Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers (2022)

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.

C-1   A P P E N D I X C Probability of Crash Severity (PSEVj) CONTENTS Chapter 1 Introduction Chapter 2 Method Chapter 3 Available Data Data from the Literature New Data Gathered Chapter 4 Estimate Unreported Crashes Chapter 5 Determine P(KA|C) Concrete Barrier Family Cable Barrier Family Metal Beam Barrier Family Other Features Chapter 6 Results and Discussion References

C-2 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers CHAPTER 1 INTRODUCTION P(KA|C), the conditional probability of a severe or fatal (i.e., KA) crash, given a collision has occurred, is needed for guideline development. NCHRP Project 15-65 has dubbed this conditional probability PSEV j and allowed the conditional probability to be for KA crashes or any other severity of interest. The development of the methodology to calculate PSEV j is documented below using KA crashes to simplify the example and improve clarity; however, the methodology could be applied to other severity levels of interest. Also documented are the data used when developing the appropriate PSEV j for these guidelines.

Probability of Crash Severity (PSEVj) C-3   CHAPTER 2 METHOD The process used with RSAPv3 to model crash severity includes the ability to account for unreported crashes and to scale crash severity by the posted speed limit (PSL). These features are desirable in roadside crash severity estimation for many reasons. Unreported crashes represent roadside safety “successes.” Capturing these unreported crashes ensures that higher severity crashes are not over-predicted. Scaling the crash severity model by PSL allows for lower speed crashes, which can be less severe, to be addressed by the model, as well as higher speed crashes where the severity could be higher. The use of PSL rather than impact speed, for example, allows for broader availability of data. Using impact speed, when it becomes widely available, to scale crash severity could be the subject of future research. The RSAPv3 crash severity model, which captures both reported and unreported crashes, is scalable by PSL and is called the equivalent fatal crash cost ratio (EFCCR). The process for developing an EFCCR for any roadside feature was documented by Ray et al. in “Method for Modeling Crash Severity with Observable Crash Data.” (Ray 2014b) Ray et al. explain that the process includes the following steps: 1. “Isolate a census of police-reported crashes with a particular type of roadside feature, ideally over a range of posted speed limits. 2. Determine the crash severity distribution for crashes that do not have events preceding the crashes with the hazard under evaluation and do not result in penetration or rollover. 3. Determine or estimate the percentage of unreported crashes and add these crashes to the reported crash severity distribution. 4. Calculate the average crash cost of the severity distribution for each posted speed limit and determine the equivalent fatal crash cost ratio (EFCCR), and 5. Adjust for speed effects by determining the equivalent fatal crash cost ratio for a baseline impact speed of 65 mi/hr (i.e., EFCCR65) for a particular hazard.” (Ray 2014b) It is desirable to maintain the ability to include unreported crashes and to scale severity by PSL much the same way they are accomplished within RSAPv3. The EFCCR procedure was extended to maintain the calculation of unreported crashes (i.e., Step 3, above) and scaling for speed (i.e., Step 5, above) while allowing for the discrete values for each level of severity to also be maintained and then utilized for a probability calculation. After completing steps one through three above, the P(KA|C) can be found by summing the total number of KA crashes in the data at each PSL level and dividing by the total number of all crashes of all severities plus the estimated unreported crashes from Step 3, as shown here: where K+A Fatal and serious injury crashes across the posted speed limits available

C-4 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers KABCO + Unreported Crashes of all reported severities plus the estimated unreported crashes from EFCCR step 3 for all posted speed limits. In this extension of the procedure, rather than calculating an EFCCR (i.e., Step 4, above), the conditional probability P(KA|C) is determined for the entire sample. Step 5 is carried forward but modified to determine the speed-weighted probability of a KA crash given a collision at a base PSL of 65 miles per hour (mph): P(KA|C)65. This is calculated using the case-weighted PSL as follows: The resulting value can be used in the guidelines and adjusted using the site-specific PSL, as follows: These same calculations can be done for each severity level (e.g., KAB, F+I). The available crash data gathered from the literature and state crash databases for the determination of these severity measures are discussed below. The data used will be first and only harmful event crashes (FOHE), where harm-inducing crash events do not precede or follow the event of interest (e.g., barrier, roll over, or cross-median). First, the available crash data are discussed, below, and then the unreported crashes are estimated. The analysis to find P(KA|C)65 is documented for use in the guidelines.

Probability of Crash Severity (PSEVj) C-5   CHAPTER 3 AVAILABLE DATA Data from the Literature NCHRP Project 22-12(03), “Recommended Guidelines for the Selection of Test Level 2 through 5 Bridge Railings” included the gathering of crash data for the concrete median barrier family. (Ray 2021) One research objective for NCHRP Project 22-12(03) included the field evaluation of the hardware as a project objective. Therefore, under NCHRP Project 22-12(03), the hardware involved in the crashes was extensively verified. The crash severity distribution by PSL with a variety of concrete barriers were gathered under that effort. The data gathered are shown in Table C-1 and Table C-2. NCHRP Project 22-27, “Update of the Roadside Safety Analysis Program” also gathered crash severity data for a variety of barriers across the full severity distribution and by PSL. (Ray 2012) These data are shown in Table C-3. The severity of a cross-median crash must also be represented. The crash data assembled under this effort and documented in Table C-3 were used to find the severity of cross-median crashes. The severity distribution is shown in Table C-4. In addition to the crash severity distribution of various barriers and cross-median crashes, the crash severity distribution of rollover crashes is necessary for those encroachment events that result in a rollover before impacting the barrier or fully crossing the median. The severity distribution for fixed objects within the median or on the roadside was also captured. The severity distribution of previously documented data collected under NCHRP Project 22-27 (Ray 2012) by PSL is shown for rollover crashes (see Table C-5), tree crashes (see Table C-6), and waterbody crashes (see Table C-7). Table C-1 New Jersey, Massachusetts, Washington, and Pennsylvania Concrete Safety Shape Barrier Crash Severity Distribution (after Ray 2021) State Barrier PSL K A B C PDO/Unk NJ TL5 Concrete 55 0 1 12 35 193 NJ TL5 Concrete 65 0 11 103 307 1395 MA 32” F-Shape 55 0 0 4 4 14 MA 32” F-Shape 65 3 4 36 17 72 MA 42” F-Shape 55 0 0 6 4 24 WA 32” Safety Shape 60 2 4 62 112 369 WA 34” Single Slope 60 0 3 20 28 127 PA 32” F-Shape 55 3 1 6 14 33 PA 32” F-Shape 65 1 0 7 28 71 PA 42” F-Shape 55 1 0 1 3 5 PA 42” F-Shape 65 0 0 4 9 33

C-6 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers Table C-2 Nebraska Crash Severity Distribution (after Ray 2021) PSL K A B C O 29" Vertical Wall 50 mph or less 0 0 0 0 0 55 0 0 0 0 0 60 0 2 2 3 9 65 0 0 1 1 2 75 0 0 0 0 0 34" Vertical Wall 50 mph or less 0 1 3 4 33 55 0 4 3 0 20 60 3 6 14 28 131 65 1 5 9 11 61 75 3 4 14 11 102 42" Vertical Wall 50 mph or less 0 0 0 0 1 55 0 3 3 2 11 60 0 0 0 1 1 65 0 0 2 2 7 75 0 1 0 0 6 32" NJ Shape 50 mph or less 0 2 4 2 46 55 0 2 1 4 19 60 0 2 4 6 36 65 0 0 4 2 19 75 1 0 1 0 14 42" NJ Shape 50 mph or less 0 0 1 3 4 55 0 1 2 6 20 60 2 0 3 9 14 65 0 0 1 0 1 75 0 0 0 0 0

Probability of Crash Severity (PSEVj) C-7   Table C-3 Barrier Crash Data Assembled Under NCHRP Project 22-27 (Ray 2012) State Barrier Type PSL K A B C O Unk WA TL3 LT Cable MB 70 0 0 6 13 220 7 WA TL3 LT Cable MB 60 0 1 9 16 312 10 WA TL3 HT Cable MB 55 0 0 1 2 11 WA TL3 HT Cable MB 60 1 0 12 16 238 2 WA TL3 HT Cable MB 70 1 1 12 16 225 3 AZ TL3 LT Cable MB 65 0 0 1 4 11 4 IA TL3 HT Cable MB 65 0 1 0 2 17 NC TL3 LT Cable MB 65 0 2 9 28 88 OR TL3 LT Cable MB 65 0 0 0 5 15 6 TX TL4 32" NJ MB 65 8 115 456 209 890 Table C-4 Washington Crash Data for Cross-Median Crashes Feature PSL K A B C O Unk CMC 70 0 1 3 5 38 0 CMC 65 7 5 7 1 15 0 CMC 60 39 27 77 87 398 3 CMC 55 385 1061 2049 913 3520 0 CMC 50 44 134 256 205 629 3 CMC 45 37 241 615 435 1569 4 CMC 40 22 114 327 315 1119 1 CMC 35 43 438 1317 1417 5656 12 CMC 30 1 14 41 41 415 3 CMC 25 4 74 311 446 1974 1 CMC 20 0 0 0 1 8 0 Table C-5 Washington Crash Data for Rollover Crashes (after Ray 2012) Feature PSL K A B C O Unk Rollover 70 36 96 670 352 731 33 Rollover 65 5 5 110 38 120 15 Rollover 60 27 125 791 542 1088 99 Rollover 55 14 75 389 257 628 62 Rollover 50 9 27 173 131 288 42 Rollover 45 1 5 38 34 58 9 Rollover 40 0 5 31 20 37 5 Rollover 35 1 11 37 35 73 19 Rollover 30 0 2 6 4 9 1 Rollover 25 0 0 8 3 8 1

C-8 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers Table C-6 Washington Crash Data for Tree Crashes (after Ray 2012) Feature PSL K A B C O Unk Tree 70 5 7 30 20 103 8 Tree 65 1 1 2 0 7 0 Tree 60 10 26 86 93 200 34 Tree 55 21 32 112 87 168 35 Tree 50 12 23 75 52 151 29 Tree 45 1 3 19 20 44 10 Tree 40 3 11 21 18 56 9 Tree 35 1 4 26 32 89 26 Tree 30 0 2 4 9 39 9 Tree 25 0 1 4 6 20 5 Table C-7 Washington Crash Data for Waterbody Crashes (after Ray 2012) Feature PSL K A B C O Unk Waterbody 70 0 0 0 1 3 0 Waterbody 65 0 0 1 0 0 0 Waterbody 60 0 0 9 5 15 1 Waterbody 55 1 2 6 4 34 3 Waterbody 50 0 3 2 6 27 1 Waterbody 45 0 0 0 0 5 0 Waterbody 40 0 1 1 0 2 0 Waterbody 35 0 0 1 1 1 1 Waterbody 30 0 0 2 0 5 0 Waterbody 25 New Data Gathered New crash data, in addition to data available in the literature, were also collected and evaluated in this study. Barrier inventories and accompanying crash data were made available by the States of Maine, Ohio, Pennsylvania, and Tennessee. The State of Washington also provided their inventory; however, much of the data available in the literature was assembled using the Washington database. A new analysis was not conducted with the Washington database to avoid double-counting of data. A much wider range of crash data was made available by Ohio with their inventory; therefore, a new analysis of the Ohio data was conducted here and the data available in the literature were not included above to avoid double-counting. Crashes that penetrate, roll over, or vault the feature (THR), or crashes that roll over after redirection on the same side of the barrier (RSS) are excluded from this severity measure such that the result will represent the severity of a single event in the overall crash sequence. This ensures that the severity measure can be confidently associated with the collision with the feature under evaluation. The crashes coded as single vehicle (SV), FOHE, and longitudinal barrier (LB) crashes were isolated from each data set. While each of these new data sets included barriers located within the median, some did not differentiate between median (i.e., double-faced) barriers and roadside (i.e., single-faced) barriers. The crash severity outcome of median and roadside barriers is assumed to be equal

Probability of Crash Severity (PSEVj) C-9   when the barrier is struck on the design-impact side. If a roadside barrier were struck from behind, however, the crash severity outcome could not be assumed to be equal to that of a median barrier because median barriers are designed to be impacted from both sides while roadside barriers are not. When impacts occurred within the median, but it could not be determined whether the vehicle impacted a barrier face, then these cases were eliminated from the data set. SV FOHE LB crashes occurring in the median or on the roadside when the vehicle impacted the barrier face were used because these crashes are the best representations of the crash outcome when events do not precede nor follow the impact with the barrier. The method used to isolate these crashes for each data set is explained here. The Ohio Highway Safety Information System data for 2003 through 2013 were used in conjunction with state-collected roadside hardware inventory provided by the Ohio Department of Transportation (ODOT). SV FOHE LB crashes were identified using EVENT1, EVENT2, EVENT3, EVENT4, F_HARM, and NUMVEH fields in the crash data. Crashes were considered to be SV FOHE LB if the NUMVEHS field was equal to 1, the F_HARM field was identified as either code ‘30’ (guardrail face) or ‘32’ (median barrier), and the EVENT1-4 fields were either blank or contained codes ‘08’ (ran off road-right), ‘09’ (ran off road-left), or ‘10’ (cross median/centerline). The ODOT hardware inventory was linked to the SV FOHE LB crashes using RTE-NBR and MILEPOST fields along with the roadside location of the hardware in the inventory. The EVENT1-4 codes ‘08’ (ran off road-right), ‘09’ (ran off road-left), and ‘10’ (cross median/centerline) were used in conjunction with the VEH_N_FROM (direction vehicle was traveling from) and VEH_N_TO (direction vehicle was traveling to) fields to determine the location and type of hardware involved in each SV FOHE LB crash. The Pennsylvania Department of Transportation (PennDOT) crash database for 2010 through 2015 was used in conjunction with the state-collected roadside guide rail inventory. SV FOHE LB crashes were identified using harmful events 1-4, First Harmful Event, and the “TOTAL_UNITS” fields. Crashes were considered SV FOHE LB crashes if “TOTAL_UNITS” = 1, First Harmful Event = 25 (hit guard or guide rail), 28 (hit concrete or longitudinal barrier), and no harmful event (blank entry) in the harmful event immediately succeeding the first harmful event. The County Number, State Route Number, Segment Number, Beginning Offset, Ending Offset, and Guide Rail Side fields were used to attach the barrier database to the SV FOHE LB crashes. The Travel Direction field was used in conjunction with the Harmful Event Side 1-4 fields to determine the location and type of hardware involved in each SV FOHE LB crash. Pennsylvania differentiates between single-sided and double-sided guardrail systems, allowing median barrier crashes on divided highways to be used. The Tennessee Department of Transportation (TDOT) crash database for 2012 through 2016 was used in conjunction with the state-collected roadside barrier inventory. SV FOHE LB crashes were identified using EVENT SEQ1-6, FIRSTHARMFULEVENTCDE, and TOTAL VEH fields. Crashes were considered SV FOHE LB crashes if: TOTAL VEH was equal to 1, FIRSTHARMFULEVENTCDE equal to "Concrete Traffic Barrier", "Guardrail Face", or "Cable Barrier", and the EVENT SEQ immediately succeeding the first harmful event is either blank or "Cross Center Line", "Cross Median", "Ran Off Road-Left", "Ran Off Road-Right", and immediately followed by a blank entry in the next EVENT SEQ field. Unlike some states, Tennessee has a First Harmful Event field separate from the event sequences. The police report has an entry for this and is listed as “First Harmful Event for the Crash”. (ACTAR 2017) It is implied that this field represents the first harmful event in the crash, not simply the first event. Zero crashes had "Cross Center Line," "Cross Median," "Ran Off Road-Left," or "Ran Off Road-

C-10 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers Right" listed as the First Harmful Event, and some instances had one of these entries in the EVENT SEQ1, followed by "Concrete Traffic Barrier," "Guardrail Face," or "Cable Barrier" in EVENT SEQ2, with one of the barrier types listed as the First Harmful Event. The COUNTY, ROUTE_NAME, BEG_LOG, and END_LOG fields were used to attach the barrier inventory database to the SV FOHE LB crashes. The TRAVELDIRCDE field along with "Cross Center Line," "Cross Median," "Ran Off Road-Left," and "Ran Off Road-Right" in the EVENT SEQ1-6 field were used to determine which barrier was involved in each SV FOHE LB crash. Tennessee also differentiates between “Median Right,” “Median Left,” and “Centerline” in the LOC_DESC field of the inventory, thus allowing median barrier crashes to be identified. The Maine Department of Transportation (MaineDOT) maintains a roadside hardware condition assessment inventory for maintenance purposes and made the inventory available for this research. The inventory can be linked with crash records which MaineDOT also made available for this research. The inventory includes the location and condition of the longitudinal barrier and includes the side of the road and the direction the barrier faces. The inventory includes the specific end treatment for the start and end of each section of longitudinal barrier as well. While the inventory includes the condition of the longitudinal barrier, it unfortunately does not provide information sufficient for this research regarding the type of barrier at each location. The objective of MaineDOT in collecting this inventory is for asset management while the objectives of this research are to positively identify each asset involved in a crash. Regrettably, there is not sufficient information about each longitudinal barrier type within the inventory to support the objectives of this research. The MaineDOT database, therefore, was not used in this analysis. This inventory, however, could prove valuable to those studying features better defined within the database (e.g., condition of the hardware and end treatment type). After isolating a census of crash data from each state database for particular hazards, the crash severity distribution can be determined for each hardware in the inventory. Ohio A total of 31,540 SV FOHE LB crashes were found during the 11-year study period in Ohio. These crashes were then linked to the roadside hardware inventory for each identified longitudinal barrier. Ohio identifies the following longitudinal barriers within the inventory: • Guardrail • 32” Jersey Barrier • 42”+ Jersey Barrier • Single Slope Barrier (i.e., either 42” or 57”) • Propriety Cable Barriers (i.e., Brifen’s wire rope safety fence (WRSF), Trinity’s Cable Safety System (CASS), Gibraltar’s Cable Barrier System, and Nucor Steel Marion’s Nu-Cable Barrier) • Other A review of the ODOT guardrail standard drawings shows that guardrail is a generic reference for w-beam barriers and that w-beam is the standard guardrail used in Ohio. (ODOT 2013) When guardrail is referenced within the ODOT inventory, W-beam is assumed to be in that location. Single slope barriers may be 42” or 57”. (ODOT 2017) There were zero reported SV FOHE LB crashes with the inventoried propriety barriers. There were two property-damage-only (“O”) severity SV FOHE crashes with the barrier

Probability of Crash Severity (PSEVj) C-11   inventoried as “other.” There were 28,141 SV FOHE LB crashes where the type of barrier involved could not be positively identified using the Ohio inventory. The full severity distribution for SV FOHE LB crashes where the barrier type could be identified using the Ohio inventory is summarized in Table C-8 through Table C-11. Those crashes that could not be associated with a particular barrier type are summarized in Table C-12. Table C-8 Ohio SV FOHE LB Crash Counts by Posted Speed Limit: W-Beam PSL K A B C O 65 2 13 122 104 1081 60 1 5 38 32 299 55 4 25 117 85 817 50 1 0 4 4 64 45 1 1 14 8 88 40 0 0 1 4 34 35 0 2 8 4 60 30 0 0 0 0 0 25 0 0 2 1 6 Table C-9 Ohio SV FOHE LB Crash Counts by Posted Speed Limit: 32” Jersey Barrier PSL K A B C O 65 0 5 15 12 60 60 0 0 5 6 34 55 0 1 5 3 41 50 0 0 1 1 3 45 0 0 1 0 2 40 0 0 0 0 0 35 0 0 0 0 0 30 0 0 0 0 0 25 0 0 0 0 0 Table C-10. Ohio SV FOHE LB Crash Counts by Posted Speed Limit: 42”+ Jersey Barrier PSL K A B C O 65 0 0 3 1 12 60 0 1 4 2 19 55 0 0 0 1 8 50 0 0 0 0 1 45 0 0 0 0 1 40 0 0 0 0 1 35 0 0 0 0 0 30 0 0 0 0 0 25 0 0 0 0 0

C-12 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers Table C-11 Ohio SV FOHE LB Crash Counts by Posted Speed Limit: Single Slope Barriers PSL K A B C O 65 0 1 3 5 54 60 0 0 2 2 5 55 0 1 2 3 8 50 0 0 0 0 3 45 0 0 0 0 0 40 0 0 0 0 0 35 0 0 0 0 0 30 0 0 0 0 0 25 0 0 0 0 0 Table C-12 Ohio SV FOHE LB Crash Counts by Posted Speed Limit: Unable to Associate with a Barrier Type PSL K A B C O 65 9 201 1513 1307 10270 60 11 120 634 813 4119 55 9 144 728 650 4544 50 5 35 173 205 933 45 0 14 49 36 427 40 0 4 26 20 142 35 0 17 94 82 652 30 0 0 3 0 2 25 1 6 14 10 119 Pennsylvania A total of 5,903 SV FOHE LB crashes were found during the 6-year study period in Pennsylvania. These LB crashes were then linked to the roadside hardware inventory for each identified longitudinal barrier. Pennsylvania identifies the following longitudinal barriers within the inventory: • Strong Post Cable • Weak Post Cable • Strong Post W-Beam with Rub Rail and Offset Bracket • Strong Post W-Beam with Offset Bracket • Strong Post W-Beam • Weak Post W-Beam • Strong Post W-Beam, Double-Faced • Weak Post W-Beam, Double-Faced • Weak Post Box Beam • Concrete Safety Shape

Probability of Crash Severity (PSEVj) C-13   • Intermediate Bulk Container (IBC) Barrier • Propriety Cable Barriers (i.e., WRSF, CASS, Blue Systems AB’s Safence, Gibraltar’s Cable Barrier System, and Nucor Steel Marion’s Nu-Cable Barrier) • Other Each of these systems is further explained in the PennDOT Shoulder and Guide Rail Condition Survey Field Manual, Publication 33. (PennDOT 2017) The results are shown in Tables C-13 through C-23. The proprietary cable systems have been summarized as one in Table C-24, an individual breakdown by system and crash severity is provided here: • There were two reported SV FOHE LB crashes with the CASS one property-damage-only (PDO) one Unk • There were three reported SV FOHE LB crashes with the WRSF three PDO • There were 13 reported crashes with the Safence one “A” one “B” three “C” six PDO one Unk • There were 62 reported crashes with the Gibraltar cable barrier 1 “B” 10 “C” 48 PDO 3 Unk • There were two reported crashes with the Nu-Cable one “C” one PDO SV FOHE LB crashes that were associated with the inventory code “Other” barrier are shown in Table C-25. All identified crashes were associated with the roadside inventory. Table C-13 Pennsylvania SV FOHE LB Crash Counts: Strong Post Cable PSL K A B C O Unk 70 0 0 0 0 0 0 65 0 0 0 0 0 0 60 0 0 0 0 0 0 55 0 0 1 0 9 0 50 0 0 0 0 0 0 45 0 0 0 2 12 0 40 0 0 0 3 11 0 35 0 0 0 1 2 3 30 0 0 0 0 3 1

C-14 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers 25 0 0 0 0 2 1 20 0 0 0 0 0 0 15 0 0 0 0 0 0 Table C-14 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: Weak Post Cable PSL K A B C O Unk 70 0 0 0 0 0 0 65 0 0 0 0 0 0 60 0 0 0 0 0 0 55 0 0 1 1 27 0 50 0 0 0 0 0 0 45 0 0 0 0 0 0 40 0 0 0 0 0 0 35 0 0 0 0 0 0 30 0 0 0 0 0 0 25 0 0 0 0 0 0 20 0 0 0 0 0 0 15 0 0 0 0 0 0 Table C-15 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: Strong Post W-Beam with Rub Rail and Offset Bracket PSL K A B C O Unk 70 0 0 0 0 0 0 65 0 0 1 4 12 0 60 0 0 0 0 0 0 55 0 1 2 16 69 14 50 0 1 1 2 8 1 45 0 0 1 9 33 10 40 0 1 2 8 27 0 35 0 2 4 7 22 9 30 0 0 0 0 6 0 25 0 0 1 2 5 5 20 0 0 0 1 0 0 15 0 0 0 0 0 0

Probability of Crash Severity (PSEVj) C-15   Table C-16 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: Strong Post W-Beam with Offset Bracket PSL K A B C O Unk 70 0 0 0 2 5 0 65 1 3 11 79 342 19 60 0 0 0 0 0 0 55 4 25 78 210 753 87 50 0 3 0 11 42 5 45 4 23 34 98 283 42 40 0 3 19 53 168 20 35 3 5 13 38 148 25 30 0 1 3 6 24 2 25 0 0 5 8 23 4 20 0 0 0 1 0 1 15 0 0 0 0 0 0 Table C-17 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: Strong Post W-Beam PSL K A B C O Unk 70 0 0 0 0 0 0 65 0 0 0 0 0 0 60 0 0 0 0 0 0 55 0 0 0 3 18 0 50 0 0 0 0 0 0 45 0 0 0 0 3 0 40 0 0 0 0 5 0 35 0 0 0 3 6 2 30 0 0 0 0 2 1 25 0 0 1 0 2 1 20 0 0 0 0 0 0 15 0 0 0 0 0 0

C-16 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers Table C-18 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: Weak Post W-Beam PSL K A B C O Unk 70 0 0 0 0 3 0 65 0 2 9 42 357 8 60 0 0 0 0 0 0 55 0 7 15 65 330 27 50 0 0 1 6 25 5 45 0 2 8 16 91 6 40 0 3 2 4 27 4 35 0 0 3 4 19 4 30 0 0 0 0 2 0 25 1 0 0 0 3 0 20 0 0 0 0 0 0 15 0 0 0 0 0 0 Table C-19 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: Strong Post W-Beam, Double-Faced PSL K A B C O Unk 70 0 0 0 0 0 0 65 0 0 0 0 0 0 60 0 0 0 0 0 0 55 0 2 2 11 32 7 50 0 0 0 1 1 0 45 0 0 1 1 2 1 40 0 0 0 1 5 0 35 0 0 2 2 2 1 30 0 0 0 0 0 0 25 0 0 0 0 0 0 20 0 0 0 0 0 0 15 0 0 0 0 0 0

Probability of Crash Severity (PSEVj) C-17   Table C-20 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: Weak Post W-Beam, Double-Faced PSL K A B C O Unk 70 0 0 0 0 0 0 65 0 0 0 0 0 0 60 0 0 0 0 0 0 55 0 0 0 6 52 8 50 0 0 0 0 0 0 45 0 0 0 0 0 0 40 0 0 0 0 0 0 35 0 0 0 0 0 0 30 0 0 0 0 1 0 25 0 0 0 0 0 0 20 0 0 0 0 0 0 15 0 0 0 0 0 0 Table C-21 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: Weak Post Box Beam PSL K A B C O Unk 70 0 0 0 0 0 0 65 0 0 0 0 0 0 60 0 0 0 0 0 0 55 0 0 0 1 15 5 50 1 0 0 0 0 0 45 0 0 0 1 5 0 40 0 0 0 0 0 0 35 0 0 0 0 5 2 30 0 0 0 0 0 0 25 0 0 0 0 0 0 20 0 0 0 0 0 0 15 0 0 0 0 0 0

C-18 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers Table C-22 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: Concrete Safety Shape PSL K A B C O Unk 70 0 0 0 0 0 0 65 0 2 9 28 79 6 60 0 0 0 0 0 0 55 1 7 44 158 523 79 50 0 0 12 45 89 22 45 0 1 4 18 66 14 40 0 0 1 20 59 6 35 0 0 3 17 22 7 30 0 0 0 0 2 0 25 0 0 1 2 2 1 20 0 0 0 0 0 0 15 0 0 0 0 0 0 Table C-23 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: IBC Barrier PSL K A B C O Unk 70 0 0 0 0 0 0 65 0 0 0 4 14 1 60 0 0 0 0 0 0 55 0 0 0 1 4 1 50 0 0 0 0 0 0 45 0 0 0 0 0 0 40 0 0 0 0 0 0 35 0 0 0 0 0 0 30 0 0 0 0 0 0 25 0 0 0 0 0 0 20 0 0 0 0 0 0 15 0 0 0 0 0 0

Probability of Crash Severity (PSEVj) C-19   Table C-24 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: Propriety Cable Systems (All Combined) PSL K A B C O Unk 70 0 0 0 0 0 0 65 0 1 1 4 9 1 60 0 0 0 0 0 0 55 0 0 1 10 51 4 50 0 0 0 0 0 0 45 0 0 0 0 0 0 40 0 0 0 0 0 0 35 0 0 0 0 0 0 30 0 0 0 0 0 0 25 0 0 0 0 0 0 20 0 0 0 0 0 0 15 0 0 0 0 0 0 Table C-25 Pennsylvania SV FOHE LB Crash Counts by Posted Speed Limit: Other PSL K A B C O Unk 70 0 0 0 0 0 0 65 0 0 0 0 1 0 60 0 0 0 0 0 0 55 0 1 0 0 5 1 50 0 0 1 0 0 0 45 0 0 1 1 2 1 40 0 0 0 1 1 0 35 0 0 0 1 1 0 30 0 0 0 0 0 0 25 0 0 0 0 0 0 20 0 0 0 0 0 0 15 0 0 0 0 0 0 Tennessee A total of 2,881 SV FOHE LB crashes were found during the 5-year study period (2012– 2016) in Tennessee. These crashes were then linked to the roadside hardware inventory for each identified longitudinal barrier. Tennessee identifies the following longitudinal barriers within the inventory: • Jersey Barrier • W-Beam • Cable Barrier—Gibraltar NCHRP350 • Cable Barrier—Nu-Cable

C-20 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers The severity distributions for each barrier type are shown in Tables C-26 through C-29. There were 124 reported SV FOHE LB crashes with the Jersey Barrier. There were 72 reported SV FOHE LB crashes with the W-Beam. There were 18 reported SV FOHE LB crashes with the Gibraltar cable barrier including zero fatal and one serious injury crash. There were 37 reported crashes with the Nu-Cable system including zero fatalities and two serious injuries. The combined severity distribution of these proprietary cable systems is shown in Table C-28. Of the reported crashes, 2,630 were not associated with the hardware inventory, the distribution of which is shown in Table C-29. This large number of unassociated crashes is due to the police crash reports often excluding the run- off-road direction, which prevented identification of the barrier involved. Table C-26 Tennessee SV FOHE LB Crash Counts by Posted Speed Limit: Jersey Barrier PSL K A B C O Unk 70 0 2 12 11 38 1 65 0 0 1 1 9 1 60 0 0 0 0 0 0 55 0 1 2 9 32 2 50 0 0 0 1 1 0 45 0 0 0 0 0 0 40 0 0 0 0 0 0 35 0 0 0 0 0 0 30 0 0 0 0 0 0 25 0 0 0 0 0 0 Table C-27 Tennessee SV FOHE LB Crash Counts by Posted Speed Limit: W-Beam PSL K A B C O Unk 70 0 0 1 1 23 0 65 0 0 0 1 9 0 60 0 0 0 0 1 0 55 1 1 2 6 15 0 50 0 0 0 0 0 0 45 1 0 1 3 3 0 40 0 0 0 0 1 1 35 0 0 0 0 0 0 30 0 0 0 0 0 0 25 0 0 0 1 0 0

Probability of Crash Severity (PSEVj) C-21   Table C-28 Tennessee SV FOHE LB Crash Counts by Posted Speed Limit: Proprietary Cable Barrier (All Combined) PSL K A B C O Unk 70 0 2 1 1 22 0 65 0 1 0 3 21 0 60 0 0 0 0 2 0 55 0 0 0 0 2 0 50 0 0 0 0 0 0 45 0 0 0 0 0 0 40 0 0 0 0 0 0 35 0 0 0 0 0 0 30 0 0 0 0 0 0 25 0 0 0 0 0 0 Table C-29 Tennessee SV FOHE LB Crash Counts by Posted Speed Limit: Unable to Associate with a Barrier Type PSL K A B C O Unk 70 0 6 29 59 367 9 65 1 9 27 53 370 6 60 0 1 1 4 27 0 55 4 29 97 188 998 29 50 0 2 6 8 76 0 45 0 1 13 12 71 1 40 0 1 4 8 26 1 35 0 0 4 5 39 2 30 0 2 1 2 21 1 25 0 0 0 0 9 0

C-22 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers CHAPTER 4 ESTIMATE UNREPORTED CRASHES Crash reporting thresholds vary by state, with some states only requiring reports when there is an injury, while others require a monetary threshold to be exceeded. It has long been recognized that police-reported crash data underreport lower severity crashes. “These low- severity crashes represent roadside design successes since the vehicle was able to encroach onto the roadside or median without causing an injury.” (Ray 2014b) When the EFCCR approach was developed, it included a step for estimating unreported crashes to account for this bias. Unreported crashes have been studied in several research studies, including the FHWA Pole Study, NCHRP Report 490, and NCHRP Report 638. (Mak 1980; Ray 2003; Sicking 2009) In his National Highway Traffic Safety Administration Technical Report on “The Economic Impact of Motor Vehicle Crashes, 2000,” Blincoe estimated that for all types of highway crashes, nearly half (i.e., 48%) of all PDO crashes and a little over 20% (i.e., 21.42%) of injury crashes are not reported. (Blincoe 2002) It has been found that the unreported crash rate is different for different types of roadside objects. For example, 77% of concrete barrier crashes were unreported while 34% of low-tension (LT) cable barrier crashes were unreported. (Fitzpatrick 1999; Hammond 2008) Building on a model developed by Nilsson (Nilsson 1982), Ray et al. estimate the percentage of non-injury crashes (PNIC) by comparing crashes at two speeds, as follows: This expression allows the unobserved percent of non-injury crashes to be estimated based on the number of observed injury crashes. (Ray 2014b) Next, the percentage of unreported and PDO crashes that is either known or assumed at the base speed of 65 mph is used to extrapolate to all other speeds. When the estimate produces no negative crash estimates, the estimate is balanced and has reached the maximum likelihood estimate of total crashes for the data set. The data found in the literature and assembled from the asset inventories of Ohio, Pennsylvania, and Tennessee represent validated data where the type of hardware or roadside feature involved in the crash can be confirmed. Summaries of the total number of reported SV FOHE LB crashes for the concrete barrier family, the cable barrier family, the beam barrier family, and other median features are shown in Tables C-30 through C-33. The maximum likelihood estimate (MLE) of unreported crashes for each data set resulted in an assumed ratio of injury crashes to total crashes at 65 mph, expressed as a percentage. This MLE of the percentage of injury crashes is also shown in Tables C-30 through C-33.

Probability of Crash Severity (PSEVj) C-23   Table C-30 Maximum Likelihood Estimate of Injuries for Concrete Barrier Family State and Barrier Observed Crashes MLE % INJ MA 32” 154 45.45 PA 32” F-Shape 164 33.64 All 32” F-Shape 318 40.16 MA 42” F-Shape 34 41.05 PA 42” F-Shape 56 28.26 All 42” F-Shape 90 28.26 All F-Shape 408 38.24 NE 29” Vertical Wall 20 43.11 NE 34” Vertical Wall 471 17.93 NE 42” Vertical Wall 40 10.73 All Vertical Wall 531 17.57 OH Single Slope 89 14.28 WA 34” Single Slope 178 0.00 All Single Slope 267 14.28 TN Jersey 124 16.66 OH 32” Jersey 195 25.14 NE 32” Jersey 169 9.37 TX 32” TL4 Jersey MB 1,678 0.00 OH 42” Jersey 54 15.51 NE 42” Jersey 67 43.34 All 32” Jersey 2,042 9.38 All 42” Jersey 121 27.77 All Jersey 2,163 9.38 All Concrete 7,325 16.74

C-24 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers Table C-31 Maximum Likelihood Estimate of Injuries for Cable Barrier Family State and Barrier Observed Crashes MLE % INJ WA TL3 HT Cable 541 10.02 IA TL3 HT Cable 20 15.00 PA HT Cable 82 23.27 TN HT Cable 55 13.26 All HT Cable 698 10.32 WA TL3 LT Cable 594 6.65 AZ TL3 LT Cable 20 25.00 NC TL3 LT Cable 127 30.70 OR TL3 LT Cable 26 19.23 PA LT Cable 29 9.63 All LT Cable 796 6.65 All HT and LT Cable 1,494 8.62 Table C-32 Maximum Likelihood Estimate of Injuries for Beam Barrier Family State and Barrier Observed Crashes MLE % INJ PA Strong Post W-Beam w/Rub Rail and Offset Bracket 287 26.00 PA Strong Post W-Beam with Offset Bracket 2,737 20.65 PA Strong Post W-Beam 47 19.95 PA MB Strong Post W-Beam 74 38.79 All PA Strong Post W-Beam 3,145 20.97 OH W-Beam 3,052 18.22 TN W-Beam 72 6.89 PA Weak Post W-Beam 1,101 12.67 PA MB Weak Post W-Beam 67 12.69 All PA Weak Post W-Beam 1,168 12.67 All Strong Post W-Beam 6,269 10.77 PA Box Beam 35 6.65 All Weak and Strong Post W-Beam 7,437 9.85

Probability of Crash Severity (PSEVj) C-25   Table C-33 Maximum Likelihood Estimate of Injuries for Non-Barrier Median Features State and Feature Observed Crashes MLE % INJ WA Cross-Median Crash 23,928 16.51 WA Rollover 7,439 51.87 WA Tree 1,922 30.90 WA Waterbody 144 21.55

C-26 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers CHAPTER 5 DETERMINE P(KA|C) Carrigan and Ray explained in “Practitioner’s Guide to the Analysis of In-Service Performance Evaluation Data” the relationship between absolute risk, portions, percentages, and probability of crashes with roadside hardware. The absolute risk of KA crashes can be found by summing the total number of KA crashes in the data for a hardware category and dividing by the total number of all crashes of all severities for that same category, as shown here (Carrigan 2016): where KA = Severe and fatal injury crashes KABCOU = Crashes of all reported severities including U for unknown severities. If unreported crashes have been studied, these unreported crashes should also be included in the denominator. The keen observer will note that absolute risk is simply a proportion. When multiplied by 100, it is also a percentage. When absolute risk is defined this way, it is also the probability of observing a KA crash given all crashes, P(KA|C). The absolute risk calculation is a point estimate calculated from a sample of the population of interest. Since the absolute risk of the entire population (p) is unknown, the estimated absolute risk ( ) from the sample is used and expressed with a confidence interval to allow inferences to be made on the larger population. (Kean 1999; Yale 2015) A confidence interval is much more useful than a p-value, as it provides a range of values that the entire population is likely between. For example, if the 95 percentile confidence bounds are provided for an absolute risk estimate, this can be interpreted as: “based on the sample data, we are 95% confident that the ‘true’ absolute risk of a KA crash with the hardware studied is between x and y.” The probability of observing a value outside of the area is less than 0.05 (i.e., 1-0.95=0.05). (Carrigan 2016) Common confidence levels and the corresponding z-values are shown in Table C-34.

Probability of Crash Severity (PSEVj) C-27   Table C-34 Published z-values for Normal Distribution (Carrigan 2016) Confidence Level z 0.70 1.04 0.75 1.15 0.80 1.28 0.85 1.44 0.90 1.645 0.92 1.75 0.95 1.96 0.96 2.05 0.98 2.33 0.99 2.58 The confidence interval is calculated for proportional data such as absolute risk using the following equation: Where: = Absolute risk calculated from the sample. z = Number of standard deviations away from the mean (see Table C-34). n = Sample size. The analysis and recommendations for P(KA|C) for use in the guidelines are discussed below. Concrete Barrier Family P(KA|C) as a proportion and the 95% confidence interval are shown for each barrier within the concrete barrier family in Figure C-1. P(KA|C) for each concrete barrier is shown on the y-axis. The diamond markers represent the point estimate of P(KA|C) with a concrete barrier. The bars extending above and below the diamond markers are the 95% confidence intervals. Based on the sampled data, we are 95% confident that the ‘true’ P(KA|C) for concrete barriers is within the range shown by the bars for each marker.

C-28 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers Figure C-1 P(KA|C) for concrete barrier family with 95% confidence intervals. Notice how the range represented by the bars within the F-Shape family overlap each other. This indicates that there is not a statically observed difference in these data between, for example, the 32” F-Shape barrier in Massachusetts and the 42” F-Shape barrier in Pennsylvania. The same is true for each concrete shape family shown in Figure C-1. A review of the combined estimates for each concrete shape grouping was performed to assess any observable difference and eliminate the individually insignificant findings, as shown in Figure C-2. Note that the y-axis of Figure C-2 has been extremely exaggerated to allow for interpretation of these small differences. Each concrete barrier shape overlaps with at least one other shape shown. There is, therefore, not an observable difference between all the barriers, but only between pairs of barriers. On the other hand, each barrier type is significant on its own. There are two options, one more restricting than the other: (1) the guidelines could adopt separate measures for each barrier, or (2) the guidelines could adopt a single crash severity measure for the entire concrete family. Adopting a single measure for concrete was preferred here due to the lack of significant difference between systems and because these data are based on NCHRP Report 350 barriers, while this severity measure will be applied to MASH barriers. As MASH barriers are implemented, this single measure will help to ensure that the crash severity of concrete barriers is estimated with the greatest confidence and smallest range. Of course, individual states can adopt the severity measure appropriate to their state in their state-specific guidance. The value P(KA|C)concrete = 0.0159, 95% CI [0.0139, 0.0178] has been adopted for the concrete family. 0.0000 0.0500 0.1000 0.1500 0.2000 0.2500 M A 3 2" F sh ap e PA 3 2" F sh ap e M A 4 2" F sh ap e PA 4 2" F sh ap e A ll 3 2" F sh ap e A ll 4 2" F sh ap e A ll F sh ap e N E 2 9" v er tic al w al l N E 3 4" v er tic al w al l N E 4 2" v er tic al w al l A ll v er tic al w al l O H S in gl e Sl op e W A 3 4" S in gl e… A ll S in gl e Sl op e T N J er se y O H 3 2" J er se y N E 3 2" J er se y T X 3 2" T L 4… O H 4 2" J er se y N E 4 2" J er se y A ll 3 2" J er se y A ll 4 2" J er se y A ll J er se y P( K A | C )

Probability of Crash Severity (PSEVj) C-29   Figure C-2 P(KA|C) limited by concrete shape with 95% confidence intervals. Cable Barrier Family P(KA|C) as a proportion and the 95% confidence interval are shown for each barrier within the cable barrier family in Figure C-3. As before, the diamond markers represent the point estimate of P(KA|C) with cable barriers, and the bars represent the 95% confidence interval. Based on the sampled data, we are 95% confident that the “true” P(KA|C) for cable barrier is within the range shown by the bars for each cable system. Recall the proprietary high-tension (HT) systems were combined into a single HT category by state for this analysis. 0.0000 0.0100 0.0200 0.0300 0.0400 0.0500 All Fshape All vertical wall All Single Slope All Jersey P( K A | C )

C-30 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers Figure C-3 P(KA|C) for cable barrier family with 95% confidence intervals. No fatal or severe (K or A) LT cable barrier crashes were observed in the Arizona, Oregon, and Pennsylvania data. The North Carolina data included two A-level injuries and Washington observed one A-level injury. There were no fatal crashes observed in the LT cable data. The HT cable systems were previously combined, which could explain the tight confidence range in this large sample. Figure C-4 exaggerates the y-axis and limits the figure to the combined analysis of HT and LT cable. While the probability of a severe or fatal injury given a crash, P(KA|C), with an LT system does appear to be less than with an HT system, the evidence is inconclusive. It is therefore recommended that crash severity for both HT and LT cable systems be presented simply as crash severity with cable barrier in the guidelines. The value P(KA|C)cable 0.0050, 95% CI [0.0021, 0.0079] has been adopted for the cable family. As MASH barriers penetrate the market and in-service performance evaluations (ISPEs) are completed, this value may be reevaluated. -0.1000 -0.0500 0.0000 0.0500 0.1000 0.1500 0.2000 W A T L 3 H T C ab le IA T L 3 H T C ab le PA H T C ab le T N H T C ab le A ll H T C ab le W A T L 3 L T C ab le A Z T L 3 L T C ab le N C T L 3 L T C ab le O R T L 3 L T C ab le PA L T C ab le A ll L T C ab le P( K A | C )

Probability of Crash Severity (PSEVj) C-31   Figure C-4 P(KA|C) for cable system groups with 95% confidence intervals. Metal Beam Barrier Family P(KA|C) as a proportion and the 95% confidence interval are shown for each barrier within the metal beam barrier family in Figure C-5. The point estimate of P(KA|C) is shown using diamond markers with corresponding numeric values on the y-axis. Based on the sample data, we are 95% confident that the ‘true’ P(KA|C) for each marker is within the range shown by the bars for each metal beam barrier studied. The Pennsylvania data provides a wide spectrum of the many types of W-beam installed within Pennsylvania. Notably, the Pennsylvania inventory captures each of these different beam systems. In addition to Pennsylvania data, Ohio and Tennessee results are also shown. The Ohio and Tennessee inventories do not include multiple types of metal beams, either because the state standardizes on a single system or because the inventory does not distinguish between the different metal beam systems. As with the other crash data, there is an underlying assumption that these data represent the crash severity of NCHRP Report 350 systems and that the crash severity data for NCHRP Report 350 systems can be extended to MASH systems until additional data become available. Notice, in Figure C-5, that the Pennsylvania strong post data have been combined into a single category and are shown next to the Ohio and Tennessee data to facilitate review. Like Figure C-4 in the previous analysis, Figure C-6 shows the combined analysis of all the strong post W-beam and the only available weak post data from Pennsylvania. There is no difference in the crash severity between weak post and strong post w-beam, accepting that the results for the weak post, as shown on the extremely exaggerated y-axis, have confidence intervals that overlap those for the strong post. -0.0020 0.0000 0.0020 0.0040 0.0060 0.0080 0.0100 0.0120 0.0140 0.0160 All HT Cable All LT Cable P( K A | C )

C-32 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers Figure C-5 P(KA|C) for metal beam barrier family with 95% confidence intervals. Figure C-6 P(KA|C) for metal beam barrier family with 95% confidence intervals. -0.0200 -0.0100 0.0000 0.0100 0.0200 0.0300 0.0400 0.0500 0.0600 0.0700 0.0800 P( K A | C ) 0.0000 0.0020 0.0040 0.0060 0.0080 0.0100 0.0120 0.0140 All PA Weak Post Wbeam All strong post wbeam P( K A | C )

Probability of Crash Severity (PSEVj) C-33   One approach for providing P(KA|C) for the metal beam barriers would be to combine the weak post and strong post systems as done for the cable and concrete barrier families. Another approach would be to simply use the strong post value and not provide for weak post in the guidelines. There is little difference in the ultimate result, and there is practicality in offering a value in the guidelines that encompasses both systems. It is therefore recommended to present a single value for the metal beam family which combines weak post and strong post systems. The value P(KA|C)beam = 0.0084, 95% CI [0.0073, 0.0094] was adopted for the metal beam family. As with the other longitudinal barriers, this value should be reexamined as ISPEs of MASH systems become available. Other Features P(KA|C) as a proportion and the 95% confidence interval are shown in Figure C-7 for the other features studied. As previously stated, the diamond markers represent the point estimate of P(KA|C), and reference should be made to the y-axis for the corresponding value. Based on the sampled data, we are 95% confident that the ‘true’ P(KA|C) for each feature is within the range shown by the bars for each category. The cross-median crash value shown was used in the development of these guidelines to represent CMC, P(KA|C)CMC = 0.0451, 95% CI [0.0441, 0.0461]. The values for rollover and fixed objects such as trees cannot be distinguished statistically. A single value for interaction with a non-designed roadside feature (NDRF) has therefore been adopted in these guidelines. P(KA|C)NDRF = 0.0589, 95% CI [0.0549, 0.0629]. Figure C-7 P(KA|C) for median features with 95% confidence intervals. 0.0000 0.0100 0.0200 0.0300 0.0400 0.0500 0.0600 0.0700 WA Cross Median Crash WA Rollover WA Tree WA Waterbody P( K A | C )

C-34 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers CHAPTER 6 RESULTS AND DISCUSSION The forgoing statistical analysis presented in this document provided the analysis and justification for each recommended grouping of barriers and features. Throughout the analysis, the 95% confidence intervals were presented to ensure high confidence in the recommended groupings. The resulting recommended groupings and P(K|C) 65 have been summarized in Table C-35. Probability levels for fatal crashes (K); fatal and serious crashes (KA); fatal, serious, and observed injury crashes (KAB); and fatal and any level of injury crashes (F+I) are also included to allow for flexibility as guideline development continues. The resulting values can be used in the guidelines and adjusted using the site-specific PSL, as follows: Table C-35 Recommended Groupings of Barriers and Median Features for Guideline Development Feature Observed Crashes MLE % INJ Nr+Nu P(Sev|C)65 K KA KAB F+I Cable 1,494 8.62 2,257 0.0009 0.0050 0.0297 0.0849 Metal Beam 7,437 9.85 28,494 0.0013 0.0084 0.0369 0.0895 Concrete 7325 16.74 15913 0.0021 0.0159 0.0810 0.1667 CMC 26,928 16.51 179,033 0.0098 0.0451 0.1290 0.1938 NDRF 9,361 50.10 13,484 0.0142 0.0589 0.3138 0.4836 Table C-35 includes the Nr+Nu estimate for each grouping. Any confidence interval level can be determined using the z-values shown above in Table C-34. Recall the equation presented above for calculating confidence levels: where = P(Sev|C) z = Number of standard deviations away from the mean (see Table C-34). n = use value in Nr+Nu column The values for P(KA|C)65 are based on observable police-reported crashes and adjusted to account for unreported crashes based on the model of crash severity discussed above. These severity measures are then standardized at a base PSL of 65 mph and can be adjusted for site- specific speeds. Using P(KA|C)65 to estimate crash severity in a conditional probability model

Probability of Crash Severity (PSEVj) C-35   such as the guidelines being developed provides a systematic methodology based on observed data and established crash severity relationships. REFERENCES ACTAR, Traffic Crash Reports and Overlay Forms, Accreditation Commission for Traffic Accident Reconstruction. https://actar.org/pdf/tn_rep1.pdf. Accessed March 13, 2017, 2017. Blincoe, L.J., A. G. Seay, E. Zaloshnja, T. R. Miller, E. O. Romano, S. Luchter, R. S. Spicer, and Administration National Highway Traffic Safety, "The Economic Impact of Motor Vehicle Crashes, 2000," 2002. Carrigan, Christine E., and Malcolm H. Ray, "Practitioner’s Guide to the Analysis of In-Service Performance Evaluation Data," Presented at 95th Annual Meeting of the Transportation Research Board, Washington, DC, 2016. Fitzpatrick, Michael S., Kathleen L. Hancock, and Malcolm H. Ray, "Videolog Assessment of Vehicle Collision Frequency with Concrete Median Barriers on an Urban Highway in Connecticut," Transportation Research Record, No. 1690, 1999. Hammond, Paula and John R. Batiste, "Cable Median Barrier: Reassessment and Recommendations Update," 2008. Kean, Confidence Interval for a Population Proportion, Kean University, Hillsdale, NJ. Mak, K. K., and R. L. Mason, "Accident Analysis - Breakaway and NonBreakaway Poles Including Sign and Light Standards Along Highways: Technical Report," Southwest Research Institute National Highway Traffic Safety Administration, San Antonio, TX, 1980. Nilsson, Göran, "Effects of speed limits on traffic accidents in Sweden," 1982. ODOT, Single Slope Barrier Details, Ohio Department of Transportation, Columbus, OH, 2017. ODOT, Midwest Guardrail System Details, Ohio Department of Transportation, Columbus, OH, 2013. PennDOT, Shoulder and Guide Rail Condition Survey Field Manual, Pennsylvania Department of Transportation, Harrisburg, PA. https://www.dot.state.pa.us/public/PubsForms/Publications/PUB%2033.pdf. Accessed February 28, 2017, 2017. Ray, Malcolm H. and Christine E. Carrigan, NCHRP Web-Only Document 307: Recommended Guidelines for the Selection of Test Levels 2 Through 5 Bridge Rails, Transportation Research Board, Washington, DC, 2021. Ray, Malcolm H., Christine E. Carrigan, and C. A. Plaxico, NCHRP Project 22-27, "Roadside Safety Analysis Program (RSAP) Update, Appendix B: Engineer's Manual RSAP," Transportation Research Board, Washington, DC, 2012.

C-36 Selection and Placement Guidelines for Test Level 2 Through Test Level 5 Median Barriers Ray, Malcolm H., Christine E. Carrigan, and Chuck A. Plaxico, "Method for Modeling Crash Severity with Observable Crash Data," Transportation Research Record: Journal of the Transportation Research Board, No. 2437, 2014b. Ray, Malcolm H., J. Weir, and J. Hopp, NCHRP Report 490: In-Service Performance of Traffic Barriers, Transportation Research Board, Washington DC, 2003. D. L. Sicking, D. L., K.A. Lechtenberg, and S. Person, NCHRP Report 638: Guidelines for Guardrail Implementation, Transportation Research Board, Washington, DC, 2009. Yale, Confidence Intervals, Yale University, New Haven, CT. http://www.stat.yale.edu/Courses/1997-98/101/confint.htm. Accessed April 27, 2015.