Motorcycle Crashes into Traffic Barriers: Factors Related to Serious and Fatal Injuries (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.

Roadway Characteristics Associated with Motorcycle Crashes into Longitudinal Barriers and the Influence on Rider Injury 9.1 Introduction Motorcyclists in the United States are more than 30 times more likely than passenger car occupants to be fatally injured in traffic crashes (NHTSA 2013). As early as 1981 in the seminal Hurt study (Hurt, Ouellet and Thom 1981a), motorcycle impacts with fixed objects have been implicated as posing an especially high risk to motorcycle riders. These fixed object crashes include impacts into longitudinal traffic barriers such as W-beam guardrails and concrete barriers. Although a motorcyclist impacting a barrier is a relatively infrequent event, previous research indicates that these crashes often result in severe injury consequences, even compared to other motorcycle crash types (Daniello and Gabler 2011a). In the United States, motorcycle- to-guardrail crashes account for more than 40% of all vehicle-to-guardrail fatalities, more than any other single vehicle type, despite motorcycles representing only 2% of the vehicle fleet (Gabler 2007). Several researchers have described the development, implementation, or evaluation of motorcycle-to-barrier crash countermeasures to mitigate the injury consequences of these crashes (Koch and Schueler 1987; Ellmers 1997; Mulvihill and Corben 2004; Janssen et al. 2005). These countermeasures include specially designed longitudinal traffic barriers as well as products intended to retrofit existing barriers (e.g., guardrail post protection). Previous research sug- gested that installation of these countermeasures is only cost effective at locations susceptible to this crash type (Koch and Schueler 1987; Domham 1987). Little is known regarding the specific roadway conditions and roadway alignment, such as horizontal curve radius, most frequently associated with this crash type. A better understanding of the roadway characteristics associated with motorcycle-to-barrier crashes is needed and would aid designers in determining the most effective locations to implement motorcycle-to-barrier crash countermeasures. Much of this chapter is provided in Gabauer (2016).This work is reprinted by permission of Taylor & Francis Ltd. on behalf of Taylor & Francis Group, LLC and The University of Tennessee. Text is reproduced largely verbatim from this work. The figures contained hereafter are not provided in the publication. 9.2 Objective The purpose of this study was to examine police-reported motorcycle impacts into longitu- dinal traffic barriers to (1) determine specific roadway and roadway alignment characteristics associated with these crashes and (2) investigate the influence of these characteristics on resulting rider injury. C H A P T E R 9 91  

92 Motorcycle Crashes into Traffic Barriers: Factors Related to Serious and Fatal Injuries 9.3 Background and Previous Research 9.3.1 Motorcycle-to-Barrier Impacts, Rider Injury, and Related Crash Test Procedures A majority of previously published literature regarding motorcycle impacts into barriers has focused on characterizing the resulting occupant injuries (Ouellet 1982; Bryden and Fortuniewicz, 1986; Koch and Brendicke, 1988; Hell and Lob 1993; Gibson and Benetatos 2000; ACEM 2004; Bambach et al. 2012; Daniello and Gabler 2012). On the whole, these studies suggested that motorcycle-to-barrier crashes result in more severe rider injury and an increased rider fatality risk when compared to all motorcycle crashes or motorcycle crashes involving only an impact with the ground. A study by Savolainen and Mannering (2007) provided a quantification of this increased injury risk, finding that guardrail impact decreases the probability of a minor or no-injury crash by 17%. A smaller number of researchers investigated the influence of barrier type on injury risk (Gabler 2007; Mulvihill and Corben 2004; Candappa et al. 2005; Daniello and Gabler 2011b) and generally found a slightly increased injury risk associated with metal beam barriers compared to concrete barriers. Full-scale crash test procedures in the United States (Ross et al. 1993; AASHTO 2009), in Europe (CEN 2010), and in Australia/New Zealand (Standards Australia 1999) are used to assess the crashworthiness of longitudinal barriers prior to field installation. Since motorcyclists have traditionally represented a small portion of the overall vehicle fleet, the current U.S. proce- dures address only passenger vehicle (e.g., car and light truck) impacts to longitudinal barriers. Researchers have developed motorcycle-to-barrier crash test procedures (Duncan et al. 2000; Berg et al. 2005a; Peldschus et al. 2007; Garcia et al. 2009), and a testing standard currently exists in Spain (UNE 2008) and in the European Union (CEN 2012). These existing motorcycle test procedures generally specify two impact configurations: an upright rider/motorcycle impacting a barrier at an angle and a sliding rider impacting the barrier headfirst, also at an angle. Injury risk in these tests is based primarily on the measured response of an ATD. 9.3.2 Roadway Characteristics Associated with Motorcycle-to-Barrier Impacts Several previous studies provided an indication of roadway characteristics associated with motorcycle impacts into longitudinal barriers. Using French crash data from the late 1970s and early 1980s, Quincy et al. (1988) found that motorcycle impacts into barriers occurred more frequently on urban roadways and were overrepresented at entrance and exit ramps. Approxi- mately 54% of these crashes occurred on ramps, but this location type represented only 5% of the total roadway length. Of 22 motorcycle-barrier crashes in Germany, Domham (1987) notes that not one of these crashes occurred on a horizontal curve with the smallest radii. Ellmers (1997) indicated that mountainous rural primary and secondary roads were critical to safety for this crash type. Gibson and Benetatos (2000) presented findings from fatal motorcycle crashes occurring in NSW from 1998 through 1999. Based on eight barrier impact fatalities, the most frequent scenario (4 of 8) was the rider losing control on a right-hand bend and exiting the roadway to the left, followed by an impact to the barrier located on the roadside. Using the police speed estimate and the speed limit in the area of the crash, the authors noted that motorcycle- to-barrier crashes occurred at speeds above 60 km/h (38 mph). Berg et al. (2005a) reported on 57 motorcycle-barrier crashes that were investigated in Germany and found that the majority of crashes occurred within curves (53% left, 7% right) with the remaining 40% occurring on straight roadway sections. More recently, Jama et al. (2011) investigated 77 fatal motorcycle-to- barrier crashes in Australia and New Zealand. A vast majority (81%) of these crashes involved a horizontal curve with an approximately equal distribution of right and left curves.

Roadway Characteristics Associated with Motorcycle Crashes into Longitudinal Barriers and the Influence on Rider Injury 93   While the studies summarized earlier provided some insight into roadway characteristics of motorcycle-to-barrier crashes, the findings tended to be general in nature (e.g., horizontal curve presence versus horizontal curve radius) and primarily anecdotal. A single recommendation was found with respect to horizontal curvature (Elliot et al. 2003) that suggests potential motorcycle- barrier countermeasures are appropriate on horizontal curves with a radius less than 250 meters (820 ft). No other quantitative roadway geometric data was found in the available literature specific to the motorcycle-to-barrier crash mode. 9.3.3 Effects of Roadway Geometry and Characteristics on Motorcycle Crash Frequency and Severity Other previous motorcycle crash research, primarily focused on either single-vehicle motor- cycle or all motorcycle crashes, provides additional insight regarding the influence of roadway geometric characteristics. Similar to the previous motorcycle-to-barrier research, however, a majority of these studies include only generic roadway geometry characteristics. The majority of these studies support the notion that single motorcycle crashes are more likely to occur in locations with horizontal curvature, vertical curvature, and vertical grade as summarized in Table 9-1 below. At least two studies provided detailed roadway alignment for motorcycle crashes, although not specific to motorcycle-to-barrier impacts and focused on rural two-lane highway crash locations. Schneider et al. (2009) examined various roadway, operator, environmental, and vehicle factors affecting the severity of horizontal curve crashes on Texas rural two-lane high- ways. Data included 5 years of police-reported crash data coupled with roadway data, a total of Author [Reference] Location/Data Years/Data Type Roadway Alignment-Related Findings Li et al. (2009) Taiwan/2000–2002/Police- reported crashes linked to hospital and death records Fatality risk on non-level, non-straight roadways was significantly increased for motor vehicle occupants but not for motorcycle occupants. Motorcycle crash victims had a higher odds of fatality (OR: 1.09), but this was not statistically significant. Savolainen and Mannering (2007) Indiana/2003–2005/Police- reported crashes with rider training data Single motorcycle crashes on horizontal curves decrease the probability of a minor or no-injury crash by 8%. For multi-vehicle motorcycle crashes, horizontal and vertical curve presence increased incapacitating injury by 45 and 81%, respectively. Preusser et al. (1995) U.S./1992/FARS Approximately 70% of run-off road fatal motorcycle crashes occurred on curves compared to 21% for all other fatal motorcycle crashes. Kim et al. (2002) HI/1986–1995/Police- reported crashes with linked hospital records Single motorcycle crash > 5 times more likely when horizontal curve present and ~1.4 times more likely when vertical curve present. Serious and fatal injuries 1.5 times and nearly 2 times more likely on curved roads. Quddus et al. (2002) Singapore/1992–2000/ Police-reported crashes Probability of fatality increases by ~72% when a bend is present. Narrow roads, sharp turns, and blind corners not found to be statistically significant with respect to injury severity. Sharp turns are found to increase motorcycle damage levels. No numerical values are indicated to define a “sharp” curve. ACEM (2004) 5 European countries/1999– 2000/In-depth crash investigations All motorcycle crashes were found to be overrepresented in curves (30% occurring on horizontal curves) compared to passenger vehicle crashes (21% on horizontal curves). Samaha et al. (2007) U.S./1992–2004/Weighted sample of police-reported crashes Motorcycle crashes occurring away from a junction were found to be 1.7 times more likely to be fatal than those crashes occurring within an intersection. Table 9-1. Summary of previous motorcycle crash study findings related to roadway alignment.

94 Motorcycle Crashes into Traffic Barriers: Factors Related to Serious and Fatal Injuries 10,029 motorcycle crashes. Horizontal curves were split into three categories based on radius: “small” curves with radius less than 500 ft, “large” curves with radius greater than 2,800 ft, and “medium” curves with radius between the small and large ranges. There were a total of 354 motorcycle crashes with 62 occurring on “small” curves, 241 occurring on “medium” curves, and 51 occurring on “large” curves. Compared to all other vehicle types, fatal and incapacitating motorcycle injuries were 604%, 628%, and 568% more likely on “small,” “medium,” and “large” horizontal curves, respectively. Non-incapacitating injuries were 73 to 98% more likely than for operators of all other vehicle types, and “large” curves had the highest increase in non- incapacitating injury risk. Schneider et al. (2010) investigated roadway geometry effects on single-vehicle motorcycle crash occurrence using crash and roadway inventory data from Ohio. Data included 225 single-vehicle motorcycle crashes occurring on Ohio rural two-lane highways between 2002 and 2008. Roadway characteristics found to have a statistically significant effect on motorcycle crashes were (1) horizontal curve length, (2) horizontal curve radius, (3) distance relative to the curve end-points, (4) roadway shoulder width, and (5) total segment average daily traffic (ADT). Longer, higher speed curves and smaller radius curves were found to increase the frequency of motorcycle crashes on a particular segment. Curves were found to influence crash risk on adjacent tangent sections for up to 300 ft, but crash risk decreased as a motorcyclist moved further from a curved section. For every percentage point increase in total ADT, motorcycle crash frequency was estimated to increase by 0.43%. In addition, roadway sections with shoulders less than 6 ft in width were found to increase motorcycle crash risk by approximately 50%. In addition to these alignment-related findings, researchers have also identified several other roadway factors that affect the occurrence and/or severity of motorcycle crashes. Shankar and Mannering (1996) found that single-vehicle motorcycle crashes occurring on interstates increase the likelihood of disabling and possible injury, and that wet pavement increases the likelihood of property damage and possible injury. Kim et al. (2002) found that single-vehicle motorcycle crashes were approximately three times more likely when an oily/wet road surface was present, 1.5 times more likely in rural areas, and approximately five times more likely when a roadway surface defect was present. For single-vehicle motorcycle crashes, Savolainen and Mannering (2007) found wet pavement and intersection crashes less likely to result in no injury, with a 77% and 29% higher chance of no injury, respectively. Similarly, posted speed limits over 50 mph were found to decrease the probability of a minor or non-injury single-vehicle motorcycle crash by 10%. Li et al. (2009) noted that motorcycle fatality risk decreased in urban areas and on city streets while it increased on highways. 9.4 Methodology The overall approach for this study was to use state-level police-reported crash data linked with roadway data to investigate the characteristics of crashes involving a motorcycle impacting a longitudinal barrier. Two additional data subsets, all single-vehicle motorcycle crashes and multiple-vehicle motorcycle crashes, were used as comparison groups for the motorcycle-to- barrier cases. Statistical models were developed to examine the influence of various roadway characteristics, particularly alignment, on resulting rider injury while accounting for other potential confounding factors. All data processing and statistical analyses for this study were performed using SAS V9.2. 9.4.1 Data Sources and Case Selection Procedures Data for the study was obtained from HSIS, a nine-state database maintained by the FHWA that contains linked crash, roadway inventory, and traffic volume data (FHWA 2011). Of the nine U.S. states with HSIS data available, only five have roadway information that includes both

Roadway Characteristics Associated with Motorcycle Crashes into Longitudinal Barriers and the Influence on Rider Injury 95   horizontal curvature and vertical grade information: Michigan, Utah, Washington, Ohio, and Illinois. For the present study, data was selected only from Washington and Ohio. Michigan (Council et al. 2001) and Utah (FHWA 2000) participation in HSIS ended in 1997 and 2000, respectively. For both of these states, detailed roadway alignment data is available until 1994 and has been excluded from further analysis due to the age of the limited data available. Illinois has more recent data available, but the alignment information is only collected for “potentially substandard” curves (Council and Mohamedshah 2009a). As the alignment information represents an incomplete dataset, Illinois data also was excluded. Ohio HSIS Case Selection and Data Preparation HSIS data from Ohio used in this study includes all motorcycle crashes occurring from 2000 through 2011. Data was available from 1997 through 1999 but was excluded from analysis as a new crash reporting form was introduced in Ohio in 2000 that represented major changes from the previous Form 46. An initial analysis of the 2000 through 2011 data indicated that a small portion, less than 1.2% of all motorcycle crashes, involved motorized bicycles. Due to the small number of motorized bicycle crashes, these cases were excluded from further analysis. To prepare the data for analysis, accident and vehicle data tables were first merged by crash year (ACCYR) and case number (CASENO). The available crashes were then divided into three data subsets: 1. Single-vehicle motorcycle crashes involving one or more longitudinal barrier impacts (SVLB) 2. All single-vehicle motorcycle crashes (SV) 3. Multi-vehicle crashes involving at least one motorcycle (MV) The SVLB subset is of primary interest, while SV and MV crashes primarily serve as com- parison groups. The vehicle type variable (VEHTYPE) was used to exclude motorized bicycles, and the number of vehicles variable (NUMVEHS) was used to distinguish between single- (NUMVEHS = 1) and multi-vehicle (NUMVEHS > 1) crashes. Longitudinal barrier crashes were selected using all of the available sequence of event variables (EVENT1 through EVENT4). For the purpose of this study, a longitudinal barrier crash was defined as one or more impacts into a guardrail face, guardrail end, or median barrier (EVENT = 30, 31, or 32). This barrier impact could occur in any one (or more) of the four event sequences recorded. Note that MV subset was limited to data only on the crash-involved motorcycles and may include crashes where a longitudinal barrier was impacted. For each data subset and crash year combination, PROC SQL was used to merge the combined accident and vehicle data with the associated roadway, curve, and grade tables by matching the county route variable (CNTY_RTE) and ensuring that the milepost was between the beginning and end of the road/curve/grade segment. A similar procedure was used to merge the accident/ vehicle data with the angle point table except that the milepost equaled the angle point milepost; Ohio DOT defines an “angle point” as any sharp angle horizontal curve with degree of curvature exceeding 90 degrees (Council and Mohamedshah 2007). This data was then merged with the available occupant data by ACCYR, CASENO, and vehicle number (VEHNO). Washington State HSIS Case Selection and Data Preparation Data provided from HSIS included all motorcycle, scooter, and moped crashes occurring in Washington from 1993 through 1996 and 2002 through 2011. A complete set of data from Washington was not available between 1996 and 2002 primarily due to state budgetary constraints during that time (Council and Mohamedshah 2009b). For consistency with the Ohio data available, only the 2002 through 2011 Washington data were used in the analysis. Initial analysis of this data indicated that scooters and mopeds represent approximately 1.7% of all Washington crashes available and, as a result, have been excluded from further analysis.

96 Motorcycle Crashes into Traffic Barriers: Factors Related to Serious and Fatal Injuries Preparation of the Washington data was nearly identical with that of the Ohio data, including merging of the accident and vehicle data tables followed by dividing the data into the same three subsets. The only difference is the presence of a ramp file and the absence of an angle point file. Longitudinal barrier crashes were selected using the available object struck variables (OBJECT1 and OBJECT2). The Washington barrier-related object struck codes differ somewhat from those present in the Ohio data and generally separate barriers by type (guardrail or concrete barrier) and impact location (barrier end, barrier face). Any impact to either a concrete barrier or guardrail was considered a longitudinal barrier impact (OBJECT = 31, 32, 33, 34, 35, or 36). PROC SQL was then used to merge the crash, vehicle, roadway, curve, grade, and ramp information into a single table by matching the road inventory variables (RD_INV, ROAD_INV, CURV_INV, and GRAD_INV) and ensuring that the milepost was between the beginning and end of the road/ curve/grade segment. This data was then merged with the available occupant data by ACCYR, CASENO, and VEHNO. 9.4.2 Data Analysis and Model Development Characterization of Motorcycle Crashes into Longitudinal Barriers For both states, descriptive statistics related to the roadway, crash, and rider were generated for all three crash subsets. Roadway characteristics included horizontal curvature, vertical grade, number of lanes, median presence/width, shoulder width, posted speed limit, functional classi- fication, and average annual daily traffic (AADT). Mean values were reported for horizontal curve radius, vertical grade, speed limit, median width, and AADT. Non-tangent sections were categorized into two groups (radius ≥ 820 ft or < 820 ft) based on the sole recommendation for motorcycle-to-barrier crash countermeasures found in Elliot et al. (2003). Although the mean grade was reported for both states, Ohio reports only data on grades greater than or equal to 3%. Two vertical grade categories were created using the 3% grade stipulation present in the Ohio data. Number of lanes was used to group the data into three categories: two lanes or less, greater than two but less than four, and more than four lanes. Roads were classified as either divided or undivided; Ohio had a separate variable indicating this while the left shoulder data (LSHL_WD2) was used in Washington. Shoulder width was grouped into three categories: less than 2 ft, greater than or equal to 2 ft but less than 10 ft, and greater than or equal to 10 ft, based loosely on the AASHTO (2011) shoulder width guidelines. Posted speed limit was divided into two groups based on the AASHTO (2011) distinction between high- and low-speed design. Functional classification was reported in aggregate for urban and rural areas, and the area type distribu- tion was reported separately. Crash characteristics included location and roadway surface condition. Crash location was categorized into three categories: intersection or intersection- related, non-intersection, or other. The “other” category included driveways, private property, and unknown location types. Roadway condition included dry, wet, and other, which included snow, ice, and sand, among others. Occupant characteristics included helmet usage, gender, mean age, and police-reported injury severity. The distributions of these characteristics are compared to all state-specific HSIS data where appropriate. Data was also provided on the number of MV crashes involving a barrier impact and the distribution of barrier type struck for the Washington data. Box and whisker plots were generated to further investigate numeric roadway characteristics such as horizontal curve radius, vertical grade (Washington only), and AADT. Plots were gener- ated for each data subset as well as all HSIS available data for the most recent year of data for a particular state. As AADT tends to change more frequently than roadway alignment, the distri- bution of AADT for all years and all roads in the state was reported. T-tests were used to compare the means of the independent sample combinations within each state (e.g., SV compared to MV crashes and SVLB compared to MV crashes).

Roadway Characteristics Associated with Motorcycle Crashes into Longitudinal Barriers and the Influence on Rider Injury 97   Statistical Model Development Using the suitable HSIS SVLB cases from both states, a binary logistic regression model was developed to predict rider injury severity based on roadway alignment characteristics, while accounting for potential crash and occupant confounding factors. Injury was categorized as either severe (fatal or incapacitating) or nonsevere (non-incapacitating, possible, and no injury). Unknown, missing, or non-traffic fatalities were excluded. Two different horizontal curve variables were used in the model development. The first categorized curved sections based on radius as described above: tangent section, radius ≥ 820 ft, or radius < 820 ft. The second normalized the radius using the recommended minimum horizontal curve radius based on Ohio (ODOT 2012) or Washington (WSDOT 2012) geometric design standards. The minimum radius was selected using a design speed equal to the posted speed limit and using the largest permissible superelevation, as superelevation data was not universally available in the HSIS roadway data. For curves with posted speed less than that of the tangent sections of the same road, the lesser speed was used in the normalized radius calculation. Vertical grade was cate- gorized as either less than 3% or greater than 3%. AADT was divided into four categories: < 2,500 vehicles per day (vpd); 2,500 to 9,999 vpd; 10,000 to 49,999 vpd; and > 50,000 vpd. Other explanatory roadway characteristics included shoulder width (< 2 ft; ≥ 2 and < 10 ft; and 10+ ft), posted speed (≤ 45 mph and > 45 mph), and divided/undivided. Confounding occupant and crash factors included occupant age (≤ 25 and > 25 years), helmet usage (worn, not worn, or unknown), and road surface condition (dry, wet, or other). ORs were used to determine the influence of roadway characteristics on injury severity in this crash type as well as to quantify the effects of the possible confounding factors. 9.5 Results 9.5.1 Characterization of Motorcycle Impacts to Barriers Of the 30,454 motorcycle crashes available for analysis, there were 1,511 single-vehicle motorcycle-to-barrier crashes involving 1,691 occupants. These crashes represented approxi- mately 5% of all available motorcycle crash cases, 4.5% of Ohio crashes, and 6.2% of Washington crashes. All single-vehicle motorcycle crashes comprised 43% of all motorcycle crashes (41% of Ohio crashes and 48% of Washington crashes). Table 9-2 provides a more detailed summary of the available data. Note that Table 9-2 only includes crashes with matching roadway data. There was a small portion, less than 1% of all cases, excluded from further analysis as no matching roadway data was available. There were differences in mean horizontal curve radius and vertical grade between states. Mean horizontal curve radius for crashes in Washington State was higher than in Ohio across all data subsets. For all reported roadway data, the mean horizontal curve radius was 2,494.7 ft and 664.5 ft for Washington and Ohio, respectively. Excluding ramps, approximately 63% of Washington SVLB crashes occurred on curved sections compared to 41% of SV crashes and 21% of MV crashes. In Ohio, approximately 19% of SVLB crashes occurred on curved sections com- pared to 12% of SV and 3.6% of MV crashes. Of the horizontal curve crashes, 61% of Washington SVLB and 74% of Ohio SVLB crashes were on curves with radius less than 820 ft. Approximately 45% of SVLB crashes in Washington occurred on vertical grades in excess of 3% compared to 29% and 22% for Washington SV and MV crashes, respectively. A similar trend is observed in Ohio with 20, 15, and 7% of SVLB, SV, and MV crashes occurring on grades in excess of 3%. For all roadway sections, the mean grade was 7.1% in Ohio and 1.8% in Washington. Considering only the grades equal or greater than 3%, the mean vertical grade in Washington was 4.6%. Approximately 20% of Washington SVLB crashes occurred on horizontal curve sections with a 3% grade or higher compared to 10.9% of SV crashes and 5.3% of MV crashes. In Ohio, 8.7% of

98 Motorcycle Crashes into Traffic Barriers: Factors Related to Serious and Fatal Injuries Variable Available Data Washington Ohio SVLB SV MV SVLB SV MV All Vehicles 556 4372 4659 955 8856 12,567Occupants 599 4415 4703 1092 10,229 14,432 Geometric, Roadway and Crash Characteristics (by involved vehicle) Horizontal Alignment Mean radius [ft] 1157 2235 3450 668 694 810 Radius < 820’ 162 632 197 135 799 268 Radius ≥ 820’ 102 801 706 48 296 189 Tangent Section 157* 2094* 3355* 772 7761 12,110 Vertical Alignment Mean grade [%] 2.72 2.02 1.65 7.7† 7.7† 6.3† < 3% 240 2469 3135 766 7549 11,699 ≥3% 195 999 865 189 1307 868 Lanes 1 – 2 lanes 262 1935 1565 473 5349 5815 3 – 4 lanes 225 1852 2103 264 2481 5539 More than 4 lanes 69 585 991 218 1026 1213 Median Undivided 363* 3009* 3769* 493 6192 9850 Divided 58* 550* 489* 462 2664 2717 Mean Width [ft] 14.1 16.4 12.4 34 38.5 32.2 Shoulder Width Less than 2 ft 65 935 1740 105 1626 5132 ≥2 and < 10 ft 360 2415 1759 466 4746 4745 10+ ft 131 1022 1158 336 1950 1880 Unknown/Missing 0 0 2 48 534 810 Posted Speed Mean [mph] 51.3 50.3 46.4 53.8 50.7 42.9 ≤ 45 mph 243 1959 2499 181 2463 7571 > 45 mph 313 2412 2160 759 6259 4650 Not Stated 0 1 0 15 134 346 Area Type Rural 247 1883 1212 478 4930 3673Urban 309 2489 3445 476 3912 8862 Roadway Functional Class Principal Arterial 238 2263 3239 501 3893 7025 Minor Arterial 124 828 788 139 1676 3368 Collector 59 468 229 314 3269 2137 Local Road 0 0 0 0 4 5 Traffic Mean ADT [vpd] 34,925 34,021 45,971 33,396 20,929 21,298 Location Intersection 49 950 2007 83 1499 6548 Non-Intersection 504 3261 2005 860 7166 4742 Other/Unknown 3 161 647 12 191 1277 Road Surface Dry 505 3812 4285 886 8057 11,964 Wet 37 407 350 60 581 493 Other 14 153 21 8 160 47 Rider and Injury Characteristics (by involved occupant) Helmet Helmet worn 498 3947 4100 509 4630 5347 Helmet not worn 1 38 53 500 4862 7304 Unknown/Missing 57 387 506 83 737 1781 Gender Male 486 3898 4249 899 8250 11,671 Female 104 424 308 185 1877 2307 Unknown/Missing 9 93 146 8 102 454 Age Mean [years] 38.8 40.8 41.5 38.9 40.9 41.2 Injury Fatal 39 136 164 101 369 517 Incapacitating 137 722 689 435 2731 3038 Non-Incapacitating 257 2023 1481 411 4665 4743 Possible Injury 113 957 1122 57 927 1864 No Injury 43 473 1070 60 1293 3654 Unknown 10 104 177 28 244 616 † Ohio only reports vertical grades equal to or greater than 3%, * Ramp crashes excluded. Table 9-2. Summary of Washington (2002–2011, inclusive) and Ohio (2000–2011, inclusive) HSIS data.

Roadway Characteristics Associated with Motorcycle Crashes into Longitudinal Barriers and the Influence on Rider Injury 99   SVLB crashes occurred on curved sections with grades at 3% or higher compared to 4.4% of SV and 0.85% of MV crashes. Approximately half of SVLB crashes in both states occurred on roadways with two lanes or less. A higher percentage of Washington SVLB crashes (40%) occur on roadways with more than four lanes compared to Ohio SVLB crashes (23%). Of all Washington roadways, 55% were two lanes or less and 11% were more than four lanes. For Ohio, 65% of roads were two lanes are less and 6% were more than four lanes. Average median width was larger for all Ohio data subsets compared to Washington. For all divided roadways, the average median width was 37.2 ft in Ohio and 16 ft in Washington. A larger portion of Washington SVLB crashes (86%) occurred on undivided roadways compared to 52% of Ohio SVLB crashes. Note that the Washington median data in Table 9-2 does not include any ramp crashes. There were 135, 813, and 401 ramp SVLB, SV, and MV crashes, respectively. The right shoulder width distribution was similar between states for corresponding data subsets. Washington did have a higher proportion (65%) of SVLB crashes that occur on roadways with shoulders between 2 and 10 ft wide compared to Ohio (49%). Mean posted speed limit was slightly higher for SVLB crashes in both states. A smaller portion (56%) of Washington SVLB crashes occurred on roadways with a speed limit greater than 45 mph compared to 80% for Ohio SVLB crashes. Approximately half of SVLB crashes in both states occur in rural areas. For SVLB crashes, Washington minor arterials were over- represented (20% of road sections but 29% of SVLB crashes) as were Ohio principal arterials (42% of road sections but 53% of SVLB crashes). SVLB crashes occurred on higher AADT roadways in Ohio, while MV crashes had the highest AADT values in Washington. The road surface and location distributions were similar for corresponding data subsets. The vast majority of SVLB crashes in both states occurred at non-intersection locations and on dry roadways. A small portion of the MV crashes contain longitudinal barrier impacts not included in the SVLB data subset. In Ohio, 162 MV crashes had at least one barrier impact, which represents approximately 1.35 of MV crashes and 0.8% of all Ohio motorcycle crashes. Washington was similar with 116 MV crashes that involved at least one barrier impact, representing approximately 2.5% of MV crashes and 1.3% of all Washington motorcycle crashes. With the Washington data, it was possible to discern barrier type; approximately 62% of SVLB crashes involved one or more metal barrier impacts compared to 37% impacting one or more concrete barriers. Less than 1% struck both a metal and concrete barrier. In terms of rider characteristics, almost all of the Washington SVLB occupants were helmeted, while less than 50% of Ohio SVLB riders were helmeted. The mean age across the data subsets was similar in both states, with SVLB crashes having a lower mean age. Distribution of gender was approximately equal for all data subsets with male occupants generally more than 80% of involved occupants. A total of 6.5% of Washington SVLB and 9.2% of Ohio SVLB involved occupants were fatally injured. Fatal injury rates for the SV and MV crashes were between 3% and 3.6%. SVLB crashes also had the lowest proportion of no injury reported (7% in Washington and 5.5% in Ohio). Figure 9-1 and Figure 9-2 show the distribution of horizontal curve radius and AADT, respectively, for the SVLB, SV, and MV crashes in each state. Figure 9-3 shows the vertical grade distribution for each Washington data subset; Ohio was excluded since the available data does not differentiate grades less than 3%. T-tests indicate a statistically significant difference (p < 0.0001 in all cases) in mean AADT, horizontal curve radius, and grade between Washington SVLB and MV crashes and between Washington SV and MV crashes. T-tests for the Ohio data indicate statistically significant differences in curve radius (p < 0.001) between SVLB and MV crashes and SV and MV crashes. The difference in mean AADT was statistically significant (p < 0.0001) between Ohio SVLB and MV crashes but not significant (p = 0.391) between Ohio SV and MV crashes.

100 Motorcycle Crashes into Traffic Barriers: Factors Related to Serious and Fatal Injuries 9.5.2 Statistical Model Results A binary logistic regression model was developed to predict rider injury using the 1,170 cases with data available for all variables of interest. The developed model had a C-statistic value of 0.69, which provides a single numerical value of how well the model distinguishes between the response variable, in this case, presence of severe rider injury. The OR values obtained from the binary logistic regression, along with the 95% confidence bounds, are summarized in Table 9-3. Note that the OR shown is with respect to the group indicated in the comparison group column. Statistically significant effects are those where the 95% confidence bounds do not bracket the value of 1.0. Rider helmet use, age, and alcohol involvement were found to have a statistically significant effect on rider injury severity. Not wearing a helmet was found to increase the odds of severe injury by a factor of 2, while the involvement of alcohol increased the odds of severe injury by a factor of 3. Aside from the road surface condition at the time of the crash, the only statistically 0 1000 2000 3000 4000 5000 6000 SVLB SV MV All Curves 0 10000 20000 30000 40000 50000 60000 70000 80000 SVLB SV MV All Curves Ho riz on ta l C ur ve R ad iu s [ ft ] Ho riz on ta l C ur ve R ad iu s [ ft ] 0 50000 100000 150000 200000 250000 300000 SVLB SV MV All Roads, All Years Av er ag e An nu al D ai ly T ra ffi c [v pd ] 0 50000 100000 150000 200000 250000 300000 SVLB SV MV All Roads, All Years Av er ag e An nu al D ai ly T ra ffi c [v pd ] Figure 9-1. Distribution of horizontal curve radius by data subset for Ohio (left) and Washington (right). Figure 9-2. Distribution of AADT by data subset for Ohio (left) and Washington (right).

Roadway Characteristics Associated with Motorcycle Crashes into Longitudinal Barriers and the Influence on Rider Injury 101   significant roadway characteristic was whether the roadway was divided. Motorcycle-to-barrier crashes occurring on a divided road were nearly 2 times as likely to result in a severe injury as those occurring on undivided roadways. There was some evidence of increased rider injury risk in curves where the radius was less than the recommended minimum, grades were less than 3%, posted speeds were greater than 45 mph, shoulder widths were between 2 and 10 ft, and AADT was less than 10,000 vpd; these results, however, were not found to be statistically significant. A nearly identical model (full results not shown) was developed using the horizontal curve radius categories in place of the normalized curve radius shown in Table 9-3. In this model, horizontal curves with radius less than 820 ft were found to increase the odds of severe rider injury by a factor of 1.2 (95% CI: 0.82–1.64) when compared to tangent sections. A similar increased odds 0 5 10 15 20 25 30 SVLB SV MV All Roads Ve rti ca l G ra de [% ] Figure 9-3. Distribution of vertical grade by data subset for Washington. Variable Type Parameter Value Comparison Group Odds Ratio 95% CI Roadway Normalized horizontal curve radius < 1 Tangent Section 1.31 0.92 – 1.9 ≥ 1 Tangent Section 0.96 0.63 – 1.5 Vertical grade < 3% ≥ 3% 1.28 0.95 – 1.7 Posted speed >45 mph ≤ 45 mph 1.32 0.97 – 1.8 Roadside shoulder width ≥2 and < 10 ft < 2 ft 1.11 0.72 – 1.7 10+ ft < 2 ft 0.91 0.53 – 1.6 AADT < 2,500 vpd > 50,000 vpd 1.09 0.62 – 1.9 2,500 - 9,999 vpd > 50,000 vpd 1.55 0.89 – 2.7 10,000 - 49,999 vpd > 50,000 vpd 1.00 0.67 – 1.5 Configuration Divided Undivided 1.81 1.16 – 2.8 Crash Surface condition Dry Wet 3.78 2.09 – 6.8Other Wet 2.86 0.80 – 10.2 Rider Helmet use No Yes 2.04 1.56 – 2.7 Unknown Yes 1.45 0.81 – 2.6 Age > 25 years ≤ 25 years 1.43 1.03 – 2.0 Alcohol Yes No 2.99 2.10 – 4.3 Table 9-3. Summary of OR results for the rider injury binary logistic regression model.

102 Motorcycle Crashes into Traffic Barriers: Factors Related to Serious and Fatal Injuries was noted for curves with radius of 820 ft or greater (OR: 1.26, 95% CI: 0.82–1.95). The odds ratios for the remainder of the effects remained nearly unchanged as well as the C-statistic (C = 0.692). 9.6 Discussion Based on the data available from both states, SVLB crashes accounted for approximately 5% of all motorcycle crashes, which is consistent with previous studies using Maryland (Daniello and Gabler 2012) and Australian (Jama et al. 2011) data. The proportion of fatal SVLB for these crashes varied from 6.5% in Washington to 9.25% in Ohio, which is comparable to previous studies, albeit on the lower end. A study (Daniello and Gabler 2011b) using data from North Carolina, Texas, and New Jersey reported a 9%, 14%, and 16% fatality percentage, respectively, for motorcycle-to-barrier crashes. For the current study, the SVLB fatality percentage was between 2 and 2.5 times that of SV and MV crashes. A vast majority of the motorcycle-to-barrier impacts were single-vehicle events, with less than 20% involving an impact with another vehicle prior to or after barrier impact. With respect to roadway alignment, SVLB crashes were found to occur on smaller radii horizontal curves and higher mean grades compared to MV crashes. For horizontal curves, this difference was more prominent in Washington. While differences in horizontal curve radius and vertical grade means existed between the two states, it appears that these differences were in part due to road network differences. The mean horizontal curve radius for all HSIS roadways reported in Ohio was approximately 25% of the mean reported in Washington. Likewise, the mean grade in Ohio was higher than that for all reported Washington roadways even after correcting for the reporting differences. Note that the mean grade reported, however, did not incorporate grade length. Similar to Quincy et al. (1988), ramp crashes in Washington were found to be overrepresented, although to a lesser extent. Approximately 24% of SVLB crashes occurred on ramps while these ramps represented 13.5% of HSIS-reported road mileage com- pared to Quincy et al. (1988), which found 54% of crashes occurring on ramps representing 5% of road mileage. No HSIS data was available to discern ramp crashes in Ohio. Although the general trend in both states was that SVLB crashes were overrepresented on curved sections, there were large differences between states. A majority (63%) of the SVLB crashes in Washington occurred on curves, while 19% of SVLB crashes in Ohio occurred on curved sections. Strictly applying the 820-foot curve radius recommendation (Elliot et al. 2003), motorcycle-to-barrier crash countermeasures would only be present in 38% and 14% of motorcycle-to-barrier crashes in Washington and Ohio, respectively. The combination of horizontal curvature and vertical grade does appear to influence the occurrence of SVLB crashes as SVLB crashes occurred at least twice as frequently in these areas compared to SV and MV crashes. Despite this overrepresentation, less than one-fourth of Washington SVLB and less than one-tenth of Ohio SVLB crashes occurred on these sections. Several other roadway characteristic differences were notable and further suggested differences between states for this crash mode. While a vast majority of Washington SVLB crashes occurred on undivided roadways, Ohio SVLB crashes were split approximately evenly among divided and undivided roadways. Also, more than three-fourths of Ohio SVLB crashes were on roads with speed limits greater than 45 mph compared to roughly half in Washington. SVLB crashes in both states occurred on higher AADT roadways compared to all SV crashes, but in Washington MV crashes occurred on the highest AADT roadways. Note that the AADT represents all vehicle types, not just motorcycles. Similarities did exist, however, in several roadway and crash charac- teristics. A vast majority of the SVLB crashes available in this study occurred at non-intersection locations in dry roadway conditions. More than two-thirds of these crashes occurred on roadways

Roadway Characteristics Associated with Motorcycle Crashes into Longitudinal Barriers and the Influence on Rider Injury 103   with four lanes or less and roughly half occurred on arterial roadways with an approximately even split between rural and urban areas. In terms of rider characteristics, the gender distribution was consistent between states and across the data subsets. The difference in helmet usage between states was likely an indication of the differences in helmet laws in each state; Washington has a mandatory helmet law while Ohio does not. The mean age of riders involved in SVLB crashes was lower than those involved in either SV or MV crashes. Based on the results of the logistic regression models, rider characteristics were found to be the most important in predicting injury severity. Not wearing a helmet, alcohol involvement, and older occupants significantly increased the risk of serious injury by factors of 2, 3, and 1.5, respectively. The only statistically significant roadway characteristic found was roadway configuration, with divided roadways increasing rider serious injury risk by a factor of nearly 2. In contrast to the Savolainen and Mannering (2007) findings for single-vehicle crashes, the current study finds wet pavement SVLB crashes less severe than those occurring on dry or other roadway conditions. There was evidence that the presence of a horizontal curve increased the risk of serious rider injury, although this risk was approximately the same for curve radii greater than or less than 820 ft. Curves with normalized radius less than 1 also demonstrated an increase in rider severe injury risk but this increase was not present for curves where the normalized radius was greater than or equal to 1. None of the horizontal curve results, however, were statisti- cally significant. AADT less than 10,000 vpd, shoulder widths between 2 and 10 ft, posted speed limits greater than 45 mph, and vertical grades less than 3% were found to mildly increase severe rider injury risk, although these effects were not statistically significant. 9.7 Conclusions This study provides an analysis of roadway and specific geometric characteristics associated with motorcycle-to-barrier crashes in two states based on a total of 1,511 crashes occurring in Washington and Ohio. Motorcycle impacts with barriers were found to be overrepresented on horizontal curves and on sections with grade in excess of 3% in comparison to all SV motor- cycle and all MV motorcycle crashes. Similar to previous studies, these crashes were found to be overrepresented on ramp sections. Based on the available curvature data, however, the sole recommendation for placing motorcycle-to-barrier crash countermeasures on curves with radius less than 820 ft may not be prudent in U.S. states as less than 40% of these crashes occur on these curves. Although there were a number of similarities in motorcycle-to-barrier roadway characteristics between the two analyzed states, large differences were found in areas, including roadway configuration (e.g., divided/undivided) and posted speed limit. Rider characteristics, such as helmet usage and alcohol involvement, were found to have a larger influence on injury severity in comparison to associated roadway characteristics. Whether or not the roadway was divided was found to be the roadway characteristic having the largest influence on rider injury. The developed models suggest that horizontal curves, vertical grades less than 3%, posted speed limits greater than 45 mph, and traffic volumes less than 10,000 vpd increase rider injury risk, although these results were not statistically significant.