|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-HRT-04-091
Date: August 2004
Signalized Intersections: Informational Guide
PDF Version (10.84 MB)
PDF files can be viewed with the Acrobat® Reader®
CHAPTER 6 — SAFETY ANALYSIS METHODS
TABLE OF CONTENTS
LIST OF FIGURES
LIST OF TABLES
6.0 Safety Analysis Methods
Decisions made at each step in the roadway life cycle are no longer dominated solely by mobility considerations; they must now be made in a context-sensitive manner and must address accessibility. At the planning stage, the need for and purpose of the project are examined. Safety, mobility, cost, and the preservation of resources must all be carefully considered in the design, operation, management, and rehabilitation phases of a roadway project. Quantifying the impacts of tradeoffs made at each step in the process is an essential element of the context-sensitive, flexible design process.
The explicit consideration of safety and the development of quantifiable safety information have greatly enhanced the flexible design process. The availability of quantifiable safety tools has helped elevate the role of safety in the design decision process, such that it may be considered on a par with competing demands such as mobility, sustainability, aesthetics, environmental impact, and cost.
Efforts to eliminate unnecessary cost while preserving needed functionality in the delivery of road infrastructure have led to the application of value analysis techniques to road designs. A key aspect in assessing the true societal value of a road infrastructure investment, over its entire useful life, is its safety performance. Safety performance has begun to take its place alongside operational performance (level of service), environmental, and financial performance.
An increased focus on the societal costs of road trauma has drawn attention to the proportion of injuries and fatalities attributable to designs in-service that, while perhaps conforming to prevailing minimum standards, are less than optimally safe. This recognition has led, in turn, to the concepts of design domain, road safety audit, and an explicit consideration of road safety at each step in the roadway life cycle ¾ in other words, in planning, functional design, detail design, construction, operations, management, maintenance, and rehabilitation. Further, it has sharpened the focus of applied research and has highlighted the need for advances in the road safety body of knowledge to be considered in setting road transportation policy, and in maintaining the currency of standards.
A key outcome of this increasingly holistic view of road safety is the growing recognition that the safety implications of each step in the roadway life cycle are inextricably linked to those that preceded it and those that follow. Decisions taken at one stage in the process are constrained by those made previously, and those taken at subsequent stages are likely to pose constraints and/or have substantive impacts in the future.
6.1 Balancing Safety and Mobility
Some countermeasures to improve safety at a signalized intersection do so at the expense of mobility. In certain cases, the operational disbenefits may be too great. The traffic engineer needs to understand the tradeoff between safety and mobility.
Furthermore, mobility means more than just the movement of motor vehicles. It includes providing mobility for bicyclists, transit, and pedestrians of all abilities. Some mobility measures fall outside the range of tradeoffs. For example, ADA improvements required by law are not subject to traditional constraints of volume of use, cost, and warranting criteria. However, these improvements should be folded into the considerations in the evaluation and selection of improvements.
Access management is another example of the tradeoff between safety and mobility. Many intersections in commercial areas allow access via driveways within the functional area of the intersection, leading to unnecessary conflict. Closure or relocation of these driveways, the addition of medians, and the prohibition of certain turns can all improve safety at the expense of overall mobility. In older intersections, residential or commercial driveway access often remains within the intersection, once of little concern in the past, but now a major source of conflict due to the amount of traffic passing through.
Consideration of the competing goals of safety and mobility are needed for operations and safety of a signalized intersection. A typical problem traffic engineers face is the need to accommodate turning vehicles while recognizing when a particular movement may be too dangerous to allow or may have a significant impact on overall intersection operations. Using multiphase signal operation instead of one with only permissive left turns aids drivers in safely making left turns at the expense of lengthening the overall signal phase. Right turns may be allowed during the red phase at the expense of increasing pedestrian and bicycle collisions.
Some means of providing for turning traffic may do so without sacrificing mobility. Exclusive turn lanes, offset lanes, and channelization remove turning traffic from through lanes. Other solutions for accommodating left-turning vehicles may be signal coordination through the provision of gaps in oncoming traffic. At some intersections, the safety record, the operations, or the accessibility of the intersection to pedestrians and bicyclists may be so poor that an intersection redesign may be considered. This may mean realigning an approach, closing a leg, reducing curb radii or lane widths, using an innovative intersection design that allows for left turns at another location, or total grade separation.
The following sections will detail the four-part process for a safety evaluation of a signalized intersection.
6.2 Selection of an Intersection
In selecting an intersection for a detailed safety analysis, the key questions are:
Based on what is observed at the intersection in terms of its overall safety experience, the traffic engineer can make a decision to proceed with a more detailed causal evaluation of the safety performance, as detailed later. If it appears that safety concerns are present, operational and design problems will likely also exist, given the relationships among poor level of service, design deficiencies, and a poor safety record. This section will discuss various statistical techniques for determining the relative safety of an individual location as compared to other locations and highlight the merits of each.
The collision history of a signalized intersection is the key indicator of its safety performance, and is the focus of the remainder of this section. Statistical techniques for evaluating the collision performance vary from the most basic to the complex. They may compare the safety performance of a single signalized intersection to another group of similar intersections, or serve as a screening tool for sifting through a large group of sites and determining which site has the most promise for improvement.
Many jurisdictions carrying out a review of safety at a signalized intersection will usually have a collision database which provides information on the location, time, severity, and other circumstances surrounding each collision reported by police or the parties involved. Collision data in this form can provide the traffic engineer with a quick assessment of safety at a location. However, the traffic engineer should keep the following in mind in the analysis of safety using collision data.
First, the engineer needs to be aware of what constitutes a collision in the available data. Collisions can be classified as reportable or nonreportable. Nonreportable collisions always involve no injury and usually involve vehicle damage below a certain threshold. Collisions may also be considered nonreportable if they take place on private property, as in a shopping mall parking lot or on a driveway. In other instances, they may involve a vehicle striking another parked vehicle and leaving the scene of the accident. The traffic engineer needs to be aware that nonreportable collisions may not show up in the collision database. If they are not recorded, their absence may possibly mask an underlying safety concern (figure 43).
The definition of a reportable collision varies from jurisdiction to jurisdiction. Usually, all collisions involving personal injury and those with damage exceeding a selected threshold may be classified as reportable. Occasionally the threshold dollar value used to classify property-damage-only collisions changes, and this may be a problem if the review period crosses through such a change and if property-damage-only collisions are being included in an analysis. This is one reason some jurisdictions will only consider injury and fatal collisions in their safety assessments. However, by doing so, they will have excluded a majority of collisions that actually occurred at a site, masking valuable information that may be useful at the diagnosis stage.
Second, in some instances, a collision may be self-reported. In some jurisdictions, when the collision is property damage only, the police will not investigate, and the parties involved are required to report the collision on their own. Because information is self-reported and no information is collected on-scene by police, collisions may contain inaccurate or erroneous information. Because of the increased potential for low-quality information, detailed analysis involving self-reported collision information needs to be treated with caution.
Last, users of collision data should be aware of how the location of a collision is recorded. Adjacent intersections can be a convenient reference point for coding collisions. It is not uncommon to code a collision as occurring at an intersection when it actually occurred at a location nearby, such as an adjacent parking lot, a driveway, or immediately upstream or downstream of the intersection (figure 44). The potential for error in coding the location of a collision should be understood. The traffic engineer should be aware of the procedures for coding locations and be satisfied that sufficient error checking of data has been done.
Once data are available, the most common method of assessing safety at a location is through comparing its collision history to other similar sites. If the traffic engineer is comparing the safety performance of a location to another set of locations, these need to be carefully thought through. At the very least, the other locations should also be signalized and have the same number of approaches as the site being examined; sites with different traffic control devices and layouts can be expected to have differing levels of safety. Surrounding land use will also have a significant effect on collision frequency, with intersections in urban areas having a different collision profile than intersections in rural areas. Finally, comparisons with sites that are located in other jurisdictions may be tainted by differing collision-reporting thresholds, enforcement, predominant land use, vehicle mix, road users, climatic conditions, or other unknown factors; results of such a comparison should be tempered with caution.
With these in mind, different methods of using collision data to assess safety are discussed in the following sections, highlighting their benefits and drawbacks. The different methods to be discussed are:
6.2.1 Collision Frequency
Traditionally, traffic engineers used (and many still use) a frequency-based method of identifying and evaluating the safety of a site.(64) Past observed collision frequencies at a site may be used to compare and rank the site with collision frequencies at a group of similar locations. Many jurisdictions produce a top-10 list of the intersections producing the highest collision frequency in their jurisdictions and concentrate all of their efforts at reducing collisions at these sites. Recently, one insurance company released a list of the “worst” intersections in the United States, according to internal insurance data.(65)
Collision frequency may also be used to screen candidate sites for improvements. The collision frequency at the site may be compared to the average collision frequency for a reference population to calculate a potential for improvement. However, relatively short periods of time, such as one year of collision data, should never be used as the basis for a safety intervention. Because collisions are relatively rare events, a high collision frequency in any given year at a particular intersection may be simply a random fluctuation around a much lower long-term average at the site. In the next year or series of years, the collision frequency may drop, without any safety intervention at all. This phenomenon is referred to as regression to the mean. Regression to the mean may be minimized by using data collected over a longer period of time, such as 3 years or more, when evaluating the site. Site selection based on multiple years of collision data will provide a truer picture of the collision profile of the intersection and avoid errors that can result from looking at collision history over a short period.
Apart from regression to the mean, there are several other disadvantages to using collision frequency as the sole means of evaluating safety at a location. First, a high collision frequency may not necessarily mean that a site is truly in need of safety improvement. It is known that sites with higher volumes will have a higher collision frequency than sites with lower volumes. Therefore, sites ranked simply by collision frequency will invariably end up with higher volume sites at the top of the list. Second, the method does not address the severity of collisions at the site. Failing to consider severity may result in the identification of sites with high numbers of minor collisions, while ignoring sites with fewer but more severe collisions. The approach results in a failure to identify sites at which the public has greater risk of injury or death.
6.2.2 Collision Rate
Collision rates are an improvement over measuring just collision frequency: They allow a measure of the risk road users face because rates consider exposure. Collision rates are calculated by dividing the collision frequency for a period of time by the estimated average annual daily traffic (AADT) of vehicles entering for all approaches in that time period. Collision rate provides an improved yardstick for comparison to other sites. As with collision frequency, a collision rate for a location undergoing a safety assessment may be compared to similar intersections (signalized, same number of legs, same range in AADT). The intersection may be ranked to produce a top-10 list, or a threshold value may be used above which a detailed safety analysis is warranted. Using a collision rate will account for the effect that volume has on collision frequency.
However, using a simple collision rate to screen locations has several disadvantages. First, using a collision rate to rank sites that have different volumes requires the assumption that collision frequency and volume have a linear relationship, but research suggests that this is not the case. Lower volume sites tend to experience a higher collision rate. Ignoring this fact means that low volume sites may appear less safe than their higher volume counterparts. Second, collision rates, as with collision frequency, do not consider collision severity. Sites with a high collision rate may have relatively few casualty (fatal and injury) collisions. Last, as collision rates are calculated from collision frequency, which fluctuate around a long-term average and experience regression to the mean, a site might be ranked high on a list due to a recent period with an unusually higher number of collisions. If collision rates are being used to screen out candidate sites for safety improvements, it is recommended that a collision rate be calculated for a longer time period (3 to 5 years).
6.2.3 Combined Collision Frequency and Rate Method
Weaknesses in the collision frequency and collision rate methods of selecting candidate intersections for further collision analysis may be somewhat overcome by an approach that combines both methods. In this approach, intersections with both a high collision frequency and a high collision rate may be candidates for a more detailed safety diagnosis.
An example of this is presented in figure 45. The collision rate (per million entering vehicles) and the 5-year average collision frequency for 10 sites were plotted. Threshold values were selected to separate out the sites with the highest combined collision frequency and rates. Sites to be considered for further safety analysis were those with an average yearly collision frequency of greater than 30 collisions/year and a collision rate of 1.50 collisions per million vehicles entering the intersection. Based on this, one candidate intersection was selected. In the above example, the threshold values were arbitrarily chosen. The traffic engineer may also consider using a value that is twice the average collision frequency and rate as the threshold.
6.2.4 Collision Severity Method
In the above discussion, sites were considered for further analysis if the collision frequency, rate, or a combination of the two was particularly high. As identified, a weakness with these methods is that they do not consider the severity of the collisions involved. The collision severity method considers the distribution of collision severity for each site under consideration. A typical approach is through use of the Equivalent Property Damage Only (EPDO) index. It attaches greater importance, or weight, to collisions resulting in a serious injury or a fatality, lesser importance to collisions resulting in a moderate or slight injury, and the least importance to property-damage-only collisions. A weighting factor suggested in the 1999 ITE publication Statistical Evaluation in Traffic Safety Studies suggests the following severity breakdown (table 28), based on values used by the U.S Department of Transportation and the American association of Motor vehicle Administrators. (64)
Table 28. Suggested weighting for collision severity method.
Depending on local considerations, the above weighting system may be modified to reflect actual values in terms of cost, such as property damage, lost earnings, lost household production, medical costs, and workplace costs. A comparison with similar intersections (signalized, same number of legs, same range of AADT) may be done by calculating the EPDO index for similar sites to the one being considered. The EPDO index will explicitly consider the severity breakdown of collisions, providing greater weight to fatal and injury collisions over property-damage-only collisions. The traffic engineer should be aware, however, that because the severity of a collision is associated with higher speeds, signalized intersections on roads with a higher operating speed, such as in a rural location, will likely have a higher EPDO index than in an urban area. This may result in a bias that emphasizes higher speed locations. In addition, as with rankings based on collision frequency and rate, regression to the mean will be an issue if the study period chosen is short.
6.2.5 Critical Collision Rate
The critical collision rate method has been widely used among traffic engineers. It represents the expected collision rate of locations with similar characteristics (in this case, the same traffic control device). The critical rate is calculated based on the system-wide average collision rates for intersection or road sections of a similar characteristic. If the actual collision rate is greater than the critical rate, the deviation is probably not due to chance, but to the unfavorable characteristics of the intersection or road section. The method considers the collision rate of a location, allows for comparison with other similar sites, and incorporates a simple statistical test to determine whether the collision rate is significantly higher than expected. The statistical test is based on the assumption that collisions have a poisson distribution.
The critical collision rate method is more robust than using collision frequency or collision rate alone, as it provides a means of statistically testing how different the collision rate is at a site when compared to a group of similar sites. The desired level of confidence may be varied depending on the preference of the user.
Disadvantages of using this method are that it still does not consider the severity of the collisions and assumes that traffic volume and collisions have a linear relationship. In addition, this approach does not consider regression to the mean.
6.2.6 Risk Analysis Methods
The concept of risk analysis involves the determination of collision risk using collision and volume data. Existing safety levels are evaluated at defined roadway locations within the jurisdiction. The collision risk is calculated at each specific site (local risk), across all sites of a specific group (area risk), and across the entire jurisdiction (global risk). Collisions of different severities (property damage only, injury, and fatal) are weighted according to the EPDO index. By combining the results of the local, area, and global risk calculation, locations can be compared and ranked according to their relative risk and the potential for collision mitigation. Sites with the highest risk score would be candidate locations for further safety diagnostics.
Risk analysis methods are robust in that they consider the exposure (volume) and severity of the collisions occurring at the site. The local risk, area risk (risk among locations of a similar nature), and global risk (across the entire jurisdiction) are considered. However, they still assume that the relationship between collision frequency and volume is linear, and they do not consider regression to the mean.
6.2.7 Safety Performance Functions
A safety performance function (SPF) is an equation that presents the mathematical relationship between collision frequency and volume based on a group of intersections with similar characteristics (e.g., signalized, same number of legs). When collision frequency and volume are plotted, an equation can be developed that is represented by a line that is the best fit possible through the various points. Generally, SPFs demonstrate that the expected number of collisions increases as traffic volume increases, and an SPF is curvilinear rather than a straight line. Because the line that plots an SPF is curved, the rate (rise/run) varies continuously along the curve. An SPF typically shows that higher volume sites have a lower collision rate than do lower volume sites.
A simple example of an SPF is illustrated in figure 46. The blue points represent individual intersections with their respective average yearly collision frequency and AADTs. A curve can be drawn through the points representing the best fit. The green point above the curve represents an intersection that is performing worse than predicted.
Advantages of using such a method are that the potential for safety improvement is more accurately calculated, and that it acknowledges that the relationship between collision frequency and volume is not a straightforward linear one. Disadvantages are that this method is relatively complex and still does not acknowledge the random variation of collisions.
6.2.8 Empirical bayes Method
Each of the above methods only considers past collision history by either ranking and selecting a candidate location for further collision analysis, or determining whether a particular intersection under study has a collision problem. Using collision history alone is flawed, because the frequency of collisions from year to year will randomly fluctuate about a long-term average (regression to the mean). Improved methods have evolved that identify high-risk sites that may benefit from remedial treatment(s), particularly the empirical bayes (EB) method. Many jurisdictions are already employing the EB method.
The EB method calculates expected collision frequencies through a combination of observed and estimated collision frequencies. The estimated collision frequencies are derived through the development of an SPF curve. The SPF relates the level of safety of an intersection to traffic volume and other relevant geometric factors. The function estimates the expected number of collisions based on traffic volume and other characteristics; it is expressed in collisions/year for intersections.
The pivotal concept upon which contemporary methods for conducting proper road safety evaluations depend is the EB method. It is superior to traditional methods because it:
Although the development of SPFs is a relatively new area of road safety research, they have been implemented successfully for measuring the safety of road locations. Sites can be ranked to determine which is experiencing the highest number of collisions based on actual collision counts, and to determine its expected collision performance.
The above section has detailed various methods of assessing the safety of a location through consideration of its collision history and comparison with other similar sites. Care must be taken to ensure that the site is being compared with sites that should have a similar level of safety (i.e., sites with a traffic signal and the same number of legs). Simpler methods such as collision frequency and rate may provide a simple and quick way of diagnosing a potential safety problem, but should be used with caution. The traffic engineer may consider using the critical collision rate method or the collision severity method as these provide a more balanced assessment of safety. Developing an SPF, either on its own or for use in applying to the EB method, is a much more sophisticated method of evaluating safety at a location. A summary of the relative merits and drawbacks of each method is presented below in table 29.
Table 29. Common methods of assessing safety at a location.
6.3 Identification of Potential Problems
The previous section discussed different tools that may be used to select a candidate intersection for a safety evaluation. At a certain point, the traffic engineer will conclude, based on past collision history, that there is a safety concern and a significant potential for safety improvement at the location in question. It should be noted that some traffic engineers may have completely bypassed the entire first step of this process (in determining candidate intersection for safety improvements), because they have been asked to carry out a safety analysis of an intersection due to:
This section will discuss how the traffic engineer may correctly diagnose what types of safety problems/issue may be present at a location. Diagnosis of a particular safety concern can then lead to appropriate treatment, as discussed in part III.
6.3.1 Safety Diagnosis
In conducting a safety diagnosis at a signalized intersection, the traffic engineer seeks to understand and identify causal factors of collisions within the functional boundaries of the intersection. All information gathered needs to be thoroughly reviewed and analyzed with the objective of identifying whether there exists one or more opportunities to improve safety at the location. The stages in carrying out a safety diagnosis are:
6.3.2 Assemble Collision Data
Collision data used for diagnosing safety at a signalized intersection should represent at least 3 years of collision data. It should include all collisions reported as occurring at or within the intersection’s sphere of influence. If available, the original police reports should be used to gather anecdotal comments written by police officers at the collision scene and firsthand accounts of the collisions based on involved parties and eyewitnesses (figure 48). Using either the original police reports or collision data taken from a database, a collision diagram should be prepared, providing a pictorial representation of the collision types, severity, movements, and involved approaches.
All collisions occurring within the intersection should be included in the analysis, as well as those occurring immediately upstream and downstream. Collisions occurring further upstream may be included in the analysis, if it is found that operations at the intersection in question contributed to the collision. Rear-end collisions commonly occur well upstream of a signalized intersection due to vehicle queuing. Queuing may occur because of an operational deficiency at the signalized intersection.
The collision diagram will help the traffic engineer quickly ascertain:
As highlighted earlier, the accuracy of safety diagnosis detailed in this chapter will depend on the quality of the collision data. Collision reporting systems may contain incorrect or missing information because:
Before beginning a detailed collision analysis, the traffic engineer should be aware of the strengths and weaknesses of collision data collected in the jurisdiction. Questions to consider are:
6.3.3 Analyze/Diagnose Collision Data
To correctly diagnose safety issues at an intersection, a detailed collision analysis is required, cataloguing the different characteristics of collisions at the intersection. The purpose of this analysis is to determine if any collision characteristics (e.g., collision type) at the study location are “abnormal” compared to these expected for signalized intersections having the same number of approaches elsewhere in the jurisdiction in question. By finding a collision characteristic that is abnormally elevated for a location, the traffic engineer can begin to pinpoint the cause of a collision.
The standard police collision report includes a number of collision descriptors that are collected by police and maintained in a database. These range from information on the driver, the vehicle, the road, and the environment. Many of these characteristics could help in determining causal factors; however, an exhaustive diagnosis of every collision descriptor may not be warranted to establish probable causes. As a primary review, collision characteristics that the traffic engineer should consider analyzing are:
It is recommended that the traffic engineer prepare a table or set of tables that shows the typical collision profile of signalized intersection that is representative of the jurisdiction for each of the above-listed characteristics.
A number of statistical tests are available to determine whether the proportion of a characteristic found at a specific site is the same as that found in a group of similar sites. Identification of abnormal trends can lead toward possible solutions. To ensure that the determination of overrepresentation is valid, appropriate statistical techniques should be employed. The chi-square method is suggested.
The chi-square test is a measure of the differences between measured and expected frequencies at location. The collision data at the subject location provide the measured frequency; the aggregate collision data from a large number of similar intersections provide the expected frequency. If the subject intersection displays a measured frequency that is greater than the expected frequency, then use of the chi-square test and reference to standard statistical tables will allow the analyst to determine whether the difference is likely a random variation or a real difference. Real differences, also called statistically significant differences, are an indication that the subject location has a site-specific deficiency that may be causing this trend.
The chi-square test can be calculated using equation 3.
The chi-square test is not reliable when the expected frequency is less than 5. The expected frequency is determined by multiplying p and n. If p*n is less than 5, then another statistical test is required. Confidence levels for the chi-square test are shown intable 30.
Table 30. Chi-square test values and corresponding confidence levels.
An example of how the chi-square test may be used to determine overrepresentation is provided in the following example. A chi-square analysis was carried out to test whether collisions during wet road surface conditions are over-represented at a signalized intersection.
x = Wet road surface collisions at site (18)
n = Total collisions at site (50)
p = Percent of wet road surface collisions (25.4 percent) typically found at signalized intersections in the municipality.
Chi-square values above 3.84 indicate a 95.0 percent confidence level in the result. In the example above, the chi-square value is 2.96, which is less than 3.84. Therefore, there is not enough evidence to suggest that wet road surface collisions are over-represented at the site.
6.3.4 Determine Overrepresentation
The traffic engineer may determine overrepresentation using either the chi-square or expected values method. The collision characteristics should be reviewed for over representation, through comparison with collision characteristic information representing the typical experience of a signalized intersection. Possible patterns the traffic engineer may encounter are highlighted below.
An examination of collision distribution by season, day of week, or time of day may be helpful in finding patterns that relate to the general travel patterns of road users passing through the intersection. Seasonal patterns, indicating a higher-than-expected proportion of collisions occurring during a particular time of year, may coincide with an influx of unfamiliar drivers to an area—as may be the case in resort areas and/or areas with a significant number of tourist attractions. Day of week and time of day patterns should be examined. Morning/afternoon weekday overrepresentation may suggest collision patterns related to commuting traffic (coinciding with the morning and afternoon rush hours). A late night/early morning/weekend overrepresentation may suggest problems with drunk drivers.
Overrepresentation in collision severity will highlight a location that has an unusually high proportion of fatal and/or injury collisions. A higher proportion of fatal and/or injury collisions may suggest a problem with higher operating speeds.
Collision type, together with a collision diagram, can greatly aid the traffic engineer in diagnosis. A cluster of rear-end collisions on a particular approach may suggest a slippery pavement surface, a large turning volume inadequately serviced by existing lanes, or poor visibility of signals. Unfortunately, most collision reporting systems provide a very basic classification of collision type. For diagnosis purposes, this may not be especially helpful. For instance, one jurisdiction may group all collisions involving a turning movement together, whereas collisions involving left-turning vehicles being struck by an opposing vehicle may be of particular interest to the traffic engineer. The traffic engineer may wish to consider using more finely developed subcategories of collision types, such as detailed in figure 49, to aid in diagnosis. As well, the collision diagram will be an invaluable tool for isolating the combination of movements and/or approaches involved in the abnormal collision type.
Further analysis of such characteristics as vehicle type, driver/pedestrian condition, and/or apparent driver action may be required to provide further symptoms of the collision occurrences.
The end product of the above analysis will be a set of characteristics that is identified as being over-represented. A collision diagram, as discussed previously, may narrow the problem to a particular approach, and should at the very least assist in searching for the causes of the collisions and identifying patterns.
In some instances, however, no specific statistically significant overrepresentation will be found. This does not mean that the location is free of any safety concern. In these cases, the predominant collision type should be the focus of the problem statement (as confirmed through field observations).
6.3.5 Conduct Site Visit(s)
To supplement the analysis and diagnosis using collision data, a site visit or series of site visits should be undertaken. Before initiating site visit(s), the study team should be aware of:
The purpose of the site visit is to gather additional information that can aid in pinpointing potential underlying cause or causes of the abnormal collision patterns (figure 50). The site visit should be undertaken to:
One of the key tasks the study team will wish to conduct during the site visit is a positive guidance review.(8) a positive guidance review uses an indepth knowledge of human factors and the driving task to screen roadways for:
Each of the above may contribute to the occurrence of driver error and collisions.
Information deficiencies occur when information that the driver needs to carry out the driving task safely is missing. An example may be inadequate signage/pavement marking for a designated right-turn lane that traps drivers intending to proceed straight. Attempts to move over to the through lane can cause queuing and possible rear-end and sideswipe conflicts.
Expectancy violations occur when a driver encounters a traffic control or roadway design that conflicts with his or her expectations. The traffic engineer should structure expectancies about treatments at similar locations.(66) The key to effective expectancy structuring is uniformity and standardization. Standard devices that are inconsistently applied can create expectancy problems for drivers. a prime example of this is the use of a left-hand exit amidst a series of right-hand exits. Positive guidance seeks to address this expectancy violation through clearly communicating to the driver that a left-hand exit is ahead.
Workload issues occur when the driver is bombarded with too much information, increasing the likelihood of error. This may occur at an intersection with an abundance of signage, pavement markings, traffic signals, and pedestrian and bicycle activity. All of the above may be further complicated if the operating speed on the approaches is high, giving the driver even less time to sort through and comprehend what to do to get safely through the intersection and on to the intended destination. The traffic engineer should seek to reduce the complexity of the information the driver receives at the intersection or spread information by using advance signs.
Although positive guidance techniques are generally applied to the driving task, these concepts and tools can easily be considered from the perspective of all road users. Positive guidance is a holistic approach that treats the roadway, the vehicle, and the driver as a single, integrated system. It recognizes drivers as the information gatherers and decisionmakers within the system and focuses attention on assuring that they get the information they need, when they need it, in a form they can understand, in time to make rapid, error-free decisions and take appropriate actions. When this occurs, the system functions most effectively, and the driving task is successfully accomplished. Creating and sustaining a supportive information environment on the roadway is the goal of positive guidance.
In conducting a positive guidance review, the analyst attempts to view the roadway through the eyes of an average driver, postulating what the driver’s perceptions, interpretations, expectations, and actions might be. This is done to formulate theories and possible explanations regarding the cause or causes of previous or potential conflicts and/or collisions.
Positive guidance normally focuses on low-cost, information-oriented improvements that can be implemented quickly, either as solutions in and of themselves, or as interim improvements until a more definitive solution can be achieved. It may also identify the need for additional investigation, in the form of conventional engineering analysis, to support theories regarding the contributory causes of collisions, and to justify mitigation measures.
6.3.6 Deciding on Further Analysis
The outcome of the site visit, including a positive guidance review, should be a clear understanding of the safety problem occurring at the locations in terms of its underlying causes (field observations that explain the reason for that particular problem) and its link to any subsequent effects (an unusual pattern/high incidence of a specific collision group).
6.3.7 Conducting Further Studies
Further studies should provide a better understanding of the level of safety experienced by various users passing through the intersection by means of direct observational methods. The study team will need to spend an extended period on site observing traffic patterns and the interaction between the roadway/roadside elements and the drivers, pedestrians, and bicyclists.
The study team may choose to a traffic conflict analysis that will provide a generic overview of driver, pedestrian, and bicyclist interactions over the course of a few days.(67) The traffic conflict analysis technique involves the systematic observation and reporting of traffic conflicts, or “near collisions,” and an assessment of the degree of severity for each conflict. When two or more road users approach the same point in time and space, one or both must take evasive action to avoid a collision. At this point, one of two events may occur. If the evasive action is unsuccessful, then a collision occurs. If the evasive action is successful, a traffic conflict occurs. In general, the presence of a significant number of traffic conflicts indicates operational deficiencies and, possibly, collision potential.
The severity of the traffic conflict is measured by the summation of two scores assigned by the observer: the time-to-collision score and the risk-of-collision score. The time-to-collision score is a measure of the time before a collision would have occurred had no evasive action been taken, and is a function of the travel speeds, trajectories, and separation of the vehicles involved. The risk-of-collision score is a subjective measure of the collision potential and depends on the perceived control that the motorist had over the traffic conflict event. Factors such as maneuvering space and severity of the evasive action taken are considered in the assignment of the risk-of-collision score.
Upon completion of the traffic conflict analysis study, the various conflicts are drawn onto a diagram of the intersection, as with the collision diagram. Questions the traffic engineer may want to investigate are:
A full traffic conflict analysis for all movements at the intersection may not be necessary if a specific behavior is suspected as being a problem based on the site visit and/or collision analysis. Engineers may choose to study a specific behavior such as red light running, lane encroachments, failing to check blind spots while turning, failing to yield, stopping beyond the stop bar, blocking the crosswalk, blocking the intersection, illegal parking, or disobeying turn prohibitions.
The end result of the safety diagnosis (including the analysis of collision data, the site visits, and further studies, if necessary), will be a clear understanding of any abnormal collision patterns occurring at the site, along with the underlying and probable cause for these patterns.
6.3.8 Defining the Problem Statement
A set of one or more clear problem statements should be developed. The problem statement(s) are developed on the basis of the collision analysis (i.e. evidence of over representation among a collision subgrouping) and should be supported through the site visit and any further analysis. It should clearly state a probable “cause” and observed “effect.”
The problem statement helps clearly define safety concerns at the location. Circumstances associated with these safety concerns may be mentioned along with possible causal factors. The problem statement may be multifaceted and encompass the physical and/or operational attributes of the intersection, road user behavior and/or actions, environment and/or temporal conditions, as well as transitory or peripheral events. In many instances, the study team will identify several problems or issues.
6.4 Identification of Possible Treatments
6.4.1 List Possible Treatments
Using the problem statement(s) developed by the study team, possible treatments may now be identified. Possible treatments will be all measures listed that are likely to decrease the frequency or severity of collisions identified as exhibiting an abnormal pattern (overrepresentation).
In part III of this guide, the reader will find treatments organized into five broad groups:
For each treatment, there are references to possible collision groups that are likely to be positively affected through a treatment’s implementation. At signalized intersections, the four collisions groups most commonly identified as a cause for concern are:
Table 31 presents possible causes and treatments for each of these types, along with the appropriate chapter. Note that treatments involving enforcement and education are not discussed in part III of this report.
The material presented here by no means represents the limitation of possible treatments. It gives some indication of the range of options that could be selected, but is not fully comprehensive. It is not possible to develop a complete list of all potential collision treatments, because new tools and techniques for improvement traffic safety are constantly being developed and adopted. It is important that the study team not limit itself to existing lists or tables of treatments. The team should consider a wide range of treatments (including those based on local practice) that may be beneficial, particularly when the collision pattern identified represents a unique situation.
Over the course of the above collision diagnostic analysis, site visits, and field analysis, the traffic engineer may have identified treatments that are of little cost and without question beneficial to improving safety at the intersection. Such treatments may relate to repairing sidewalks, removing sight obstructions, reapplying faded pavement markings, and relocating or adding new signs. These may be implemented without going through the process described below.
Table 31. Collision types commonly identified, possible causes, and associated treatments.
6.4.2 Screen Treatments
The study team likely will generate a long list of treatments that may have been identified in this guide, based on local practice or representative of a unique situation identified at the intersection through the collision analysis, site visits, and subsequent studies carried out in the field. To narrow the options, the study team may consider screening the treatments. One method of screening proposed treatments is to develop a matrix where each treatment is given a score within different categories, based on the consensus among study team members. The individual score categories may be as follows:
Scoring each treatment allows the study team to quickly determine which treatments are expected to have a positive or negative effect on the intersection. The long list of potential treatments then can be reduced to a short list of viable treatments. Based on a threshold score decided upon among the study team, the treatments may then be screened and those scoring poorly may be discarded.
The ability to evaluate the safety impacts of a treatment is paramount to implementing an intersection improvement plan. Information is needed on whether the treatment under consideration is effective in reducing collisions. Most treatments proposed in part III of this guide have some published material that provides a quantitative estimate of effectiveness. For other treatments in part III, no research was found that provided any quantifiable estimate of safety benefits. Before any further consideration as to be applicability of a treatment can occur, the study team will need to decide whether they have a quantifiable estimate of the expected results of a treatment available. If they do, they can proceed with the steps described below. If not, they should carefully consider whether the treatment should be implemented.
6.4.3 Selecting a Collision Modification Factor or Study Finding
Generally speaking, quantitative estimates of the effectiveness of a treatment may be developed from:
CMF is a term that is widely used in road safety engineering. It may also be referred to as an accident modification factor (AMF). A CMF is the ratio of expected collision frequency at a location with a treatment divided by the expected collision frequency at the location without the treatment. If the expected collision frequency with a treatment is 9 and the expected collision frequency without the treatment is 12, then the CMF is 9/12 = 0.75.
Traffic engineers should be careful not to confuse the term CMF with CRF, which stands for collision reduction factors. A CRF is the portion of collisions that will be reduced if a treatment is applied. The CRF is easily determined, being 1 minus the CMF value. Using the above example, if the CMF is 0.75, then the CRF is 1 – 0.75 = 0.25. The expected reduction in collisions that would come from application of the treatment would be 0.25 x 12 = 3.
Many State jurisdictions have developed reference lists of CMFs to help them choose an appropriate treatment for an intersection improvement plan. In some cases, very little or no documentation exists showing how these CMFs were derived. Some state authorities are currently using CMFs developed from in-house projects; others use CMFs developed by other transportation authorities or based on published research.
Part III of this guide reports study findings taken from a variety of sources. These findings either reported a change in collision frequency or rate as part of a cross-sectional study, a before-after study, or by more sophisticated methods. Each study finding was reviewed in terms of:
In general, findings were discarded that appeared unreasonable, were outdated, used overly simplistic methods, or were based on research carried out outside of North America, unless no other finding was available for the treatment in question. The results are presented as the expected change in collision frequency (and are expressed as a percentage). A study finding of 50 percent means that there is expected to be a reduction of 50 percent in the number of collisions occurring after the application of the treatment the study finding describes. Each CMF or study finding in part III of this guide is referenced. In applying a CMF or finding was to determine the expected outcome of implementing a treatment, the user is urged to review the source material from which the CMF or study finding was derived, to determine its applicability to his or her specific project. Readers may wish to use their own CMFs or the results of another study finding known to them, should they believe that it is more accurate or better reflects conditions occurring at the location in question.
6.4.4 Quantify Safety Benefit
The target benefit of any treatment is a reduction in the frequency or severity of collisions. Assumptions regarding potential benefit of a treatment must be realistic. The collision frequency (or collision frequency of a specific group of collisions) cannot be driven below zero. To quantify the safety benefit of implementing a treatment, the estimated collision reduction that will be connected with the implementation of the treatment must be determined. If a treatment is successful in eliminating or reducing the severity of collisions that would have been expected without the treatment, then the benefits can be attributed to the treatment.
When two treatments are considered and each has a quantifiable safety benefit, a common way to express the combined safety benefit is to multiply both values. For example, treatment a might have a CMF of 0.90, and treatment B might have a CMF of 0.80. Combined, the two treatments should have an expected benefit of (CMF A (0.90) x CMF B (0.80)) of 0.72.
Usually, treatments can only be expected to be successful when applied to a particular target group of collisions. For example, the installation of protected left-turn phasing on one approach should substantially reduce left-turn collisions involving that particular approach, but cannot be expected to affect left-turn collisions on any other approach.
Treatments can also have undesirable effects that need to be considered in evaluating their overall benefit. For example, the installation of right-turn channelization may reduce collisions involving right-turning vehicles and possibly rear-end collisions on a particular approach, but may increase collisions involving pedestrians. If the treatment is to be applied, both positive and negative consequences need to be considered.
The potential collision reduction from a treatment is determined by multiplying the expected number of collisions by the percentage reduction that the treatment is expected to have. The expected number of collisions (total or by severity) may be assumed to be the same as in the period before the treatment, but a much more refined method would be to develop an estimate of the expected number of collisions based on SPF curves or the EB method.
Placing an economic value on collisions, by severity, is a common practice in quantifying the safety benefits of a treatment. There are several ways of arriving at societal cost (such figures are available from FHWA and various state transportation agencies).
Calculating the safety benefit of a treatment means multiplying the expected collision reduction by severity (property damage, injury, and fatal) by applicable society cost figures. A means of expressing the calculation of the safety benefit of the treatment is as follows:
The end result of the above calculations will be a list of treatments with associated societal benefits.
As an example: A multilane signalized intersection has been diagnosed as having a safety problem associated with a particular approach. Adding a right-turn lane is being considered as a possible treatment. Calculation of the safety benefit involves determining the product of the yearly average number of collisions, the societal benefit, and the estimated reduction in collisions grouped by collision type (table 32). The total societal benefit is calculated to be $66,000.
Table 32. Example calculation of safety benefit of adding a right-turn lane.
Source for estimated reductions: (68)
In certain situations, no CMF or study findings will be available for a particular treatment. A qualitative assessment of safety risk at the considered treatment may be undertaken. Safety risk is used to determine the relative severity of each issue or problem, and is a function of the exposure of the different road users; the probability of a crash occurring under the geometric, environmental, and traffic characteristics; and the likely consequences of a crash. This concept is further explored in a 2002 paper by de Leur and Sayed.(69)
Exposure. Exposure typically is measured in terms of traffic volume, including passenger vehicles, trucks, pedestrians, and bicyclists. Exposure is also expressed in distance traveled or in time spent on the roadway. With increased exposure, the more a person is involved in road traffic, the more likely it is that the person will be involved in a collision.
Probability. The probability of a crash measures the degree of certainty that a particular event would occur. Probability is dependent upon design parameters, traffic operations, time periods, environmental conditions, and traffic characteristics.
Consequence. The consequence refers to the severity of an injury sustained by a person involved in an accident. Severity is measured in terms of property damage collisions, injuries, or fatalities.
6.4.5 Selected treatment(s)
As a result of the above, the traffic engineer will have selected a set of treatments, some having an associated dollar value indicating the societal benefit in terms of collision avoided. These treatments should all be carried forward for the improvement plan development discussed in the next section.
6.5 Improvement Plan Development
The stages in the improvement plan development are:
Findings from the safety analysis should be carried forward in coordination with the findings of the operational analysis; the two components together can be carried forward into an improvement plan.
Treatments that have quantifiable safety benefits, based on the results of the numerical analysis, may be considered with some degree of confidence using a cost-benefit analysis. The cost of implementing the treatment should be considered in terms of construction, operating, and maintenance costs, if applicable, and then compared to the benefits in terms of the societal cost of collisions avoided.
A treatment with no quantifiable safety benefit also can be evaluated in a cost-benefit analysis, but the results should be treated with caution. The study team should recognize the reduced confidence in a treatment’s ability to effect an improvement in safety. The following questions may help the team decide whether to implement such a treatment:
In developing an improvement plan, the study team will need to refer to the findings of the operational analysis. The operational analysis provides recommendations that address the identified operational problems occurring at the intersection. These recommendations may be in agreement with the recommended treatments of the safety analysis.
Keywords: Signalized Intersections, Intersection Safety, Intersection Design, Intersection Performance, Intersection Treatments
TRT Terms: Roads–Interchanges and intersections, Roads–Interchanges and intersections–Safety measures, Traffic signs and signals, Signalized intersections, State of the art reports, Best practices