Identification of High Pedestrian Crash Locations

CHAPTER 2. LITERATURE REVIEW

TOOLS OR METHODS AVAILABLE

Approaches to Problem Identification

An initial step in reducing the frequency of pedestrian crashes is identifying where they occur or where there is a concern they are likely to occur. Once locations with a large number of pedestrian crashes or with a safety concern for pedestrians have been identified, appropriate treatments can be selected and installed. Identification of locations can be accomplished using one of the following approaches:

• A reactive approach (also known as a “traditional approach”) is generally based on an observed or historical crash pattern that suggests a hot spot (sometimes called a “black spot”) and the need for a countermeasure to improve safety at that location.
  
  
• A proactive approach uses models—such as crash prediction models—to determine the expected number of crashes for locations within a set region. These estimates can be used to prioritize the locations that may potentially need treatments. The approach is called “proactive” because it addresses a potential risk rather than an experienced problem. An example of a proactive approach is the systemic approach. The FHWA website notes the following about the systemic approach:

  …involves widely implemented improvements based on high-risk roadway features correlated with specific severe crash types. The approach provides a more comprehensive method for safety planning and implementation that supplements and compliments traditional site analysis. It helps agencies broaden their traffic safety efforts and consider risk as well as crash history when identifying where to make low cost safety improvement locations.⁽³⁾

The 2016 National Cooperative Highway Research Program (NCHRP) Synthesis 486 identified the current process for identifying locations with local road safety concerns. The synthesis considered all crash types and was not limited to pedestrians.⁽⁴⁾ The objective of the synthesis was to document State programs and practices that address local road safety. The most frequent response from State DOTs concerning problem identification was a combination of both reactive and proactive methods. The survey indicated the following as the most frequently applied criteria for prioritizing local safety projects:

• Cost–benefit analysis (28 States).
• Crash history (26 States).
• Available funding (25 States).

The survey found that fatal and serious injury crash numbers and crash rates were the major performance measures used. Table 1 provides additional findings from the synthesis.

Techniques

General Knowledge

In some cases, a location for consideration of improvements could be identified through recommendations made by local law enforcement or from citizens’ comments or complaints. School districts may contact the city requesting assistance with crossings near their schools. Cities with traffic-calming programs may also identify locations of concern for pedestrians based on evaluations of drivers’ operating speeds. A sidewalk and curb ramp inventory is another city effort that could identify locations of concern. While these techniques are not dependent on crashes, they can provide an earlier indication of locations where additional evaluation is needed.

Suggested Techniques From the Highway Safety Manual

Several techniques are used to identify HCLs, including the following list of performance measures from the Highway Safety Manual (HSM) (see table 4-2, pp. 4–9, in volume 1):⁽⁵⁾

• Average crash frequency.
• Crash rate.
• Equivalent property-damage-only average crash frequency.
• Relative severity index.
• Critical rate.
• Excess predicted average crash frequency using the method of moments.
• Level of service of safety.
• Excess expected average crash frequency using safety performance functions (SPFs).
• Probability of specific crash types exceeding threshold proportion.
• Excess proportions of specific crash types.
• Expected average crash frequency with empirical Bayes (EB) adjustments.
• Equivalent property-damage-only average crash frequency with EB adjustments.
• Excess expected average crash frequency with EB adjustments.

The most common methods include identifying intersections or midblock crossings with the highest number of crashes in a specific time period (i.e., frequency) or the highest number of crashes after adjusting for exposure (i.e., crash rate).

Given the limited number of pedestrian crashes, other data or approaches may be necessary to appropriately identify the locations where pedestrian safety is a concern. For example, identifying zones prone to pedestrian crashes—rather than a single intersection or marked crosswalk—could identify crashes where pedestrians are not crossing at a specific intersection or at a signed and marked midblock crossing.

Zones

Because of the limited number of pedestrian crashes at a specific location, zones can be used to identify a land area where improvements may be needed. In 1998, the National Highway Traffic Safety Administration (NHTSA) developed a guide that describes what zoning is and explains how to design and use pedestrian safety zones to increase the efficiency and effectiveness of pedestrian safety programs.⁽⁶⁾ “Zoning” is defined as a relatively small geographic area where a relatively large proportion of the problem occurs. The NHTSA guide highlights examples of successful pedestrian safety zone programs and includes charts, checklists, and the following steps to define and use zones:

• Step 1. Select the crash problem.
• Step 2. Map the pedestrian crashes.
• Step 3. Define zones.
• Step 4. Calculate the efficiency measure and select final zones.
• Step 5. Evaluate zones and identify resources.
• Step 6. Select program activities.
• Step 7. Implement program activities.
• Step 8. Monitor program activities.

NCHRP Synthesis 295

NCHRP Synthesis 295 presents a discussion about the methodology for identifying hazardous locations.⁽⁷⁾ The synthesis was published in 2001 and contains dated material; however, the authors discuss key concerns that are still relevant. They make the following points:

• Techniques based on high crash counts or rates may have regression-to-the-mean (RTM) issues.
  
  
• The EB approach was developed to overcome RTM. EB estimates the expected number of crashes at a site rather than the crash count.
  
  
• Another approach is to rank sites by their potential for safety improvement.
  
  
• A common thread in recent research is a departure from the use of crash rates.

The authors also comment that current procedures try to overcome the difficulty of the nonlinear relationship between crashes and traffic volume (crash rates usually decrease with traffic volume, and therefore, sites with low volumes tend to be selected if crash rate is used as a selection criterion by itself) by requiring a minimum crash count for a site to be flagged. The authors note that the extent to which this refinement overcomes the problem is unclear because counts (and rates) are subject to random fluctuation.

United States Road Assessment Program

The United States Road Assessment Program (usRAP) has the following goals and objectives as documented on the usRAP website:⁽⁸⁾

• To support iRAP [International Road Assessment Programme] and its vision of a world free of high-risk roads, by ensuring that ALL highway authorities have access to a robust and cost-effective tool for data-driven safety analysis.
  
  
• To provide a unique, innovative, and proactive means of enhancing the safety of the built road environment, so that hazardous designs and risky locations can be assessed and corrected even before a death or serious injury occurs.
  
  
• To reduce death and serious injury on U.S. roads through a program of systematic assessment of risk that identifies major safety shortcomings, which can be addressed by practical road improvement measures.
  
  
• To ensure that limited resources are efficiently allocated so that the greatest safety gains can be made within existing constraints.
  
  
• To forge partnerships among those responsible for a safe road system.

usRAP has a risk-mapping protocol to demonstrate which roads have the highest and lowest risk of fatal and serious injury crashes. The usRAP protocol includes four basic risk maps based on the following safety performance measures:

• Map 1—crash density (fatal and serious injury crashes per mile).
  
  
• Map 2—crash rate (fatal and serious injury crashes per 100 million vehicle miles [veh mi] of travel).
  
  
• Map 3—crash rate ratio (the ratio of the fatal and serious injury crash rate for an individual road section to the average crash rate for similar roads).
  
  
• Map 4—potential crash savings (the annual number of fatal and serious injury crashes that would be reduced if the crash rate for an individual road section could be lowered to the average crash rate for similar road sections).

The road sections on each risk map are color coded to indicate the risk level for fatal and serious injury crashes on that road section. The following distribution is used:

• Highest risk—5 percent of highway system length, shown in black.
  
  
• Medium-high risk—10 percent of highway system length, shown in red.
  
  
• Medium risk—20 percent of highway system length, shown in yellow.
  
  
• Medium-low risk—25 percent of highway system length, shown in light green.
  
  
• Lowest risk—40 percent of highway system length, shown in dark green.

According to the American Automobile Association website, risk maps have been prepared for Iowa, Michigan, Florida, and New Jersey. The following States were added in phase 3: Illinois, Kentucky, New Mexico, and Utah.⁽⁹⁾ Highway agencies interested in having additional maps generated can seek information from the developer of usRAP. Examples of risk maps include those created using Kentucky data for speed-involved crashes, alcohol-involved crashes, aggressive-driving crashes, and lane-departure crashes. These maps were developed as part of phase 3 of the research project.⁽¹⁰⁾ Phase 3 also explored pedestrian star ratings. The authors concluded that the crash frequencies for pedestrian and bicycle crashes proved to be too low to assess the star ratings.

The data needed to develop the four basic usRAP risk maps are listed in table 2 . Only crashes that involve fatalities or serious injuries are used in risk mapping. Other typical approaches include (1) using mainline roadways and excluding interchange ramps, (2) combining both directions of travel on a divided highway even if the highway agency linear referencing system (LRS) treats the two directions of travel as separate segments, and (3) normally conducting the analysis using crash data for a 5-yr period.

Safety Analyst

Safety Analyst is a set of software tools used for highway safety management. The software automates the six main steps of the highway safety management process: network screening, diagnosis, countermeasure selection, economic appraisal, priority ranking, and countermeasure evaluation.⁽¹¹⁾ As noted on the website, “Safety Analyst can be used to proactively determine which sites have the highest potential for safety improvement, as opposed to reactive safety assessments done conventionally,” and it implements the procedures from part B (“Roadway Safety Management Process”) of the HSM.^(11,5) The program is designed for more than just identifying HCLs, and its applicability to identifying high pedestrian crash locations would need additional investigation. The minimum data elements are listed in table 3 .

Prioritizing Improvements Using ActiveTrans Priority Tool

NCHRP Report 803 is a guidebook that presents the ActiveTrans Priority Tool (APT), a step-by-step methodology for prioritizing pedestrian and bicycle improvements along existing roads.⁽¹²⁾ The APT is intended to be used by planners and other agency staff charged with managing a pedestrian or bicycle prioritization effort. It is designed to encourage practitioners to prioritize pedestrian and bicycle improvement locations by establishing a clear prioritization process that meets the following criteria:

• Responsive to agency/community values: Transportation agencies often make decisions based on a defined set of goals or values of the communities they serve.
  
  
• Flexible: Rather than being a rigid, one-size-fits-all tool, the APT is flexible and allows practitioners to choose the most appropriate approach that reflects agency/community values and resource availability.
  
  
• Transparent: The APT is designed to facilitate transparency by breaking the prioritization process down into a series of discrete steps, each of which can be easily documented and explained to the public.
  
  
• Responsive to the unique needs of pedestrians and bicyclists.

Crash Prediction or Planning Models/Systemic Approach

An approach for being proactive is to use crash prediction models to determine the expected number of crashes at a location and to use this information to prioritize the locations that may potentially need treatments. FHWA has suggested several potential risk factors that States may want to consider. These factors, which are not specific to pedestrian crashes, are listed in table 4 .⁽¹³⁾

A current NCHRP project (NCHRP 17-73) will develop a process for the following:⁽¹⁴⁾

• Conducting systemic safety analyses for pedestrians using analytical techniques to identify pedestrian activities (including behavior), roadway features, and other contextual risk factors (e.g., land use) that are associated with pedestrian crashes.
  
  
• Identifying appropriate and cost-effective systemic pedestrian safety improvements to address their associated risk factors.
  
  
• Enabling transportation agencies to prioritize candidate locations for selected safety improvements based on risk.

The NCHRP 17-73 research results will aid transportation agencies in more effectively allocating resources for pedestrian safety improvements.⁽¹⁴⁾ The research builds on element 1 of the FHWA Office of Safety’s Systemic Safety Project Selection Tool and focuses on existing countermeasures within the 4E framework—Education, Enforcement, Engineering, and Emergency response—and will not include developing a software solution.

Jermprapai and Srinivasan provided an example of using crash prediction models by applying their models to calculate the expected number of crashes for census block groups within Florida that had no observed crashes.⁽¹⁵⁾ They noted that if safety assessments are made purely based on crash history, all the locations with zero observed crashes should be deemed equally safe. Their results are summarized in table 5 .

The predictive model highlighted that there is significant variability in crash risk across these locations because of differences in land use and socioeconomic patterns. They noted that these types of findings could be used to allocate statewide safety funds and prioritize safety projects. They concluded with the following:

It is envisioned that future studies will develop similar models for other states to determine the transferability of the empirical findings from the models in this study. Further, the predictive ability of the models could be improved with additional explanatory variables and advanced, flexible modeling methods. With additional explanatory variables, the challenge lies in finding consistent spatial data at a fine spatial resolution for the entire state and the effective treatment of multicollinearity among the variables. An important future effort would also seek to perform rigorous comparative assessments of the application of alternate methods to quantify the benefits obtained from advanced models.⁽¹⁵⁾

Grembek et al. developed a systemic approach for identifying sites where there is potential for significant reductions in pedestrian and bicyclist injuries in California.⁽¹⁶⁾ They applied their method to a 16.5-mi section of San Pablo Avenue (State Route 123) in the San Francisco East Bay.

The authors developed the Pedestrian Systemic Monitoring Approach for Road Traffic Safety (PEDSMARTS). The procedure involves the following steps:

1. Estimate the number of a specific type of crash at a specific type of facility.
   
   
2. Present the data in a matrix to map the distribution of crash types across all facility types.
   
   
3. Identify the systemic hot spots in the matrix.
   
   
4. Identify the corresponding countermeasures for each cell to identify which one could be implemented for the specific crash type occurring at the specific location type.
   
   
5. Select the appropriate countermeasure that can be implemented throughout all similar locations.

The authors recommended that the California DOT use the method proposed in the research to study pedestrian/bicycle crashes on State arterial roadways.

Whether the suggested PEDSMARTS method is transferable to other regions would need to be investigated.

Integrated Method That Includes Crashes, Conflict, and Subjective Risk

Pešić et al. discuss a proactive method of identifying and ranking danger spots for pedestrians that incorporates the following:⁽¹⁷⁾

• Objective risk based on the analysis of traffic crashes.
• Potentially objective risk based on the subjective conflict technique.
• Subjective risk based on the analysis of participants’ attitudes to traffic.

Further, using this method to rank identified spots helps officials set priorities in allocating resources.

Iterative Process to Group 1-Mi Segments

A 2003 paper discussed a study conducted to provide the framework for the systematic identification of pedestrian HCLs on the Florida highway system as part of the Highway Safety Improvement Program (HSIP).⁽¹⁸⁾ The authors noted that the then-current methodology to identify HCLs as part of Florida’s HSIP did not identify the location where most pedestrian crashes occur. The study considered roadways with continuous sidewalks on both sides in Miami-Dade County using data from 1997 to 1999. The methodology used crash data available from the Florida DOT and did not consider pedestrian exposure data. It used an iterative process to group 1-mi roadway segments with similar roadway characteristics, such as the presence of sidewalk and facility type. The iterative process was needed because different groupings may be needed before a Poisson process can model the actual pedestrian crash frequencies. The study had to omit the crashes within Miami Beach because of significantly higher levels of pedestrian activity in the area. The authors noted that a larger data sample may allow the grouping of a significant number of segments with much higher levels of pedestrian activities. They also noted that development type (i.e., urban or suburban) may be an additional criterion for further segregation of the data. With a confidence level of at least 90 percent and for 1-mi segments, a pedestrian crash frequency was called “abnormally high” for four-lane divided facilities with three or more crashes and for six-lane divided roadways with four or more crashes. From these thresholds, 22 1-mi segments were identified as pedestrian HCLs.

Geographic Information System

A geographic information system (GIS) is a digital system intended to code, display, and analyze data in relation to their geolocation (i.e., data with a geodetic system of coordinates). A GIS manipulates layers containing different types of data. For safety analyses, GIS tools allow displaying and analyzing a layer with crash data on a map. Depending on the purposes, crashes can be analyzed in terms of their relation to each other or in relation to other layers containing different types of data, such as road infrastructure, census tract, and land use. The complexity of the types of analyses that can be performed with GIS data varies significantly depending on the purpose of the analysis.

Pulugurtha et al. used the following steps with a commercial GIS software program to identify high pedestrian crash zones:⁽¹⁹⁾

• Step 1. Geocode pedestrian crash data.
• Step 2. Create a crash concentration map.
• Step 3. Identify zones and their shapes and sizes.

In their 2013 report, Scopatz et al. assessed the state of the practice of GIS tools used for safety by State and local agencies.⁽²⁰⁾ An important element moving toward widespread use of a GIS is the Moving Ahead for Progress in the 21st Century Act legislation signed in 2012 requiring statewide base maps and an LRS that include all public roads by June 2014. At the time of that report, about two-thirds of the States had base maps for local roads, but it was not clear what proportion had an LRS in place. The most widely available data to be used in a GIS are crash records, traffic volume, and roadway inventory information.

Although several agencies use GIS tools for visualization of crashes on a Web platform, the most common type of further analysis is the identification of hot spots for crash frequency. Scopatz et al. report that 33 States were implementing HSM methods, 12 were implementing Safety Analyst, and 6 were implementing the Interactive Highway Safety Design Model.⁽²⁰⁾

Visualization Tools

As identified earlier, the simplest and most widely used types of GIS tools are those that provide crash visualizations. In general, crashes can be displayed using variables as filters or as criteria to display the data. Several agencies offer visualization tools for their crash data using a Web-based interface. An example of such maps is available from several sources. For example, the Florida DOT shows its pedestrian and bicycle crashes at https://fdotewp1.dot.state.fl.us/TrafficSafetyWebPortal/FivePercent/LocationList.aspx.

Visualization and analysis tools are available to analyze crashes. In addition to tools specifically for processing layers of data prior to safety analyses, modules offer the construction of the following four types of risk maps to visualize crash data:

• Crash density map—highlights road segments by the proportion of severe crashes.
  
  
• Crash rate map—highlights road segments by the magnitude of their crash rates (crashes per average annual daily traffic (AADT), vehicle miles traveled, or other measures of exposure).
  
  
• Crash rate ratio map—highlights segments by the relative size of their crash rates with respect to the mean of other segments of similar characteristics.
  
  
• Potential crash savings map—displays the difference between the crashes per mile in a segment and the average number of crashes per mile at a group of segments of similar characteristics.

While the first two types of maps are basically visualization tools that also yield some ranking of hot spots, the last two maps carry a further analysis of the data to provide a ranking based on a comparison of each segment and other similar segments.

Advanced GIS Analyses

A GIS with evaluation tools allows safety analyses beyond visualization, depending on the amount of data available. This section highlights a subset of those possible analyses.

Identifying Intersection-Related Crashes

One basic analysis consists of identifying intersection-related crashes for further analysis. If an inventory of intersections is available, the main task is identifying an area around the intersection and selecting those crashes within that area for further analysis. Research has shown that an average distance of 300 ft around the intersection is appropriate for developing SPFs or crash frequency prediction equations.⁽²¹⁾ If no intersection inventory is available, a layer of intersections can be generated based on the points of intersection between georeferenced road segments. These intersection locations can then be overlapped with layers of crashes of interest.

Relationships With Exposure Measures

The relationship between pedestrian and bicyclist crashes and exposure measures is still the subject of active research. Contrasted with AADT for other types of crashes, there is no consensus on what measure of exposure is best to incorporate in these types of analyses, mostly because of availability and strength of correlation.

Development of Safety Routes

A state-of-the-practice evaluation in Alabama presents a variety of GIS-based analyses. One interesting technique identified in that document is the development of “safety routes,” potentially useful for defining school routes.⁽²²⁾ This procedure combines the LRS defined for the road inventory layer and the geolocations of crashes. The procedure develops different paths based on the shortest path between two nodes of interest and the recommended path based on reduced probability of crashes.

NHTSA’s Advancing Pedestrian and Bicyclist Safety: A Primer for Highway Safety Professionals

NHTSA recently published Advancing Pedestrian and Bicyclist Safety: A Primer for Highway Safety Professionals.⁽²³⁾ The report provides examples for geocoding crashes.In the geocoding crash method, crashes are coded with geographic location such as latitude and longitude and are presented in map format and/or spatially analyzed in relation to factors of interest, such as roadway types, destinations (e.g., schools), and sociodemographic information.

Conflicts

The behaviors of motorists, bicyclists, and pedestrians at intersections could be an indication of the potential for crashes at those conflict areas. “Conflict” is described as a sudden action taken by either party to avoid a collision. “Avoidance maneuvers” are defined as any change in speed or direction in response to the presence of another party. A conflict exists when it is believed that the absence of a change in action would result in a collision, while an avoidance maneuver would not necessarily result in a collision. An example of an avoidance maneuver is when a pedestrian changes course to walk around a vehicle.

In 2006, Carter et al. presented a pedestrian intersection safety index that can prioritize intersection crossings given macro-level site characteristics.⁽²⁴⁾ The analysis incorporated behavioral data in the form of conflicts and avoidance maneuvers and subjective data in the form of expert safety ratings. The final model included consideration of the presence of traffic control signal or stop-sign controlled crossing, the number of through lanes, 85th-percentile speed of the street being crossed, traffic volume on the street being crossed, and predominant lane use in the surrounding area.

There is ongoing debate about the validity of using near misses as a surrogate for vehicle crashes. Guo et al. analyzed near crashes (termed “safety critical events”) in naturalistic driving data and concluded that near crashes could be used as a surrogate for actual crashes.⁽²⁵⁾ Knipling challenges the “near-crash as surrogate for crash” by arguing that near crashes and crashes are two different types of events that have only a weak correlation.⁽²⁶⁾ Knipling does concede that, with proper sampling and validation, certain types of near crashes can be used as crash surrogates.

Interactive Pedestrian Crash Map Websites

Numerous city and State DOTs are making pedestrian crash data available online through interactive maps. These online maps allow anyone with access to a Web browser to perform a variety of functions for pedestrian crashes, including the following:

• Zoom out to get a citywide view, or zoom in to get a street- or intersection-specific view.
  
  
• Display the crash density by grouping crashes that occur close to each other.
  
  
• View pedestrian crashes in the context of nearby land use, socioeconomic attributes, or planned street or highway improvements.
  
  
• Filter/subset the crashes by year, crash severity, crash cause, and other crash attributes (e.g., driver or pedestrian gender, day of week, road agency ownership).
  
  
• Click on an individual crash or HCL to get additional information.

Several examples of these interactive pedestrian crash maps are shown and described in this section.

FHWA shows pedestrian fatal crashes for 2009 to 2013 on its HEPGIS site: http://hepgis.fhwa.dot.gov/fhwagis/ViewMap.aspx?map=Annual+Fatal+Crashes|Pedestrian+Fatal+Crashes+2009-2013.

California has funded the development of a Transportation Injury Mapping System (TIMS) to provide data and mapping analysis tools and information for traffic-safety-related research, policy, and planning. The TIMS is available at https://tims.berkeley.edu/. Several tools are available, such as interactive maps and the ability to query collision data, including pedestrian-involved collisions.

With consideration for the Vision Zero initiative, Vision Zero ATX, a local nonprofit organization, created a color-coded map that indicates each fatality’s mode of travel. The map is available at http://www.visionzeroatx.org/austin-fatality-map/. It does not group nearby crashes to show crash density but does allow the user to filter by several attributes, including gender, control of road, time of crash, and day of week.

As part of its Vision Zero initiative, Portland, OR, shows all traffic fatalities and serious injuries on a map that displays fatality and serious injury crash density. This map shows 10 yr of crash data (2005 through 2014), displays density with different-sized symbols, and allows users to click on a specific crash for additional information and view pedestrian crashes separately from other crashes. Additionally, when viewing the pedestrian-only crash tab, this map highlights (with a yellow line) the 20 pedestrian high crash corridors as determined by the city’s crash analysis. The map is available at http://pdx.maps.arcgis.com/apps/MapSeries/index.html?appid=cf122cd3b4ef46f0ac496b2d61d554e9.

As part of its Vision Zero initiative, Los Angeles, CA, displays its high injury network for people walking and biking who are killed or seriously injured in the context of other variables such as a community health and equity index. The Los Angeles DOT also used this high injury network to prioritize Safe Routes to School (SRTS) projects. While SRTS prioritization preceded Vision Zero, the city noted that, interestingly, the top 50 schools are all within 0.25 mi of the high injury network. The high injury network in Los Angeles can be viewed at http://visionzero.lacity.org/high-injury-network/.

As part of a pedestrian safety initiative called WalkFirst, San Francisco, CA, displays high pedestrian injury corridors and intersections in a map-based interface. The WalkFirst site categorizes these high injury corridors and intersections by crash profiles (i.e., the causal crash factor, such as low nighttime visibility, a left-turning vehicle at a signalized intersection, alcohol use, etc.) so that a user can click on a certain crash profile to see where that causal factor contributes to a high crash density. The interactive map also uses color-coded lines to indicate crash density on these high injury streets and is available at http://walkfirst.sfplanning.org/index.php/home/streets.

In a separate analysis of pedestrian safety in Los Angeles, CA, the Los Angeles Times identified the 817 “most dangerous intersections” in Los Angeles County using the California Highway Patrol’s Statewide Integrated Traffic Records System ( http://graphics.latimes.com/la-pedestrians-how-we-did-it/). The Los Angeles Times analysis considered the following three factors in designating dangerous intersections:

• The total number of pedestrian collisions.
• The proportion of collisions that involved a pedestrian.
• The proportion of pedestrian accidents that were fatal.

The resulting map shows the density of dangerous intersections, not the density of actual pedestrian crashes. The intent of the graphical map was to show, from a regional perspective, the location of problem areas (zones with multiple intersections) with respect to pedestrian safety. For more information, see http://graphics.latimes.com/la-pedestrians/.

The North Central Texas Council of Governments, the metropolitan planning organization (MPO) for the Dallas–Fort Worth region, has developed static pedestrian crash maps that are available online. The maps show individual crash locations overlaid on a color-coded density map and include pedestrian and bicyclist crashes from 2010 through 2014. An example is available at http://www.nctcog.org/trans/sustdev/bikeped/BikePedCrashInfo.asp.

The Massachusetts DOT displays an interactive map of “Top Crash Locations” for pedestrians and bicyclists. The map groups nearby pedestrian crashes into intersection clusters based on the number of crashes occurring, with the top 200 ranking intersections displayed with a different-colored symbol. Several different types of cluster displays can be viewed, depending on the desired year range and type of crash (pedestrian or bicyclist). The map is available at http://gis.massdot.state.ma.us/maptemplate/topcrashlocations.

The Houston Chronicle has compiled pedestrian crash data that are displayed in an interactive map. The map can be used to show individual crashes and associated attributes, such as road conditions, weather conditions, and contributing factors. A heat map that groups together nearby pedestrian crashes is also viewable and uses color coding to indicate relative crash density. Filters also allow the user to focus on a selected crash severity or contributing crash factor. These maps are available at http://www.houstonchronicle.com/default/media/Pedestrian-accident-map-252958.php.

PREVIOUS STUDIES

Chicago, IL

Chicago used descriptive and spatial analyses in 2011 to identify crash trends.⁽²⁷⁾ As documented in the NHTSA report, the city evaluated injury frequency and severity for different age groups (e.g., children and seniors) as well as for various crash types and contributing environmental factors.⁽²³⁾ The spatial analysis presented crash density citywide by ward as well as near schools. The findings will be used to develop the Chicago Pedestrian Plan, identify engineering treatments throughout the city, and aid ongoing pedestrian safety education efforts.

Oregon

The NHTSA report states that the Oregon Department of Transportation (ODOT) Traffic-Roadway Section decided to focus limited resources on locations that have the greatest potential for crash reductions.⁽²³⁾ In 2013, ODOT set out to match infrastructure countermeasures with potential locations for improvements by identifying key patterns of behavior and roadway conditions that cause locations to be high risk. The NHTSA report noted that “this approach is promising, but ODOT expressed that the analysis was constrained by the limited availability of roadway information such as bicyclist and pedestrian volumes, the presence of a crossing treatment, presence of a turn lane, driveway activity, and sight distances.”⁽²³⁾

San Francisco, CA

San Francisco’s WalkFirst (see http://walkfirst.sfplanning.org/) combines public engagement with technical and statistical analysis of where and why pedestrian collisions occur on San Francisco’s streets and updates the knowledge of effectiveness and costs of various engineering measures. The first phase included identifying high-demand and high-risk corridors and intersections based on a history of severe or fatal injuries. The crash analysis found that just 12 percent of San Francisco streets account for over 70 percent of all severe and fatal crashes.

North Carolina

As documented in the NHTSA report, the North Carolina DOT Division of Bicycle and Pedestrian Transportation provides a suite of tools (see http://www.ncdot.gov/bikeped/researchdata/) designed to help practitioners understand pedestrian and bicycle crash issues, including the following:⁽²³⁾

• A database of crash-typed data that can be queried.
  
  
• An interactive map.
  
  
• Detailed summary reports on pedestrian and bicyclist crash trends and common crash types.

The NHTSA report states that these tools are used to inform decisionmaking when developing the State Highway Safety Improvement plan and by local agencies receiving planning grants.⁽²³⁾

San Francisco, CA; Las Vegas, NV; and Miami, FL

FHWA awarded grants to San Francisco, Las Vegas, and Miami in 2002 to examine pedestrian crashes and then to evaluate pedestrian safety countermeasures within the identified high crash zones. The goal of the project was “to demonstrate how a city could improve pedestrian safety by performing a detailed analysis of its pedestrian crash problem, identifying and evaluating high crash locations, observing factors such as driver and pedestrian behavior, and deploying various lower cost countermeasures tailored to the site.”⁽²⁸⁾ The HCLs were identified by reviewing police reports.⁽²⁹⁾

Factors Associated With Pedestrian Crashes

Understanding those roadway, traffic control devices, and traffic factors that affect pedestrian crash severity or are associated with pedestrian crashes could lead to an approach for identifying potential locations that might have a crash concern.

A 2015 study identified significant factors affecting pedestrian crash injury severity at signalized and unsignalized intersections using a mixed logit model.⁽³⁰⁾ The study used 3 yr of pedestrian crash data from Florida (2008–2010) and included 3,038 pedestrian crashes. Of those crashes, 2,360 occurred at signalized intersections, and 678 occurred at unsignalized intersections. The study only included State roads, and the authors recommended additional research for local roads.

For signalized intersections, the following were associated with higher pedestrian severity risk:

• AADT.
• Speed limit.
• Percentage of trucks.
• Very old pedestrians (age ≥ 80 yr).
• At-fault pedestrians.
• Rainy weather.
• Dark lighting conditions.

The authors provided the following examples:

• A 1-percent higher truck percentage increases the probability of severe injuries by 1.37 percent.
• A 1-mph higher speed limit increases the probability of severe injuries by 1.22 percent.

At unsignalized intersections, the following were associated with higher pedestrian severity risk:

• A pedestrian walking along a roadway.
• Middle-aged and very old pedestrians (age ≥ 80 yr).
• At-fault pedestrians.
• Vans.
• Dark lighting conditions.
• Higher speed limit.

Standard crosswalks were associated with a 1.36-percent reduction in pedestrian severe injuries for unsignalized intersections. (The standard crosswalk was shown in a figure to be a crosswalk with two transverse lines. The crosswalk marking type was a binary categorical variable: 1 when transverse lines were present and 0 otherwise.) This variable was not found to be significant for signalized intersections.

Haleem et al. reviewed several previous studies that conducted pedestrian injury severity analyses.⁽³⁰⁾ Table 6 and table 7 summarize the variables that were found to be significant in the various studies.

In 2014, Jermprapai and Srinivasan reported on a planning-level model for assessing pedestrian safety.⁽¹⁵⁾ They noted that their review of the literature found that past research had focused on specific cities or counties with census tracts as the unit of analysis. Their study used a larger study area (the entire State of Florida) at a finer spatial resolution (census block groups rather than tracts). Crash data from 2005 to 2009 and land use data from the entire State of Florida were used in developing four models: total crashes, severe and fatal crashes, fatal crashes, and nighttime crashes. Their results generally reaffirmed other studies concerning the relationship between crashes and socioeconomic, land use, and transportation characteristics. One of their findings was that a low-income location in a higher income county is riskiest. They also concluded that locations with a larger volume of conflicting vehicular and pedestrian movements make the locations riskier. Table 8 through table 10 summarize the findings from the Jermprapai and Srinivasan study for severe and fatal crashes along with their findings from a review of the literature.⁽¹⁵⁾

Previous Evaluation of Methods

Vasudevan et al. evaluated the following methods used to identify pedestrian HCLs:⁽⁵⁷⁾

• Crash index based on crash frequency method. The HCL with the highest number of crashes is ranked 1, and the remaining HCLs are then ranked in a descending order of their numbers of crashes.
  
  
• Crash index based on weighted frequency method. This method was based on the severity and frequency of pedestrian crashes and accounted for the density of crashes.
  
  
• Crash index based on vehicular traffic volume. This index is defined on the basis of the number of pedestrian crashes and the average daily traffic (ADT).
  
  
• Crash index based on pedestrian age group. This index considered the number of crashes for each age group, the population for each age group, and weights to consider the pedestrians’ inexperience and other characteristics.
  
  
• Sum-of-the-ranks (SR) method. The SR method combines the previous methods into a single rank value.
  
  
• Crash score (CS) method. The CS method considers crash frequency, weighted factors, traffic volume, and pedestrian exposure that are normalized.

The authors used data for 30 pedestrian HCLs in Las Vegas to illustrate the methods and to conduct statistical analyses using Spearman’s correlation coefficient of ranking. Crash data between 1996 and 2001 were available. They concluded that, although any of the methods could be used to identify pedestrian HCLs, the simple-frequency-based or weighted-frequency-based methods would be adequate as a first step.

Pulugurtha et al. used Las Vegas, NV, crash data from 1998 to 2002 to evaluate several of the methods listed previously along with a GIS method.⁽¹⁹⁾ The GIS method was used to help quantify the concentration of crashes and reduce the degree of subjectivity involved in identifying high crash zones. It resulted in 22 linear zones and 7 circular zones. Pulugurtha et al. evaluated the following methods for those 29 zones:⁽¹⁹⁾

• Crash frequency.
• Crash frequency based on severity.
• Crash density based on area.
• Crash rate based on vehicular volumes.
• Crash rate based on population.
• SR method.
• CS method.

The authors concluded that, because the rankings obtained for each zone were relatively consistent for the SR and CS methods compared to the individual methods, the composite methods are more robust. They recommended the CS method over the SR method, since it also helps identify the cause of the safety problem.