develop the 17-88 ED, as no one existing data source provides all the encroachment-related information required to fulfill the objectives of the project.

The data sources used to populate the 17-88 ED are described briefly below:

• Second Strategic Highway Research Program (SHRP2) Naturalistic Driving Study (NDS) and Roadway Information Database (RID). The SHRP 2 NDS was a recent NDS led by the Transportation Research Board of the National Academies of Science (Dingus et al. 2015). In SHRP 2, over 3,300 private vehicles were equipped with cameras, radars, and other sensors to collect over 3 years of data, including a total of 6,650,519 trips and nearly 50 million miles of driving. The SHRP 2 NDS captured a substantial number of roadside encroachments, both with and without collisions.

  The SHRP 2 dataset includes a second supplemental database of the roadway characteristics that can be linked by event location to each SHRP 2 encroachment discussed above. This linkage allows the determination of the effects of roadway characteristics on the nature and frequency of roadside encroachments. Data elements in the RID include horizontal curvature (radius, length, and direction of curve, i.e., left or right), vertical grade, cross-slope/super-elevation, lane configuration (number, width, and type, e.g., turn or acceleration lanes), shoulder type, presence of curb, presence of longitudinal barriers (e.g., guardrails), and presence of rumble strips.

  The SHRP 2 NDS/RID data can be used to primarily characterize passenger vehicle encroachments that do not result in any collision, unreported or police reported. To a lesser extent, the SHRP 2 NDS/RID data can be used to investigate passenger vehicle encroachments resulting in a collision across the range of severities, from minor unreported crashes to more serious police-reported collisions following an encroachment.

  NCHRP 17-43 Road Departure Database. The NCHRP 17-43 database is comprised of 1,581 police-reported road departures extracted from the National Automotive Sampling System/Crashworthiness Data System (NASS/CDS) for years 2011 to 2015. NCHRP 17-43 contains full reconstructions and trajectories for these cases through analysis of CDS crash scene diagrams and scene photos recorded by the NASS/CDS crash investigator (Riexinger and Gabler 2020). The NCHRP 17-43 dataset can be used to determine characteristics of encroachments resulting in a police-reported crash for passenger vehicles.

• State Crash and Maintenance Data. Police-reported crash data, roadside inventory files, and roadside hardware maintenance files for three states, Iowa, Washington, and Tennessee, were examined to provide encroachments to further populate the 17-88 ED. The crash data provides the frequency and severity of encroachments that resulted in a collision or rollover for all vehicle types. When linked with the inventory and crash data, the maintenance data can provide the frequency of crashes that required repairs but were not reported to police. Unlike the other datasets, the availability of roadway characteristic information across the network of state-maintained roadways allowed selection of a sample of roadways in each of the three states that ensured a variety of roadway conditions.

  The state data can be used to determine unreported and police-reported encroachment frequency and characteristics across all vehicle types.

• Motorcycle Crash Causation Study (MCCS) and NCHRP Project 22-26 In-Depth Motorcycle-Barrier Crashes. The FHWA MCCS is the most recent comprehensive data collection effort investigating the causes of U.S. motorcycle crashes (Nazemetz et al. 2019). The MCCS database is comprised of 351 on-scene crash investigations and 702 control cases collected from case years 2011 to 2016 in Orange County, CA. For crash cases, investigations were conducted by experienced motorcycle crash data collection experts and included interviews of crash-involved parties and collection of crash, environmental, and injury data. Control cases were comprised only of characteristics of non-crash involved motorcycle riders and passengers based on interviews of motorcyclists passing the crash location at a later date but under similar conditions. As the MCCS includes all types of motorcycle crashes, only the MCCS cases involving encroachment are considered for inclusion in the 17-88 ED.

  NCHRP Project 22-26 was tasked with identifying factors that contribute to serious and fatal injury in motorcycle collisions with traffic barriers (Gabler et al. 2021). As part of the research effort, the research team collected in-depth crash data and detailed rider injury data for 21 motorcycle-barrier crashes where at least one of the riders was admitted to a Level 1 trauma center. All the crashes occurred between years 2010 to 2016, inclusive, and took place at the Wake Forest University catchment area in either North Carolina or Virginia. All NCHRP 22-26 in-depth crashes involved an impact with a barrier situated outside the traveled way, so each NCHRP 22-26 case can be considered for inclusion in the 17-88 ED.

  The MCCS and NCHRP 22-26 cases can be used to determine characteristics of encroachments resulting in a police-reported crash involving a motorcycle.

• Large Truck Crash Causation Study (LTCCS). The LTCCS provides detailed information for approximately 1,000 large truck crashes that occurred in the U.S. between 2001 and 2003 (FMCSA et al. 2006). The heavy vehicles included in LTCCS include both single-unit and multiple-unit (e.g., tractor trailer) type vehicles, all with gross vehicle weight in excess of 10,000 pounds. Similar to the NCHRP 17-43 database, the available scene diagrams and information collected by the LTCSS researchers provide the ability to determine the vehicle trajectory for each case. Similar to MCCS, the LTCCS includes all types of heavy vehicle crashes. Only the subset of LTCCS cases involving encroachment was considered for inclusion in the 17-88 ED.

  The LTCCS can be used to determine characteristics of encroachments resulting in a police-reported crash for heavy vehicles.

Each source database contains data from potentially different time periods. Table 1 presents the time frames of the source databases used to populate the 17-88 ED and shows the coverage of the vehicle type and encroachment severity coverage associated with each data source. As previously discussed, note that the state datasets are the only source data that allowed for a direct control of the included roadway types. The in-depth studies (NCHRP 17-43, MCCS, NCHRP 22-26, and LTCCS) selected cases based on crash characteristics irrespective of roadway and/or roadside characteristics. Similarly, the SHRP 2 data are limited to the roadways that the study participants traveled during the study period. While there will be variation in roadway and roadside characteristics for the in-depth database and SHRP 2 cases, the variation in these cases is dictated solely by the available cases. For the state datasets, however, a sample of state-maintained routes

within in each state was selected to ensure a breadth of roadway types and associated characteristics.

Methods to Populate the 17-88 ED

Although the 17-88 ED is comprised of data from other existing sources, the existing data sources do not necessarily contain the same data or even the same data elements in the same format. Also, there are instances where additional data were collected/determined from the available source data to populate a 17-88 ED data element. Generating a single database with uniform data elements and attributes required differing levels of transformation of the available source data. The process for transforming the available source data into data stored in the 17-88 ED fell into one of three categories: (1) direct from source, (2) derived from source, or (3) researcher determined. Each of these approaches are described in more detail in Table 2 with associated examples.

For each 17-88 ED data element, a table is provided to indicate the method and source variable(s) used to translate the available source data into the data ultimately coded into the 17-88 ED. The notes column of the table provides more specific detail on how the source data were transformed, such as which data element attributes were combined to match the 17-88 ED format, or limitations on data availability.

Table 2. Summary of Data Collection Methods Used to Populate the 17-88 ED

17-88 ED Inclusion Criteria by Data Source

Each source database was examined to select suitable cases for inclusion in the 17-88 ED. Inclusion criteria for cases from each data source are briefly described in the sections below.

SHRP 2 NDS and RID

The research team examined 863,280 trips in SHRP 2 from all advanced vehicle trips. In total, 215,993 hours of driving were analyzed traversing 6.49 million miles. Advanced vehicle trips were defined as those with at least one valid data point of steering wheel position and turn signal status. Advanced vehicles are newer and provide more consistent reporting of speedometer speed used in this study. The examined SHRP 2 data include information about encroachments from both normal trips and trips with events such as crashes and near-crashes.

The longitudinal position of the vehicle during each event was determined by integrating the speed of the vehicle over time, starting at a position of zero feet at the beginning of each trip. The SHRP 2 study recorded both the speed as displayed on the speedometer and the speed as measured by the GPS receiver. The speed measured by the vehicle was used, as it was more accurate and was reported at a higher frequency (10 Hz) than the GPS data (1 Hz).

We then computed the minimum distance to lane boundary (DTLB) between the left and right sides of each vehicle using the distance from the centerline of the vehicle to the lane markings and the width of the vehicle. A diagram of DTLB is given in Figure 3. Vehicle width was determined from the Canadian Vehicle Specifications Database (CVSD) (CARSP 2019). The “Overall Width” of the vehicle was used. In 43% of trips, the width of the subject vehicle was directly known and used to compute DTLB. In the remaining cases, the SHRP 2 vehicle could not be identified in the CVSD, and the median width of all vehicle types in SHRP 2 was used (5 feet, 11 inches). For trips where the vehicle width was known, the computed DTLB value was more accurate.

coded in the dataset. The NCHRP 17-43 database codes additional information based on the scene diagram and photographs available for each NASS/CDS case. The data collection methods are described in a later section.

The NCHRP 17-43 database focuses on single-vehicle run-off-road crashes to support improvement of roadside crash safety. NASS/CDS cases were considered for inclusion in the NCHRP 17-43 database if only one vehicle was involved and the crash type was “drive off road.” Cases were excluded from the sample if the departure occurred after the driver successfully avoided another crash, if the driver lost control of the vehicle, if there was an impact before the departure, or if the driver was in an intersection. The remaining cases are single-vehicle, drift out of lane, run-off-road crashes. Each crash was manually inspected before extracting the relevant data. Most of the excluded cases at this stage either did not have a scaled scene diagram or were missing necessary trajectory points. The resulting dataset contains a total of 1,581 cases representing 510,154 drift out of lane, single-vehicle roadway departure crashes in the U.S., which were added to the NCHRP 17-43 database (Table 3).

Table 3. Case Selection Criteria for the NCHRP 17-43 Database

State Crash and Maintenance Data

Police-reported crash data, roadside inventory files, and roadside hardware maintenance files for three states, Iowa, Washington, and Tennessee, were collected and examined to identify potential unreported and police-reported encroachments to be included in the 17-88 ED. Representative roadways from each participating state were selected using the following general procedure:

1. Categorize state-maintained roadways in each state based on the Highway Safety Manual (HSM) framework/roadway subtypes shown in Table 4.
2. For each road subtype, evaluate the ranges of roadway characteristics for the longest routes for that specific roadway subtype. Primary roadway characteristics examined include traffic volume, posted speed limit, shoulder width, median type/width (if applicable), lane width, and rumble strip presence.
3. Select representative routes for each roadway type category considering the ranges of roadway characteristics determined in Step 2.

crash data for the previous 10 years along with roadway and traffic data for Iowa are available through the Iowa DOT open GIS portal. The Iowa crash data include all crashes across Iowa and are not limited to crashes occurring on state-maintained Iowa roadways. The research team has obtained 6 years of crash data, from 2012 through 2017, to match the obtained maintenance information.

Available crash data were used to select the police-reported crashes on the representative Iowa routes to be included in the NCHRP 17-88 encroachment database. For the 2012 through 2017 period, there were a total of 315,646 crashes involving 544,213 vehicles across the entire state on all roadways. On the 11 selected Iowa representative routes, there were a total of 54,163 crashes involving 91,778 vehicles. These 11 Iowa routes represent approximately 17% of the total crashes occurring statewide in Iowa. Considering only state-maintained roadways, there were 129,342 crashes involving 218,812 vehicles occurring from 2012 through 2017. The 11 Iowa routes selected represent approximately 42% of the crashes occurring on state-maintained routes in Iowa.

Prior to the identification of run-off-road crashes, the crashes were categorized as single-vehicle or multiple-vehicle crashes using the number of vehicles variable (vehicles) in the crash table. A single-vehicle crash was defined as vehicles = 1, while a multiple-vehicle crash was defined as vehicles > 1. Selection of run-off-road crashes were then completed for each group separately.

The Iowa crash data for each year on the selected routes were combined and then merged with the available vehicle data for each crash. A total of 22,757 vehicles were involved in single-vehicle crashes on the selected Iowa routes from 2012 through 2017. Unlike the Washington and Tennessee data, the Iowa data have a specific sequence of events (SOE) code pertaining to the presence of a phantom vehicle (i.e., a non-contact vehicle). A total of 215 of the single-vehicle crashes had a phantom vehicle coded in one of the SOE, which represents approximately 1% of the available single-vehicle crashes. Since the Iowa data do not code phantom vehicles separately in the vehicle file (as do Washington and Tennessee), there were no phantom vehicle records to exclude from the available single-vehicle crashes. Information on the presence of a phantom vehicle will be used, however, to determine if another vehicle influenced the “single vehicle” crash to populate the “AnotherVehicle” variable present in the NCHRP 17-88 database.

Each of the 22,757 vehicles involved in a single-vehicle crash were examined further to determine if inclusion in the NCHRP 17-88 encroachment database was warranted. The Iowa crash data do not have a single specific variable indicating whether a vehicle departed the roadway and also do not contain a data element indicating the crash location in relation to the roadway. As a result, only the SOE (seqevents1 - 4) variables were used in the selection process. The following inclusion criteria were used to select appropriate cases:

1. Run-off-road code present anywhere in the four possible SOE variables.
2. OR fixed object code present anywhere in the four SOE variables.

Examination of the available data revealed that “run-off-road” is not always included in the SOE; for example, a vehicle that leaves the roadway and impacts a guardrail could be coded with one SOE as “collision with guardrail” or with two SOE variables as “ran off the road” followed by “collision with guardrail.” Given this, either a “run-off-road” code present in the SOE variables or

a fixed object code present was permissible. Application of the three selection criteria resulted in 12,509 single-vehicle crashes identified as candidates to include in the NCHRP 17-88 database.

The fixed object-only crashes (meeting only criterion 2 above) were examined further to ensure these were crashes that should be included in the NCRHP 17-88 database. This was to ensure that fixed object impacts occurring while the vehicle remains on the roadway were not included, such as overheight trucks impacting a bridge/overpass underside. No fixed object cases were excluded. The 12,509 crashes represent approximately 55% of the available single-vehicle crashes. Of the 10,248 omitted single-vehicle crashes, animal strikes (7,637 cases) accounted for approximately 75% of the omitted cases. Vehicle rollover on the roadway (370 cases), other non-collision events (367), and non-motorist strikes (347) account for the majority of the remaining excluded cases. These four crash types account for approximately 85% of the single-vehicle cases excluded. Note that approximately 8% (822 cases) of the 10,248 omitted single-vehicle crashes had “not reported” or unknown SOE information.

For multiple-vehicle crashes (vehicles > 1), the case selection process was similar to the single-vehicle crashes and focused only on the vehicle-level SOE variables so that only crash-involved vehicles that departed the traveled way were included. As an example, consider a crash where vehicle 1 strikes vehicle 2. Vehicle 2 then remains on the roadway while vehicle 1 departs the roadway and impacts a guardrail. In this case, only vehicle 1 would be appropriate to include in the NCHRP 17-88 database.

The initial inclusion criterion for vehicles involved in a multiple-vehicle crash was that the vehicle had a run-off-road or fixed object coded in any of the four possible SOE variables. Using this criterion, there were 2,214 vehicles involved in a multiple-vehicle crash, which represents approximately 3% of the 69,021 vehicles involved in multiple-vehicle crashes on the selected Iowa routes from 2012 through 2017. All vehicles with run-off-road coded in one or more SOE variables were included by default. Vehicles with one or more fixed objects struck but no run-off-road coded in the SOE variables were examined further to exclude any impacts only with objects present only on the roadway; no cases were excluded. The result was 2,214 vehicles selected.

The multiple-vehicle crashes were examined in more detail to identify crossover crashes. As the Iowa SOE data have specific categories for crossing centerline and crossing median, these were used directly to identify crossover crashes.

Washington State Representative Routes and Data

A total of 12 routes were selected from Washington State for inclusion, as summarized in Table 7. This includes the total length of each route as well as a single year representation of traffic flow in units of MVMT. For each route, Table 7 shows the proportion of the total roadway length in each general HSM roadway type category.

general, the selected rural undivided and urban divided roadways have lower overall crash rates compared to all similarly classified roadways across the state.

The Washington State DOT maintains an electronic database of approximately 10 years of maintenance records for various roadside features that the agency is responsible for maintaining. The research team has obtained Washington State DOT maintenance records from years 2010 through 2020. Police-reported crash data matched with detailed roadway and traffic data for Washington State is available through the Highway Safety Information System (HSIS). Washington State HSIS data have been obtained from 2005 through 2018. Note that 2018 is currently the most recent HSIS year available. Police-reported crash and corresponding maintenance data are thus available for a 9-year period, 2010 through 2018. An examination of all the provided Washington State DOT maintenance data, however, revealed very few maintenance entries for years 2010 and 2011, with less than 20 total maintenance entries for all maintained devices for both years combined. For more recent years of maintenance data, there were between 500 and 2,000 entries for all maintained devices. As there were so few entries for 2010 and 2011, maintenance data from these years were excluded from further analysis. As a result, a 7-year period was selected for analysis, years 2012 through 2018.

Available HSIS crash data were used to select the police-reported crashes on the representative routes to be included in the NCHRP 17-88 encroachment database. For the 2012 through 2018 period, there were a total of 354,684 crashes involving 703,450 vehicles across the entire state on state-maintained roadways. On the 12 selected representative routes, there were a total of 192,187 crashes involving 388,779 vehicles. These 12 Washington State routes represent approximately half of the total crashes occurring statewide on roadways maintained by Washington State.

Prior to the identification of encroachment crashes, the crashes were categorized as single-vehicle or multiple-vehicle crash using the number of vehicles variable (numvehs) in the crash table. A single-vehicle crash was defined as numvehs = 1 while a multiple-vehicle crash was defined as numvehs > 1. Selection of run-off-road crashes were then completed for each group separately.

The HSIS crash data for each year on the selected routes were combined and then merged with the available vehicle data for each crash. A total of 65,903 vehicles were involved in single-vehicle crashes on the selected Washington routes from 2012 through 2018. Many of these single-vehicle crashes, however, involved a phantom vehicle, that is, a vehicle identified in the available vehicle data but not actually struck. In these cases, the crash-level data indicate a single-vehicle crash (i.e., numvehs = 1) but merging in the associated vehicle-level data for the crash results in more than one vehicle listed. While identified, these phantom vehicles do not have any SOE or other information present. Excluding these vehicles resulted in 44,801 crash-involved vehicles. Information on the presence of a phantom vehicle was used, however, to determine if another vehicle influenced the “single vehicle” crash to populate the “AnotherVehicle” variable present in the NCHRP 17-88 database.

Each of the 44,801 vehicles involved in a single-vehicle crash were examined further to determine if inclusion in the NCHRP 17-88 encroachment database was warranted. As the Washington HSIS data do not have a specific variable indicating whether a vehicle left the roadway, a combination of four variables was used in the selection process: Accident type (Acctype), impact location

(Impact), object struck (Object1, Object2), and SOE (Event1, Event2, Event3, Event4). The following inclusion criteria were used to select appropriate cases:

1. Acctype = 33, 60, 61, 62 (strikes appurtenance, ran into ditch, ran into water, ran over embankment).
2. OR Acctype = 50 (rollover) and location NOT on road.
3. OR Acctype = 34 (other object) and location NOT on road.
4. OR run-off-road code present anywhere in SOE. Note that the Washington HSIS data contains up to four “events” for each vehicle.

The “Acctype,” “Impact,” and “Object1/2” variables are coded for the entire crash. These three variables can be used to reliably identify cases of interest in this instance since these crashes involve only a single vehicle (i.e., the entire crash applies directly to the only involved vehicle). The “Impact” variable identifies the location of the crash relative to the roadway (e.g., specific lane, shoulder, median, or beyond shoulder). Only the shoulder/median/beyond shoulder locations were considered for inclusion based solely on this variable. Application of the four selection criteria resulted in 34,272 single-vehicle crashes identified as candidates to include in the 17-88 ED.

Examination of the available data revealed that “run-off-road” is not always included in the SOE; for example, a vehicle that leaves the roadway and impacts a guardrail could be coded with one SOE as “collision with fixed object” or with two SOE variables as “ran off the road” followed by “collision with fixed object.” There were 560 additional single-vehicle crashes with an object struck in the SOE but not captured by the four selection criteria listed above. All but 14 of these were included in the dataset, bringing the single vehicle impact total to 34,818 cases. The 14 excluded impacts were primarily overheight trucks impacting a bridge/overpass underside (11 cases). Another crash involved an impact with a reversible lane gate, and two others involved an impact with only debris in the road. The 34,818 crashes represent approximately 78% of the available single-vehicle crashes. Of the 9,983 omitted single-vehicle crashes, animal strikes (4,426 cases) and pedestrians struck (1,328 cases) together accounted for more than half of these. Non-collision fire (1,051 cases), vehicle rollover on the roadway (1,046 cases), other object strike (936 cases), and bicyclist strike (526 cases) account for the majority of the remaining excluded cases. These six crash types account for approximately 93% of the single-vehicle cases excluded.

For multiple-vehicle crashes (numvehs > 1), the case selection process was similar to the single-vehicle crashes but focused on the vehicle-level SOE variables so that only crash-involved vehicles that departed the traveled way were included. The initial inclusion criterion for vehicles involved in a multiple-vehicle crash was that the vehicle had a run-off-road or fixed object coded in any of the four possible SOE variables. Using this criterion, there were 10,478 vehicles involved in a multiple-vehicle crash, which represents approximately 3% of the 322,860 vehicles involved in multiple-vehicle crashes on the selected Washington routes from 2012 through 2018. All vehicles with run-off-road coded in one or more SOE variables were included by default. Vehicles with one or more fixed objects struck but no run-off-road coded in the SOE variables were examined further to exclude any impacts only with objects present only on the roadway. A total of 49 vehicles were excluded based on object struck: 29 vehicles struck the underside of a bridge, 12 struck only miscellaneous debris in road, five struck lane control gates, two struck fallen rock/trees in the roadway, and one struck construction material. A significant number of the fixed object-only

vehicles (1,283 total) had no object1 or object2 listed; these cases were also excluded since we cannot be sure the vehicle did in fact leave the roadway and impact a fixed object. The result was 9,146 vehicles selected.

The multiple-vehicle crashes were examined in more detail to identify cross median crashes. The HSIS data do not have a specific variable identifying cross median crashes. Vehicles that departed the roadway and collided with a vehicle in the opposing direction were identified using a combination of the crash-level vehicle direction variables (v1dircde and v2dircde should be in opposing directions) as well as a run-off-road coded in at least one of the four vehicle event sequence variables (event1, event2, event3 or event4). Only the vehicle that departed the roadway was included in the 17-88 ED.

Tennessee State Representative Routes and Data

A total of 11 routes were selected from Tennessee for inclusion, as summarized in Table 9. This includes the total length of each route as well as a single year representation of traffic flow in units of MVMT. For each route, Table 9 shows the percentage of the total roadway length in each general HSM roadway type category.

Table 10 provides a comparison of the crash characteristics for the NCHRP 17-88 selected Tennessee routes to all similarly classified routes in the entire state. For the selected Tennessee routes of a given route type/subtype, the average crash rate was determined by summing the crash rate multiplied by the corresponding MVMT for each selected route and then dividing by the total MVMT for the same routes. A similar procedure was followed to determine the weighted ratio of fixed object crashes to total crashes. For each route type, the rightmost columns in Table 10 show the crash rate and fixed object to total crash ratio for all of the similarly classified Tennessee routes available.

represent approximately 30% of the crashes occurring on state-maintained routes in Tennessee. Although this proportion was lower than the routes selected for Washington State (the selected Washington routes represented approximately half of the crashes across all state routes), Tennessee has approximately twice the number of unique state routes and twice the total state-maintained road mileage of Washington State.

Prior to the identification of run-off-road crashes, the crashes were categorized as single-vehicle or multiple-vehicle crashes using the number of vehicles variable (totalvehicles) in the crash table. A single-vehicle crash was defined as totalvehicles = 1 while a multiple-vehicle crash was defined as totalvehicles > 1. Selection of run-off-road crashes were then completed for each group separately.

The Tennessee crash data for each year on the selected routes were combined and then merged with the available vehicle data for each crash. A total of 36,862 vehicles were involved in single-vehicle crashes on the selected Tennessee routes from 2015 through 2019. Similar to the Washington State data, there were single-vehicle crashes present that involved a phantom vehicle (i.e., a vehicle identified in the available vehicle data but not actually struck). Although phantom vehicles were present in the Tennessee data, they were less frequent compared to the Washington data; approximately 2% of Tennessee vehicles were involved in single-vehicle crashes compared to nearly one third of vehicles in Washington crashes. The reason for this discrepancy is not known but could be linked to police reporting differences between the two states. In the phantom vehicle cases, the crash-level data indicate a single-vehicle crash (i.e., totalvehicles = 1), but merging in the associated vehicle-level data for the crash results in more than one vehicle listed. While identified, these phantom vehicles do not have any SOE or other information present. Excluding these vehicles resulted in 36,091 crash-involved vehicles. Information on the presence of a phantom vehicle was used, however, to determine if another vehicle influenced the “single vehicle” crash to populate the “AnotherVehicle” variable present in the NCHRP 17-88 database.

Each of the 36,091 vehicles involved in a single-vehicle crash were examined further to determine if inclusion in the NCHRP 17-88 encroachment database was warranted. Similar to Washington, the Tennessee crash data do not have a single specific variable indicating whether a vehicle departed the roadway. As a result, two variables were used in the selection process: roadway relation (relationtoroadwayfirst) and SOE (event_seq1-10). The following inclusion criteria were used to select appropriate cases:

1. Relationtoroadwayfirst = 2, 3, 4, 5, 6, 7, 8, 10 (shoulder, median barrier, roadside left, roadside right, off roadway/unknown, parking lane, gore, outside trafficway).
2. OR run-off-road code present anywhere in the 10 possible SOE variables.
3. OR fixed object code present anywhere in the ten SOE variables.

The “Relationtoroadwayfirst” variable is coded for the entire crash and indicates the location of the crash relative to the primary roadway. This variable can be used to reliably identify cases of interest in this instance since these crashes involve only a single vehicle, that is, the entire crash applies directly to the only involved vehicle. Similar to Washington State, “run-off-road” is not always included in the SOE; for instance, a vehicle that leaves the roadway and impacts a guardrail could be coded with one SOE as “collision with guardrail” or with two SOE variables as “ran off the road” followed by “collision with guardrail.” Given this, either a “run-off-road” code present

in the SOE variables or a fixed object code present was permissible. Note that an accident type variable was also used in the Washington data to help identify crashes where a vehicle departed the roadway. Although Tennessee crash data do contain crash type variable at the crash level, the Tennessee variable is much less detailed than the corresponding Washington variable. The Tennessee crash type variable categorizes all single-vehicle crashes into a single category (i.e., no collision with vehicle). As the variable offered little additional information that could be used to aid in identifying suitable cases, it was not used in the selection process. Application of the three selection criteria resulted in 26,036 single-vehicle crashes identified as candidates to include in the 17-88 ED.

The fixed object-only crashes (i.e., meeting only criterion 3 above) were examined further to ensure these were crashes that should be included in the 17-88 ED. All but 10 of these were included in the dataset, bringing the single vehicle impact total to 26,026 cases. The 10 excluded impacts were overheight trucks impacting a bridge/overpass underside.

The 26,026 crashes represent approximately 72% of the available single-vehicle crashes. Of the 10,065 omitted single-vehicle crashes, animal strikes (5,171 cases) and other non-fixed object struck (1,794 cases) together accounted for approximately 70% of the omitted cases. Pedestrian strike (528), vehicle rollover on the roadway (406 cases), thrown/falling object (396 cases), and bicyclist strike (111 cases) account for the majority of the remaining excluded cases. These six crash types account for approximately 84% of the single-vehicle cases excluded. Note that approximately 8.6% (864 cases) of the 10,065 omitted single-vehicle crashes had missing or unknown SOE information.

For multiple-vehicle crashes (totalvehicles > 1), the case selection process was similar to the single-vehicle crashes but focused on the vehicle-level SOE variables so that only crash-involved vehicles that departed the traveled way were included.

The initial inclusion criterion for vehicles involved in a multiple-vehicle crash was that the vehicle had a run-off-road (including cross median or cross centerline) or fixed object coded in any of the 10 possible SOE variables. Using this criterion, there were 18,305 vehicles involved in a multiple-vehicle crash, which represents approximately 6% of the 300,568 vehicles involved in multiple-vehicle crashes on the selected Tennessee routes from 2015 through 2019.

For Washington, all vehicles with run-off-road coded in one or more SOE variables were included by default. Vehicles with one or more fixed objects struck but no run-off-road coded in the SOE variables were examined further to exclude any impacts only with objects present only on the roadway. In the process of applying the same process to the Tennessee data, the research team discovered that the Tennessee SOE is identical for each vehicle in multiple-vehicle crashes, that is, it represents a crash-level variable rather than the vehicle-level data indicated in the available Tennessee crash data documentation.

Motorcycle Crash Causation Data

Only on-scene crash cases from MCCS were considered for inclusion in the 17-88 ED. The control case data include only information on riders passing the crash location and not involved in an encroachment event (i.e., no trajectory data is available). For the on-scene crash cases, the other vehicle count variable in the Crash Form table was first used to identify single-vehicle crashes.

Then, the Crash Configuration variable in the Crash Form table was used to identify run-off-road crashes. Cases were included if the crash configuration fell in one of the two following categories: 1) motorcycle running off roadway, no other vehicle involvement or 2) motorcycle running off roadway in collision avoidance with other vehicle.

NCHRP 22-26 Project In-Depth Motorcycle-Barrier Crash Cases

NCHRP Project 22-26 was tasked with investigating injury mechanisms in motorcycle-to-barrier crashes. As part of the research effort, the research team collected in-depth crash data and detailed rider injury data for 21 motorcycle-barrier crashes where at least one of the riders was admitted to a Level 1 trauma center. All the crashes occurred between years 2010 to 2016, inclusive, and took place at the Wake Forest University catchment area in either North Carolina or Virginia. The case summary for each NCHRP 22-26 case was manually inspected to determine control loss status and which side of the road the motorcycle departed. Of the 21 cases, 12 were tracking crashes and nine were control loss crashes (Table 11).

Table 11. NCHRP 22-26 Inclusion Criteria and Final Target Population

LTCCS

Only single-vehicle road departure crashes are eligible for inclusion in the 17-88 ED. Road departure crashes in the LTCCS dataset were identified based on the CrashCode variable, which is equivalent to the ACCTYPE variable in NASS/CDS. The selected cases could have departed the road to the left or right drift-out-of-lane, or control loss (CrashCodes 1-10). There were 163 single-vehicle road departure crashes. Each crash was individually inspected to ensure that the encroachment was marked on a scaled scene diagram. The encroachment analysis relies on the quality of the CDS crash scene diagram. There were nine diagrams that were not drawn to scale, and the encroachments were not extracted. There were 12 cases excluded because the truck did not depart the road. There were 10 scene diagrams that were missing critical features such as impacted objects or trip locations. There were two crashes excluded because the scene diagrams incorrectly included features such as a second tractor. There were six cases removed because the truck tipped over on the road while navigating an intersection. After applying the case selection criteria, a total of 124 large-truck crashes had completed trajectories. These trajectories include crashes for single-unit medium/heavy trucks and large tractor-trailers.

Remarks:

This variable specifies the year, month, and day that the event occurred. All the in-depth crash sources (NCHRP 17-43, MCCS, NCHRP 22-26, and LTCCS), the police-reported crashes from each state, and the SHRP 2 data have event date information corresponding to the actual event based on the data available from the corresponding source. For unreported crashes from the state data sources, however, the date information corresponds to the roadside hardware device repair date. The matching of police-reported crash and available maintenance data allows an identification of unreported crashes, but the exact occurrence date of these unreported events is not known. Coding the date information as “unknown” is reserved for cases where the date information is missing from the source data (e.g., all the calendar day values for NCHRP 17-43 cases are coded as “unknown” since the calendar day of the event is not available in NASS/CDS and, thus, not available in NCHRP 17-43). Similarly, the MCCS dataset does not record the specific calendar day of the event.

Remarks:

This variable specifies the day of the week the event occurred. Similar to the year/month/day information, the day information for unreported crashes from the state data sources corresponds to the roadside hardware device repair date.

Remarks:

This variable specifies what time of day the event occurred. For unreported crashes, hour and minute are coded as unknown.

Remarks:

This variable describes the attentiveness of the driver at the time of the event. The variable definitions above are adapted from the NASS. If more than one condition is presented in the database of interest, the first condition listed will be coded. For the state datasets, presence of any distraction-related factors triggers the coding of “distracted.”

Remarks:

This variable describes any evasive maneuver that the driver took during the event.

Remarks:

This variable describes the injury severity of the driver according to the KABCO Injury Scale.

Remarks:

This variable describes the weather conditions during the event.

Remarks:

This variable describes the roadway lighting during the event.

Remarks:

This variable captures the type of vehicle involved in the case.

Remarks:

This variable specifies if the event occurred in a construction work zone.

Remarks:

This variable describes the condition of the roadway surface. This may differ from the condition of the roadside; for example, the road may have normal conditions, but the roadside is covered in snow.