Data Needs for the MASH

AASHTO’s MASH provides guidelines for crash testing of both permanent and temporary roadside safety features and recommends evaluation criteria to assess test results (AASHTO 2016). Specifically, MASH prescribes crash test conditions to evaluate roadside safety hardware. The test conditions were chosen to be representative of a practical worst-case crash, typically defined to be the 85^th percentile of real-world crash severities. Table 13 presents a list of the data elements that would be needed for a potential update of these crash test conditions.

Note that LOS varies temporally just as traffic varies with time (e.g., LOS could be closer to F during rush hour and then closer to A throughout the late-night hours on the same roadway). AADT has been used as a proxy for traffic, but some researchers have suggested that the amount of traffic presented at the time of encroachment may have a strong influence on encroachment probability and could not readily be modeled (Roque and Cardosa 2016).

Combined Set of Data Needs

Table 15 presents a combined set of data needs assembled from the preceding data elements either (a) collected by Hutchinson and Kennedy (Hutchinson 1962; Hutchinson and Kennedy 1966) and Cooper (1980, 1981), (b) necessary for potential updates to the RDG and MASH, or (c) necessary to conduct project-specific analyses on the nature of roadside encroachments. The data element names and descriptions have been adjusted for uniformity and grouped into five broad categories: case identifiers, roadway data elements, roadside data elements, trajectory elements, and driver/vehicle elements.

3.2.3 Data Sources

Data to populate the NCHRP 17-88 encroachment database were obtained from U.S. national and state crash databases coupled with data from NDS and specialized databases of roadway characteristics. Data sources for the study include:

• NCHRP 17-43 Road Departure Database. The NCHRP 17-43 database is comprised of 1,581 police-reported roadside departures extracted from the NASS/CDS case years 2011 to 2015. NCHRP 17-43 has developed full reconstructions and trajectories for these cases through analysis of crash scene diagrams and scene photos recorded by the NASS/CDS crash investigator.
• NASS/CDS Database. NASS/CDS is a national stratified sample of 4,000 to 5,000 crashes that were investigated each year until 2015 by NHTSA at up to 27 locations throughout the United States. For a crash to be included in NASS/CDS, at least one of the vehicles in the

• crash had to be towed from the scene. NASS/CDS only investigates crashes involving passenger vehicles and light trucks. NASS/CDS provides extensive investigation of vehicle crashworthiness, occupant injury, and occupant restraint performance, but it contains only limited data on the unique characteristics of ROR crashes (i.e., barrier design, slope, and roadway geometrics). NCHRP Project 17-43 collected these specialized roadside characteristics through a supplemental data collection program.
• State Data. Washington, Iowa, and Tennessee have agreed to share state crash data files, roadside inventory files, and roadside hardware maintenance files in support of the current research efforts to develop a next-generation encroachment database. Our objective was to link the crash, inventory, and maintenance data through keys such as route number-milepost location or GPS location identifiers. The crash data alone can provide the frequency and severity of encroachments that resulted in a collision or rollover. 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.
• Motorcycle Crash Causation Study (MCCS). The FHWA MCCS database was used as the basis for motorcycle encroachments and motorcycle road departure trajectories (Nazemetz et al. 2016). This database is comprised of 351 injury crashes, of which 47 cases are single-vehicle crashes.
• LTCCS. The LTCCS was used as the basis for large truck encroachments and large truck road departure trajectories (FMCSA et al. 2006). The LTCCS provides detailed information for approximately 1,000 large truck crashes that occurred in the U.S. between 2001 and 2003. An initial scan of the available data indicates 124 total roadway-departure crash events.
• NCHRP 22-26 Motorcycle Departure Database. The NCHRP 22-26 database is comprised of 22 motorcycle-barrier crashes in North Carolina. To be eligible for NCHRP 22-26 data collection by Virginia Tech researchers, the case needed to involve a motorcycle impact with a roadside barrier where the motorcyclist was transported to a Level 1 trauma center. NCHRP 17-88 has developed full reconstructions and trajectories for these cases through analysis of crash scene diagrams and scene photos recorded by the NCHRP 22-26 crash investigators.
• SHRP 2 NDS. The SHRP 2 was an NDS led by TRB of the National Academies of Science. From 2010 to 2013, SHRP 2 equipped over 3,300 private vehicles with cameras, radars, and other sensors to collect over 3 years of data that included 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 current study analyzed a subset of these data as a source of non-collision encroachments and collisions across the range of severities from minor unreported collisions (e.g., sideswipes of guardrails) to serious crashes into roadside hazards.
• SHRP 2 Roadway Information Database (SHRP 2 RID). The SHRP 2 data provided a second supplemental database of the roadway characteristics that can be linked by event location to each SHRP 2 road departure discussed above. This linkage allowed the project team to determine 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, such as turn or acceleration lanes), shoulder type, presence of curb, presence of longitudinal barriers (e.g., guardrails), and presence of rumble strips.

Table 16 presents the time frames of the databases used to populate the NCHRP Project 17-88 encroachment database:

Table 16. Data Sources for the NCHRP 17-88 Encroachment Database

3.2.4 Summary of Data Availability

State Data

A total of three state agency partners had been identified and agreed to participate in NCHRP Project 17-88: (1) Washington State Department of Transportation (DOT; Maintenance Operations Office), (2) Iowa DOT (Office of Highway Support, Office of Traffic and Safety, and Office of Analytics), and (3) Tennessee DOT (Safety Data Section and Asset Management Office).

The primary purpose of the state data was to identify police-reported ROR crashes and unreported ROR crashes for a variety of roadway geometric and traffic conditions. Table 17 provides a high-level summary of the data needed for the study along with the potential data sources. In addition to data to determine reported and unreported ROR crashes, data were needed to characterize the associated roadway and traffic conditions, distinguish between vehicle types, and determine occupant injury severity. The state data available for the project can be generally classified as crash, inventory, or maintenance data. The following sections provide more details on the required police-reported crash data, roadway/roadside inventory data, and maintenance data.

Table 17. High-level Summary of Data Needed and Potential Data Sources

Crash Data

Table 18 lists the required police-reported crash data, which can be classified into general crash, vehicle, or occupant information. Brief notes accompany each data element required. Crash time

and location data elements were critical because they allow a link to any available roadway and roadside inventory data as well as available maintenance data. Crash sequence of events information was required to separate ROR crashes from other crash modes present in the data. Vehicle type information was required to distinguish passenger vehicles (cars, pickups, and SUVs) from other specific vehicle types of interest (motorcycles, heavy vehicles, and buses). Occupant injury data were required to support analysis of the encroachment characteristics in tandem with the occupant injury outcomes.

Table 18. Summary of Police-reported Crash Data Needed

In general, the available crash data for all three states were adequate for use in the current study. Iowa and Tennessee had all the required information, and Washington State was missing only information to indicate crash configuration (e.g., front, side). While crash configuration cannot easily be collected via alternate means, the missing data did not affect the determination of unreported crashes, only the ability to analyze the collected data by crash configuration type. Since Iowa and Tennessee had these data available, we considered this deficiency minor.

Inventory Data

Table 19 lists the required roadway and roadside feature inventory data. The roadway data were primarily intended to support an analysis of the nature and frequency of encroachments as a function of relevant roadway (e.g., horizontal alignment, vertical alignment, speed limit) and roadside characteristics (e.g., shoulder type/width, curb presence). Some of the required roadway data were also present in the available crash data (e.g., posted speed limit and qualitative roadway alignment descriptors), which helped verify appropriate linkage between available crash and roadway data. Roadside inventory data should ideally provide device location that includes lateral offset from the travel lane and the device type. At a minimum, the roadside inventory data should have included traffic barriers, end terminals, crash cushions, and sign/light supports. Data on exact device type and crash test level (if applicable) were sought and obtained when available; however, these data were not necessarily a requirement for estimating unreported crashes, provided the

maintenance records linked in some form to the available inventory data. The goal was to estimate the number of unreported crashes and focus on roadside devices that had associated maintenance data available. Combining these two data sources provided a means of estimating unreported crashes. There were other objects present on the roadside (e.g., trees, mailboxes), but without associated maintenance data for that object, it was not possible to estimate the associated number of unreported crashes with that object.

Table 19. Summary of Roadway and Roadside Inventory Data Needed

In general, the available roadway and roadside inventory data from the three partner agency states were adequate to support identification of unreported crashes with roadside devices across various roadway characteristics, especially when considering the data available from all states in combination. As an example, Iowa lacks detailed horizontal and vertical alignment data, but Washington State and Tennessee had detailed horizontal and vertical alignment data present. Table 20 summarizes the missing inventory data from the three partner agency states and an associated priority level for obtaining each remaining element.

Table 20. Summary of Missing Inventory Data Elements from Washington and Iowa

Roadway segment LOS was estimated using simplified procedures developed by Margiotta and Washburn (2017). The procedures are based on the Highway Capacity Manual (HCM; TRB 2016) procedures but employ a number of assumptions, primarily the K-factor value, to simplify the LOS computation for various roadway segment types.

To collect device offset, the research team developed a method to extract roadside hardware lateral offset distance using Google Street View images. Using available roadside hardware inventory data with known GPS location information, a developed MATLAB program fetches the associated Google Street View image. The program allows the user to change the orientation of the view to ensure the lane width and roadside hardware device are both visible. Using the image, the user indicates the endpoints of a known distance, typically lane width available from the roadway inventory data, directly on the image. The user then indicates the endpoints of a distance to be measured, such as the lateral offset of the roadside hardware device. The developed program uses the known distance to compute the lateral offset to the roadside hardware device shown. The method has been used to estimate longitudinal barrier lateral offset along selected routes in Washington State, specifically the lateral offset measurements for the available w-beam guardrail, cable barrier, and impact attenuator inventory information. Based on approximately 15 local sites, the lateral offset measured using this method was within 12% of the actual lateral offset distance.

The team attempted a similar approach to visually categorize roadside slope based on the original RDG slope cutoffs for traversable (1V:4H or flatter), non-recoverable (between 1V:4H and 1V:3H), and critical (steeper than 1V:3H). Using the same local sites mentioned above, the results

derived from this method were checked against the actual slope measured in the field. While the method was reasonable for very flat slopes and very steep slopes, the intermediate classification was inconsistent. Given the limited resources of the project and the extensive simulation for roadside slopes and roadside ditches conducted as part of NCHRP Project 17-55 (Sheikh et al. 2019) and NCHRP Project 16-05 (Sheikh et al. 2020), the team made no further efforts to collect roadside slope as part of this project.

Maintenance Data

Table 21 lists the required state maintenance data. For each roadside hardware type, the location, date, and activity description (e.g., to distinguish damaged hardware repair from other maintenance activities) was needed to link the maintenance data with associated crash and inventory records. Depending on the state agency procedures, maintenance and inventory data could have been combined. If available in a given state, existing links between documented roadside hardware repairs and police-reported crashes were used.

Table 21. Summary of Maintenance Data Needed

In general, the available maintenance data for Washington State were the most complete for use in the current study, including identified matches to available police-reported crashes. Ultimately, the maintenance data from Iowa did not provide sufficient detail to identify unreported crashes with reasonable confidence. While the Iowa maintenance records provided some indication of equipment and materials needed for the repair, there was no direct linkage to a specific roadside device, and often repairs to multiple devices in a roadway segment were grouped into a single maintenance record. For Tennessee, the data were sufficient only with regard to sign repairs.

In-Depth Crash Data

Four different in-depth crash datasets were included in the NCHRP 17-88 database: NCHRP 17-43, LTCCS, MCCS, and NCRHP 22-26. The NCHRP 17-43 database cases were extracted from NASS/CDS, a dataset collected by NHTSA. LTCCS was also collected by NHTSA, while MCCS was collected by FHWA. The NCHRP 22-26 database was collected by Virginia Tech.

• right side of the road, over-corrected, and then departed on the left side of the road, the roadside encroachment conditions table would include two entries for this incident, that is, one record for the right-side departure and one record for the left-side departure.
• Event Table. The event table presented in Table 25 will provide an array of coordinates and angles of recorded event(s) from the point of departure to the object or objects struck. Cases will describe the object(s) struck and the coordinates of impact events, along with general and specific horizontal damage that occurred during the collision. For example, if a vehicle departed the road, impacted with two trees, and then struck a building coming to final rest, the vehicle’s path off-road would be recorded as four events: (1) impacted the first tree, (2) impacted the second tree, (3) struck a building, and (4) the vehicle coming to a final rest. For this study, a vehicle only departing, partially departing, or re-entering does not count as an event, as we are only analyzing crash event(s) that occur during the encroachment.
• Encroachment Trajectory Table. The encroachment trajectory table presented Table 26 in will provide an array of coordinates from the point of departure either to the object struck or to the point of roadway re-entry. In cases in which multiple objects are struck after departure, a separate trajectory record will be included between each object struck. For example, if a vehicle departed the road, sideswiped a tree, and then struck a utility pole before coming to final rest, the vehicle’s path off-road would be recorded as a sequence of three trajectories: (1) from the point of departure to the tree, (2) from the tree to the utility pole, and (3) from the utility pole to the point of final rest.

The key fields “Source,” “Case Identifier,” “Vehicle Number,” and “Component” are provided in each table to allow these five tables to be linked. In addition, the case table provides the “Data Source” element, which allows linkage to the source of the underlying data (i.e., NASS/CDS, LTCCS for large trucks, MCCS for motorcycles, state data by state name, and SHRP 2). This mechanism allows the analysis to link in, for example, the detailed description of each occupant (age, gender, and belt use) and the resulting injury outcomes by body region in NASS/CDS, LTCCS, and MCCS.

NASS/CDS, LTCCS, and MCCS provided many of the data elements needed for this project. However, these databases were primarily focused on crashworthiness and injury outcomes. Specialized roadside-oriented data elements (e.g., roadside geometry and run-off road trajectories) were not collected. These data were extracted, however, by examining and analyzing the crash scene diagrams and crash site photos. This process was fairly time consuming, but the research team has used this approach successfully to construct the NCHRP 17-43 database. Our plan was to use this same methodology to mine roadside elements for large trucks from the LTCCS and for motorcycles from the MCCS. In the discussion that follows, encroachment data extracted from LTCCS scene diagrams and photos will be denoted by “LTCCS-ED.” Encroachment data extracted from MCCS scene diagrams and photos will be denoted by “MCCS-ED.”

The state data shown in the tables that follow were obtained by merging the state accident data, state maintenance data, and state inventory data. The state data were primarily used to determine the frequency of unreported roadside departures. The state data did not include trajectories or reconstructions needed for updates to the RDG or MASH. The research team considered using police accident reports, which frequently have sketches of the crash. However, only in rare cases (typically fatal crashes) are these sketches to scale and suitable for use in an update to the RDG or MASH.

(Bonneson et al. 2012); the associated freeway types have been included in Table 30. Although not explicitly present in Table 30, there was some degree of access control considered in this classification scheme, as the freeways and associated ramps typically have full access control while the other roadway classes generally have either partial or no access control.

Table 30. Roadway Types Considered in the HSM Predictive Methodology

The current HSM also does not include specific procedures for predicting ROR crashes. Procedures specific to ROR crashes were developed under NCHRP Project 17-54 (Carrigan and Ray 2018). Similar to the original HSM procedures, the proposed draft procedures involve ROR SPFs specific to different roadway types. The currently proposed NCHRP Project 17-54 ROR SPFs follow the same overall scheme depicted in Table 30 with four different ROR SPFs (i.e., rural undivided, rural divided, urban undivided, and urban divided).

The roadway characteristics relevant to the RDG, RSAP, and HSM (i.e., Table 28, Table 29, and Table 30) were then compared to the roadway, traffic, and roadside related data elements present in the proposed NCHRP 17-88 data element catalog. This comparison allowed the identification of any data elements present in the proposed NCHRP 17-88 data element catalog but not listed (or implied) in Table 28, Table 29, or Table 30. The identified data elements could also be considered in the roadway selection process and are listed in Table 31.

Table 31. Potential Additional Roadway Characteristics to Consider in the Roadway Segment Selection Process

3.3.3 Evaluation of Potential Route Selection Schemes

The research team considered several route selection schemes for potential use in the current project. These schemes are listed below along with a brief description of each, followed by a discussion of possible associated advantages and disadvantages.

1. Random selection of segments – Randomly selected segments from all available segments within a particular state.
2. Random selection based on ROR crash characteristics – Randomly selected segments from a subset of all available segments. Determine the subset of segments for each state based on ROR crash frequency (e.g., a high number of ROR crashes).
3. Random selection within primary roadway categories – Categorize all the available roadway segments within a state using the HSM roadway classifiers shown in Table 30. Randomly selected some number of roadway segments from each of the 12 (excluding ramps) distinct roadway types listed.
4. Selection within primary roadway categories – Categorize all the available roadway segments within a state using the HSM roadway classifiers shown in Table 30. Selected some number of routes from each of the 12 distinct roadway types (excluding ramps) listed. Use all roadway segments for each selected route.

While random selection of roadway segments (Option 1) would produce a representative sample of roadway sections within each state, the method would not necessarily produce the requisite range of roadway, traffic, and roadside characteristics sought in the current project. Selection based on ROR crash characteristics (Option 2) would likely generate the highest number of encroachment data points to add to the NCHRP 17-88 encroachment database but would bias the included roadway segments to those that have a higher ROR crash experience. Random selection within the primary roadway categories (Option 3) would ensure representation across a variety of

roadway types but still may not necessarily produce the requisite range of roadway, traffic, and roadside characteristics sought.

A practical issue with any of the random selection of segment options (Options 1 – 3) relates to the acquisition of associated maintenance data. Each state maintains a database, or multiple databases, that must be queried, generally by route number, to generate the needed maintenance data. Any type of random selection would likely generate many different route numbers, and potentially some number of sections from all available routes, to be queried by the state DOT personnel (i.e., the research team does not have access to the entire dataset). In addition, roadway segments are generally very short; for instance, the approximately 300 miles of Washington State Route 2 is comprised of more than 1,800 roadway segments, each with homogeneous characteristics. Based on the initial matching of crash and maintenance data, longer segments are preferable, as the crash and maintenance location may differ slightly for matches; that is, the match could be missed if it is technically on a different segment and that different section was not selected for inclusion.

Selection of roadway segments within the primary roadway categories (Option 4) ensured representation across a variety of roadway types, minimized the number of routes to be queried for associated maintenance data, and allowed the selection of a combination of roadways that provide an adequate range of roadway, traffic, and roadside characteristics. As each of the three states will have a different spectrum of roadway, roadside, and traffic characteristics, this method would allow routes to be selected from each of the states to maximize the range of characteristics included in the NCHRP 17-88 database.

3.3.4 Use Selection Scheme to Select Roadway Segments

Using the Option 4 selection method described above, the research team selected the roadway segments as follows:

1. For each state, categorize all available roadway segments based on the HSM classification scheme presented in Table 30 (i.e., using land use and roadway configuration). Determine count of different route numbers and associated mileage of roadways that fall into each possible category (excluding ramps). A single route designation may span multiple classifications, so the mileage associated with each classification will provide additional context for the selection process.
2. Assess the ranges of secondary variables (i.e., all other widely available variables not used to determine HSM classification) for roadway sections in each category. Select route numbers for inclusion.

Table 33 illustrates how the composite list of variables (originally shown in Table 32) was incorporated into the route selection process. The primary variables were used in Step 1 to determine route classification. For each state, one route representing each of the possible 12 categories was selected. Ramps selected for inclusion were a census of the ramps for the associated controlled access freeways selected in each state. The secondary variables were used in Step 2 to ensure an adequate range of characteristics were captured by the routes ultimately selected for inclusion in the project. The first priority in the selection of routes in Step 2 was to span the needed posted speed and traffic volumes needed to update the relevant RDG table, shown in Table 28. Secondary priorities were to maximize the range of the other available secondary variables (i.e., other than posted speed and traffic volume), provide adequate coverage of encroachment modifier

ranges in RSAP (Table 29), and provide additional data in gap areas identified by the critical literature review, such as traffic volumes greater than 40,000 vehicles per day. Tertiary variables are those not widely available at the time of route selection and thus were not used directly to select routes from all three states.

Table 33. Roadway, Traffic, and Roadside Characteristics Used in the Roadway Selection Process

3.3.5 Selection Results

The described methodology resulted in the selection of the representative routes listed in Table 34. For each roadway subtype, the selected route is listed along with total length (classified as that specific roadway subtype) and MVMT for a 1-year period. Note that many available routes have portions classified as more than one roadway subtype. Routes marked with an asterisk (*) indicated the route had a non-zero length classified as the listed roadway subtype but was selected as a representative route for a different roadway subtype. As an example, Washington SR 2 was selected as the representative RU2L2W route but has portions classified as 2U/3T, 4U/5T, R4D, R4F, U4D, and U4F. As a result, SR 2 appears with an asterisk in all cases except the RU2L2W subtype. Also note that certain states had very limited or no routes available for certain roadway subtypes; for example, Washington State had no available U10F routes, so only routes from Tennessee and Iowa were selected for this subtype.

Additional details on the selection of representative routes for each roadway subtype can be found in Appendix B. For each roadway subtype, the Appendix contains a full report of the available

routes and associated characteristics from each of the three states and a rationale for the selection of the representative routes.

For each roadway subtype, Table 35 summarizes the ranges of operational and geometric roadway characteristics for the selected routes across all three states. Overall, there was a wide range of traffic volumes from less than 10,000 vehicles per day to over 200,000 vehicles per day. Similarly, there was a wide variation in posted speed limit, shoulder width, and median width.

3.3.6 Iowa State Representative Routes and Data

A total of 11 routes were selected from Iowa for inclusion, as summarized in Table 36. The table includes the total length of each route and a single year representation of traffic flow in units of MVMT. For each route, Table 36 shows the proportion of the total roadway length in each general HSM roadway type category.