Considering Safety in the Transportation Planning Process
Chapter 4: Data and Information for Considering Safety in the Transportation Planning Process
Many transportation planning activities require large quantities of data, such as land-use, demographic, and traffic data. Timely and accurate crash data, when combined with other system data such as traffic volumes, can help transportation planners incorporate safety into the transportation planning process. This combined data, referred to as a crash information system, can be analyzed to provide useful information for transportation planning.
This chapter contains two parts. Part one addresses the fundamentals of crash data and presents information on obtaining and collecting crash data and organizing it into crash information systems. Part two describes the analysis of crash data and presents information on the tools, methodology, and application of the results of crash data analysis.
Part One: Fundamentals of Crash Data
Most metropolitan and state transportation planners are not involved in collecting or maintaining crash data, although some planners use crash data or the end products of crash data analysis (for example, location reports). However, understanding the process of crash data collection and maintenance can increase planners' awareness of and appreciation for transportation safety.
The Crash Reporting Process
The starting point for crash data is the initial crash. The law enforcement agency in the jurisdiction where the crash occurs is usually called to the scene to investigate. Depending on the severity of the crash, the reporting officer may fill out a report detailing the particulars of the crash. Almost all states report crashes resulting in fatalities or injuries. In addition, property-damage-only crashes are also reported if they exceed a legally defined reporting threshold. In most states, the property damage threshold is between $500 and $1,000.
For crashes meeting the reporting threshold, the responding officer conducts an investigation which varies depending on the severity of the crash. On a standard form, the officer records information on the circumstances, including drivers and vehicles involved and environmental and roadway conditions. In addition, information on the location and any traffic control in effect at the time of the crash is recorded.
The reporting officer may also try to ascertain and record the cause or contributing factors of the crash such as failure to observe traffic control, vehicle defects, or impairment. However, crashes may have multiple causes, all of which may not be apparent to the reporting officer. Depending on the reporting requirements of the jurisdiction, the crash report may also contain statements of witnesses and involved parties.
The reporting officer also records any injuries, usually making this assessment visually. Jurisdictions use various types of scales to describe injuries. A commonly used injury scale is the KABCO scale. Using this scale, the reporting officer assesses the injuries by using one of the following five designations: fatally injured (K); incapacitating injury (A); injury, not incapacitating (B); possible injury (C); and property damage only (O). Another common injury scale is the Maximum Abbreviated Injury Scale (MAIS). Using this scale, the reporting officer rates the injuries from one to six. One is for a minor injury, while five is for a critical injury and six is for an immediate fatality.
For crashes resulting in fatalities and injuries or when malfeasance is suspected, additional investigation may be required. Usually, special crashes such as heavy vehicle, transit, or hazardous material crashes also require additional information.
Crash ReportingThe organizational arrangements and crash-reporting thresholds are different for each state.
|
Crash Databases
Federal law requires each state to maintain a crash database for monitoring the safety of the transportation system. Once the crash report is filed, most enforcement agencies send a copy of the report to the state agency responsible for compiling the state's crash database, which varies by state but may be the department of motor vehicles (DMV), state police, or DOT. Most enforcement agencies also retain a copy of the crash report and maintain their own file or crash database. For example, the Honolulu Police Department maintains its own crash database in addition to sending the crash reports to the Hawaii DOT.
When the central state office receives crash reports from enforcement agencies, it compiles the information into a database. In some states, this may involve additional coding of the crash report. For example, in the State of Michigan, the central office instead of the reporting officer codes the crash type on the basis of the crash narrative and diagram.
A more precise location may also be assigned to the crash. For some states, this may mean assigning a map coordinate system to the crash. For other states, it may involve simply converting the transportation names into appropriate route numbers. The method of location assignment depends on the highway location reference system used by the state.
All statewide data is maintained on computer database systems, which usually requires the central office to perform some form of data entry, and data is often subjected to quality control and error checking procedures. These centralized databases are paid for, in part, by federal funding, and the federal government provides guidance for collecting and maintaining the data. For example, states are required to send data on all fatalities to the National Highway Traffic Safety Administration (NHTSA) for inclusion in the Fatality Analysis Reporting System (FARS), a national fatal crash database. Similarly, each state is required to report commercial heavy vehicle and bus crashes to the Federal Motor Carrier Safety Administration (FMCSA) through the SAFETYNET computer reporting system.
Sharing Data with Planning Partners
If data is needed from the centralized database, it is the responsibility of the state to determine if and how the data will be disseminated to planning partners. Many states are concerned about liability problems with crash data, and as such are reluctant to share the data with the planning partners. Some states accommodate for this by conducting crash data analysis for the planning partners so that they do not have to release the data. Other states, such as Connecticut, share the data freely with all planning partners. This is discussed further at the end of this chapter.
Potential Problems in Crash Data
Once crash data is received from the managing agency, users must consider the limitations of the dataset before conducting a crash analysis. The scope of data collected (reporting thresholds, jurisdictional boundaries, years of available data) should be understood. In addition, crash data is subject to errors from numerous sources of potential errors in the crash reporting process. These dataset errors and limitations in scope must be assessed in the context of crash-analysis needs. The following paragraphs describe some potential sources of errors and aspects of a crash dataset that must be considered.
Coding Errors
Even though police officers are trained in collecting crash data, they can make recording errors at the crash scene because of additional crash-site demands or time pressures to respond to other events. In addition, most crash reports are collected on paper forms and must be coded and digitizing for entry into the crash database. During the coding and digitizing stages, data entry specialists can make errors.
Underreporting
Not all transportation crashes within a jurisdiction are reported; that is, not all crashes are recorded on a crash report form and entered in the database of the jurisdiction. This is called underreporting. Drivers involved in single vehicle crashes are less likely to report the crashes. Elvik and Mysen compared 49 studies in 13 countries and showed that the reporting of injuries in official accident statistics was incomplete at all levels of injury severity (1). The mean reporting level of crashes in all countries was found to be 95 percent for fatalities, 70 percent for serious injuries requiring hospitalization, 25 percent for injuries in which the patient was treated as an outpatient, and 10 percent for very slight injuries.
Crashes involving some transportation system users are less likely to be reported than others. Underreporting is especially a problem for motor vehicle crashes involving pedestrian and bicycle crashes. Elvik and Mysen found that crash reporting was lowest for crashes involving cyclists. This is especially a problem when the crashes occur away from intersections at locations such as at driveways and across sidewalks. These crashes may result in very severe injuries to the pedestrian or bicyclist, but possibly because there is often minimal damage to the motor vehicle, the crash will go unreported.
Incorrect Assessment of Variables
Other common reporting errors on motor vehicle crash forms occur in the reporting officer's assessment of variables; the officer may incorrectly assess a condition such as the weather or simply fill in the wrong value for the field. For example, in a study of Honolulu motor vehicle crashes, the weather conditions recorded in crash reports were compared with actual National Weather Service data on rainfall (2). Only about a third of all wet days as defined by the National Weather Service were recorded in the weather conditions of the crash reports. Although analysts could obtain the necessary information from the National Weather Service data instead, it indicates that other variables not as easily verified may be reported incorrectly.
Location Accuracy
The actual location of a crash is, itself, prone to numerous reporting errors. Police officers are trained to record the name of the street on which the crash occurred and the intersecting street or the nearest cross street that can be used for reference. However, in a study of Honolulu crashes, a variety of geographical errors were found that cut across different data elements (3, 4). These included differences in the way primary and reference streets were identified or in the direction of the crash (north, south, east, west) from the reference intersection. The errors involved using non-standard terminology, misspelling street names, abbreviating street types (for example, listing "5th" when the street could be "5th Street" or "5th Avenue"), using place names for locations, using local slang for locations, and miscoding names during digitizing (5). Differences were also found between the use of common street names and state route names with mileposts (mile markers); many police officers will use common street names rather than standardized state route names. A standardized state route name will allow crashes sites to be identified precisely, to a hundredth or even a thousandth of a mile (if the officer records that level of detail). On the other hand, if common street names are used, typically crashes have to be allocated to the nearest intersection.
In addition, locating crashes requires that street names in crash reports be matched against a base map with standardized street names. In a geographic information system (GIS), this process is called geocoding (or geo-referencing). Because necessary geographical information is missing from the motor vehicle crash reports, numerous errors are introduced during the geocoding process. For example, in the Honolulu study, crashes were identified by the nearest intersection because common street names (e.g. Chain Bridge Road) rather than standardized names (e.g. Route 123) were predominately used for the primary sites in the crash reports. The geocoding process involves matching as many crashes as possible including as much information as possible. Additional matching is obtained by ignoring (relaxing) certain information. Thus, when full street names (number, direction, street name, and street type) were required for both the primary and the reference streets, only 46.1 percent of all crashes were matched; however, the accuracy of these matches was 100 percent (5). When the street type (or suffix) was relaxed (that is, "avenue", "street", "drive" and so forth were ignored), an additional 37.5 percent of the crashes were matched, but the accuracy of the additional matches dropped to 96.5 percent. Further, when the direction of the street (North, South, East, West) was ignored, an additional 2.5 percent of the cases were matched but the accuracy of the additional matches dropped to 90.5 percent. Finally, when the street name and number were relaxed, an additional 7.8 percent of the cases were matched but the accuracy of the additional matches dropped to 52.5 percent.
Timeliness
The time from the actual crash until the crash report's full inclusion in the state-maintained crash database may range from a few months to a year or more. The actual time until inclusion varies by agency and depends on the agency's process. In addition, users of the data may receive a set of crash data only once a year from the maintaining agency. Meanwhile, conditions at some of the locations in the analysis may have changed since the crashes occurred. For example, in a study to determine countermeasures for run-off-the-road crashes in Maryland, the study team found that shoulder rumble strips had been installed at almost all the freeway locations after the crashes had occurred. Therefore, the study team concluded that it was inappropriate to consider other countermeasures until a further analysis of the crash data was performed for a period following the rumble strip installation.
Uniformity
Unfortunately, crash data lacks uniformity among the states and potentially within a state or urban area. Police departments may use different crash reporting forms or procedures. For planning agencies such as an MPO that spans multiple states, merging the various state databases may provide many challenges. These agencies may find more advantages in keeping the databases separate and conducting an individual analysis for each state.
Crash Information Systems: Linking Crash Data With Other Transportation Data
Crash data analysts often require additional information for more in-depth studies. For example, to calculate crash rates, analysts also need traffic volume data. When crash data is combined with other useful data for safety analysis, the result is a crash information system that can combine or link crash data with traffic volume, roadway inventory, and land-use data. This system allows crashes and locations to be analyzed in the context of the surrounding environment.
Different units and agencies collect traffic volume, roadway inventory, and land-use data. City, county, and state governments and MPOs collect traffic volumes. Road inventory items are collected and managed by multiple state and local agencies including local government and state DOTs, local government public works and planning departments, or a number of different state agencies. Similarly, land-use information may be collected from different sources such as local or state county tax assessment reports (for parcels), local zoning data, local building and demolition permits, local land-use plans, state data sources for areas under state jurisdiction, and even aerial and satellite photography (for land density and coverage maps, for example). The multiple data sources pose numerous consistency and currency problems. Assembling all this information for the planning area may be difficult but would add to the quality and depth of crash analysis. In Southern California, for example, SCAG is the MPO for the region; the planning area covers five counties with more than 170 separate jurisdictions. Because the data of any one jurisdiction represents only a piece of the larger picture, the data must be merged to provide a coherent multi-jurisdiction perspective.
In southeast Michigan, an integrated crash information system is maintained by SEMCOG. SEMCOG's system uses a GIS program to integrate data sources as diverse as road inventory characteristics, traffic counts, and land use. These data sources all provide information that can help to analyze particular crashes. For example, SEMCOG used the land use inventory data in GIS to conduct a spatial analysis on deer crash occurrences and the surrounding land use attributes (for example, forest, commercial, residential). Where possible, standardized data is used in the GIS program. There are some standardization issues, however. For example, traffic counts conducted by local governments have wide differences in the way they are implemented.
Linking data reported in different ways creates another problem. Traffic crash data is identified by specific locations, usually represented by points on GIS. Traffic volume data is usually measured over specific links of a highway network. On GIS, this data can be represented by line segments or, occasionally, by points (to measure, for example, the mid-point of a segment). Road inventory data (types of roads, lanes, bridges, rail lines) can be represented as lines or points. On the other hand, traffic analysis zones (TAZs), the basic analysis unit of travel-demand forecasting, are represented by zones or polygons on a map. Land-use data is frequently represented by zones but can also be represented by lines (block faces) or points (specific buildings). All data types that may be included in a crash information system can be measured in different ways and represented in different geometrical units. Added to this is the complexity involved in translating these data sources into the same geographic coordinate system (for example, state plane coordinates). Linking all this data requires a complicated set of programs and routines. In addition, other tools and programs must be linked to this data so that analysis can be performed.
Part Two: Crash Data Analysis, Tools, and Techniques
This section discusses crash data analysis in the transportation planning process. A brief overview describes how the analysis is conducted. Tools that are available to assist transportation planners in analyzing crash data are also identified, including GIS, which is highlighted, software, national databases, and useful reports.
Conducting Crash Data Analysis-A Brief Overview 1
Safety improvement needs are addressed in two stages: identification and detailed safety analysis (6). One of the predominant uses of crash data is for the first stage-identification of hazardous locations. After locations are identified, a detailed safety analysis is conducted to determine if the sites can be improved by transportation investment and, if so, what improvements are needed. These hazardous locations, also referred to as "hot spots," "black spots," or "priority investigation locations," then become candidates for transportation programming.
Crash data analysis can also be used to identify region-wide safety programs that transcend one location or facility. For example, a region may be experiencing pedestrian crashes at multiple transit stops. Because the crashes are occurring at more than one transit stop, the problem may never appear in a hazardous location analysis. A planner may use crash analysis to identify the problem and develop a program to address the issue.
Classifying Locations
A location may be defined as a roadway segment or an intersection. It also may be defined as a single spot such as a curve or a transit stop. For state-level analysis, a location may be described broadly as a whole corridor or roadway. Generally, locations are defined as either spots or sections. A spot is a single location where many crashes occur, such as an intersection, an access driveway to a commercial center, or a railroad-highway grade crossing. A section is a length of highway, usually homogeneous, and can be as short as a half mile to several miles. When an aspect of the roadway changes, a new section or spot begins. For example, a new section begins when a lane is dropped, the shoulder width changes, or the type of pavement changes. Roadway inventory databases are used to parcel roadways into sections and spots. If a roadway inventory database is unavailable, uniform section lengths can be used in the analysis. (Most high crash location software systems can apply a "floating" section length to identify segments of roadway with high crash frequencies.)
Most state DOTs have defined state-maintained highways with a standard linear referencing system using control and section numbers. Each state highway is identified by one or more control numbers that, in turn, are subdivided into section numbers. The section numbers are then subdivided into mile points, usually to a hundredth or a thousandth of a mile. Crashes can then be allocated to small segments along a highway or grouped into larger segments (for example, tenths of a mile). In cities and counties, on the other hand, most local roads are known by their street names. Because most crashes are not identified by addresses, but by major streets and reference streets, many crashes in urban areas must be allocated to intersections (that is, the intersection of the main street where the crash occurred and the nearest reference street). This coding convention becomes important when interpreting the results, especially when using the crash data to identify hazardous locations.
Identifying and Ranking Hazardous Locations
Identifying hazardous locations requires that many different aspects of the location's crash history be considered including crash frequency, severity, rate, nature, and environment. Because available transportation funds are limited, hazardous locations must be ranked in order to identify those sites most in need of safety remediation. Intersections and sections are ranked separately. The importance and use of the different types of ranking are explained in the following paragraphs.
Frequency: A basic identification of a hazardous location is by the absolute frequency of crashes occurring. For example, an analyst may compare the total crashes occurring at each intersection in an area over a 3-year period. The intersections would be ranked according to crash frequency. The analyst may decide that the top 10 intersections need safety remediation. However, this will only identify the intersections where the most crashes are occurring. Potentially, one of these intersections is experiencing very minor crashes. Perhaps the intersection is heavily congested and rear-end crashes commonly occur, but the crashes result only in property damage or minor injuries.
Fatal and Injury Crash Frequency: It is more advantageous to try to prevent crashes that result in injuries or fatalities than it is to prevent crashes that result in property damage. A similar frequency analysis could be conducted in the context of injury to identify hazardous sites. For example, the transportation planner may want to compare intersections by the frequency of fatal crashes occurring at each. However, fatal crashes are relatively rare events.
Equivalent Property-Damage-Only Frequency: The analyst may want to consider using a weighting scheme to represent fatal crashes and injury crashes as their monetary equivalent in property-damage-only crashes; that is, an incapacitating injury crash may have the same negative impact on the safety of the intersection that 25 property-damage-only crashes would have. All crashes occurring at each intersection are converted to the equivalent property-damage-only (EPDO) crashes. The result is a weighted frequency of crashes at each intersection. The analyst would use this new weighted frequency to identify the top 10 intersections in need of safety remediation. However, this still may not identify those sites most in need of safety remediation. Potentially, although the equivalent crash frequency at an intersection is high, the traffic volume at the intersection may also be very high. Typically, locations or segments with the most crashes are those that also have high traffic volumes. Freeway segments, for example, will usually have among the highest numbers of crashes because of their very high volumes. An analysis of the 10 locations with the most crashes for Honolulu in 1990 showed that 8 of the 10 locations were on freeways. Because of the high volumes on the segments, inevitably a high number of crashes will occur even though the risk of crashes is much smaller than for other types of roads. Conversely, non-freeway highways and local roads in general will have much smaller crash totals because their traffic volumes are much lower.
Crash Rates: Analyzing each location using a crash rate will take into account the traffic volume. The crash rate expresses the frequency of crashes at a location in the context of the exposure. For most crash analysis, the exposure will be some measure of the traffic volume at the location. Crash rates on highway segments are often expressed as the frequency of crashes per 100 million vehicle miles. Crash rates at intersections or other spots are often expressed as the frequency of crashes per million entering vehicles. Similar to crash frequency calculations, many different crash rates can be calculated including total crash rate, fatal crash rate, fatal and injury crash rate, and equivalent property damage crash rate.
Even ranking locations by crash rates can be problematic. A location on a low volume facility with only one crash could rank high because of the very low exposure. However, using limited funds to improve a location with only one crash may not be the most effective use of funds and may not prevent future crashes. Some agencies rank crashes by using a combination of frequency and crash rates.
Hazardous locations also can be identified using the classical statistical method, rate quality control method, and Empirical Bayes method, among others. These methods rank locations by applying statistical distributions.
Sites With Promise: A Change in FocusA new school of thought has emerged in the safety analysis community that the emphasis should be shifted from hazardous location identification to locating sites that are in need of safety remediation and have the potential to be improved cost-effectively. These locations are referred to as "sites with promise." The basis of this idea is that a site does not have to be unduly hazardous for there to be the opportunity to prevent crashes cheaply. This is a shift in emphasis from only funding improvements at locations that are the most hazardous, to funding improvements at locations where crashes can be prevented in a cost-effective manner. A program for the identification of sites with promise would rank sites by at least five criteria in the interest of efficiency and fairness. A procedure is explained in Dr. Ezra Hauer's paper, "Identification of Sites with Promise," published in Transportation Research Record 1542(6). |
Further InformationThis section only provides a brief description some methods of crash data analysis used to identify hazardous sites. However, many publications are available for engineers and planners that describe the process. The following are some notable publications: Manual of Transportation Engineering Studies, Institute of Transportation Engineers, Washington, D.C., 1994. Highway Safety Engineering Studies Procedural Guide, USDOT, Washington, D.C., 1981. Higle, J.L. and M.B. Hecht, A Comparison of Techniques for the Identification of Hazardous Locations, Transportation Research Record 1238, National Research Council, Washington, D.C., 1989. Transportation Engineering: An Introduction. Khisty, C.J. and B.K. Lall, Prentice Hall, Upper Saddle River, New Jersey, 1990. |
Targeted Crash Types or Conditions
Another crash analysis method entails identifying certain crash types (such as run-off-the-road, heavy vehicle, or pedestrian crashes) or crash conditions (such as wet pavement and nighttime crashes) of interest. This method can be useful when funds are available for safety remediation in a targeted improvement type or when there is public support for reducing a certain crash type, such as a those involving pedestrians. To improve pedestrian safety, the analyst may identify locations (particular intersections, sections, or areas) where pedestrian crashes are occurring.
When analyzing targeted crash types or conditions, transportation planners may use specialized forms of exposure data. For example, for crash analysis of pedestrians at intersections, the crash rate may be calculated using the number of pedestrians crossing the intersection as exposure. For crash analysis of heavy vehicles on highway segments, the crash rate may be calculated based on truck VMT.
Over representation
Another variation on searching crash data for high crash locations is the identification of overrepresentation. An analyst can identify certain crash characteristics (nighttime, wet roadway, alcohol-involved) that are proportionally over represented at locations in a transportation network. This is a useful method for applying a targeted countermeasure. For example, if funds are available to install highway lighting, an analyst can identify locations where a large proportion of the crashes occur at night. The funds can be used to install lights where they would prevent the most crashes. In addition, this method may be useful in diagnosing crash causes. For example, an area that is experiencing a large proportion of wet roadway crashes may need improved pavement, an overlay, or milling.
Hot Spots Versus Hot Zones: The Problem of Spatial Autocorrelation
Problems with hot-spot analysis could occur when it is assumed that observed high crash locations are spatially independent. Often an isolated hot spot is more of a hot zone (a cluster of streets that have similar crash frequency).
For example, in 1990, approximately 19,000 crashes were reported on the island of Oahu (the City and County of Honolulu). Exhibit 4.1 identifies the 10 intersections with the most crashes. The crash frequency at these 10 locations varied from 100 to 191. The symbol size is not proportional to the number of crashes. While their crash frequency identifies these discrete locations high-incident hot spots, the map can be misleading. Inspecting the crash frequency in the vicinity of these locations revealed that there was also a high crash frequency on the roadways around these spots. Many of these high crash locations have crash frequencies that are not that much higher than the surrounding road segments.
Exhibit 4.1
10 Highest Crash Frequency Intersections in Honolulu
This condition is statistically called spatial autocorrelation and indicates that geographic entities (in this case, intersections) located close to one another have similar crash levels. Spatial autocorrelation between the number of crashes (and presumably traffic volume) means that they are not really independent locations, but part of a larger complex. The similarities may be due to the similarities in traffic volume (that is, adjacent street segments carry the same traffic) as well as spillover of crashes from adjacent streets. (8, 9)
Shifting the analysis to highway segments does not solve the problem. The same problem of spatial autocorrelation would be true for highway segments with many crashes (hot links). Adjacent segments are liable to have similar characteristics, such as the numbers of crashes, traffic volume, number of lanes, bordering land uses, and even similar types of drivers (8). Without considering the crash experience of adjacent segments, any identification of a high accident segment is liable to be only part of the crash problem.
Another analysis plotted clusters of crashes in central Honolulu. Most of the clusters encompass several street segments; a couple fall along one single street, but others capture parallel streets. In other words, instead of isolated hot spots, the clusters show a collection of streets where there is a high concentration of crashes (hot zones).
GIS as a Tool for Crash Data Analysis
Many of the newer safety information systems use GIS software. GIS is a computerized mapping system that allows multiple data sources to be linked using a common coordinate system and to be displayed graphically. Information can be layered in GIS to produce detailed descriptions of conditions and to conduct analysis of relationships among variables. Most commercial GIS applications have associated database and analytical functions that can be linked to the graphic display of data.
Advantages of using a GIS-Based Crash System
A GIS-based crash system has many advantages over a traditional tabular database. The predominant advantage of a GIS-based system is the ability to visually represent crashes; that is, crashes can be represented geographically on a map, allowing the analyst to visually inspect roadway and roadway junctions. The geographic relationship between crashes can be seen. Other advantages of GIS include the ability to efficiently link crash data with other types of information and to easily manipulate spatially referenced information.
In GIS, crash data is maintained separately from roadway data (and from other data layers). The data is linked through a relational database system using the geographical coordinate system as the means of linking different variables. All types of information included in the database have their geographical coordinates encoded as a field and can be displayed as objects. Thus, a highway segment can be represented as a line in the database and a crash can be represented as a point. In turn, different types of highway segments can be displayed as separate databases (called layers), and different types of crashes can also be displayed as separate layers. For example, to display crashes on a highway, the crashes are drawn on the screen separately from the highway segments. By controlling the drawing order and color, the analyst can show a map of the crashes on the highways.
In GIS, high crash locations can be identified through coordinate analysis. Crashes are referenced in GIS by their X/Y coordinates, not by the highway on which they occurred. (Since crashes are usually references by the segment or intersection where they occurred, the crash data must usually be translated to get it into this form when it is input into the GIS program.) Using the coordinate system, the location can be defined narrowly (for example, exact X/Y coordinates) or more broadly (for example, all crashes within 100 yards of a particular X/Y coordinate).
Relationships between objects are developed in GIS by conducting spatial adjacency operations. For example, to link a crash to a particular highway segment, the analyst selects the highway segments and then selects crashes that have coordinates that fall along the highway. The GIS program will make the linkage by examining all crashes having coordinates that coincide with selected highway segments or that fall within a certain distance of the coordinates of the highway segments. Those crashes that fit the criteria are kept while those crashes that do not fit the condition are ignored. Because GIS is a visual representation of the database, the crashes are automatically identified on a map by the operation. (Example GIS output is provided in the previous section.)
Identifying particular types of roadway elements or subtypes of crashes is simplified with GIS. The roadway elements of interest are selected for display using GIS. The crashes occurring on roadways with those elements are also identified and displayed. For example, to identify the number of crashes occurring on two-lane rural highways, the analyst (1) selects all two-lane rural highways from the roadway inventory database and (2) uses the selected roadway segments within GIS to select crashes that occurred on these segments. The GIS operation can identify all crashes that occurred on roadways with those elements or within a specified distance of the highway segment.
Examples of Crash Analysis Operations Using GIS
There are many examples of GIS use in crash data analysis. A few examples that illustrate GIS capabilities are discussed in the following paragraphs.
Analysis by Type of Roadway
Exhibit 4.2 displays the location of all crashes occurring on one-way streets in an area of Honolulu. The one-way streets are shown as red lines on the map, and the crashes occurring on the one-way streets are shown as black dots. The analysis revealed that, as expected, the crashes were concentrated in the most built-up part of Honolulu. In addition, the crashes occurred on streets that traverse commercial areas. While fewer crashes occur on some one-way streets in Honolulu (particularly downtown), other one-way streets have a high concentration of crashes (for example, in Waikiki at the bottom right). In short, a GIS-based crash system can facilitate identification of subsets of the data and can quickly display the data.
Exhibit 4.2
One-Way Street Crashes in Honolulu: 1990
Location of Crashes on One-Way Streets
Although similar analyses could also be performed with a traditional crash database system, GIS allows the user to conduct this analysis quickly and with less effort. The analysis in the Honolulu example took approximately 15 minutes to prepare.
Analysis by Crash Characteristics
Using GIS, an analyst can identify certain characteristics of crashes (such as nighttime, wet roadway, and alcohol-involved crashes) that are proportionally overrepresented at locations on a roadway network. For example, Exhibit 4.3 displays locations where crashes occurred on wet days in Houston (red points). Of these crashes, the locations where proportionately more crashes occurred on wet days than on dry days are shown as green ellipses. This type of analysis can help to identify a possible deficiency in the roadway environment. For this example, a detailed review may be needed at those locations where proportionately more crashes occurred on wet pavement. The roadway at those locations may need drainage improvements or a friction overlay added to the pavement.
Exhibit 4.3
High-risk Wet Crash Hot Spots in Houston: 1998
Crash Locations for Wet Days Relative to All Crashes
Using GIS for Crash Data Analysis:
|
Density Analysis
Even more complicated spatial operations can be conducted with a GIS-based crash information system. Surface maps can be calculated that allow crash densities to be seen over an entire highway segment. For example, Exhibit 4.4 displays the distribution of crashes along Farm-to-Market Road (FM 1093) in Houston. The crashes are displayed as red points. FM 1093 is a major state-managed arterial running from the west loop of Interstate 610 to Fort Bend County. The heavy concentration of crashes in the eastern half of this arterial is mostly related to the much heavier volumes in the central part of Houston (to the right of the map). A density surface was created that overlays a fine grid on the arterial and calculates the density of crashes in each cell. The density of crashes is then related to a density surface of VMT. The result is crashes are reported in the context of their exposure. Exhibit 4.5 displays the relative density of crashes to VMT. As seen, the highest rate of crashes (crashes relative to VMT) is at the far eastern part of the arterial. Not only is the volume of crashes higher, but the crash rate is much higher as well.
Exhibit 4.4
Houston Crashes: 1998
Crash Locations on FM 1093
Exhibit 4.5
Crash Risk Along FM 1093
Density of Crashes Relative to Density of VMT
Kernel Density Estimate
These examples illustrate the possibilities for crash analysis using a GIS-based system. Many other examples can be shown that link crashes to numerous factors that correlate with the crashes, such as crashes in commercial areas, temporal variations in the crash risk along particular highways, or crashes around particular venues such as stadiums. The ability to link diverse variables is one of the strengths of GIS. Analysts can use this powerful tool to easily identify factors involved in crashes as well as subsets of crashes and to link crash information to traffic volume, roadway inventory, land use, and many other characteristics. The ability to display the results of the analysis immediately make this tool particularly appealing because patterns can quickly be detected and unusual concentrations of crashes identified.
Software for Safety Analysis
GIS Safety Analysis Tools
FWHA developed a set of programs to perform spot/intersection analysis, cluster analysis, strip analysis, sliding-scale evaluations, and corridor analysis in GIS. Packaged together as GIS Safety Analysis Tools, the software is available free of charge from FHWA. One of the goals of distributing the software is to encourage safety engineers and others in state and municipal DOTs and MPOs to explore the capabilities of GIS-based highway safety analysis tools. The software also includes pedestrian and bicycle analysis tools to select safe routes to schools, assess the bicycle compatibility of roadways, and define high pedestrian crash zones (11).
GIS Software
Many GIS software programs are commercially available. The appropriate software depends on the size and preference of the jurisdiction. Travel forecasting software used by transportation planners is often integrated with GIS programs.
PBCAT
PBCAT is the Pedestrian and Bicycle Crash Analysis Tool (see Exhibit 4.6). Developed by the Highway Safety Research Center for the FHWA, PBCAT software is intended to assist state and local planners and engineers in analyzing pedestrian and bicycle crashes. The software includes a database that analyzes pedestrian and bicycle crashes and helps the user to identify problems and potential countermeasures. The software also includes a user's manual with examples.
Exhibit 4-6
Example of PBCAT Software Analysis Input Screen
PBCAT has the following capabilities:
- Quickly determines the crash type through a series of on-screen questions about the crash, crash location, and maneuvers of the parties involved.
- Customizes the database in terms of units of measurement, variables, and location referencing as well as imports/exports data from and to other databases.
- Produces a series of tables and graphs defining various crash types and other factors associated with the crashes such as age, gender, and light conditions.
- Links recommended countermeasures to specific bicycle and pedestrian crash types and related resource and reference information.
- Provides user-friendly, online instructions and help features, including examples, along with a user's manual.
|
The PBCAT software and User's Manual are available free of charge from the FHWA. The software can be ordered online at the Pedestrian and Bicycle Information Center web site at: www.walkinginfo.org/pbcat |
Highway Safety and Monitoring Software
Highway Safety and Monitoring Software (HISAM) is computer software developed under an FHWA research contract to assist local jurisdictions in developing, monitoring, and evaluating their highway safety programs. The software aids in database development and accident analysis. It also allows for integration between data files. Users can identify high accident locations based on accident frequencies, accident rates, or EPDO indexes and rates. The software provides the framework for a highway safety system for jurisdictions under 500,000 people (12).
|
The HISAM software and other applicable software can be ordered through the McTrans, Center for Microcomputers in Transportation at the University of Florida. McTrans is a software distribution center originally established by FHWA and now independently operated. McTrans is available on the Internet at: http://mctrans.ce.ufl.edu/ |
Other Crash Analysis Software
Many commercial software programs are available that can assist traffic engineers, transportation planners, and traffic safety specialists in identifying high crash locations and crash patterns. These programs can rank high accident locations and analyze the locations for patterns that help in developing solutions. Some of the programs can also be integrated with GIS programs.
|
Applicable software can be ordered through the FHWA-designated software distribution center, PC-TRANS. PCTRANS is available on the Internet at: http://www.kutc.ku.edu/pctrans/ |
CHSIM: Tools on the Horizon
The Comprehensive Highway Safety Improvement Model (CHSIM) is being developed for FHWA. CHSIM, which will consist of a set of software tools, will assist in identifying safety improvement needs and in developing a systemwide program of site-specific projects to maximize highway safety improvement. CHSIM will have five specific computerized analytical tools that will accomplish the following highway safety management steps:
- Network screening to identify sites with promise
- Diagnosis of safety problems at specific sites
- Selection of appropriate countermeasures
- Economic appraisal of candidate improvements
- Priority rankings for candidate improvements
The development of CHSIM began in April 2001. Interim tools are scheduled to be available in 2004 and final tools in 2006.
Useful National Databases
Nationally, many transportation agencies conduct analyses and produce reports on transportation safety that could be useful to transportation planners. In addition, national crash databases are available that can be used by planners either to acquire information about their own areas or state or to make comparisons.
National Center for Statistics and Analysis Databases
NHTSA's National Center for Statistics and Analysis (NCSA) provides both data and analysis on the nature, causes, and injury outcomes of crashes. NCSA data and reports are nationally representative. Two of NCSA's databases, the Fatality Analysis Reporting System (FARS) and the General Estimates System (GES), are useful to transportation planners.
Fatality Analysis Reporting System
FARS contains data on a census of fatal traffic crashes within the 50 states, the District of Columbia, and Puerto Rico. It includes all crashes involving motor vehicles traveling on a roadway customarily open to the public that result in the death of a person within 30 days of the crash, either an occupant of a vehicle or a non-occupant, such as a pedestrian.
NHTSA has a cooperative agreement with an agency in each state government to provide information in a standard format on fatal crashes. Data is collected on more than 100 different data elements. The system has been operational since 1975 and has collected information on more than 989,451 motor vehicle fatalities. Much of this data is directly available on the Internet where NHTSA maintains a direct query database (13), at the following URL: http://www-fars.nhtsa.dot.gov/.
General Estimates System
GES is calculated from a nationally representative sample (called the National Automotive Sampling System [NASS]) of police-reported motor vehicle crashes of all types, from minor to fatal (14). NASS/GES, which began operation in 1988, was created to identify traffic safety problem areas, provide a basis for regulatory and consumer initiatives, and form the basis for cost and benefit analyses of traffic safety initiatives. The information is used to estimate how many different kinds of motor vehicle crashes take place and what happens when they occur.
To be eligible for the GES sample, a crash must be recorded on a police crash report; involve at least one motor vehicle traveling on a traffic way; and result in property damage, injury, or death. These accident reports are chosen from 60 areas that reflect the geography, roadway mileage, population, and traffic density of the United States. GES data collectors make weekly visits to approximately 400 police jurisdictions in the 60 areas across the United States, where they randomly sample about 50,000 reports each year.
Highway Safety Information System
FHWA developed and maintains the Highway Safety Information System (HSIS). HSIS is a roadway-based system used for the study of highway safety. It provides data on accident, roadway, and traffic variables from a group of select states and can be used to associate the risk of crashes with roadway and traffic variables. The data is already being collected by the states for the management of their highway systems. Currently, HSIS contains data from California, Illinois, Maine, Michigan, Minnesota, North Carolina, Utah, and Washington. The states send the data to the HSIS laboratory where the data is subjected to quality-control procedures.
HSIS is primarily used in support of the FHWA safety research program and as input to program and policy decisions, although it is also available to analysts conducting research for NCHRP, university researchers, and others involved in the study of highway safety. Researchers define specific requests for the data and extracts of files are developed by HSIS staff. Full state data files are not available from HSIS because of agreements with the states. Transportation planners could request extracts from the files if they are studying specific safety problems.
Federal Motor Carrier Safety Administration
The U.S. DOT provides motor carrier safety information online through the Analysis and Information Online web site. The web site provides access to three online databases: SafeStat, Crash Profiles, and Program Measures. It also provides useful analysis reports on motor carrier safety.
The most applicable of the online databases to transportation planners is Crash Profiles. Crash Profiles summarizes crash statistics for large trucks and buses involved in fatal and non-fatal crashes. The database merges data on fatal crashes from the FARS database and information on non-fatal crashes from the Motor Carrier Management Information System (MCMIS) crash file. Each state is represented in the database with a profile that provides information on the vehicle, driver, environment, carrier, location and other circumstances of heavy vehicle crashes. The database can be used for an individual state or combined for the nation. Transportation planners can use the database to identify heavy vehicle safety problems in specific geographical areas. National statistics can also be used for comparison. Direct file extracts from the MCMIS crash file can also be requested from FMCSA.
|
The FMCSA Analysis and Information Online website can be accessed at: http://ai.volpe.dot.gov/mcspa.asp |
National Transit Database
The Federal Transit Administration collects annual information from urbanized area recipients of federal funds in a wide number of areas, including safety and security. The 2002 report cycle introduces a new and heightened safety and security database, requiring agencies to report safety information on a monthly or quarterly basis, depending on their size. The new reports collect data on an incident level (over a certain threshold) requiring agencies to report all major events (fatalities and major injuries) as well as a wide variety of security related information. The new database will assist FTA and industry users in identifying safety and security trends, and finding solutions to recurrent issues.
Further InformationThe National Transit Database is available through the FTA website: http://www.fta.dot.gov |
Useful Publications
The following publications provide information on or examples of safety analysis.
SEMCOG Traffic Safety Manual
The MPO for southeast Michigan, SEMCOG, has developed a manual for planners, engineers, and other agencies involved with traffic safety (for example, law enforcement) that describes a comprehensive approach to traffic safety analysis. It includes information on collecting crash data, maintaining a crash database, identifying high crash locations, choosing appropriate solutions, and performing a benefit/cost analysis. Although the manual was developed for distribution to SEMCOG's planning partners, it should be a valuable resource for other transportation planners (15).
Further InformationInformation on obtaining a copy of the SEMCOG manual is available through the following website: http://www.semcog.org |
Implementation of GIS-Based Highway Safety Analyses: Bridging the Gap
FHWA has published a report that discusses the integration of GIS in safety analysis. The report is an educational document for safety engineers and GIS professionals that provides information on the following (16):
- The benefits that GIS technology offers in general analyses, including display, spatial, and network evaluations, as well as cell-based modeling. The applications from the already-developed GIS Safety Analysis Tools are discussed as examples.
- A description of how historical safety data (crashes and roadway inventory) is acquired, why such data is collected as linear referenced data, and how linear referenced data is different from spatial data. Definitions of common route systems are provided along with illustrations to show how each is different.
- General background information on Linear Location Referencing Systems (LLRS or LRS), which includes an explanation of routes and their measures, common types of LRS, how linear referencing methods (LRMs) are used to locate crashes and roadway inventory, and how GIS uses LRS to locate linear features.
- A general understanding of how GIS manages network data and how in GIS route data is different from roadway network data. The impact of resolution, scale, and route calibration is discussed as they relate to data accuracy.
- A detailed discussion of the process of integrating GIS and safety data, including the need to plan for the integration and development of the GIS network and route system, and the processing of LRS data within GIS.
Further InformationThe report would be a useful resource to a state or metropolitan planning agency interested in using GIS as a tool to assist in highway safety analyses. The report is available from FHWA or on the Internet at: http://www.tfhrc.gov/safety/pubs/1039.pdf |
NCHRP 295: Statistical Methods in Highway Safety Analysis
A synthesis of statistical methods used in highway safety analysis was conducted for NCHRP. The report summarizes the current practice and research on statistical methods in highway safety analysis, including statistical methods used for the identification of hazardous locations and the development and evaluation of countermeasures. The synthesis could be a useful resource to transportation planners and other transportation professionals interested in highway safety analysis (17).
Further InformationThe synthesis report is available from the Transportation Research Board or on the Internet at: http://www.nationalacademies.org/trb |
Institutional Issues
Reliability of Crash Data
This chapter highlights some of the known problems with police-reported crash data-problems that can reduce the reliability of safety analyses. Innovations in crash data reporting, location identification, and database management are improving the quality of crash data. As these innovations and improved technologies become more widely used, crash data quality will increase.
Combining Data from Multiple Sources
Many MPOs span more than one local or state jurisdiction, and they must combine data from multiple agencies to conduct crash analysis of the entire metropolitan area. Because the crash reporting process is not nationally uniform, the data may not be easily merged; it may be in different formats or contain different elements. In addition, the data needed for analysis may not cover the same timeframe because the timeliness of available data varies by jurisdiction. In these situations, it may be advantageous to conduct a separate analysis for each jurisdiction in the metropolitan area.
Liability
Many state DOTs are concerned about the tort liability implications of crash data and, therefore, may be reluctant to share crash data with their planning partners. However, many states such as Connecticut and North Carolina look past this concern. These states expect that as more professionals have access to crash data, more safety remediation measures can be taken.
Many state DOTs have developed operational procedures, referred to as risk management programs, to decrease their tort liability. Conducting routine identification and remediation of hazardous highway safety locations is a component of a risk management program (18).
Personnel Resources
Software such as GIS programs can enhance capabilities and decrease the time needed to conduct safety analysis. Most traditional databases can be downloaded to commercially available database or spreadsheet software. However, staff must be trained (1) in using the programs to their full capability and (2) in basic safety analysis methodologies. Transportation planning personnel must balance the time needed to train for and conduct the analysis with their other responsibilities.
References
(1.) Elvik and Mysen, Incomplete accident reporting: Meta-analysis of studies made in 13 countries, Transportation Research Record. 1999.
(2.) Levine, Ned, Karl E. Kim, and Lawrence H. Nitz, Daily fluctuations in Honolulu motor vehicle accidents. Accident Analysis & Prevention. 27 (6), 1995.
(3.) Levine, Ned, Karl E. Kim, and Lawrence H. Nitz, Spatial analysis of Honolulu motor vehicle crashes: I. Spatial patterns. Accident Analysis & Prevention, 27 (5), 1995.
(4.) Levine, Ned, Karl E. Kim, and Lawrence H. Nitz , Spatial analysis of Honolulu motor vehicle crashes: II. Generators of crashes. Accident Analysis & Prevention, 27 (5), 1995.
(5.) Levine, Ned and Karl E. Kim, The spatial location of motor vehicle accidents: A methodology for geocoding intersections. Computers, Environment, and Urban Systems. 22 (6). 1999.
(6.) Hauer, E. Identification of Sites with Promise. Transportation Research Record 1542, National Research Council, Washington, D.C., 1996.
(7.) Manual of Transportation Engineering Studies. Published by the Institute of Transportation Engineers, Washington, D.C., 1994.
(8.) Pierce, B. K. and J. G. Kinateder, Incorporation geographic correlation when sampling a transportation network. Transportation Research Record. 1999.
(9.) Boyle, A.J. and C. C. Wright. Accident 'migration' after remedial treatment at accident blackspots. Traffic Engineering and Control. 25(5). 1984.
(10.) Accident Data Use and Geographic Information System (GIS). Presentation by Suzette Thieman, Cheyenne Area Transportation Planning Process, Cheyenne, Wyoming. Presented at the Sixth National Conference on Transportation Planning for Small and Medium-Sized Communities, Spokane, Washington, 1999.
(11.) Federal Highway Administration, GIS Safety Analysis Tools, Version 2.0 (CD format), Washington, DC, July 2000.
(12.) Harkey, D.L. and Ruiz, R. HISAM: An Accident Data Base Manager. Transportation Research Record 1238, TRB, National Research Council, Washington, D.C., 1989.
(13.) National Highway Traffic Safety Administration. 4313 EB - State Data Program. National Highway Traffic Safety Administration, U. S. Department of Transportation: Washington, DC. (www-nrd.nhtsa.gov/include/summaries/4313EB.htm). 2000.
(14.) National Center for Statistics and Analysis. Fatality Analysis Reporting System (FARS). National Center for Statistics and Analysis, National Highway Safety Transportation Administration, U. S. Department of Transportation: Washington, DC. (www.nhtsa.dot.gov/people/ncsa/fars.htm).
(15.) SEMCOG manual Southeast Michigan Council of Governments, SEMCOG Traffic Safety Manual, Second Edition, SEMCOG, Detroit, Michigan, 1997.
(16.) Smith, R.C., D.L. Harkey, and B. Harris, Implementation of a GIS-Based Highway Safety Analysis: Bridging the Gap. FHWA Task Report, FHWA-RD-01-039, U.S. DOT, Washington, D.C. 2001.
(17.) Persuad, B.N., Statistical Methods of Highway Safety Analysis. NCHRP Synthesis of Highway Practice, TRB, National Research Council, Washington, D.C., 2001.
(18.) Demetsky, M.J. and K. Yu. Assessment of Risk Management Procedures and Objectives in State Departments of Transportation. Transportation Research Record 1401, TRB, National Research Council, Washington, D.C., 1993.
Endnotes
- The primary reference for this section is the Manual of Transportation Engineering Studies (7). It is published by the Institute of Transportation Engineers, Washington, D.C., 1994. back to text for chapter 4, endnote 1.
