on segment N for alternative M, ADT = Average Daily Traffic in vehicles/day, LN = Length of segment N in miles, P(Encr) = The probability a vehicle will encroachment on the segment, P(Cr|Encr) = The probability a crash will occur on the segment given that an encroachment has occurred, P(Sevs|Cr) = The probability that a crash of severity s occurs given that a crash has occurred and E(CCs|Sevs)= The expected crash cost of a crash of severity s in dollars. An RSAPv3 analysis is composed of four major steps for assessing each alternative: • Encroachment Probability, • Crash Prediction, • Severity Prediction, and • Benefit-Cost and/or Risk Analysis. Some improvements were made to RSAPv3 in this project to specifically address the challenges of developing bridge rail selection guidelines. These upgrades are discussed below. Crash Data Unfortunately, the literature review and survey did not uncover much in the way of in-service studies or even crash studies of bridge railings so it was necessary to look for other sources of data for bridge railing performance to use in the development of the Selection Guidelines. There are some existing databases like the NCHRP22-08 and FHWA Narrow Bridge databases discussed earlier but these are now very old and, at least in the case of the NCHRP 22-08 Texas data, have been determined to have serious

83 coding problems. In any case, a great deal has changed in the types of bridge railings that are available today in comparison with the early 1990s as reflected in the NCHRP 22-08 report.[Mak94] Crash data was collected and used to populate the crash severity module of RSAPv3 for several concrete bridge railings and similarly shaped median barriers for use in this project. Crashes with concrete median barriers were reviewed to determine the severity distribution for striking the bridge railings of similar shape. Concrete median barriers were chosen because there are more miles of median barrier installed than bridge railing which maximizes the amount of crash data that is available for modeling the severity of crashes with typical median barrier and bridge railing shapes. Most concrete median barriers and concrete bridge railings use essentially the same shapes (i.e., New Jersey shape, F-shape, vertical wall, constant or single slope, etc); therefore, these roadside devices are expected to perform similarly with respect to crash severity. Crashes with bridge rails were analyzed to determine the severity distribution of crashes where the bridge railing was penetrated. All types of bridge rails were considered in this analysis because the analysis was focused on the result of penetration, not the probability of penetration. The type of bridge rail would not impact the outcome after penetration since the outcome is a function of the land use characteristics around and under the bridge (i.e., presence of other transportation facilities, urban or rural areas, etc.). Crashes with embankments and water hazards were also considered as a reference point. The following sections briefly describe the data used in these analyses. The computational steps and results of the analyses are discussed throughout this report in the relevant sections. New Jersey Median Barrier Crash records for the New Jersey Turnpike were requested for 2003 through 2009 from Rutgers University. Rutgers maintains a database of crashes throughout New Jersey which are linked to road geometrics.[Plan4S11] The ADT and percent of trucks for the New Jersey Turnpike were obtained directly from the New Jersey Turnpike Authority (NJTA). A 105-mile long section of the New Jersey Turnpike was selected for study where a TL5 concrete safety shape median barrier is used exclusively and continuously for the entire length of the highway. [NJTA1] The speed limits on this 105-mile long section were either 55 or 65 mi/hr. A total of 1,816 crashes within the 65 mi/hr zone and 241 crashes within the 55 mi/hr zone were obtained and reviewed. The severity distributions were calculated and adjusted for unreported crashes and the EFCCR65 was determined (see the section entitled “Severity” on page 135 for a discussion of the EFCCR65). The distributions for each speed are presented in Table 23. The penetrations and rollovers percentages were also determined and are presented in Table 24. Narratives of the crashes for which the vehicles appear to have penetrated or rolled over the barrier were requested and subsequently reviewed to verify that a penetration-rollover-vault (PRV) had occurred.

84 Table 23. New Jersey Turnpike Crash Severity Distribution. Barrier K A B C PDO/UNK Total Contained, Stopped or Redirected on 55 mi/hr Segments No. 0 1 12 35 193 241 % 0.00 0.41 4.98 14.52 80.08 100 Contained, Stopped or Redirected on 65 mi/hr Segments No. 0 11 103 307 1395 1816 % 0.00 0.61 5.67 16.91 76.82 100 As shown in Table 24, two instances of a barrier PRV in the 55 mi/hr zone occurred. One of these resulted in a possible injury (i.e., C), while the other resulted in property damage only crash. There were 11 instances of barrier PRV in the 65 mi/hr zone. Two of these resulted in visible injuries, one in a possible injury, five in property damage only crashes, and the injury level was unknown for three of the cases. Table 24. NJTA After Barrier Contact Behavior. Behavior 55 mi/hr 65 mi/hr # # Contained/Redirected 241 1,816 PRV 2 11 Rollover After Redirection 5 41 The number of instances where the vehicle rolled over after being redirected by the barrier were also collected and are shown in Table 24 for both of the speed zones. In the 55 mi/hr zone, five instances of redirection rollovers occurred; two had visible injuries, two had possible injuries, and one resulted in property damage only. In the 65 mi/hr zone, 41 instances occurred; 1 fatality, 1 incapacitating injury, 19 visible injuries, 11 possible injuries, 4 resulted in property damage only, and five cases where the injury level was unknown. Massachusetts Median Barrier The Massachusetts DOT crash database was also examined for 2006-2009 to identify median barrier collisions on specific sections of roadways where median barriers were recently constructed (i.e., within the past 5 or 6 years). A subsequent field review was conducted to isolate sections of roadway where 32-inch tall and 42-inch tall concrete F-shape median barriers exist absent of other types of barriers. This field review was conducted to eliminate the possibility of reviewing crash records where the reporter may have confused the type of barrier struck. After this review, 154 crashes with 32-inch barrier and 34 crashes with 42-inch barrier were identified. All of these crashes occurred on roads with posted speed limits of either 55 or 65 mi/hr. The severity distribution was

85 determined and the results and distributions are shown in Table 25. The percentages of penetrations and rollovers were determined and confirmed using available narratives of the police reports. These percentages are presented in Table 26. From Table 26 it can be seen that the 32 inch F-shape barrier had two reported instances where the vehicle penetrated, rolled, or vaulted over the barrier in the 55 mi/hr speed zones. Both of these crashes resulted in non-incapacitating injuries. This same barrier had six instances of PRV failure in the 65 mi/hr zone as well, two of which were non-incapacitating injuries and four resulted in property damage only. The 42 inch F- shape barrier had two instances of PRV failure in the 55 mi/hr zone (which was the only zone the 42 inch F-shape was installed), and both instances resulted in property damage only. Also in Table 26 is the rollover after redirection information for the different barriers within the different speed zones. No rollover after redirection cases were reported for the 32 inch F-shape barrier within the 55 mi/hr speed zones, but five were reported in the 65 mi/hr zones. Of these five crashes, three resulted in non-incapacitating injuries, one in property damage only, and one crash had an unknown level of injury. Only one occurrence of a rollover after redirection was reported for the 42 inch F-shape barrier in the 55 mi/hr zones (again, this was the only speed zone where crashes were analyzed for this F-shape barrier). Unfortunately, this single rollover resulted in a fatality. Table 25. Massachusetts Crash Severity Distribution. Barrier K A B C PDO/UNK Total Contained, Stopped or Redirected on 55 mi/hr Segments, 32" F-shape No. 0 0 4 4 14 22 % 0.00 0.00 18.18 18.18 63.63 100 Contained, Stopped or Redirected on 65 mi/hr Segments, 32" F-shape No. 3 4 36 17 72 132 % 2.27 3.03 27.27 12.88 54.55 100 Contained, Stopped or Redirected on 55 mi/hr Segments, 42" F-shape No. 0 0 6 4 24 34 % 0.00 0.00 17.65 11.76 70.59 100

86 Table 26. Massachusetts After Barrier Contact Behavior. Behavior 32" @ 55 mi/hr 32" @ 65 mi/hr 42" @ 55 mi/hr # # # Contained/Redirected 22 132 34 PRV 2 6 2 Rollover After Redirection 0 5 1 Washington State Median Barrier The Washington State crash data was examined for I-90 and I-5 with posted speed limits of 60 mi/hr where 32-inch New Jersey safety shape and 34-inch single slope concrete median barriers were used. The severity distributions of 549 cases involving 32- inch safety shape barriers and 178 cases involving single slope barriers were calculated and can be seen in Table 27. The behavior after contact was determined for both barriers and is presented in Table 28. Table 27. Washington State Crash Severity Distribution. Barrier K A B C PDO/UNK Total Contained, Stopped or Redirected on 60 mi/hr Segments, 32" Safety Shape No. 2 4 62 112 369 549 % 0.36 0.73 11.29 20.40 67.21 100 Contained, Stopped or Redirected on 60 mi/hr Segments, 34" Single Slope No. 0 3 20 28 127 178 % 0.00 1.69 11.24 15.73 71.35 100 Table 28. Washington State After Barrier Contact Behavior. Behavior 32" Safety Shape # 34" Single Slope # Contained/Redirected 549 178 PRV 3 1 Rollover After Redirection 14 6 Table 28 shows the reported crashes that resulted in a penetration, roll or vault over the barrier (PRV). For the 32 inch safety shape barrier, three of these crashes occurred. One of these crashes resulted in a non-incapacitating injury, while the other two resulted in property damage only. For the 34” single slope barrier, the only crash of this type resulted in property damage only. Table 28 also shows the number of reported crashes that resulted in a rollover after being redirected by the two barrier types. The 32 inch safety shape barrier had 14 of

87 these crashes; including one that resulted in a fatality. Six of the remaining 13 crashes resulted in non-incapacitating injuries (level B), three resulted in possible injuries (level C), and four resulted in property damage only. The 34 inch single slope barrier had six crashes where the vehicle rolled over after being redirected by the barrier. Two of these crashes resulted in non-incapacitating injuries, one resulted in a possible injury, and three resulted in property damage only. Pennsylvania Bridge Railing The Pennsylvania Department of Transportation (PennDOT) requires “bridge railings that meet the requirements of Test Level 5 (TL5) of NCHRP Report 350, unless another test level is authorized by the District Executive.” [PennDOT11] PennDOT generally specifies a 42-inch concrete F-shape barrier as the TL5 railing, however, other PennDOT adopted railings may also be used. A TL4 32-inch concrete F-shape barrier is also a common barrier used and was also considered in this research. Crash records were reviewed from 2006 to 2010 for bridge rail crashes on interstates highways. Traffic volumes for the interstates and the roads which crossed under the interstates were found online.[PA11] Unfortunately, the percentage of trucks in the traffic was not available. The environmental features surrounding each bridge were reviewed using Google Earth.[Google11] Table 30 shows the reported PRV events for the two heights of the F-shape bridge rail in both the 55 and 65 mi/hr speed limit zones. For the 32” F-shape bridge rail in the 55 mi/hr zone, two PRVs occurred. The first of these involved a tractor-trailer where the trailer rolled over and stayed on the bridge but the tractor broke through the barrier and dropped off the bridge. The tractor fell approximately 90 ft and landed on its passenger side on a small island in the middle of Maiden Creek, resulting in a possible injury. Witnesses say the truck was traveling at or around the speed limit at the time of the crash. The second 32 inch bridge rail crash that occurred in the 55 mi/hr zone involved a passenger car that vaulted over the rail and fell approximately 80 ft and came to rest on its roof in a wooded area next to a river. This crash resulted in property damage only. The same bridge rail in the 65 mi/hr zone experienced five PRVs. The first of these involved a tractor-trailer truck where the tractor portion stayed in the traveled way of the highway, but the trailer portion broke through the bridge rail and ended up hanging over the bridge but did not fall off. This resulted in an incapacitating injury. This bridge crosses Pulaski Mercer Road (State Route 468), which experiences a 400 vehicle per day traffic volume. The second 32-inch bridge rail PRV crash that occurred in the 65 mi/hr zone involved a passenger car which mounted the railing and rode on top of the rail for 30 ft before falling off the bridge and dropping 60 ft. It came to rest on its roof, resulting in an incapacitating injury. According to witnesses, this vehicle was traveling somewhere between 50 and 55 mi/hr when the incident occurred, which is below the posted speed limit. It was later discovered that a vehicle defect caused the vehicle to hit the bridge rail.

88 The vehicle landed in an unused area under the bridge; a fortunate occurrence as this bridge spans South Fork Tenmile Creek (Route 188), a 5,000 vehicle per day state route, and a set of railroad tracks. The third PRV incident for this barrier in the 65 mi/hr speed zone also involved an incapacitating injury. In this crash, the vehicle had rolled over onto its passenger side prior to coming in contact with the bridge rail. As the vehicle hit the bridge rail it rolled over the barrier and landed in a grassy area next to Bullfrog Road (unknown average daily traffic). Table 29. Pennsylvania Crash Severity Distribution. Barrier K A B C PDO/UNK Total Contained, Stopped or Redirected on 55 mi/hr Segments, 32" F-shape Bridge Rail No. 3 1 6 14 33 57 % 5.26 1.75 10.53 24.56 57.89 100 Contained, Stopped or Redirected on 65 mi/hr Segments, 32" F-shape Bridge Rail No. 1 0 7 28 71 107 % 0.93 0.00 6.54 26.17 66.36 100 Contained, Stopped or Redirected on 55 mi/hr Segments, 42" F-shape Bridge Rail No. 1 0 1 3 5 10 % 10.00 0.00 10.00 30.00 50.00 100 Contained, Stopped or Redirected on 65 mi/hr Segments, 42" F-shape Bridge Rail No. 0 0 4 9 33 46 % 0.00 0.00 8.70 19.57 71.74 100 Table 30. Pennsylvania Bridge Rail After Barrier Contact Behavior. Behavior 32" @ 55 mi/hr # 32" @ 65 mi/hr # 42" @ 55 mi/hr # 42" @ 65 mi/hr # Contained/Redirected 57 107 10 46 PRV 2 5 0 0 Rollover After Redirection 6 4 3 1 The fourth PRV crash involving the 32 inch F-shape bridge rail in the 65 mi/hr zone resulted in a non-incapacitating injury. The vehicle struck the bridge rail and then rolled over, landing on the centerline of the road passing below the bridge. This road (Route 374) experiences roughly 1,500 vehicles per day. The last PRV crash involving the 32 inch F-shape bridge rail in the 65 mi/hr zone involved a bus filled with 42 passengers and resulted in a single non-incapacitating injury. The bus struck the barrier and rode up onto the top of the barrier and balanced there before falling approximately 100 ft into the Lehigh River below. The bus passengers were evacuated entirely while the bus was balancing on top of the bridge rail.

89 The 42” F-shape bridge rail had no PRV failures for both the 55 mi/hr and 65 mi/hr speed zones. The number of incidents for each bridge rail and speed zone where a vehicle rolled over after being redirected by the bridge rail are tabulated in Table 30. Six incidents were reported for the 32 inch rail in the 55 mi/hr speed zone. Two of the incidents resulted in fatalities, two resulted in non-incapacitating injuries and two resulted in possible injuries. For the same size rail in the 65 mi/hr speed zone, four incidents occurred. One of these incidents resulted in a fatality while the remaining three resulted in only possible injuries. The 42-inch bridge rail experienced four cases where the vehicle was reported as rolling over after being redirected; three in the 55 mi/hr zone and one in the 65 mi/hr zone. One of the events in the in the 55 mi/hr zone resulted in a non- incapacitating injury and two resulted in possible injuries. The one reported case where a rollover occurred in the 65 mi/hr zone resulted in a non-incapacitating injury. Ohio Bridge Railing Crash data for bridges for bridges in Ohio from 2005 through 2010 includes 4,600 bridge railing crashes. Ohio installs TL3 bridge rail on “all bridge structures on the National Highway System (NHS) or the State System…as defined by NCHRP Report 350,” effective October 1, 1998. The Twin Steel Tube Bridge Guardrail (Standard Bridge Drawing TST-1-99) should be used for side draining structures, which shall not be used over highways and railroads. “For bridges with heights of 25 feet or more above the lowest groundline or normal water, concrete deflector parapets should be used.” [OH11; OH11a] Therefore, the barrier shown in Figure 19 is the TL3 concrete barrier which is typically installed on NHS roadways which cross over highways.

90 Figure 19. Ohio Standard Drawing BR-1 [OH11a] The 2005 through 2010 Ohio data indicated that there were 4,560 police-reported crashes that involved bridge parapets (i.e., code 28) or bridge rails (i.e., code 29). Vehicles crossed the barrier line 28 times in 4,600 crashes or only in 0.6 percent of the crashes. Of the 28 PRV crashes, none were fatal, four involved A-level injuries (14 percent), 11 involved B-level injuries (39 percent), two involved C-level injuries (7 percent), 10 involved no injuries (36 percent) and one was of unknown severity (4 percent). In three of the cases the vehicle came to rest on a roadway that passed under the bridge and in the remaining 25 cases the vehicle came to rest in or near a body of water. Only one event involved a tractor-trailer truck. Table 31 provides a summary of the severity distribution of all reported crashes with bridge rails during the study period and Table 32 summarizes the behavior after the crash.

91 Table 31. Ohio Crash Severity Distribution. Posted Speed Limit K A B C O Total Contained, Stopped or Redirected for 42" Constant Slope 65 1 1 19 13 83 117 60 0 0 4 2 20 26 55 0 2 4 2 33 41 50 0 0 1 0 4 5 45 0 0 1 1 2 4 Ramp 0 3 2 2 11 18 Total 1 6 31 20 153 211 % 0.47 2.84 14.69 9.48 72.51 100 Contained, Stopped or Redirected for 36" Safety Shape 65 2 8 63 46 292 411 60 0 3 8 8 34 53 55 0 4 32 27 126 189 50 0 1 1 3 11 16 45 0 0 1 5 26 32 40 0 0 1 0 6 7 35 0 2 2 6 22 32 30 0 0 0 0 0 0 25 0 0 1 1 2 4 Ramp 0 7 23 16 110 156 Total 2 25 132 112 629 900 % 0.22 2.78 14.67 12.44 69.89 100 Contained, Stopped or Redirected for 42" Safety Shape 65 1 6 38 29 169 243 60 0 8 19 20 78 125 55 0 4 6 9 61 80 50 0 0 2 1 4 7 45 0 0 3 1 8 12 40 0 0 0 1 2 3 35 0 0 2 3 4 9 30 0 1 1 0 1 3 25 0 0 0 0 2 2 Ramp 0 3 16 17 62 98 Total 1 22 87 81 391 582 % 0.17 3.78 14.95 13.92 67.18 100

92 Table 32. Ohio Bridge Rail After Barrier Contact Behavior. Behavior 36” # 42” # Combo Rail # All Rail Crashes Contained/Redirected 986 873 0 4,449 PRV 5 2 0 29 Rollover After Redirection 28 27 1 82 Nebraska Bridge Rails The Nebraska DOR crash database was examined for 2007 through 2009 to identify bridge rail collisions on state and local highways, freeways and interstates in Nebraska. The review contained concrete rails and metal rails. These crashes occurred on roads with a variety of posted speed limits. The review of the Nebraska data includes 1,212 crashes on roadways with a variety of posted speed limits. The behavior could be determined for 979 of these crashes. This review included crashes with 29-inch, 34-inch, and 42-inch vertical wall type bridge rails, 32-inch and 42-inch New Jersey shape bridge rails and w-beam type bridge railings. The severity distribution of the crashes which were contained or redirected is shown in Table 33. The behavior of the crashes with these different rails is shown in Table 34.

93 Table 33. Nebraska Crash Severity Distribution. PSL K A B C O Total Contained, Stopped or Redirected for 29" Vertical Wall 50mph or less 0 0 0 0 0 0 55 0 0 0 0 0 0 60 0 2 2 3 9 16 65 0 0 1 1 2 4 75 0 0 0 0 0 0 Total 0 2 3 4 11 20 % 0.00 10.00 15.00 20.00 55.00 100 Contained, Stopped or Redirected for 34" Vertical Wall 50mph or less 0 1 3 4 33 41 55 0 4 3 0 20 27 60 3 6 14 28 131 182 65 1 5 9 11 61 87 75 3 4 14 11 102 134 Total 7 20 43 54 347 471 % 1.49 4.25 9.13 11.46 73.67 100 Contained, Stopped or Redirected for 42" Vertical Wall 50mph or less 0 0 0 0 1 1 55 0 3 3 2 11 19 60 0 0 0 1 1 2 65 0 0 2 2 7 11 75 0 1 0 0 6 7 Total 0 4 5 5 26 40 % 0.00 10.00 12.50 12.50 65.00 100 Contained, Stopped or Redirected for 32" NJ Shape 50mph or less 0 2 4 2 46 54 55 0 2 1 4 19 26 60 0 2 4 6 36 48 65 0 0 4 2 19 25 75 1 0 1 0 14 16 Total 1 6 14 14 134 169 % 0.59 3.55 8.28 8.28 79.29 100

94 Table 33. Nebraska Crash Severity Distribution. (CONT’D) PSL K A B C O Total Contained, Stopped or Redirected for 42" NJ Shape 50mph or less 0 0 1 3 4 8 55 0 1 2 6 20 29 60 2 0 3 9 14 28 65 0 0 1 0 1 2 75 0 0 0 0 0 0 Total 2 1 7 18 39 67 % 2.99 1.49 10.45 26.87 58.21 100 Contained, Stopped or Redirected for W-Beam Guardrail 50mph or less 0 2 0 0 9 11 55 0 1 2 5 11 19 60 2 1 7 7 42 59 65 0 2 5 2 27 36 75 0 0 4 3 30 37 Total 2 6 18 17 119 162 % 1.23 3.70 11.11 10.49 73.46 100 Table 34. Nebraska Bridge Rail After Barrier Contact Behavior. 29" Vertical Wall 34" Vertical Wall 42" Vertical Wall 32" NJ Shape 42" NJ Shape W- beam All Behavior # # # # # # Contained/Redirected 20 471 67 169 67 162 956 PRV 0 6 0 4 0 14 24 Rollover After Redirection 1 11 0 4 1 10 27 After Penetration Hazards The Highway Safety Information System (HSIS) is a multistate database that contains crash, roadway inventory, and traffic volume data for a select group of states. [HSIS01] Crash data from Washington State HSIS data for embankments and water hazards were reviewed. Currently, information is available for embankments in 55 and 70 mi/hr speed zones and water hazards in 55 mi/hr speed zones. Table 35 shows the severity distribution of crashes where a bridge rail was penetrated and the vehicle entered the embankment or water hazard.

95 Table 35. Severity Distributions for After Penetration Hazards Hazard K A B C PDO/UNK Total Embankment in 55 mi/hr Speed Zones No. 7 10 42 33 93 185 % 3.78 5.41 22.70 17.84 50.27 100.00 Embankment in 70 mi/hr Speed Zones No. 3 3 16 5 25 52 % 5.77 5.77 30.77 9.62 48.08 100 Water in 55 mi/hr Speed Zones No. 1 2 6 4 37 50 % 2.00 4.00 12.00 8.00 74.00 100 Summary of Crash Data This crash data has been used throughout this research effort to develop the severity models used within RSAPv3 and to validate the results obtained. The individual analyses conducted with this data are discussed in the relevant sections. Encroachment The probability that a vehicle will encroach (i.e., vehicle leaving the road) on a segment is the first event considered in a series of conditional events evaluated in RSAPv3. These conditional events include: the encroachment probability, the probability of crash given an encroachment, the severity of the crash if an object is struck and the cost of the entire crash sequence. The probability of an encroachment has been the focus of several studies in the last 40 years, however very little successful data collection on the frequency of encroachments has been accomplished. Data collected by Cooper and by Hutchinson and Kennedy have received much attention, but there are few alternate sources of encroachment data. [Cooper82; Hutchinson62] RSAPv3 uses the Cooper data. The data was re-analyzed to attempt to resolve some long-standing problems with the data in NCHRP22-27. [Ray12] The results of the re-analysis included generating baseline encroachment frequencies for two-lane undivided, four-lane and multi-lane divided highways. The base conditions for the encroachment frequencies are: • Posted speed limit of 65 mph, • Flat ground, • Relatively straight segment, • Lane width greater than or equal to 12 feet, and • Zero major access points per mile. Deviating from these base conditions requires the use of adjustment factors to calibrate the encroachment frequency to the specific site conditions. The values shown in Table 36 are the base encroachment frequencies for the base conditions.

96 Table 36. Total Encroachment Frequency by AADT and Highway Type. AADT (bi-directional) 2 Lane Undivided (encr/mi/yr) 4 Lane Divided (encr/mi/yr) One Way (encr/mi/yr) 1,000 1.2244 0.8473 0.4236 5,000 2.6514 3.5915 1.7958 10,000 1.8631 5.8435 2.9217 15,000 0.9819 7.1306 3.5653 20,000 1.3091 7.7344 3.8672 25,000 1.6364 7.8650 3.9325 30,000 1.9637 7.6779 3.8389 35,000 2.2909 7.2870 3.6435 40,000 2.6182 6.7749 3.3874 45,000 2.9455 7.6206 3.8103 50,000 3.2728 8.4673 4.2337 55,000 3.6000 9.3140 4.6570 60,000 3.9273 10.1608 5.0804 65,000 4.2546 11.0075 5.5038 70,000 4.5819 11.8542 5.9271 75,000 4.9091 12.7010 6.3505 80,000 5.2364 13.5477 6.7738 85,000 5.5637 14.3944 7.1972 90,000 5.8910 15.2412 7.6206 95,000 6.2182 16.0879 8.0439 100,000 6.5455 16.9346 8.4673 Encroachment Models for Roads Over Capacity Encroachment modeling programs like RSAPv3, RSAP and BCAP make the assumption that traffic is in a free-flow condition. In light of this assumption, showing AADT values within the selection guidelines which exceed the AADT and percent truck values where free-flow is possible would be a misrepresentation. The user of the selection tables should be aware of this assumption when using these tables. Low-Volume Encroachments The 1989 GSBR was developed using the program BCAP which NCHRP 22-08 modified into the similar program ABC. Both used a constant encroachment rate based on the Hutchison-Kennedy data. RSAP and RSAPv3, in contrast, use a variable encroachment rate based on the Cooper data. One of the consequences is that the Cooper data has a pronounced “hump” at about 25,000 vehicles/day for divided highways and a pronounced trough at 40,000 vehicles/day. After 40,000 vehicles/day the expected number of encroachments increases monotonically. RSAPv3 generally calculates the

97 mid-life number of encroachments and then uses that value in calculating the expected crash costs. If the mid-life ADT turns out to be on the top of the “hump” the encroachments would be overestimated for the entire life and if the mid-life ADT occurs at the bottom of the “trough” the encroachments would be underestimated. In order to avoid this problem, which only happens at low AADTs, the selection guidelines have been developed using a procedure which calculates the number of encroachments at 10 equally spaced times over the life and then takes the average of these values to estimate the encroachments at the mid-life. This is a more realistic estimate of the average encroachment rate over the life of the project, however, a traffic growth rate must be assumed for the development of these guidelines. Annual Traffic Growth Many traffic engineering sources like the Highway Capacity Manual use or recommend a default traffic growth rate of 2 percent. Similarly, the 1989 AASHTO GSBR assumed a 2 percent traffic growth as well. The selection guidelines have been developed by calculating the expected number of encroachments at 10 points during the life with an assumed 2 percent growth rate and then averaged to find the mid-life number of encroachments. A procedure to change these assumptions has been provided within the selection process. Traffic Mix Considerations The GSBR used the Federal Highway Administration (FHWA) 13-vehicle classification system in developing its guidelines. According to Harwood et al., 5-axle tractor-trailer trucks (i.e., Class 9) alone account for 46.1 percent of the trucks on the nation’s highway as measured by the vehicle miles traveled. [Harwood03] Two-axle single-unit trucks (i.e., Class 5) account for 29.5 percent. These two types of trucks alone, then, account for more than 75 percent of the vehicle miles traveled by trucks in the U.S.. After Class 5 and 9, the next highest class is three-axle single-unit trucks (i.e., Class 6) at 5.3 percent and all other classes (i.e., Classes 7, 8, 10-13) each account for less than 2 percent of the vehicle miles traveled and in many cases less than one percent. In fact, some of the higher classes of trucks (e.g., multi-trailer trucks -- Classes 11 through 13) are either not allowed at all, are allowed only by special permit or restricted to particular routes in many states. Clearly the single-unit truck and tractor-trailer truck used in Report 350 and MASH are good representative vehicles for trucks since they account for the majority of vehicle miles traveled. All classes of single-unit trucks comprised about 40 percent of the truck vehicle miles traveled and 60 percent were accounted for by a variety of tractor- trailer trucks. Generally speaking, there are on average about 1.5 times as many tractor- trailer trucks in the average traffic stream as single-unit trucks. The rollover and penetration algorithm used in RSAPv3 was found to be relatively sensitive to the properties of the heavy vehicles (i.e., weight, dimensions and

98 c.g. location). An unloaded Class 9 tractor-trailer truck, for example, weighs about 27,000 lbs whereas a fully loaded Class 9 truck weighs 80,000 lbs and in some states (e.g., Maine) may even weigh more than 100,000 lbs. In addition to selecting the most representative trucks, it is also necessary to select truck properties that accurately reflect the mix of loading conditions experienced on the nation’s roadways. Figure 20 shows a cumulative axle weight distribution of the truck classes from the new Mechanistic- Empirical Pavement Design Guide (MEPDG). [MEPDG04] This data was used to develop the vehicle weights shown in Table 37. The 22,000-lbs single-unit truck used in MASH appears to be a good representation of the 85th percentile two-axle single-unit truck whereas the 80,000-lbs tractor-trailer is more representative of the 95th percentile for Class 9 truck weights based on the MEPDG. [MEPDG04] An understanding of the vehicle properties of each vehicle class is needed for the analyses, however, obtaining vehicle properties for vehicles which have not been crash tested would require estimation. Also, the methods used in RSAPv3 to estimate the probability of penetration and rollover are unlikely to work well with vehicles with multiple articulations. Rather than include all 13 vehicle classes in developing the recommendations it seemed more reasonable to focus on the vehicle types with the highest proportion of vehicle miles traveled. Classes 11 through 13 represent a very small percentage of truck VMT and operate only on selected roadways so there is little harm in ignoring them for general-purpose guidelines. Table 37. Percetile of Gross Truck Weights for Classes 5 and 9. After [MEPDG04] Percentile Gross Weight (lbs) Class 5 Single- Unit Trucks Class 9 Tractor- Trailer Trucks 15th 6,754 35,350 50th 11,942 48,822 85th 21,885 62,069

99 Figure 20. Mean Axle Load by Vehicle Classification. The capacity and rollover algorithms in RSAPv3 are more sensitive to the c.g. height and weight of the vehicle than the particular class so it appears to be more important to represent the range of weights and c.g. locations for the most common Classes than to represent the average condition of all 13 Classes. The following eight vehicle types were used in developing the selection guidelines: 1. Passenger car, 2. Pickup truck, 3. Light Class 5 single-unit truck (i.e., 15th percentile), 4. Average Class 5 single-unit truck (50th percentile), 5. Heavy Class 5 single-unit truck (85th percentile), 6. Light Class 9 combination tractor-trailer truck (15th percentile), 7. Average Class 9 combination tractor-trailer truck (50th percentile), and 8. Heavy Class 9 combination tractor-trailer truck (95th percentile). Vehicles 1, 2, 5 and 8 were specifically chosen because they are essentially MASH test vehicles and, therefore, maintain a link to crash testing specifications. The percent of trucks was varied between zero and 40 percent as was done in the 1989 GSBR. Within the defined percent of trucks, the split was 67 percent tractor-trailer trucks and 33 percent single-unit trucks. Within the defined percentage of passenger vehicles, the split 0 10 20 30 40 50 60 70 80 90 100 0 10000 20000 30000 40000 50000 Cu m m ul at iv e Di st rib ut io n Mean Axle Load (lbs) 4 5 6 7 8 9 10 11 12 13

100 used was 75 percent passenger cars and 25 percent pickup trucks. This strategy results in the traffic mix for each percentage of trucks shown in Table 38. Table 38. Vehicle Mix Used in RSAP to Develop the Guidelines. Percent Trucks Passenger Cars Pickup Trucks Light Tractor Trailer Average Tractor Trailer Heavy Tractor Trailer Light Single Unit Average Single Unit Heavy Single Unit 0 75.00 25.00 0.00 0.00 0.00 0.00 0.00 0.00 5 72.50 22.50 1.00 2.00 0.35 0.50 0.65 0.50 10 70.00 20.00 2.00 4.00 0.70 1.00 1.30 1.00 15 67.50 17.50 3.00 6.00 1.05 1.50 1.95 1.50 20 65.00 15.00 4.00 8.00 1.40 2.00 2.60 2.00 25 62.50 12.50 5.00 10.00 1.75 2.50 3.25 2.50 30 60.00 10.00 6.00 12.00 2.10 3.00 3.90 3.00 35 57.50 7.50 7.00 14.00 2.45 3.50 4.55 3.50 40 55.00 5.00 8.00 16.00 2.80 4.00 5.20 4.00 In addition to the mix and weight properties discussed above, the inertial and geometric properties used for the various vehicle classes are also important input parameters. While the vehicle weight distributions for each class are relatively easy to estimate from weigh-in-motion data, other important data like the location of the center of gravity are very difficult to determine for the in-service fleet. The same truck loaded will have very different properties when it is unloaded so the properties chosen for each type of vehicle need to reflect the appropriate proportion of vehicle miles traveled. The vehicle properties used for the vehicle mix recommended in Table 38 are shown in Table 39. The c.g. height recommended for the tractor-trailer corresponds to the effective c.g. of the trailer, since the trailer rolling over the barrier is generally what pulls the tractor over as well. The effective c.g. is somewhere between the overall c.g. of the vehicle and the c.g. of the trailer. For tractor-trailer trucks, the “weight” listed in Table 39 refers to the weight of the King Pin axle rather than the whole vehicle weight. Tractor- trailer trucks are articulated and crash test data indicates that the second impact with the King Pin axle is generally the most demanding. A King Pin axle weight of 22,000 lbs corresponds to an 80,000 lbs tractor-trailer truck whereas a 6,800-lbs King Pin axle weight corresponds to an empty tractor-trailer truck. The 4.2-ft center of gravity height is based on the crash test measured value of 80,000-lbs tractor-trailer trucks and the 3.4-ft height for the empty tractor-trailer truck is based on the empty-weight location of the center of gravity based on a finite element model. The heavy tractor-trailer in Table 39 corresponds closely to the MASH 36000V vehicle. The upper bound value (trailer c.g.), which resulted in slightly more rollover barrier incidents (i.e., this method is conservative) was used in the analysis. The overall c.g. and the “effective” c.g. used in the generation of the guidelines are listed below.

101 Table 39. Vehicle Properties Used in RSAPv3. RSAPv3 VEHICLES FHWA Vehicle CLASS WEIGHT LENGTH WIDTH C.G. Long. C.G. Hgt Lbs ft ft ft ft Passenger Vehicles 2 3,300 15.00 5.40 6.00 2.00 Pickup Truck 3 5,000 19.75 6.50 8.50 2.30 Light Tractor Trailer 8-9 16,000 48.00 8.50 12.00 4.80 Average Tractor Trailer 8-13 22,250 48.00 8.50 20.00 4.80 Heavy Tractor Trailer 8-13 37,500 48.00 8.50 20.00 6.00 Light Single-Unit Truck 5 6,800 35.00 7.77 12.50 3.40 Average Single-Unit Truck 6 12,000 35.00 7.77 12.50 3.40 Heavy Single-Unit Truck 7 22,000 35.00 7.77 12.50 4.20 The weights shown for the single-unit trucks in Table 39 correspond to the total weight of the truck. The weight and c.g. height for the heavy single-unit truck corresponds to the MASH 10,000S vehicle. The light single-unit truck was based on properties of the same vehicle but in an unloaded condition and average load condition, respectively. The light and average SUT values were determined using a finite element model. Truck Trajectories RSAPv3, like RSAP and BCAP, bases trajectories on information collected from passenger vehicle encroachments. In the case of BCAP and RSAP, the distributions of encroachment speeds and angles were used to create straight-line trajectories. RSAPv3 takes a more realistic approach where actual vehicle trajectories measured in NCHRP 17- 22 were used.[Mak10, Ray12] This allows for a much richer representation of the vehicle trajectory since driver reactions and side-slope conditions are implicitly included in the trajectories. Assuming that trucks and passenger vehicle share the same encroachment characteristics does not seem reasonable for three reasons: 1. The handling and acceleration/deceleration properties of trucks are very different than passenger cars. 2. Trucks may leave the roadway with different speeds and angles than passenger vehicles. 3. Trucks may encroach at a different rate than passenger vehicles. Heavy Vehicle Encroachment Angle Unfortunately, there is no database of heavy vehicle trajectories which includes information relative to departure angle and speed, encroachment rates or trajectories.

102 Using passenger vehicle trajectories would include some trajectories that are clearly difficult to attain for heavy vehicles. Passenger vehicles are smaller, have better braking and acceleration characteristics as well as different inertial properties. BCAP recognized this fact and used the following simple point-mass procedure for limiting the possible encroachment angles based on the vehicle type, offset from the road, available friction and encroachment speed: [AASHTO89] 𝜃 = cos [1 − 𝑆 𝑓 𝑔𝑉 ] where Θmax = The maximum likely encroachment angle in degrees, So = The vehicle offset from the edge of the traveled way in feet, fmax = The maximum available coefficient of friction, g = The gravity constant (i.e., 32.2 ft/s2) and V = The encroachment velocity in ft/s. Table B2 in the AASHTO GSBR presents available friction coefficients of 0.60 for most single-unit trucks and 0.45 for most tractor-trailer trucks. [Mak93] The MEPDG indicates that on two-lane in one direction cross-sections 90 percent of trucks travel in the right most lane so the most common offset value in the above equation would be 6 feet (i.e., half the typical lane width).[MEPDG04] The NCHRP 17-22 data contains 787 trajectories that were included in the RSAPv3 trajectory tables. The maximum achievable encroachment angle using the equation above was compared to the actual encroachment angle for each of these 787 trajectories. If the actual encroachment angle was greater than the maximum achievable encroachment angle for single-unit trucks or tractor-trailer trucks it was excluded from the heavy vehicle analysis. For single-unit trucks, 315 trajectories were found where the actual encroachment angle was less than the maximum achievable and 253 trajectories were found for tractor-trailer trucks. The maximum encroachment angle in the trajectory databases used for both single-unit trucks and tractor-trailer trucks is 32 degrees. Coincidentally, BCAP limited all encroachment angles to 36 degrees. The excluded trajectories represent high-angle, high-speed passenger vehicle encroachment trajectories that would be highly unlikely for trucks. Having different trajectory tables for each type of vehicles is easily accomplished in RSAPv3. Passenger vehicle trajectories (i.e., passenger cars and pickup trucks) are taken from the TrajectoryGrid2 worksheet whereas single-unit truck and tractor-trailer truck trajectories are taken from TrajectoryGrid3 and TrajectoryGrid4, respectively. TrajectoryGrid3 and TrajectoryGrid4 are limited to those trajectories that satisfy the side- friction criteria discussed in the previous paragraphs. Heavy Vehicle Encroachment Rate In addition to considering heavy vehicle trajectory differences, the encroachment frequency differences were also examined. RSAPv3 uses the so-called Cooper data to

103 model the vehicle encroachment frequency. The vehicle type is unknown in this dataset since the data was based on tire marks, however, data collectors were instructed to focus on passenger vehicles. There has been a long-held assumption that heavy vehicles leave the roadway at the same rate as all vehicles. There is no known database of heavy vehicle encroachments and trajectories, however, some organizations have collected heavy vehicle paths and heavy vehicle crash statistics.[NHTSA13;FMCSA12] This research examined the assumption that heavy vehicles encroach onto the roadside at the same rate as passenger vehicles by analyzing a national sample of run-off-road crashes and a detailed regional sample. The results of each are comparable and challenge the assumption that trucks encroach at the same rate. The national crash data examined for this analysis includes: the Federal Motor Carrier Safety Administration (FMCSA) national dataset of commercial truck crashes from 2002 through 2011; the U.S. Department of Transportation National Highway Safety Administration (NHTSA) annual Traffic Safety Facts reports for the years 2002 through 2010; and the FHWA Highway Statistics website for 2002 through 2010.[NHTSA12; FMCSA12; FHWA12c] The detailed regional dataset used for this analysis includes crash and traffic records for 100 miles of the New Jersey Turnpike from 2005 through 2008.[ Plan4S11] NATIONAL DATA The NHTSA Traffic Safety Facts is a nationwide database of different crash types, including run-off road crashes. The data is presented by crash location in relation to the roadway. The available fields are: “On Roadway,” “Off Roadway,” “Shoulder,” “Median,” “Other/Unknown,” and “Total”. The fields “Off Roadway,” “Shoulder,” and “Median” were used to represent run-off road crashes, while the “Total” field was used to represent all crashes.[ NHTSA13] The FMCSA database contained only truck and bus crashes. Event IDs that were labeled as “Non-collision ran off road”, “Non-collision overturn (rollover)”, “Non- collision cross median/centerline”, or “Collision involving fixed object” were considered to be run-off road crashes. FMCSA data for the years 2002-2011 was used.[ FMCSA12] Traffic counts by vehicle classifications were obtained from the FHWA Highway Statistics Series. “Single-Unit 2-Axle 6-Tire or More and Combination Trucks” were used for trucks, “Buses” was used for buses, and “All Motor Vehicles” was used with the NHTSA Traffic Safety Facts data. Beginning in 2007 states were required to report motorcycle data and FHWA implemented a new methodology to calculate traffic counts, therefore, some of the VMT values given for the different years may not fit into the trend lines that adjacent years fit into. This is apparent in Table 40 and should be kept in mind when interpreting the results.[ FHWA12c]

104 REGIONAL DATA The New Jersey Turnpike has a continuous TL5 concrete median barrier from milepost 1.2 to 104.7. Since the median barrier is continuous and relatively close to the left shoulder it is similar to a direct measure of primary and opposing left encroachments. Crash data for left exiting vehicles that were reported as striking the “Concrete Traffic Barrier” under any of the four possible sequence of events fields were considered in the analysis. The frequency of these events was determined for all crashes on this section of highway and for heavy vehicle crashes. Heavy vehicles were defined as one of the following: “Bus/Large Van (9 or more seats)”, “Recreation Vehicle”, “Single-Unit (2 axle)”, “Single-Unit (3+ axle)”, “Single-Unit Truck w/Trailer”, “Tractor Double”, “Tractor Semi-Trailer”, “Truck Tractor (Bobtail)”, “Tractor Triple”, and “Other Truck”. Crashes were then separated by year and the milepost where the crash occurred. The highway segments were defined by the known traffic volumes for given mileposts, not geometrics. The hundred million vehicle miles traveled (100 MVMT) was calculated for each segment and year, using the following equation: 100 𝑀𝑉𝑀𝑇 = (𝐴𝐴𝐷𝑇)(365)(𝑠𝑒𝑔𝑚𝑒𝑛𝑡 𝑙𝑒𝑛𝑔𝑡ℎ 𝑖𝑛 𝑚𝑖𝑙𝑒𝑠)(#𝑦𝑒𝑎𝑟𝑠)100,000,000 where: AADT = Annual Average Daily Traffic, and 100 MVMT = Hundred Million Vehicle Miles Traveled. After the vehicle miles traveled was calculated for each segment, the crash rate was calculated for each segment and year for both heavy vehicles and all vehicles as follows: 𝐶𝑅 = #𝑐𝑟𝑎𝑠ℎ𝑒𝑠(100 𝑀𝑉𝑀𝑇) where: CR = Crash Rate. RESULTS The vehicle miles traveled data collected by the FHWA are shown in the second through fourth columns of Table 40. Table 40 also contains the number of crashes and the crash rates for truck, bus, and truck/bus for the years 2002 through 2010. These national values represent all possible crashes from vehicles which may have encroached from all possible directions (i.e., primary right, opposing right, primary left, opposing left) while the NJTA data only include left encroachments. A dramatic decrease in crash rates can be seen from the years 2006 – 2007, presumably caused at least in part by the methodological changes for reporting MVMT data implemented in 2007.

105 Table 40. National Traffic Volumes, Crashes, and Crash Rates by Year. Year Trucks 100 MVMT Bus 100 MVMT Truck & Bus 100 MVMT Truck and Bus ROR Crashes Bus Only ROR Crashes Truck Only ROR Crashes # ROR Crash/ 100 MVMT # Bus ROR Crash/ 100 MVMT # Truck ROR Crash/ 100 MVMT 2002 2,146 68.45 2,214 24,483 11.06 683 9.98 23,800 11.09 2003 2,179 67.82 2,247 27,102 12.06 772 11.38 26,330 12.08 2004 2,208 68.01 2,276 32,234 14.16 902 13.26 31,332 14.19 2005 2,225 69.80 2,295 33,000 14.38 964 13.81 32,036 14.40 2006 2,225 67.83 2,293 31,754 13.85 946 13.95 30,808 13.85 2007 3,042 145.16 3,187 33,051 10.37 1,025 7.06 32,026 10.53 2008 3,107 148.23 3,255 36,954 11.35 1,471 9.92 35,483 11.42 2009 2,880 143.58 3,024 29,658 9.81 1,295 9.02 28,363 9.85 2010 2,866 137.89 3,004 28,026 9.33 997 7.23 27,029 9.43 Avg 2,542 101.86 2,644 30,696 11.61 1,006 9.88 29,690 11.68 Table 41 was developed from NHTSA FARS/GES data as a representation of run- off-road crash rates for all vehicles. Table 41. NHTSA FARS/GES Crash Data and Crash Rates. Year 100 MVMT Total Crashes “All Crash” Crash Rate (per 100 MVMT) Crashes “Off Road” Crashes “Shldr” Crashes “Med” Total ROR Crashes ROR Crash Rate (crashes per 100 MVMT) 2002 28,555 6,316,000 221.19 790,000 25,000 122,000 1,116,000 39.08 2003 28,902 6,328,000 218.95 865,000 34,000 127,000 1,143,000 39.55 2004 29,648 6,181,000 208.48 969,000 34,000 156,000 1,128,000 38.05 2005 29,894 6,159,000 206.03 913,000 59,000 154,000 1,151,000 38.50 2006 30,144 5,973,000 198.15 859,000 65,000 143,000 1,067,000 35.40 2007 30,311 6,024,000 198.74 948,000 53,000 150,000 1,126,000 37.15 2008 29,765 5,811,000 195.23 919,000 46,000 163,000 1,159,000 38.94 2009 29,535 5,505,000 186.39 954,000 32,000 157,000 1,026,000 34.74 2010 29,665 5,419,000 182.67 946,000 33,000 137,000 937,000 31.59 Avg 29,602 5,968,444 201.76 907,000 42,333 145,444 1,094,778 37.00 Table 42 contains a summary of the ROR crash rates obtained from both the FMCSA (i.e., heavy vehicle ROR crashes) and NHTSA FARS/GES data sets (i.e, all vehicle ROR crashes). The last row contains a simple comparison of the crash rates defined as:

106 𝐶𝑅𝐶𝑅 = 𝑇𝐶𝑅𝑀 where: CRT = Crash Rate of Trucks and Buses (crashes / 100 MVMT), CRA = Crash Rate of All Vehicles (crashes / 100 MVMT), and TCRM = Truck/Bus Crash Rate Multiplier (dimensionless). Table 42. National Heavy Vehicle Crashes Per 100MVMT. Year ROR Bus ROR Truck ROR Truck and Bus ROR All Vehicles TCRM (CRTB/CRA) 2002 9.98 11.09 11.06 39.08 0.28 2003 11.38 12.08 12.06 39.55 0.30 2004 13.26 14.19 14.16 38.05 0.37 2005 13.81 14.40 14.38 38.50 0.37 2006 13.95 13.85 13.85 35.40 0.39 2007 7.06 10.53 10.37 37.15 0.28 2008 9.92 11.42 11.35 38.94 0.29 2009 9.02 9.85 9.81 34.74 0.28 2010 7.23 9.43 9.33 31.59 0.29 Avg 9.88 11.68 11.61 37.00 0.31 This analysis suggests that the long-held assumption that heavy vehicles encroach onto the roadside at the same rate as all vehicles is false. In fact, heavy vehicles appear to encroach at an average rate of approximately one-third of all vehicles. REGIONAL ANALYSIS A review of a regional sample of data was conducted to validate the findings from the national average statistics. The regional sample chosen had detailed traffic volumes, roadway inventory data, and only one hazard was considered (i.e., concrete median barrier). The results for the New Jersey Turnpike analysis are shown in Table 43, Table 44, Table 45, and Table 46 for the years 2005, 2006, 2007, and 2008, respectively. Only primary left and opposing left encroachment directions were considered and the segments are defined solely by available traffic volumes. The truck/bus crash rates of the New Jersey Turnpike appear to be approximately 28 percent of the crash rates for all vehicles which is similar to the 0.31 from the nationwide analysis. This analysis supports the national analysis finding that heavy vehicles have a lower encroachment rate than “all vehicles.”

107 Table 43. NJ Turnpike Ran Off Road Left Crash Rates, 2005. Link and Posted Speed Limit Mileposts Miles Total Crash # Truck Crash # ADT %T 100 MVMT CRA (All) CRT (Trk/Bus) 1 - 2 (65) 1.2 - 12.9 11.7 26 0 45830 15.3 1.96 13.28 0.00 2 - 3 (65) 12.9 - 26.1 13.2 51 2 49177 15.5 2.37 21.52 5.45 3 - 4 (65) 26.1 - 34.5 8.4 31 2 58486 15.3 1.79 17.29 7.29 4 - 5 (65) 34.5 - 44 9.5 45 2 73616 15.1 2.55 17.63 5.17 5 - JCT (65) 44 - 51 7 33 2 79833 14.9 2.04 16.18 6.60 JCT - 7 (65) 51 - 53.3 2.3 20 0 109671 15.5 0.92 21.72 0.00 7 - 7A (65) 53.3 - 60 6.7 62 3 120474 16.2 2.95 21.04 6.29 7A - 8 (65) 60 - 67.6 7.6 57 2 132809 16.7 3.68 15.47 3.25 8 - 8A (65) 67.6 - 73.7 6.1 62 4 137157 16.2 3.05 20.30 8.08 8A - 9 (65) 73.7 - 83.3 9.6 61 6 160390 15.4 5.62 10.85 6.95 9 - 10 (65) 83.3 - 88.1 4.8 31 1 202341 13.6 3.55 8.74 2.07 10 - 11 (65) 88.1 - 90.6 2.5 18 0 187670 13.7 1.71 10.51 0.00 11 - 12 (65) 90.6 - 95.9 5.3 38 0 224591 14.0 4.34 8.75 0.00 12 - 13 (65) 95.9 - 97.3 1.4 7 0 235830 14.4 1.21 5.81 0.00 12 - 13 (55) 97.3 - 99.9 2.6 39 0 235830 14.4 2.24 17.43 0.00 13 - 13A (55) 99.9 - 101.6 1.7 31 0 250812 14.8 1.56 19.92 0.00 13A - 14 (55) 101.6 - 104.7 3.1 33 1 223337 15.3 2.53 13.06 2.59 Average: 26 1.47 148697 15.1 2.59 15.27 3.16

108 Table 44. NJ Turnpike Ran Off Road Left Crash Rates, 2006. Link and Posted Speed Limit Mileposts Miles Total Crash # Truck Crash # ADT % T 100 MVMT CRA (All) CRT (Tks/Bus) 1 - 2 (65) 1.2 - 12.9 11.7 35 0 46789 15.7 2.00 17.52 0.00 2 - 3 (65) 12.9 - 26.1 13.2 40 1 50085 15.8 2.41 16.58 2.62 3 - 4 (65) 26.1 - 34.5 8.4 34 1 59345 15.5 1.82 18.69 3.55 4 - 5 (65) 34.5 - 44 9.5 56 0 74484 15.4 2.58 21.68 0.00 5 - JCT (65) 44 - 51 7 20 0 80861 15.2 2.07 9.68 0.00 JCT - 7 (65) 51 - 53.3 2.3 15 0 110908 15.7 0.93 16.11 0.00 7 - 7A (65) 53.3 - 60 6.7 49 2 121384 16.3 2.97 16.51 4.14 7A - 8 (65) 60 - 67.6 7.6 68 1 133548 16.8 3.70 18.36 1.60 8 - 8A (65) 67.6 - 73.7 6.1 60 1 137794 16.3 3.07 19.56 2.00 8A - 9 (65) 73.7 - 83.3 9.6 51 1 160763 15.5 5.63 9.05 1.14 9 - 10 (65) 83.3 - 88.1 4.8 49 3 203277 13.7 3.56 13.76 6.13 10 - 11 (65) 88.1 - 90.6 2.5 26 1 189709 13.8 1.73 15.02 4.19 11 - 12 (65) 90.6 - 95.9 5.3 50 1 228790 14.0 4.43 11.30 1.61 12 - 13 (65) 95.9 - 97.3 1.4 7 0 240350 14.4 1.23 5.70 0.00 12 - 13 (55) 97.3 - 99.9 2.6 47 5 240350 14.4 2.28 20.61 15.23 13 - 13A (55) 99.9 - 101.6 1.7 43 5 255416 14.8 1.58 27.13 21.28 13A - 14 (55) 101.6 - 104.7 3.1 43 3 226013 15.4 2.56 16.81 7.62 Average: 41 1.47 150580 15.2 2.62 16.12 4.18

109 Table 45. NJ Turnpike Ran Off Road Left Crash Rates, 2007. Link and Posted Speed Limit Mileposts Miles Total Crash # Truck Crash # ADT %T 100 MVMT CRA (All) CRT (Trk/Bus) 1 - 2 (65) 1.2 - 12.9 11.7 24 1 47325 16.0 2.02 11.88 3.08 2 - 3 (65) 12.9 - 26.1 13.2 40 1 50906 16.2 2.45 16.31 2.52 3 - 4 (65) 26.1 - 34.5 8.4 38 3 60384 15.9 1.85 20.53 10.21 4 - 5 (65) 34.5 - 44 9.5 55 1 75790 15.7 2.63 20.93 2.42 5 - JCT (65) 44 - 51 7 38 1 81964 15.5 2.09 18.15 3.08 JCT - 7 (65) 51 - 53.3 2.3 16 0 112538 16.1 0.94 16.94 0.00 7 - 7A (65) 53.3 - 60 6.7 49 4 122467 16.6 2.99 16.36 8.05 7A - 8 (65) 60 - 67.6 7.6 68 3 134428 17.1 3.73 18.24 4.69 8 - 8A (65) 67.6 - 73.7 6.1 38 5 138625 16.7 3.09 12.31 9.73 8A - 9 (65) 73.7 - 83.3 9.6 90 2 161555 15.8 5.66 15.90 2.23 9 - 10 (65) 83.3 - 88.1 4.8 45 0 204774 13.9 3.59 12.54 0.00 10 - 11 (65) 88.1 - 90.6 2.5 30 0 193081 14.0 1.76 17.03 0.00 11 - 12 (65) 90.6 - 95.9 5.3 41 0 232567 14.2 4.50 9.11 0.00 12 - 13 (65) 95.9 - 97.3 1.4 10 0 244033 14.4 1.25 8.02 0.00 12 - 13 (55) 97.3 - 99.9 2.6 53 3 244033 14.4 2.32 22.89 8.97 13 - 13A (55) 99.9 - 101.6 1.7 39 4 258419 14.9 1.60 24.32 16.69 13A - 14 (55) 101.6 - 104.7 3.1 39 2 227390 15.5 2.57 15.16 5.01 Average: 42 1.76 152369 15.5 2.65 16.27 4.51

110 Table 46. NJ Turnpike Ran Off Road Left Crash Rates, 2008. Link and Posted Speed Limit Mileposts Miles Total Crash # Truck Crash # ADT %T 100 MVMT CRA (All) CRT (Trk/ Bus) 1 - 2 (65) 1.2 - 12.9 11.7 21 0 45640 15.4 1.95 10.77 0.00 2 - 3 (65) 12.9 - 26.1 13.2 43 1 48889 15.5 2.36 18.26 2.73 3 - 4 (65) 26.1 - 34.5 8.4 36 0 58134 15.4 1.78 20.20 0.00 4 - 5 (65) 34.5 - 44 9.5 39 1 73248 15.3 2.54 15.36 2.57 5 - JCT (65) 44 - 51 7 31 3 78947 15.1 2.02 15.37 9.85 JCT - 7 (65) 51 - 53.3 2.3 16 1 109200 15.7 0.92 17.45 6.95 7 - 7A (65) 53.3 - 60 6.7 48 1 118768 16.2 2.90 16.53 2.12 7A - 8 (65) 60 - 67.6 7.6 56 2 130011 16.8 3.61 15.53 3.30 8 - 8A (65) 67.6 - 73.7 6.1 46 1 134028 16.3 2.98 15.41 2.05 8A - 9 (65) 73.7 - 83.3 9.6 49 2 155581 15.5 5.45 8.99 2.36 9 - 10 (65) 83.3 - 88.1 4.8 40 1 197707 13.6 3.46 11.55 2.12 10 - 11 (65) 88.1 - 90.6 2.5 10 1 187069 13.9 1.71 5.86 4.23 11 - 12 (65) 90.6 - 95.9 5.3 38 4 225401 14.0 4.36 8.71 6.55 12 - 13 (65) 95.9 - 97.3 1.4 8 0 236947 14.4 1.21 6.61 0.00 12 - 13 (55) 97.3 - 99.9 2.6 62 3 236947 14.4 2.25 27.57 9.27 13 - 13A (55) 99.9 - 101.6 1.7 34 2 251121 15.1 1.56 21.82 8.51 13A - 14 (55) 101.6 - 104.7 3.1 35 2 221602 15.6 2.51 13.96 5.12 Average: 36 1.47 147602 15.2 2.56 14.70 3.98

111 Table 47. Truck/Bus Crash Rate Multipliers, by Year and Link. Link and Posted Speed Limit Mileposts TCRM 2005 TCRM 2006 TCRM 2007 TCRM 2008 1 - 2 (65) 1.2 - 12.9 0.00 0.00 0.26 0.00 2 - 3 (65) 12.9 - 26.1 0.25 0.16 0.15 0.15 3 - 4 (65) 26.1 - 34.5 0.42 0.19 0.50 0.00 4 - 5 (65) 34.5 - 44 0.29 0.00 0.12 0.17 5 - JCT (65) 44 - 51 0.41 0.00 0.17 0.64 JCT - 7 (65) 51 - 53.3 0.00 0.00 0.00 0.40 7 - 7A (65) 53.3 - 60 0.30 0.25 0.49 0.13 7A - 8 (65) 60 - 67.6 0.21 0.09 0.26 0.21 8 - 8A (65) 67.6 - 73.7 0.40 0.10 0.79 0.13 8A - 9 (65) 73.7 - 83.3 0.64 0.13 0.14 0.26 9 - 10 (65) 83.3 - 88.1 0.24 0.45 0.00 0.18 10 - 11 (65) 88.1 - 90.6 0.00 0.28 0.00 0.72 11 - 12 (65) 90.6 - 95.9 0.00 0.14 0.00 0.75 12 - 13 (65) 95.9 - 97.3 0.00 0.00 0.00 0.00 12 - 13 (55) 97.3 - 99.9 0.00 0.74 0.39 0.34 13 - 13A (55) 99.9 - 101.6 0.00 0.78 0.69 0.39 13A - 14 (55) 101.6 - 104.7 0.20 0.45 0.33 0.37 Average: 0.20 0.22 0.25 0.28 SUMMARY The regional analysis found that, on average, heavy vehicles experienced an encroachment rate that was 28 percent of the rate for all vehicle types. The national analysis found heavy vehicle ROR crashes occur at a rate of 31 percent of the all vehicle run-off-road crashes. When modeling run-off-road crashes, it appears necessary to reduce the number of heavy vehicle encroachments by about 30 percent to account for the smaller likelihood of these vehicles encroaching. There are several possible reasons for this reduced likelihood of heavy vehicles encroaching. First, heavy vehicles are not as maneuverable as passenger vehicles and they have much more restrictive acceleration and deceleration capabilities. Second, heavy vehicles are operated by trained professional drivers who must operate their vehicles in accordance with specific requirements including hours of rest available to the driver. While the causes of this reduced encroachment rate are speculative, the data examined demonstrates that the number of encroachments that can be expected from heavy vehicles is only about 30 percent of passenger vehicles.

112 Encroachment Adjustments for Site-Specific Characteristics The 1989 AASHTO GSBR includes three adjustment factors; a horizontal curvature adjustment factor, a grade adjustment factor and an adjustment factor for deck height and under-structure conditions. The adjustments for the grade and horizontal curvature from the 1989 AASHTO GSBR are shown in Table 48 and Table 49. These values were also used in RSAP 2.0.3 as well as the latest version of RSAP, RSAPv3. While NCHRP 17-54 is in the process of updating these adjustments, the values shown in Table 48 and Table 49 are the best available data at the present time. These adjustment factors should be used until suitable replacements can be made. Table 48. Grade (Fgrade) Adjustment Factor. Grade Fgrade <= -6 2 -6 2 -4 1.5 -2 1 >= -2 1 Table 49. Horizontal Curve (Fhcurv) Adjustment Factor. Degree of Curvature Radius of Curvature Fhcurv <= -6 -9545 4.0 -5 -1145 3.0 -4 -1430 2.0 -3 -1910 1.0 0 ∞ 1.0 3 1910 1.0 4 1430 1.3 5 1145 1.7 >= 6 955 2.0 RSAPv3 also includes adjustment factors for the number of lanes, lane width, posted speed limit and the access density. These adjustment factors have been incorporated into the selection guidelines. The 1989 AASHTO GSBR bridge height adjustment to adjust for the severity of the under-bridge conditions was applied in the 1989 AASHTO GSBR inappropriately to the number of expected encroachments since the increase in severity should only apply to those cases where the vehicle penetrates the bridge railing. The GSBR applies the factor to all crashes regardless of whether a penetration occurred or not. In RSAPv3, the condition of the area under the bridge is accounted for by a special edge hazard. The severity of crossing the special edge is only calculated and included for those cases where the vehicle crosses the special edge by penetrating, rolling over or vaulting over the

113 bridge railing. The area under the bridge is accounted for in the selection tables and described in full in a later section. It is not treated as an encroachment adjustment factor. Bridge Shoulder Offset The 1989 GSBR and NCHRP 22-08 provided the four offset distances (i.e., bridge shoulders) in the selection guidelines: • 0-3 ft (nominal 1 ft) • 3-7 ft (nominal 4 ft) • 7-12 ft (nominal 8 ft) • >12 ft (nominal 12 ft) The effect of shoulder width on crash severity has been one of the more interesting features of this research. BCAP, ABC and the 1989 AASHTO GSBR all assume that encroachments follow a straight path and are in a state of constant deceleration. Under these assumptions, any trajectory has a lower severity (i.e., slower speed and same angle) the farther the vehicle travels so wider shoulder widths always result in reduced crash severity. RSAPv3, as discussed above, uses actual trajectories collected in NCHRP 17-22. These trajectories are sometimes straight, sometimes curved to the left, sometimes curved to the right and sometimes have compound curvatures. These various trajectory curvatures are due to the driver’s response to leaving the roadway. In early RSAPv3 runs it was discovered that the crash severity for a particular trajectory sometimes increased as the shoulder width increased. The reason is that some trajectories leave the road with the angle relative to the roadway increasing as the trajectory travels further off the road. If, for example, the vehicle left the roadway with a speed and angle that were close to indicating rollover or penetration with a one-ft offset, the speed and angle at a four-ft offset sometimes was enough to indicate penetration or rollover. Simply stated, the vehicle sometimes would not penetrate or rollover when the offset was one foot but would penetrate or rollover at a higher offset because the angle was increasing. For example, when performing runs for a one foot shoulder with impacts by the average single-unit truck, RSAPv3 selected 40 trajectories that matched the site conditions (i.e., speed limit, flat side-slope, horizontal curvature and grade). Of these 40 trajectories, five penetrated or rolled over 24-inch high low-profile barrier. When the offset was increased to 4 feet, seven trajectories resulted in rollover or penetration. The reason was that the angle for some of the trajectories was increasing so for some trajectories, a rollover/penetration did not occur at one-ft but did occur at four-ft due to the higher angle. The offset where the equivalent adjusted ADT is minimized appears to be at about 8 ft. As the offset further increases the ADT and percent truck values increase slightly as well so it appears that the 8-ft offset is the limiting value. Since the worst-case value always occurs at a shoulder offset of 8 ft, the selection guidelines have been based on the shoulder offsets at 8 ft and the shoulder offset should not be a consideration in the

114 selection. Doing so is conservative for narrower shoulders. If, for example, an agency built a bridge with a 12-ft shoulder anticipating converting it to an 11-ft lane and 1-ft shoulder in the future, the bridge railing selected would not change since it is based on the more critical 8-ft shoulder offset. Crash The probability of a collision given that a vehicle has encroached onto the roadside or median is determined in RSAPv3 by directly projecting reconstructed vehicle trajectories onto the roadside or median and determining if the trajectory intersects the position of any hazard, in this case bridge railings. Figure 21 provides a simplified representation of the steps which the crash prediction module takes to determine if a collision occurs; if a terrain rollover occurs; or if nothing happens and the encroachment results in a non-crash event.[Ray12] As shown in Figure 21, when examining a trajectory, there are many possible outcomes. For example, a trajectory may interact with a roadside hazard, then either stop in contact, penetrate, or be redirected. The probability of each of these events is calculated and the outcome of each sub-event is evaluated. For example, if the vehicle penetrates the barrier, the trajectory is followed further to determine if it interacts with another hazards or results in a rollover. If the trajectory is redirected, the redirected paths are evaluated. Predicted Penetration, Rollovers and Vaults The ability to reasonably predict the number of roll-over-the-barrier, vault the barrier and penetrate-the-barrier crashes is critical to obtaining correct crash costs and so, is also critical to obtaining correct results. In RSAPv3, a penetration implies a complete structural failure of the barrier which allows the vehicle to pass through. A rollover-the- barrier is when the vehicle rolls over the barrier and off the bridge whereas a redirection rollover is one in which the vehicle rolls over but remains on the bridge. The capacity values for TL2 through Report 350 TL four used in developing these guidelines are based on taking 1.6 times the recommended AASHTO bridge railing design loads from Table A13.2-1 of the LRFD Bridge Design Specification as shown in Table 50. [AASHTO12] The values used for MASH TL4 and TL5 are based on recommendations from a recent TXDOT study that examined what recommended values should be used for the new MASH test criteria. [AASHTO09, Sheihk11] When compared to available crash tests this tends to provide a good estimate of the lower bound of crash tested strengths when complete barrier failure might occur.

115 Figure 21. RSAPv3 Crash Prediction Module Flow Chart. 1. Select trajectories for a uniform segment. 2. Map selected trajectories onto roadside. 3. Examine trajectories for intersections with hazards. 4. Collision is detected. 4a. Probability of rollover before collision? 4b. Probability of penetrating hazard? 4c. Probability of redirection? Repeat analysis. 5. No collision detected. Probability of terrain rollover? 6. Increment forward.

116 Table 50. Bridge Railing Load Capacities. Test Level Barrier Height AASHTO LRFD BCAP ABC RSAPv3 Load Capacity (inches) (kips) (kips) (kips) (kips) MASH TL 2 24 -- -- -- 43.2 MASH TL 3 27 27 15 30 43.2 R350 TL 4 32 54 35 64 86.4 MASH TL 4† 36 80 -- -- 128.0 MASH TL 5† 42 160 55 108 256.0 † These values are being considered by but have not been adopted by AASHTO for a future revision of the LRFD Bridge Design Specifications. They are based on [Sheihk11]. NCHRP 22-08 performed a crash study of bridge railing crashes in Texas. [Mak93] As discussed in the literature review, there were numerous problems analyzing and interpreting the data as well as some serious coding issues. In the end, however, it appeared that for bridge railings installed after 1965 the percentage of all vehicle types that penetrated or rolled over the bridge railing was 3 percent (i.e., 1.1 percent penetrated and 1.9 percent rolled over the bridge railing). Since NCHRP 22-08 was completed in 1993 it is presumed that the majority of bridge railings in the TXDOT data would, at best, represent PL1 bridge railings in the AASHTO GSBR or a mixture of TL2 and TL3 railings in NCHRP Report 350 or MASH. RSAPv3 models the edge-of-bridge hazard as a line on the back side of the bridge railing. If a vehicle crosses this line for any reason, the “hazard” is consider contacted. Vehicles can cross this line by penetrating through the bridge railing, vaulting over it or rolling over it. These types of events are called PRV in RSAPv3. Consider an example case of a 60 mi/hr highway, RSAPv3 results in the penetration, roll-over-the-barrier and redirection rollover values shown in Table 51. Table 51 also compares the RSAPv3 predictions to the average values for passenger vehicles (i.e., passenger cars and pickups), the average single-unit truck and the average tractor-trailer truck. Unfortunately, there is no crash data available for the low-profile concrete bridge railing (i.e., TL2) since there is only a small inventory of that particular barrier installed. In general, RSAPv3 tends to slightly over predict roll-over-barrier and penetration collisions so the method is still conservative. Approximately 6,450 bridge rail crashes in Pennsylvania, Ohio, and Nebraska were reviewed in this study as described earlier. Only 89 of these events (i.e., 1.4 percent) were heavy vehicle crashes. Unfortunately, the crash data does not identify specific types or loadings of vehicles but if the RSAPv3 predictions are weighted by the vehicle mix percentages discussed above they can be combined into a likely penetration and rollover percentage that can be compared to the crash data collected in this project and in NCHRP 22-08 (see Table 52). Given that the TXDOT data was collected in the 1980’s it is assumed that bridges constructed after 1965 would likely have bridge railings that

117 conform to the 1989 GSBR PL1 but there was probably still relatively little inventory of PL2 or PL3. It is assumed in Table 51, therefore, that the crash data is mainly representative of PL1 railings. While the rollover and penetration percentages predicted by RSAPv3 are still higher than observed crash data indicate, they appear to be much more reasonable than the predictions of either BCAP or ABC. Given the acknowledged unreliability of the NCHRP 22-08 crash data and the relatively few heavy vehicle cases that could be obtained in the current study, the RSAPv3 PRV procedures were used in the development of these guidelines. The results are on the conservative side but appear to be reasonable given the uncertainty in the crash data. One should also consider that the capacity loads represent the low end of the spectrum – specific bridge railings could be much stronger based on the geometric and reinforcement details used.

118 Table 51. RSAPv3 Predictions of Penetration and Rollovers compared to NCHRP 22-08 TXDOT Crash Data. V eh ic le Br id ge R ai l H ei gh t ( in ) RSAPv3 Prediction NCHRP 22-12(3) Crash Data Pe ne tr at io n R ol lo ve r Ba rr ie r R ed ir ec tio n R ol lo ve r R ed ir ec te d Pe ne tr at io n R ol lo ve r Ba rr ie r R ed ir ec tio n R ol lo ve r R ed ir ec te d Passenger Car 24 0.3 0.0 2.0 97.7 27 0.3 0.0 2.0 97.7 32 0.0 0.0 2.0 98.0 36 0.0 0.0 2.0 98.0 42 0.0 0.0 2.0 98.0 Pickup Truck 24 4.3 0.0 2.0 93.7 27 4.4 0.0 2.0 93.6 32 0.0 0.0 2.0 98.0 36 0.0 0.0 2.0 98.0 42 0.0 0.0 2.0 98.0 All Passenger Vehicles 24 1.3 0.0 2.0 96.7 -- -- -- -- 27 1.3 0.0 2.0 96.7 5.0 5.6 18.8 70.6 32 4.3 0.0 2.0 93.7 0.6 1.9 5.6 95.7 36 2.5 0.0 2.0 98.0 0.1 0.3 2.7 96.9 42 0.9 0.0 2.0 98.0 0.0 0.2 3.2 96.6 Light Single-Unit Truck 24 3.4 12.5 4.0 80.2 27 3.8 2.9 3.5 89.8 32 0.0 5.0 0.0 95.0 36 0.0 0.0 0.0 100.0 42 0.0 0.0 0.0 100.0 Average Single- Unit Truck 24 8.8 7.4 3.8 80.1 -- -- -- -- 27 9.3 0.6 2.4 87.7 12.2 4.9 19.5 63.4 32 2.1 3.1 0.0 94.8 2.4 4.9 2.4 90.2 36 0.0 0.0 0.0 100.0 0.0 2.4 4.9 92.7 42 0.0 0.0 0.0 100.0 0.0 0.0 4.9 95.1 Heavy Single-Unit Truck 24 33.5 2.0 15.0 49.5 27 34.4 0.0 6.2 59.4 32 11.5 7.3 4.4 76.8 36 5.5 1.2 3.9 89.5 42 0.1 0.2 0.0 99.7 Light Tractor- Trailer Truck 24 22.7 14.3 10.7 52.4 27 23.1 5.6 10.6 60.7 32 4.5 16.2 6.1 73.2 36 2.5 13.7 4.4 79.3 42 0.0 3.0 3.3 93.7 Average Tractor- Trailer Truck 24 7.5 15.1 19.1 58.2 -- -- -- -- 27 8.0 9.2 19.0 63.7 12.2 4.9 19.5 63.4 32 1.1 14.1 11.4 73.4 2.4 4.9 2.4 90.2 36 0.0 4.9 14.2 80.8 0.0 2.4 4.9 92.7 42 0.0 2.7 3.4 93.9 0.0 0.0 4.9 95.1 Heavy Tractor- Trailer Truck 24 31.5 13.9 19.5 35.1 27 32.4 8.7 20.3 38.6 32 9.3 16.7 25.0 49.0 36 2.8 21.1 19.1 57.0 42 0.0 16.3 15.9 67.8

119 Table 52. Comparison to Crash Data of RSAPv3 predictions of Penetrating, Rolling over or Vaulting the Bridge Railing for all Vehicle Classes. Bridge Railing Height (Inches) BCAP (GSBR) ABC (NCHRP 22-08) RSAPv3 Full Mix RSAPv3 Average Mix NCHRP 22-12(3) Crash Data % % % % % 24 26.8 20.5 -- 27 20.8 14.8 17.1 32 32.7 10.1 15.0 11.9 7.3 36 7.5 3.3 2.4 42 2.8 1.8 0.0 Table 53. Comparison of 1988-1990 TXDOT Bridge Crash Data for bridges built after 1965 with RSAPv3 Predictions for MASH TL3. Vehicle Type Vehicle Retained on Bridge Vehicle Left Bridge TXDOT RSAPv3 TL3 TXDOT RSAPv3 TL3 Passenger Car 97.9 99.7 2.2 0.3 Pickup Truck 95.4 95.6 4.7 4.4 Single-Unit Truck 97.7 90.1 2.4 9.9 Tractor-Trailer Truck 92.3 82.7 7.8 17.3 Severity Once the probability of leaving the roadway and the probability of striking the bridge rail have been calculated, it is necessary to estimate the likely average severity of the crash in order to appropriately apportion the crash costs. RSAPv3 introduced the Equivalent Fatal Crash Cost Ratio (EFCCR) as a measure of crash severity. “EFCCR65 is a single, dimensionless measure of crash severity with a particular roadside feature at a baseline speed of 65 mi/hr.”[Ray12] The EFCCR65 allows for direct comparison of hazard severity between different hazards. The values are based on observable police- reported crashes and adjusted to account for unreported crashes. “Using the EFCCR65 to estimate crash severity in a conditional probability model like RSAPv3 provides a systematic methodology based on observed data and established crash severity relationships.”[Ray12] This approach removes the subjectivity of previously used crash severity models. The EFCCR can be considered the probability of a fatal injury crash given that an impact has occurred. EFCCR65 values were determined for bridge railings and for vehicles which leave the bridge. Bridge Railing Crash Severity An impact with a bridge railing is composed of several possible events each of which has its own associated severity: the impact with the bridge railing itself, the

120 potential for another harmful event if the vehicle is redirected (e.g., rolling over in the roadway or striking another barrier) or leaving the bridge structure and falling to the area below the bridge. This section deals with the first hazard – striking the bridge railing itself. The development of RSAPv3 included research on the severity of various longitudinal barriers. Table 54 shows an abbreviated list of EFCCR65 values, percent of PRVs and percent of impact-side rollover (i.e., RSS) for the longitudinal barriers of interest to this project. The EFCCR is the equivalent fatal crash cost ratio which is the average crash cost divided by the fatal crash cost. If a roadside feature has an EFCCR of 0.0035, for example, and the fatal crash cost is $6,000,000 then the average crash cost on a 65 mi/hr roadway is 0.0035∙6,000,000=$21,000. These data were obtained from several data sets as well as from the literature. Ray et al provided a summary of data sources in the RSAPv3 Engineer’s Manual. [Ray12] Table 54. EFCCR65 of Longitudinal Barriers used in RSAPv3.[after Ray12] Hazard EFCCR65 %PRV %RSS TL3 27” Vertical Wall 0.0098 TL3 27” NJ SS BR 0.0066 10.08 0.00 TL4 34" Vertical BR 0.0070 0.00 0.00 TL4 34" SS MB 0.0020 0.17 1.01 TL4 32” NJ SS 0.0042 0.06 0.29 TL4 32” F-shape 0.0087 1.38 1.37 TL5 42" Vertical BR 0.0035 0.00 0.00 TL5 42" SS BR 0.0037 0.00 0.00 TL5 42” NJ SS 0.0020 0.15 0.53 TL5 42” F-shape 0.0035 0.67 1.60 The EFCCR is based on observed crashes where the vehicle did not penetrate, over-ride or roll over the feature so it represents the ideal result of a crash (i.e., redirection or stopping in contact with the barrier). For closed-faced concrete barriers, there should be little if any difference in the severity of crashes that result in redirection. In developing the bridge railing warrants, the EFCCR65 used for all closed-face concrete barriers was 0.0035 and the percentage of PRV was set to zero since the penetration and rollover algorithms determine the appropriate value based on the barrier height and vehicle properties. The EFCCR65 value of 0.0035 was used to represent the crash severity with all test levels of bridge railings in the development of these selection guidelines. This crash severity applies only to cases where the vehicle is redirected or stops in contact with the barrier. Rollovers that occur during or after redirection are assigned an EFCCR65 of 0.0220 consistent with the usual RSAPv3 analysis. The

121 EFCCR65 appropriate for use when the vehicle penetrates, rolls over or vaults over the barrier is discussed in the next section. Bridge Railing Penetration Severity For purposes of estimating crash severity, the 1989 GSBR and NCHRP 22-08 assumed that penetrating the bridge railing resulted in a 35 ft drop. The subjective severity index (i.e., SI) method was used to rate the severity of the crashes. The 35-ft drop assumption could be modified using an adjustment factor. RSAPv3 uses a very different severity method based on observed crash data. Striking a bridge railing actually includes several possible outcomes in RSAPv3—the severity of the crash with the bridge railing itself (i.e., discussed in the last section), the possibility of being redirected into another hazard (i.e., striking another barrier or rolling over in the roadway) and penetrating or rolling over a barrier. Penetrating through or rolling over the bridge rail is represented by the edge-of-bridge hazard in RSAPv3. The severity of this hazard is a function of characteristics of the area beneath the bridge. While each bridge rail has a distinct probability of penetration which does not change with the area around the bridge, the possibility of causing harm after the penetration occurs does change after the penetration. For example, a bridge railing on a bridge in a very rural area over a stream will cause no harm to others aside from the occupants of the vehicle. On the other hand, a bridge railing on a bridge over an urban street in a heavily populated area has much more potential for causing harm. This difference in the consequences of penetrating the bridge railing is explained through three types of edge- of-bridge hazards as follows: HIGH: A high-hazard environment below the bridge includes possible interruption to regional transportation facilities (i.e., high-volume highways, transit and commuter rail, etc.) and/or damage to a densely populated area below the bridge. Penetrating the railing may limit or impose severe limitations on the regional transportation network (i.e., interstates, rail, etc.). Penetrating the railing also has the possibility of causing multiple fatalities and injuries in addition to the injuries associated with the vehicle crash itself. Nearby facilities where a collision could lead to a catastrophic loss of life such as chemical plants, nuclear facilities or water supplies should be considered high-hazard environments. A high-hazard environment is also present if penetration or rolling over the bridge railing could lead to the vehicle damaging a critical structural component of the bridge (e.g., a through-truss bridge). MEDIUM: A medium hazard environment below the bridge includes possible interruption to local transportation facilities, large water bodies used

122 for the shipment of goods or transportation of people, and/or damage to an urban area which is not densely populated (i.e., single family homes, single office buildings, etc.). Penetrating the railing would limit local transportation routes, however, detours would be possible and reasonable. Penetrating the railing has the possibility of causing at least one non-motor vehicle injury or fatality. LOW: A low-hazard environment below the bridge includes water bodies not used for transportation, low-volume transportation facilities, or areas without buildings or houses in the vicinity of the bridge. Penetrating a low hazard railing would have little impact on regional or local transportation facilities. A low hazard railing has no buildings or facilities in the area which present possible non- motor vehicle related victims of a rail penetration. Bridge rail crash data was examined in Pennsylvania, Ohio, and Nebraska. Only 38 penetrations were found in these censuses of crash data. These 38 bridge rail penetrations provide possibly the only understanding of the consequence of penetrating a bridge rail. The probability of penetration is presented above, however, the severity of a crash which does penetrate the rail was determined from this census of data. While catastrophic penetration events often are news worthy, a census of police-reported data should be used to understand the severity of these crashes to remove any bias toward more catastrophic crashes. Zero motorcycles, 26 passenger cars, and 12 heavy vehicles penetrated the bridge rails resulting in five fatal crashes, 13 A injury crashes, 5 B injury crashes, 6 C injury crashes, 8 PDO crashes and one crash of unknown severity. This results in an EFCCR of 0.1584. This value was used as the medium level hazard discussed above. A value of 0.0584 was used for the Low-level discussed above. This is equivalent to reducing the K + A crashes to one each. A value of 1 was used for the High- level, which represents absolute certainty that a fatality will be observed every time the rail is penetrated which is consistent with a catastrophic crash. Table 55. Bridge Railing Penetration Hazard Severities. Bridge Penetration Severity EFCCR65 Low 0.0584 Medium 0.1584 High 1.0000 The low severity bridge rail penetrations have a severity that is similar to an on- road rollover (i.e., 0.0220) which seems reasonable since in both cases vehicle occupants and the vehicle itself are the only cost components that are at risk.

123 Costs When conducting an RSAPv3 analysis, the crash costs of each feasible alternative are determined. A benefit-cost ratio (B/C) for each feasible alternative will be calculated with benefits in the numerator and agency costs in the denominator. Project benefits, in this case, would be defined as a reduction in crash costs between the alternatives under consideration. Project costs include the design, construction, and maintenance costs associated with each alternative. RSAPv3 determines the crash costs of each user entered roadside design alternative. The three conditional probabilities: (1) the encroachment frequency, (2) the probability of a crash given an encroachment and (3) the probability of an injury given a crash have been discussed above. The results of these analyses are converted to a monetary unit of measure for direct comparison with project costs. The B/C ratio, therefore, is unitless. The benefit-cost ratio (BCR) is defined as follows: BCRi/j= where: BCRi/j = Incremental BCR of alternative j with respect to Alternative i, CCi , CCj = Annualized crash cost for Alternatives i and j and DCi , DCj = Annualized direct cost for Alternatives i and j. For each alternative, an average annual crash cost is calculated by summing the expected crash costs for the predicted crashes. These crash costs are then normalized to an annual basis. Any direct costs, as defined by the user (i.e., initial installation and annual maintenance) are also normalized using the project life and the discount rate to an annualized basis and the BCR is calculated. The project life will therefore influence the results, regional variation in agency costs will influence the results and temporal variations in agency costs and crash costs will influence the results. The following sections discuss these influences and the costs used in the development of the selection guidelines. Project Life The 1989 GSBR assumed a 30-year life for bridges. The AASHTO Red Book generally recommends a 30-year service life for most transportation projects but the AASHTO LRFD Bridge Design Specification recommends in Section 1.2 a 75-year service life for bridge structures. [AASHTO12, AASHTO07] Replaceable portions of a bridge are sometimes assumed in the AASHTO LRFD specification to have a service life of between 30 and 50 years assuming that the component is either refurbished or maintained at the end of that period. Bridge railings are certainly a long-lived portion of a bridge structure that would generally only be replaced if the structure is being replaced or if the deck is being replaced or refurbished. Choosing a longer service life amortizes the construction cost over a longer period so higher performance railings would be cost-

124 beneficial at lower traffic volumes. Conversely, a shorter service life amortizes the construction cost over a smaller period so higher performance railings would be cost- beneficial at higher traffic volumes. In short, a long service life will result in the use of more high performance barriers and a shorter service life will result in proportionally fewer high performance barriers. A design life of 30 years as was used in the development of these selection guidelines, as was done in the AASHTO GSBR. Regional Cost Variations Construction Costs The Washington Department of Transportation (WSDOT) preformed a survey of highway agencies within the United States in 2002 to better understand all project related costs and to gauge how WSDOT costs relate to other States. WSDOT found the average construction cost nationwide was $2.3 Million per lane-mile of highway in 2002. This figure excludes “…right of way, pre-construction environmental compliance, and construction environmental compliance and mitigation.” [WSDOT09] These exclusions are quite variable by project and region, let alone state so excluding them allows the comparison to be based only on highway construction costs. Design costs, or the costs related to preparing a project for construction, are generally accepted to be approximately 10 percent of the construction costs of the project. Using the data gathered by the Washington Department of Transportation, the relative cost of each state’s construction to the national average can be determined and adjustment factors for regional variations in construction cost can be developed to adjust costs to the national average construction cost. The relative comparison of each responding state to the average value was determined and is shown in Figure 22. States with a value of one have approximately the same construction costs as the national average construction costs, while states with a relative value higher than 1.0 have construction costs which are higher than the average by the multiple shown. For example, New York has a relative cost of 3.6, therefore, it costs more than three times as much to construct a lane-mile of highway in New York than the national average. Arkansas and California both have approximately average construction costs, while states like Mississippi, Michigan and Montana have below average construction costs. The regional adjustments vary from a low of 0.44 in Mississippi to a high of 3.63 in New York State. These data show that construction costs reported from different states in the same year vary widely by region. Crash Costs Bahar collected 18 crash costs components from each state to determine the relative crash cost per state. [Baher11] These included the cost of: police, EMS, fire, emergency incident management, Medicaid, coroner, employee medical, employer cost, lost wages excluding taxes, insurance administration, at-fault liability, property damage, legal, court, roadside hardware repair, state tax loss, state welfare safety net, and

125 vocational rehabilitation. This list of cost components is essentially the same as that used by Miller in his 1988 study of nationwide crash costs. [Miller88] Bahar developed a regional crash cost application tool which includes adjustments for comprehensive crash costs from the national average to each as shown in Table 56.[Baher11] For example, Virginia has a regional crash cost adjustment of 1 indicating crash costs in Virginia are the same as the national average. On the other hand, crash costs in Washington D.C. are 1.61 times higher than the national average and those in Alaska are 64 percent of the national average. Like construction costs, therefore, crash costs also vary widely by region. Comparison The regional crash cost adjustment factors from Table 56 are shown alongside the previously introduced construction cost adjustment factors from Figure 22 in Figure 23. Also shown in Figure 23 is the ratio of the adjustment factors for crash cost to construction costs. This adjustment factor ratio is constructed in the same way a benefit/cost ratio would be: the crash cost adjustment is divided by the construction cost adjustment. One might think that these adjustments would generally cancel each other out resulting in a value near unity. In other words, if crash costs are higher in a particular state one might think that the construction costs are higher by the same proportion. In reality, as shown in Figure 23, this is not true. States such as New York, New Jersey and Hawaii have considerably higher construction cost adjustments than crash cost adjustments, while states like Mississippi, New Mexico, and North Carolina have considerably higher crash cost adjustments than construction cost adjustments. A few states do have equal crash and construction cost ratios (e.g., Massachusetts, Oregon and South Dakota) but they are the exceptions. The ratio of crash costs to construction costs is 36 percent of the national average in New York but twice the national average in Mississippi. Although both crash costs and construction costs vary by region they do not necessarily vary in the same proportion. As an example, say a national roadside safety guideline was developed using a BCR of 2 and national average construction and crash costs. The policy would result in a project whose actual regionally adjusted BCR was 2 ∙ 0.36 = 0.72 in New York and 2∙ 2.02 = 4.04 in Mississippi. The guideline would have the unintended effect of recommending a non-cost-beneficial project in New York and a very cost-beneficial project in Mississippi.

126 Figure 22. Lane-Mile Cost Comparison by State. [after WSDOT09] 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 A riz on a A rk an sa s Ca lif or ni a Co lo ra do H aw ai i Id ah o Ill in oi s K an sa s Lo ui sia na M ai ne M as sa ch us et ts M ic hi ga n M iss iss ip pi M on ta na N ew Je rs ey N ew M ex ic o N ew Y or k N or th C ar ol in a O hi o O kl ah om a O re go n So ut h D ak ot a W as hi ng to n W es t V irg in ia W yo m in g R el at iv e V al ue B as e= 1 Relative Cost/Lane Mile Base Value

127 Table 56. Crash Cost Adjustments by State to the National Average. [After Bahar11] State Adj. State Adj. State Adj. Alabama 0.80 Kentucky 0.87 North Dakota 0.88 Alaska 0.64 Louisiana 0.89 Ohio 0.92 Arizona 0.86 Maine 1.08 Oklahoma 0.80 Arkansas 0.91 Maryland 1.10 Oregon 0.87 California 1.23 Massachusetts 1.31 Pennsylvania 1.01 Colorado 0.90 Michigan 0.91 Rhode Island 1.11 Connecticut 1.57 Minnesota 1.22 South Carolina 0.79 Delaware 0.92 Mississippi 0.89 South Dakota 0.73 D.C. 1.61 Missouri 0.82 Tennessee 0.74 Florida 0.85 Montana 0.79 Texas 0.75 Georgia 0.84 Nebraska 0.94 Utah 0.89 Hawaii 1.56 Nevada 0.96 Vermont 1.10 Idaho 0.93 New Hampshire 0.77 Virginia 1.00 Illinois 0.98 New Jersey 1.28 Washington 1.09 Indiana 0.91 New Mexico 0.89 West Virginia 0.91 Iowa 0.90 New York 1.32 Wisconsin 1.02 Kansas 0.95 North Carolina 0.96 Wyoming 1.04

128 Figure 23. Regional Crash Cost and Construction Cost Adjustment Factors Relative to a Base of One. 0 0.5 1 1.5 2 2.5 3 3.5 4 A riz on a A rk an sa s Ca lif or ni a Co lo ra do H aw ai i Id ah o Ill in oi s K an sa s Lo ui sia na M ai ne M as sa ch us et ts M ic hi ga n M iss iss ip pi M on ta na N ew Je rs ey N ew M ex ic o N ew Y or k N or th C ar ol in a O hi o O kl ah om a O re go n So ut h D ak ot a W as hi ng to n W es t V irg in ia W yo m in g R el at iv e V al ue B as e= 1 Crash Costs Relative Const. Cost/Lane Mile Relative Crash Cost/Construction Cost Base Value

129 Temporal Cost Variations Further compounding the regional variations in costs are the variations of costs in time due to general economic variations. For example, the U.S. economy was robustly growing in the period between 2003 and 2008 but contracted in 2008 leading to a national recession. Construction Costs Since 2003 the FHWA has been collecting highway construction data and using it to calculate the National Highway Construction Cost Index (NHCCI). [NHCCI11] This index can be used to convert or compare current construction expenditures to other years. The NHCCI base of one is relative to the first year of data collection, 2003. When costs increase relative to the 2003 base year, the index is greater than 1 whereas when costs fall below the 2003 year values the index is less than 1. Figure 24 shows the NHCCI index for 2003 through 2012. In 2006, the NHCCI index was 1.35 meaning construction costs were 1.35 times higher than they were in 2003. In 2010, the NHCCI was 1.06 meaning construction costs had decreased almost back to the level of 2003. Figure 24 shows, as would be expected, that highway construction costs decreased by more than 20 percent between 2006 and 2010 in response to the general economic conditions at the time. Figure 24. NHCCI Index for 2003 through 2010. Crash Costs Miller et al. conducted a study in 1988 which determined the comprehensive costs of crashes related to the KABCO scale commonly used on police crash reports to describe the severity of a crash.[Miller88] Each letter of the scale equals a different severity (e.g., K for a 1.00 1.07 1.18 1.35 1.29 1.29 1.10 1.06 1.07 1.11 0.90 0.95 1.00 1.05 1.10 1.15 1.20 1.25 1.30 1.35 1.40 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 In de x Ba se Y ea r 2 00 3 = 1. 0 Year

130 fatal injury and O for a property damage only crash) and results in a different comprehensive cost. Miller noted that “these costs should be updated annually using the GDP implicit price deflator.”[Miller88] FHWA then updated this study to 1994 dollars. [FHWA09] FHWA issued a memorandum in 2008 which suggested that the GDP implicit price deflator should no longer be used to update the comprehensive costs of crashes but rather the value of statistical life (VSL) should be used instead. The memorandum notes “the relative values of injuries of varying severity were set as a percentage of the economic value of a life.” These values are still being reviewed by FHWA and the relative values may be modified in the future. In 2008, a VSL of $5.8 million was established. In 2009 the VSL was changed to $6.0 million. The value was updated in 2011 to $6.2 million and again in 2012 to $9.1 million. [FHWA08, FHWA09, FHWA11, FHWA12b] FHWA plans to periodically issue updates to the VSL rather than having users update the comprehensive costs through updates to the GDP, as suggested previously by Miller. Using the relative values of injuries established in 1988 and the 2008, 2009, 2011 and 2012 VSLs provided by FHWA, Table 57 reflects the current comprehensive cost of crashes. Table 57. Comprehensive Crash Costs by Year. Crash Severity Cost per Crash 1994 2008 2009 2011 2012 K $2,600,000 $5,800,000 $6,000,000 $6,200,000 $9,100,000 A $180,000 $401,538 $415,385 $429,231 $630,000 B $36,000 $80,308 $83,077 $85,846 $126,000 C $19,000 $42,385 $43,846 $45,308 $66,500 PDO $2,000 $4,462 $4,615 $4,769 $7,000 Comparison After adjusting both the crash and construction indexes to a base year of 2008, the first year the VSL was available, the data can be directly compared as shown in Figure 25. The triangles represent the increasing VSL over the past 5 years while the diamonds represent the variation in construction costs. While the construction cost varies with general economic conditions, the VSL monotonically increases causing the values to diverge. These diverging values will have tremendous implications to benefit-cost analysis conducted from one year to the next. For example, an alternative which was cost-beneficial in 2010 when construction costs were relatively low may not have been cost-beneficial in 2008 when construction costs were higher. These significant changes can impact the choice of a preferred alternative when using an incremental benefit-cost analysis.

131 . Figure 25. Comparison of Annual VSL to NHCCI Index Updates. While benefit-cost will always be a valuable tool for choosing among feasible alternatives, the temporal and regional variations of both crash costs and construction costs create a problem when developing national guidelines that are intended for long-term use across all regions of the country. For example, say a roadside design guideline was developed in 2008 that assumed a decision BCR of 2. By 2011 the construction costs would decrease by a factor of 0.82 but the crash cost would have increased by a factor of 1.07 so the benefit-cost of that same alternative would really be 2.6; even better than the conditions of the original policy. On the other hand, if construction prices begin to increase dramatically in the coming years to an NHCCI of 2.6 and crash cost reaches $12 million the same alternative will have a BCR of 1.5, below the threshold value of 2 used to develop the guideline. Since design alternatives generally have design lives of 20 to 30 years, it would seem highly likely that the actual BCR will change dramatically over the life of the project due to temporal and regional variations. This points out a difficulty of performing benefit-cost analyses during an economic contraction since the real construction costs will decrease but, as a political matter, the value of statistical life is not likely to ever decrease since devaluing the VSL would imply the Federal government places less value on the lives of its citizens. The 2012 values for VSL and NSCCI have been used to adjust costs from different regions and different times to a 2012 national average to develop one option presented in these selection guidelines, however, a risk-based guideline development method which is independent of these temporal and regional variations was explored and ultimately recommended for implementation. The risk-based method is 0.60 0.80 1.00 1.20 1.40 1.60 1.80 20 08 20 09 20 10 20 11 20 12 In de x Ba se Y ea r 2 00 8 = 1. 0 NHCCI Index VSL Index

132 discussed later, but first the costs and adjustments used in the development of the benefit-cost tables are presented. Bridge Railing Agency Costs Construction Costs Construction bid prices for various test level bridge railings from a variety of states were reviewed, adjusted and averaged using the NCHHI index and WSDOT study discussed above to determine the 2012 national average construction prices shown in Table 58. The 1989 AASHTO GSBR assumed the bridge railing construction costs shown on the left side of Table 58. As shown in Table 58, the ratios between the construction cost of each test level bridge railing with respect to TL3 is very similar for both the cost data acquired in this project and that used in the 1989 GSBR. It is also interesting to note that if the 1989 GSBR costs are inflated to 2012 dollars (i.e., 23 years at 4% interest) they are similar to the values obtained in this project from recent construction bid projects in a variety of states so the construction costs appear to be consistent. The construction cost for the TL2 low-profile concrete barrier and the MASH TL4 concrete barrier are estimates. There have only been a couple of TL2 low-profile barriers built so their documented cost is unreasonably high. In developing the values in Table 58 it is presumed that if the low-profile concrete barrier were built in sufficient volume the cost per linear foot would be just under the cost of a TL3 (i.e., 27” tall) concrete safety shape. While some crash tests have been performed with the new 36” tall MASH TL4 concrete barriers, there is no known installed inventory so the cost shown in Table 58 was interpolated based on the better documented costs of Report 350 TL4 and TL5 barriers. Table 58. National Average Construction Costs for Closed-Profile Concrete Bridge Railings. Performance Level GSBR Construction Cost $/LF GSBR Ratio wrt PL1 Equivalent Test Level 2012 Construction Cost $/LF Ratio wrt TL3 -- -- -- MASH/R350 TL2 100 0.9 PL1 28.80 1.0 MASH/R350 TL3 110 1.0 PL2 43.62 1.5 Report 350 TL4 165 1.5 -- -- -- MASH TL4 240 2.1 PL3 68.96 2.4 MASH/R350TL5 325 2.9 Demolition Costs A similar review of the cost to remove bridge rail was undertaken. Maryland, Colorado, Oregon and Vermont specifically identify this line item in their standard specifications. Maryland and Colorado list the 2010 weighted average bid price per linear foot as $15 and $6 respectively while Vermont lists the 2011 weighted average bid price as $10.59 per linear foot. Oregon summarized the 2009 through 2011 weighted average bid prices as $57.23 per linear

133 foot. After applying the state and annual factors then averaging the values, a resulting 2012 value of $23 per linear foot was obtained as the national average cost per linear foot for the removal and disposal of bridge rail. This cost is only applicable when considering the rehabilitation of bridges and upgrading from an existing railing to a new railing. Repair and Maintenance Costs Repair costs for bridge railings can be significant but are quite difficult to determine and often not tracked separately. One respondent to the survey conducted during this research suggested an estimate of $54 per linear foot. A review of the Texas and Oregon weighted average bid prices suggests that the 2011 concrete repair price ranges from $65 to $175 per square foot of exposed area, depending on the depth of the repair. A recent project from the State of Florida to enhance its bridge management software includes cost elements for all types of bridge construction and repair. [Sobanjo11] The data contained 176 winning bids for work on bridge barrier repairs and retrofits. The 2009 repair of railing line item ranged from $215 to $3,880 per each event, however, the railing type or material was not specified. Many states prefer concrete bridge railings because for the vast majority of crashes there will be little important barrier damage. In the rare case of a major collision involving a heavy vehicle, however, concrete bridge railings can fail catastrophically. Such catastrophic damage may including the complete structural failure of the concrete barrier itself as well as failure of the bridge deck since many concrete bridge railings are constructed integral with the deck. It appears that a reasonable price for minor repairs would be $110 per square foot. Bridge Railing Crash Costs The comprehensive costs of all vehicle crashes is discussed above and shown in Table 57 for 1994 through 2012. Zaloshnja and Miller determined the comprehensive cost of various truck crashes by truck size. [Zaloshnja06] These costs were reported by total crash cost by most severe injury and by cost of injury per victim. These costs include the following categories: “(1) medically related, (2) emergency services, (3) lost productivity (wage and household work), and (4) the monetized value of pain, suffering, and lost quality of life.” [Zaloshnja06] A summary of the findings are presented in Table 59. Table 59 includes the annual number of truck crashes by crash severity and truck type as well as the comprehensive cost of truck crashes by crash severity and truck size.

134 Table 59. Annual Number and 2005 Cost of Truck Crash by Injury Severity and Truck Type. [After Zaloshnja06] Truck Type Max Severity in Crash Annual Number of Crashes Cost per Crash Truck Type Max Severity in Crash Annual Number of Crashes Cost per Crash St ra ig ht tr uc k, no tr ai le r K 1,016 $3,136,409 Tr uc k tra ct or w ith 2 or 3 tr ai le rs K 150 $3,352,753 A 2,612 $640,494 A 1,129 $121,936 B 4,665 $198,225 B 559 $244,084 C 17,491 $62,364 C 740 $116,920 O 116,476 $13,286 O 4,976 $24,883 U* 527 $44,307 U* - - Unk 7,245 $22,114 Unk 420 $30,872 St ra ig ht tr uc k w ith tr ai le r K 162 $3,142,831 M ed iu m / h ea vy tr uc k K 87 $3,105,969 A 594 $363,436 A - - B 517 $220,440 B 259 $235,327 C 1,359 $91,530 C 455 $78,442 O 12,502 $17,295 O 3,143 $10,072 U* 20 $45,990 U* 6 $34,734 Unk 1,277 $23,396 Unk 1,767 $19,435 Bo bt ai l K 37 $3,172,568 A ll m ed iu m / h ea vy tru ck s K 4,278 $3,604,518 A 858 $381,348 A 16,035 $525,189 B 266 $173,507 B 23,955 $180,323 C 1,269 $64,324 C 40,774 $78,215 O 9,843 $19,089 O 326,121 $15,114 U* 59 $22,923 U* 1,024 $38,661 Unk 786 $22,401 Unk 21,685 $23,479 Tr uc k tra ct or W ith o ne tr ai le r K 2,825 $3,833,721 A 10,843 $437,845 U*-Injury, unknown severity B 17,688 $171,710 Unk-unknown severity C 19,461 $90,959 O 179,181 $15,749 U* 413 $33,397 Unk 10,191 $24,939

135 The FMCSA updated these costs to reflect the FHWA 2008 updated VSL. These updated 2008 costs are shown in Table 60. The cost of a fatal truck crash in 2005 was approximately $3.6M, while the 2008 estimate jumped to $7.2M which is a reflection of the increase in the statistical value of life. [FMCSA08] Knowing the relationship between the cost per crash and the cost per victim in 2005, the relationship between the 2005 cost per crash and the 2008 cost per crash, along with the published VSLs for 2008 and 2012, an estimate of the 2012 crash cost per victim was calculated. This information is also shown in Table 60. Table 60. 2005 Cost of All Truck Crashes by Injury Severity and per Victim. [After Zaloshnja06] Truck Type Max Severity in Crash Annual Number of Crashes 2005 Cost Per Crash 2005 Cost per Victim 2008 Cost per Crash 2012 Cost per Victim All medium/ heavy trucks K 4,278 $3,604,518 $3,055,232 $7,200,310 $9,575,482 A 16,035 $525,189 $325,557 $1,049,107 $1,020,746 B 23,955 $180,323 $134,579 $360,209 $421,788 C 40,774 $78,215 $62,702 $156,241 $196,516 O 326,121 $15,114 $5,869 $30,191 $18,395 U* 1,024 $38,661 $33,759 $77,228 $105,809 Unknown 21,685 $23,479 $20,540 $46,901 $64,375 Using the information from Zaloshnja and Miller an approximate comprehensive cost of the crashes investigated by NTSB or reported in the media discussed earlier in this report can be determined. [Zaloshnja06] These costs are presented to show the possible range of crash costs for these catastrophic bridge rail crashes. This comprehensive cost data is appropriate for cost/benefit analysis when evaluating the cost of crashes that might happen against the cost of safety improvements. However, the crash cost values published by the National Safety Council (NSC) are best suited for calculating the cost of crashes which have already happened. The NSC values can be used to measure the economic loss (i.e., economic impact) of crashes. The NSC values do not include what people are willing to pay for improved safety. [NSC11] The NSC values are therefore lower than the comprehensive cost values. The NSC crash costs per victim values for 2011, as well as the inflation adjusted calculated values for 2012 are shown in Table 61. Using these values and the comprehensive crash cost values, the costs of each NTSB investigated or media reported crash was determined and is shown in Table 62 and Table 63.

136 Table 61. NSC Economic Impact Crash Costs.[NSC11] Severity 2011 Cost /Victim 2012 Cost/Victim Death $1,420,000 $1,449,386 Nonfatal Disabling Injury $78,700 $80,329 Property Damage Crash (including non-disabling injuries) $9,100 $9,288 Table 62. Summary of the Cost of Bridge Rail Crashes in the Media. Crash 2012 Comprehensive Crash Costs NSC 2012 Crash Cost Additional Unknown Costs St. Petersburg, Florida, 2001 $9,823,890 $1,467,962 Glenmont, New York, 2007 $18,395 $9,288 Wiehlthal Bridge, Germany, 2004 $9,575,482 $1,449,386 $400 Million in Repairs to Structure Boston, Massachusetts, 2007 $843,576 $18,576 San Francisco, California, 2009 $9,575,482 $1,449,386 Amesbury, Massachusetts, 2011 $9,575,482 $1,449,386 Avon, Colorado, 2012 $9,612,272 $1,467,962 Syracuse, New York, 2012 $843,576 $160,658 Montreal, Quebec, 2011 $19,150,964 $2,898,772 Avellino, Italy, 2013 $364,926,406 $55,169,548 Beaverton, Oregon, 2012 $421,788 $9,288 Bronx, New York, 2012 $67,028,374 $10,145,702 Grand Prairie, Texas, 2013 $9,575,482 $1,449,386 Boston, Massachusetts, 2013 (1) $214,911 $9,288 Boston, Massachusetts, 2013 (2) $18,395 $9,288 Buellton, California, 2012 $11,018,016 $1,539,003 Galesburg, Illinois, 2013 $9,681,291 $1,458,674 Williamsburg, Kansas, 2012 $51,082,801 $7,509,756

137 Table 63. Summary of the Cost of Bridge Rail Crashes Investigated by NTSB. Crash 2012 Comprehensive Crash Costs NSC 2012 Crash Cost Additional Unknown Costs Fort Sumner, New Mexico, 1972 $183,521,293 $27,677,654 Nashville, Tennessee, 1973 $76,603,856 $11,595,088 Siloam, North Carolina, 1975 $45,050,536 $5,946,152 Martinez, California, 1976 $277,688,978 $42,032,194 New bridge constructed Houston, Texas, 1976 $76,286,894 $10,349,580 Destroyed 94 feet of bridge railing, damaged bridge deck, and column supporting overpass was sheared off Elkridge, Maryland, 2004 $38,301,928 $5,797,544 Huntsville, Alabama, 2006 $62,825,006 $7,321,033 Sherman, Texas, 2008 $167,824,682 $25,072,835 The costs of repairs to bridges are considerably higher than any other given section of highway and generally not considered in the comprehensive and NSC crash cost figures presented above. In addition to these calculated crash cost values, for example, the bridge rail crash in Wiehlthal Germany ultimately required that the entire bridge be reconstructed. [Wiehlthal04] The temporary traffic control and bridge repair costs totaled approximately $400 million. The bridge rail crash in Martinez, California [Marinez76] required that a new bridge be constructed in addition to the approximately $277 million in calculated crash costs. Sight Distance Considerations The AASHTO LRFD Bridge Design Specification (LRFD) states in Section C2.3.2.2.2 “Special conditions, such as curved alignment, impeded visibility, etc.., may justify barrier protection, even with low design velocities.” The AASHTO Policy on Geometric Design of Highways (Green Book) offers this guidance for sight distance considerations: “For stopping sight distance calculations, the height of object is considered to be 600 mm [2.0 ft] above the road surface. … The basis for selection of a 600 mm [2.0 ft] object height was largely an arbitrary rationalization of the size of object that might potentially be encountered in the road and of a driver’s ability to perceive and react to such situations.” [AASHTO01] For passing sight distance calculations, the height of object is considered to be 1,080mm [3.5 ft] above the road surface.” “It is not necessary to consider passing sight

138 distance on highways or streets that have two or more traffic lanes in each direction of travel.” [AASHTO01] When “horizontal sight distance on the inside of a curve is limited by obstructions” measurements are taken at an average of the stopping sight distance and passing sight distance height (i.e., 2.75 ft). “Such refinement on two-lane highways generally is not necessary and measurement of sight distance along the centerline of traveled way edge is suitable.” [AASHTO01] For Intersection Sight Distance, “the determination of whether an object constitutes a sight obstruction should consider both the horizontal and vertical alignment of both intersecting roadways, as well as the height and position of the object.” In making this determination, it should be assumed that the driver’s eye is 1,080mm [3.5ft] above the roadway surface and that the object to be seen is 1,080mm [3.5ft] above the surface of the intersecting road. “The recommended dimensions of the sight triangles vary with the type of traffic control used at an intersection….” [AASHTO01] It seems the LRFD and Green Book have somewhat conflicting guidance for improving visibility along the roadway, where the LRFD suggests considering a barrier if visibility is limited and the Green Book limits object height to 2 feet (24 inches) for stopping sight distance (SSD), 3.5 feet (42 inches) for passing sight distance and 2.75 feet (33 inches) for horizontal sight distance (HSD). All TL2 through TL5 bridge railings are 24 inches high or taller so even a TL2 bridge railing is a sight obstruction for stopping sight distance since the obstruction height is 24 inches. The guidelines presented herein use a shoulder width of 8 feet so if, as suggested by the AASHTO Green Book, the sight line is taken from the “centerline of the traveled way edge” and the lane width is assumed to be 12 feet, then the middle ordinate of the circular curve is 20 feet (i.e., 12+8=20). Knowing the stopping sight distance based on the design speed and the middle ordinate, the minimum radius of the horizontal curve which would not present a stopping sight distance obstruction was calculated and is shown in Figure 26 and Table 64. Also shown in Figure 31 is the Green Book recommended minimum horizontal curvature based on 10 percent super-elevation and design speed. As shown in Figure 26 and Table 64, the minimum radius imposed by the stopping sight distance obstruction of the barrier is always greater than the minimum Green Book radius from Green Book Exhibit 3-14 and the disparity increases with the design speed. This is true even if the shoulder width is increased to 14 feet or more which is probably an unreasonable shoulder width to expect on a typical bridge. Stopping sight distance is generally considered necessary at all points along the highway but the above analysis suggests that it is not possible to supply the appropriate stopping sight distance for radii less than the dashed line in Figure 31 with any bridge railing. The same curves apply for “HSD and indicate that MASH TL4 and TL5 bridge railings would limit the HSD when the radius of the curve is less than the dashed line in Figure 26. If HSD were considered in the bridge rail selection then the values shown below in Table 64 could

139 be used as minimum radii. The Green Book recommendation for the HSD object height is 33 inches. Intersection sight distance (ISD) can be impeded by bridge rail choice. Options may exist early in design to realign the intersection to provide an acceptable sight triangle. There are plans to study improving roadside safety hardware choices at bridge ends/intersection in the near future that may provide some additional options and insight to this problem (i.e., NCHRP Project 15-53). The Green Book recognizes the continual development of roadside hardware through a single statement: “highway designers should recognize the dynamic developments currently under way in the entire area of roadside design. … Highway designers should endeavor to use the most current acceptable information in their designs.” Maintaining the intersection sight triangles may necessitate stopping or modifying the barrier as the end of the ramp or bridge is approached. Figure 26. Minimum Horizontal Curve Radius Based on Barrier Obstruction to the Stopping Sight Distance Compared to AASHTO Exhibit 3-14. 0 500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 25 30 35 40 45 50 55 60 65 70 75 M in im um C ur ve R ad iu s ( ft) Design Speed (mi/hr) AASHTO Ex. 3-14 with e<=10% Min. Radius Based on Barrier Obstruction to SSD

140 Table 64. Minimum Radius and Maximum Degree of Curve Based on the HSD Obstruction from a TL4 or TL5 Bridge Railing. Design Speed Minimum Radius Maximum Degree of Curve 25 141 40.7 30 238 24.0 35 375 15.3 40 561 10.2 45 805 7.1 50 1,119 5.1 55 1,512 3.8 60 1,999 2.9 65 2,592 2.2 70 3,305 1.7 75 4,153 1.4 Analysis Methods The objective of this project was to establish selection guidelines for test level two through five bridge railings. The publication of MASH at the onset of this research required some evaluation of which crash testing procedures the barriers guidelines should be based on. A decision was made to limit the selection guidelines to MASH performance criteria, thereby having this research completely coordinated with the current literature on crash testing bridge railings. The alternatives considered in the analyses were, therefore, MASH TL2 through MASH TL5. The analyses performed to develop the selection procedures were performed using RSAPv3 release 130912XL14. [Ray12] RSAPv3 is the software implementation of a conditional probability model for estimating the number and severity of crashes based on roadway, traffic and site conditions. The model has three basic parts: • Estimating the number of vehicle encroachments (i.e., the encroachment module), • Estimating the probability of a vehicle striking a hazard if it leaves the roadway (i.e., the collision module) and • Estimating the expected crash cost or severity of a collision with a roadside object if it occurs (i.e., the severity module). Details on the methods and algorithms used in RSAPv3 functions are available elsewhere and will not be repeated here (see [Ray12]) other than those specifically related to or updated for this project. The inputs to the conditional probability model and necessary changes to RSAPv3 needed to accomplish the goals of this project include determining: • Encroachment characteristics for heavy vehicles,

141 • Encroachment characteristics over the project life at low traffic volumes (i.e., averaging out the Cooper data “humps”), • Vehicle mix characteristics, • Penetration, rollover, and vault potential for each test level of bridge railing, • The severity of a bridge rail penetration if it does occur, and • An improved understanding of how temporal and regional variations in crash costs and construction costs impact the development of national guidance development. Each of these issues has been discussed in detail in an earlier section of this report. Using these inputs and having made the necessary changes to RSAPv3, the four alternatives (i.e., MASH TL2 through TL5) were evaluated across a range of traffic volumes and heavy vehicle percentages to determine the expected annual crash cost of each alternative. Using the expected annual crash costs, a benefit-cost approach or a risk approach can be taken to generate the selection guidelines. Benefit-Cost Versus Risk Approach There are advantages and disadvantages to both methods of analysis. The benefit-cost method has the advantage that it includes both societal benefits (i.e., reduction in crash costs) and agency costs (i.e., construction, maintenance and repair) such that the benefits are maximized while making the best possible use of agency funds. The disadvantage is that since costs are explicitly included, regional and temporal variations in the cost elements can make the same solution cost-beneficial in one region and not cost-beneficial in another. Similarly, an alternative that is cost-beneficial under one set of economic assumptions may become not cost-beneficial if economic conditions change in the future. Another disadvantage to a benefit-cost approach is that the risk is not necessarily uniform so one cost-beneficial solution can have a different inherent risk than another with the same BCR. On the other hand, risk analysis sets a specific risk objective that is uniform across regions and through time such that the risk of an unacceptable event is always the same. Temporal and regional variations in either the crash or construction cost do not change the underlying risk of each alternative. The disadvantage is that the best risk-based solution may not always be cost-beneficial in every region or at every point in time. The primary advantage to a risk-based approach is that construction and maintenance costs do not affect the results so the performance goal will not change over time or from one region to another. This allows the policy decision about the appropriate level of safety that should be provided to be separated from the question of which alternative is the most economically attractive. A further advantage is that policy decisions for new construction, rehabilitation, or retrofitting would be identical using a risk-based approach. For example, a decision to install 32-inch concrete bridge rail for 20,000 vpd and 10% trucks would be set regardless of whether the barrier will be installed where no barrier currently exists (i.e., new barrier construction costs only) or for replacing an existing barrier (i.e., new barrier construction costs plus demolition and removal of an existing barrier costs). The risk-based goal remains the same for both problems.

142 Benefit-cost analysis has been a valuable tool in roadside safety for nearly 35 years and has been used to both prioritize specific projects as well as develop state and national guidelines for barrier selection and placement. One the one hand, benefit-cost analyses helps transportation agencies make the most effective use of their limited roadside safety funding. On the other hand, the level of risk implicit in these decisions is usually hidden, resulting in different levels of risk based on the time an analysis was performed and the region where the alternative was placed in service. The two tools can, in fact, be used together when guidelines are developed based on acceptable risk criteria. Benefit-cost methods can be used to determine the most economical way of achieving the desired risk level on a project-by-project basis. Agencies in different regions may chose different roadside safety alternatives based on the economic situation in their locale but the overall level of risk would be uniform throughout the nation. For these reasons, a risk method has been used in the development of the selection guidelines. While the selection guidelines are presented as a risk-based approach, the process through which the guidelines are applied is very flexible. Additional tables have been provided in Appendix A for those that prefer a cost-benefit approach. Both sets of tables and the selection process are discussed below. Developing the Selection Guidelines The most straight-forward way to develop selection guidelines would be to simply run a large number of RSAPv3 analyses and present them in table format with the construction year AADT listed in the rows and percent trucks in the columns and each entry indicating the appropriate test level. For three different highway types and three hazard environments this would have resulted in at least nine pages of tables not including additional material needed for the roadway adjustment factors like horizontal curvature and grade as well as space to explain the procedure. Adding a dozen pages to Chapter 13 of the LRFD Bridge Design Manual for bridge railing selection did not seem to be the best approach so another more concise method was developed as discussed below. The encroachment-based procedure used by RSAPv3, as discussed above, is naturally broken into three parts: probability of encroachment, probability of collision and expected severity. For bridge railings, the bridge railing is located relatively close to the edge of the traveled lane (i.e., shoulders are generally between two and 12 ft) so the probability of collision estimation in a bridge railing problem is fairly simple and is easily combined with the severity portion of the analysis. The analysis can be deconstructed into two parts – encroachment and collision/severity – such that the guidelines can be made much more concise. The problem of estimating the risk associated with observing a severe or fatal crash along a 1000-ft segment of bridge railing over its 30-year life is analogous to a coin toss problem and, therefore, follows the binomial distribution: 𝑃(𝑘) = 𝑛!𝑘! (𝑛 − 𝑘)! 𝑝 (1 − 𝑝)

143 where, P(k) = Probability or risk of observing k failures (i.e., severe or fatal crashes) in n trials, n = Number of trials (i.e., the number of encroachments over the life of the bridge) k = Number of failures (i.e., severe or fatal crashes) observed and p = Probability of a failure in any one trial. The equation above shows that if a fair coin (i.e., a fair coin is one where the probability of “tails” = probability of “heads” = 0.5) is tossed 10 times, the probability that “heads” will appear twice is: 𝑃(𝑘) = 𝑛!𝑘! (𝑛 − 𝑘)! 𝑝 (1 − 𝑝) = 10!2! (10 − 2)! 0.5 (1 − 0.5) = 0.0440 The chance of observing exactly two “heads” in a series of 10 fair coin tosses is 0.0440 or approximately 1 in 22 (i.e., 1/22~0.0440). Stated another way, if 22 people flip a coin 10 times, one of those persons should get two heads and eight tails and the other 21 will get some other combination. While the probability of observing exactly two heads is very small, it is not zero. In the case of a severe or fatal crash involving a bridge railing, the number of trials is the number of encroachments expected over the 30-year life of the bridge railing and the number of failures is the number of severe or fatal crashes that should occur in that time period. The term on the left hand side of the equation, P(k), is the risk of observing one or more crashes with a particular severity over the life of the bridge railing. Only three values are required to calculate the risk of a severe or fatal crash over the life of a bridge railing in this problem: n = The number of encroachments over the 30-year life of 1,000-ft of bridge railing, k = 1 and p = Probability of a severe or fatal crash in any encroachment. Estimate the Number of Lifetime Encroachments on 1,000-ft of Bridge Railing Input to the guidelines are the construction year AADT, anticipated traffic growth and percent of trucks in the traffic stream as well as roadway characteristics like the highway type, speed limit, horizontal curvature, grade, lane width and access density. The first step in the analysis, therefore, is to predict the number of encroachments that can be expected over the life of the bridge railing based on the highway type and AADT. As discussed earlier, the assumed life of a bridge railing is 30 years and a traffic growth of 2 percent per year was assumed in developing the guidelines. Bridges, of course, vary in length so it is necessary to select a standard unit length for the analyses. A unit length of 1,000-ft was chosen. Any length could have been used to develop the selection guidelines but the same length needs to used throughout the process since the risk criterion includes units of length.

144 RSAPv3 uses the so-called Cooper data to estimate the number of encroachments as a function of highway type and AADT. The Cooper data, which was collected over a wide area in Canada in the 1970’s, along with the statistical methods used to develop the relationships pictured in Figure 27 are discussed in much more detail in the RSAPv3 Engineer’s Manual.[Ray12] One of the interesting features of the Cooper data shown in Figure 27 are the pronounced “humps” in each of the curves. For two-lane undivided highways, the top of the “hump” occurs at about 5,000 veh/day and for four-lane divided highways it occurs at about 30,000 veh/day. The relationship is a little clearer if the encroachment rate (i.e., encroachments per million vehicle miles traveled) is examined as is shown in Figure 28. The encroachment rate is highest for two-lane undivided roadways at low volumes. The encroachment rate for all highway types decays rapidly as the lane volume increases. Eventually each curve reaches an asymptotic value; The two-lane undivided curve reaches the value of 0.0448 encroachments/MVMT for all lane volumes greater than about 8,000 veh/day and the four-lane divided curves reaches the value of 0.1160 encroachments/MVMT for all values greater than about 10,000 veh/day. Some may think the relationships shown in Figure 27 are counter-intuitive since four-lane divided highways have a larger number of expected encroachments than two-lane undivided highways at the same traffic volume regardless of whether rates or frequencies are examined. One important fact to remember in viewing Figure 26 and Figure 27 is that the Cooper data predicts encroachments not crashes. While the observational Cooper data suggests that more vehicles encroach on the roadside on four-lane undivided highways, the roadside of four-lane divided highways generally feature much more generous clearzones, roadsides with less dramatic slopes and many fewer fixed objects so while there may be more encroachments, there are fewer crashes. In developing the selection guidelines, RSAPv3 analyses were performed for the following conditions: - Two-lane undivided, four-lane divided and one-way highway types, - 1,000-ft long highway segment, - 500 veh/day ≤ AADT ≤ 110,000 veh/day, - 30 year design life, - 2 percent annual traffic growth and - Flat, tangent sections with 12-foot lanes and 8-ft shoulders. The results were then tabulated by AADT and highway type as shown later in Figure 30 and Table 68. At this point, the adjustments for roadway conditions like grade, horizontal curvature, lane width, access density, number of lanes and posted speed can be determined. The adjustments are applied to the number of expected encroachments. The result of this first step, then, is a tabulation of the expected number of encroachments over the 30 year life assuming a 2- percent growth that also accounts for the roadway characteristics at the site.

145 Figure 27. Cooper Encroachment Frequency Data [after Ray12] Figure 28. Cooper Encroachment Rate by Lane Volume. [after Ray12]

146 Estimate the Probability of a Severe or Fatal Crash in Any Encroachment The probability of a severe or fatal crash occurring in each particular collision is the next parameter that needs to be calculated for applying the binomial distribution. RSAPv3 uses the conventional police-reported crash severity (i.e., the KABCO scale) which can be converted into dollar values based on the FHWA crash cost recommendations for the particular year of interest. While the fatal crash cost (i.e., the VSL) changes from year to year as discussed earlier, the distribution of crash costs by severity does not; the proportion of each injury severity with respect to a fatal crash remains constant based on the work of Miller. [Miller88] Table 65 shows a table of crash costs by police-reported crash severity and the EFCCR represented by each severity. As shown in Table 65, an encroachment whose sum of event EFCCRs is equal to 0.0692 corresponds to a severe (i.e., A-level in the police-reported scale) injury so any encroachment resulting in an EFCCR (i.e., dimensionless crash cost) of 0.0692 or greater is a severe or fatal crash (i.e., A+K). Table 65. Police-Reported Crash Severity by Crash Cost and EFCCR. 2006 Crash Cost EFCCR K $ 2,600,000 1.0000 A $ 180,000 0.0692 B $ 36,000 0.0139 C $ 19,000 0.0073 PDO $ 2,000 0.0008 For each encroachment, RSAPv3 calculates the cumulative EFCCR for each event that occurs during the encroachment. The EFCCR is generally converted to a crash cost considering the VSL and vehicle type for each predicted crash. When only considering the risk of a crash, the EFCCR of each encroachment is not converted to crash costs, but is carried forward as an EFCCR and the cumulative probability distribution of EFCCRs is obtained similar to the example shown in Figure 29. Using the cumulative crash severity distribution in Figure 29 for a TL2 bridge railing in a low-hazard environment, an EFCCR of 0.0692 is associated with a probability of 0.9560. The probability of an EFCCR of 0.0692 or greater (i.e., the probability of a severe or fatal injury) is 1 – 0.9560 = 0.0440. If the hazard environment is high, an EFCCR of 0.0692 corresponds to a cumulative probability of 0.8552 for a TL2 bridge railing and 40 percent trucks. For a high- hazard environment, therefore, the probability of a severe or fatal injury crash on a TL2 bridge railing is 1 – 0.8552 = 0.1448. A cumulative probability distribution of the encroachment EFCCRs was developed for each test level bridge railing, each hazard environment and each percent of trucks and the cumulative probability of a severe or fatal crash was assembled in Table 65. Each entry in Table 66 is associated with a cumulative probability distribution like the example shown in Figure 29. The cumulative probability distributions, like the examples shown in Figure 29, are the conditional probability of a severe of fatal injury involving a particular test

147 level bridge railing given that a crash occurs so they are only dependent on the vehicle mix (i.e., percent of trucks) and the hazard environment. This fact is what allows the guidelines to be split into an encroachment prediction part and a crash severity prediction part. Figure 29. Cumulative EFCCR Distribution for TL2 and TL5 Bridge Railings with 40% Trucks. Unlike the fair coin toss discussed above where the probability of “failure” is 0.5, the probability of observing a failure in a bridge rail crash is very small and depends on (1) the number of trucks in the traffic mix and (2) the test level of the bridge railing. As shown in Table 66, for a given hazard environment, the probability of single encroachment resulting in a severe or fatal crash decreases as the test level increases. This would be expected with an increase in the performance level of the bridge railing. Also, for a particular test level bridge railing, the probability of a severe or fatal crash increases as the percent of trucks increases which is also as expected since more heavy vehicles in the mix make a collision with a very heavy high-energy vehicle more likely. The values in Table 66 are invariants of AADT and encroachment meaning that the probability of a single encroachment resulting in a severe or fatal crash does not depend on the traffic volume or the roadway characteristics. The AADT, traffic mix and roadway

148 characteristics influence the number of encroachments to be expected but not the severity of any particular crash. The severity is influenced by the area under the bridge (i.e., High, Medium, Low). Thus, Table 66 only needs to be generated once. The probabilities shown in Table 66 where obtained by performing RSAPv3 analyses at varying percentages of truck traffic for the following conditions: • Primary right encroachments only, • Shoulder width of 8 ft (i.e., face of bridge railing is 8 ft from the edge of the lane), • 65 mi/hr divided highways and • An encroachment increment of 500 ft on a 1,000-ft segment of roadway (Station 5+00 to 15+00) in the middle of a 2,000 section of bridge railing (Station 0+00 to 20+00). Calculating the Risk Now that the number of encroachments and the probability of a single crash resulting in a severe or fatal crash have been estimated, the risk of observing a severe or fatal crash over the 30-year life of a 1,000-ft section of bridge railing can be calculated. The risk can now be calculated using the cumulative density function of the binomial distribution with: k = The number of failures =1, n = The expected number of encroachments from the first step and; p = The probability of any particular encroachment resulting in severe or fatal crash from Table 66 calculated in the second step. A worksheet was developed listing the range of percent trucks and encroachments for each hazard environment and the risk associated with each cell was calculated. This was done for each of three risk levels (i.e., 0.005, 0.01 and 0.02). The resulting data was then formulated into tables such as those shown later in Figure 32, Figure 36 and Figure 37. These tables form the basis of the selection process as will be outlined in the next section.

149 Table 66. Probability a Collision Will Result in a Severe or Fatal Injury by Hazard Environment and MASH Test Level. Low Hazard Medium Hazard High Hazard Percent Trucks TL2 TL3 TL4 TL5 TL2 TL3 TL4 TL5 TL2 TL3 TL4 TL5 0.00 0.0203 0.0220 0.0016 0.0016 0.0560 0.0571 0.0084 0.0016 0.1251 0.1268 0.0101 0.0016 1.00 0.0207 0.0223 0.0018 0.0017 0.0563 0.0574 0.0086 0.0017 0.1254 0.1270 0.0103 0.0017 2.00 0.0210 0.0226 0.0020 0.0018 0.0567 0.0577 0.0088 0.0018 0.1257 0.1272 0.0105 0.0018 3.00 0.0214 0.0229 0.0022 0.0019 0.0571 0.0580 0.0090 0.0020 0.1260 0.1275 0.0108 0.0020 4.00 0.0218 0.0232 0.0024 0.0020 0.0575 0.0583 0.0093 0.0021 0.1264 0.1278 0.0110 0.0021 5.00 0.0223 0.0236 0.0026 0.0022 0.0579 0.0586 0.0095 0.0022 0.1267 0.1280 0.0112 0.0023 10.00 0.0244 0.0253 0.0038 0.0028 0.0599 0.0602 0.0107 0.0029 0.1285 0.1294 0.0125 0.0030 15.00 0.0268 0.0272 0.0051 0.0035 0.0623 0.0621 0.0121 0.0037 0.1305 0.1309 0.0139 0.0038 20.00 0.0295 0.0293 0.0065 0.0043 0.0648 0.0641 0.0136 0.0045 0.1327 0.1327 0.0154 0.0047 25.00 0.0325 0.0317 0.0081 0.0052 0.0677 0.0664 0.0153 0.0055 0.1352 0.1346 0.0172 0.0057 30.00 0.0358 0.0344 0.0099 0.0062 0.0709 0.0689 0.0172 0.0066 0.1380 0.1367 0.0191 0.0069 35.00 0.0397 0.0374 0.0120 0.0073 0.0746 0.0718 0.0193 0.0078 0.1412 0.1392 0.0214 0.0082 40.00 0.0440 0.0409 0.0143 0.0086 0.0788 0.0751 0.0218 0.0092 0.1448 0.1419 0.0239 0.0097