Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
123  This appendix provides an overview of quantitative safety performance, explains how quan- titative safety analysis can inform decisions, introduces the general methods for quantifying safety performance, and presents a decision process for selecting an appropriate method. Introduction to Quantitative Safety Analysis One of the most basic components of evidence-based safety is quantifying the safety per- formance of an existing or planned facility in terms of the estimated frequency and severity of crashes. Whereas traditional approaches to quantitative safety performance rely on the short- term average crash history to represent the safety performance of a given location, more reliable approaches to quantitative safety performance focus on long-term crash frequency and severity, incorporating the crash history for the location of interest when applicable, but also incorpo- rating crash history from other similar locations and adjusting for changes in traffic volume and other factors that influence safety over time. Quantifying safety performance supports decisions throughout the project development pro- cess from planning and design to operations and maintenance. Agencies can use the methods presented in this guide to estimate and compare safety performance with and without a given access management strategy or to select a preferred alternative based on the estimated safety performance of multiple alternatives. For example, transportation planners could use these methods to quantify and compare the safety impacts of different median alternatives alongside other factors such as operational and environmental impacts. Permitting departments could use these methods to quantify and consider the safety impacts of adding one or more driveways along a roadway segment as well as locating driveways with respect to the functional area of an adjacent intersection. Highway designers could use these methods to quantify and compare the safety performance of a median opening with and without a turn lane. Overview of Approaches to Quantify Safety Performance In preparing to quantify the safety performance of a given location, it is important to define the area of interest, identify the study period, determine the scenario of interest, and establish the performance measure of interest because these factors will impact data requirements and the analysis method. The location may be defined as an individual roadway segment or intersection or as an entire corridor or network. The study period is typically defined in terms of years and may include a single year or multiple years. The scenario of interest could include existing or expected future conditions. The following are examples of scenarios of interest: ⢠Existing roadway and traffic conditions, ⢠Existing roadway conditions under future traffic conditions, or A P P E N D I X B Overview of Quantitative Safety Analysis
124 Application of Crash Modification Factors for Access Management ⢠A proposed alternative that incorporates some changes to the roadway geometrics and/or operations. The typical measure of safety performance is the estimated number of crashes, but this could be in terms of specific crash types or severities. The crash type defines the manner of collision (e.g., right-angle, left-turn, rear-end, head-on, and run-off-the-road). The crash severity defines the greatest level of injury in a crash and may be categorized as fatal, injury, and property damage only (PDO) or by the KABCO scale. The Model Minimum Uniform Crash Criteria (MMUCC) 5th edition provides the following definitions for the KABCO scale (NHTSA 2017): ⢠K (fatal crash): fatal injury where death occurs within 30 days after the motor vehicle crash in which the injury occurred. ⢠A (suspected serious injury): any injury other than fatal which results in one or more of the following: severe laceration resulting in exposure of underlying tissues/muscle/organs or resulting in significant loss of blood; broken or distorted extremity (arm or leg); crush injuries; suspected skull, chest, or abdominal injury other than bruises or minor lacerations; significant burns (second and third degree burns over 10% or more of the body); uncon- sciousness when taken from the crash scene; or paralysis. ⢠B (suspected minor injury): any injury that is evident at the scene of the crash, other than fatal or suspected serious injuries. Examples include lump on the head, abrasions, bruises, minor lacerations (i.e., cuts on the skin surface with minimal bleeding and no exposure of deeper tissue/muscle). ⢠C (possible injury): any injury reported or claimed which is not a fatal, suspected serious, or suspected minor injury. Examples include momentary loss of consciousness, claim of injury, limping, or complaint of pain or nausea. Possible injuries are those which are reported by the person or are indicated by his/her behavior, but no wounds or injuries are readily evident. ⢠O (no apparent injury): a situation where there is no reason to believe that the person received any bodily harm from the motor vehicle crash. There is no physical evidence of injury and the person does not report any change in normal function. The Highway Safety Manual (1st Edition) Part C Predictive Method (AASHTO 2010) pro- vides an estimate of the expected (long-term average) crash frequency. The expected crash fre- quency is based on a weighted average of observed and predicted crashes. The following sections introduce three terms related to the Part C Predictive Method: observed crash frequency, pre- dicted crash frequency, and expected crash frequency. Observed Crashes Observed crashes are those reported and documented at a site of interest. The observed crash frequency is the number of reported crashes at a specific location during a defined time period. To estimate the current or future safety performance using the observed crashes, there is a need to compute the average observed crash frequency over some past time period. For example, consider the 5-year crash history for a four-legged, signalized intersection on an urban arterial. If there were 3, 1, 5, 4, and 6 observed crashes in the last 5 years, then the average observed crash frequency is 19 divided by 5âor 3.8 crashes per year. Figure B-1 presents four different representations of average observed crash frequency. The solid line in the figure represents the 5-year average of 3.8 observed crashes per year. The dashed lines in the figure represent the 3-year average based on three different 3-year periods. This figure helps to demonstrate the potential limitations of using the short-term observed crash fre- quency as a measure of the long-term estimated safety performance. Specifically, the short-term average is generally not a reliable measure of safety performance because annual fluctuations in crashes can produce misleading results. A longer study period is generally preferred because it
Overview of Quantitative Safety Analysis 125  provides a larger sample of crashes for analysis; however, a longer study period also increases the chances for other changes over time such as safety and operational improvements; natural degradation of surface friction and sign reflectivity; and changes in general land use and traffic patterns. As such, it is important to balance the length of the study period with the potential for other changes over time. Predicted Crashes Predicted crashes are based on a safety performance function (SPF). SPFs are equations that represent the statistical relationship between safety and roadway characteristics for a given facility type. A facility type represents a group of similar segments or intersections, typi- cally defined by geometric and operational characteristics such as traffic control (signalized, stop-controlled, uncontrolled), area type (urban or rural), number of lanes, and number of approaches. SPFs are developed by combining data from several similar locations and devel- oping a statistical relationship between crashes, traffic volume, and possibly other roadway characteristics. For example, the estimated safety performance of an urban, four-legged, sig- nalized intersection would be predicted based on the safety performance of other similar urban, four-legged, signalized intersections. Figure B-2 represents the SPF for multivehicle crashes at four-legged signalized intersections on urban and suburban arterials from the Highway Safety Manual (1st Edition) (AASHTO 2010). The individual points are the observed crashes for several locations from the same facility type, plotted against the respective traffic volume. A nonlinear function is fit through the points, which represents the SPF. The SPF can then be used to predict the crashes for any given traffic volume for the specific facility type. For example, using this SPF, the predicted crash frequency for a facility with 10,000 vehicles per day on the major road and 8,000 vehicles per day on the minor road is approximately 2.5 crashes per year because this is where the line for predicted crashes crosses the traffic volume of 8,000 vehicles per day. SPFs are typically presented as equations as shown in Figure B-3. Figure B-4 shows a sample calculation using the SPF for multivehicle crashes at four-legged signalized intersections on urban and suburban arterials from the Highway Safety Manual (1st Edition) (AASHTO 2010). This SPF requires the major road traffic volume and minor road traffic volume as data inputs. If, for the condition of interest, the major road volume is 10,000 vehicles per day and the minor road volume is 8,000 vehicles per day, then the predicted crash frequency is 2.54 crashes per year. Figure B-1. Examples of average observed crash frequency.
126 Application of Crash Modification Factors for Access Management Figure B-2. SPF for multivehicle crashes at four-legged signalized intersections on urban and suburban arterials with major road traffic volume of 10,000 vehicles per day. Figure B-4. Example SPF calculation. Figure B-3. SPF for multivehicle crashes at four-legged signalized intersections on urban and suburban arterials. Where: SPFTotal Crashes = safety performance function to predict total crashes, AADTmajor = annual average daily traffic on the major road, and AADTminor = annual average daily traffic on the minor road. SPFs represent the average safety performance of sites based on specific locations and time periods. As such, users must calibrate SPFs to reflect current local conditions. Even if trans- portation agencies have developed SPFs based on local data, it is still necessary to calibrate periodically to account for changes over time. The calibration factor is applied as a multiplier to the SPF. Calibration factors greater than 1.0 indicate the SPF is under-predicting crashes for local conditions, and there is a need to inflate the predictions. Calibration factors less than 1.0 indicate the SPF is over-predicting crashes for local conditions, and there is a need to reduce the predic- tions. There are also calibration functions that allow the value of the calibration factor to change with other variables such as predicted crashes or traffic volume. Refer to The Calibrator User Guide (https://safety.fhwa.dot.gov/rsdp/downloads/fhwasa17016updated0618.pdf) for further guidance on how to develop calibration factors and functions (Lyon et al. 2018). Expected Crashes The expected crash frequency is the weighted average of observed and predicted crash fre- quencies using the Empirical Bayes (EB) method. By combining site-specific crash history with the information from other similar locations, it is possible to improve estimates of safety performance and account for potential issues such as random fluctuations in crashes over time. If the observed crash history is not available or applicable, then the weight assigned to the
Overview of Quantitative Safety Analysis 127  observed crashes is zero and the expected number of crashes is based solely on the predicted crashes. If a reliable SPF is not available for the facility type of interest and the crash history is available and applicable, then the weight assigned to the SPF is zero and the expected number of crashes is based solely on the observed crashes. If both the observed and predicted crashes are available and applicable, then the EB method is used to compute a weighted average of the observed and predicted crash frequencies. Figure B-5 represents the EB weighted average of observed and predicted crashes. Figure B-6 represents the weight, w, used in the EB method. In these equations, the observed, predicted, and expected crashes are all in terms of crashes over the same study period. So, if the study period is 5 years, then the observed crashes should be the sum of observed crashes over the 5-year period, the predicted crashes should be the sum of predicted crashes over the 5-year period, and the expected crashes would represent the expected crashes over the 5-year period. Figure B-6. Empirical Bayes weight. Figure B-5. Empirical Bayes method. Where: w = weight used in EB method, and k = dispersion parameter associated with the SPF. The weight is based on the quality of the SPF and the number of years in the study period. The more reliable the SPF, the more weight is placed on the predicted crashes. The more years in the study period, the more weight is placed on the observed crashes. The quality of the SPF is indicated, at least in part, by the dispersion parameter, which is a measure of uncertainty in the SPF. Note that some resources report the dispersion parameter while others report the inverse of the dispersion parameter. In Figure B-6, k is the dispersion parameter where larger values of k indicate a lesser quality SPF and result in a smaller weight, w. This guide is consistent with the Highway Safety Manual (1st Edition), which reports k for the given SPFs. While the disper- sion parameter generally applies to the SPF for base conditions, it approximates the dispersion parameter for the adjusted predicted crashes. The weight is relatively insensitive to the value of k, so this approximation is not a concern. As an example, assume the facility type of interest is a four-legged, signalized inter- section on an urban arterial with a major road traffic volume of 10,000 vehicles per day and a minor road traffic volume of 8,000 vehicles per day. Continuing with the previous examples, the total observed crash frequency was 19 crashes over the 5-year study period and the predicted crash frequency from the SPF was 2.54 crashes per year, which translates to 12.7 predicted crashes over the 5-year study period if the traffic volume remained constant. The value of k is 0.39 for this SPF. The weight, w, is computed as shown, using Figure B-6 and inputting the values for k and the sum of the predicted crashes over the 5-year study period. w = + â =1 1 0.39 12.7 0.17 (continued on next page)
128 Application of Crash Modification Factors for Access Management Crash Modification Factors and Functions Crash modification factors and functions (CMFs) can be applied to observed, predicted, or expected crashes to estimate the change in safety performance associated with changing the geometric or operational conditions of the roadway. The Highway Safety Manual (1st Edition) defines CMFs as multiplicative factors used to compute the long-term average crash frequency after implementing a given strategy at a specific site. Values of CMFs represent the long-term expected change in crashes relative to a set of base conditions. Under the base conditions, the value of the CMF is 1.0. A CMF of 1.0 indicates no expected change in crashes. A CMF less than 1.0 indicates an expected reduction in crashes, and a CMF greater than 1.0 indicates an expected increase in crashes. A crash modification function is a formula used to compute the CMF for a specific site based on its characteristics. It allows the CMF to change over the range of a variable or a combination of variables. As an example, Figure B-7 provides the crash modification function for changing the signalized intersection spacing along an urban corridor that is divided by a median (Mauga and Kaseko 2010). This function shows that as the signal spacing increases from the base condi- tion, the CMF decreases, indicating a lower crash frequency. Note the resulting CMF applies to corridor-level analysis, which is covered in Chapter 5. The expected crash frequency is computed using the EB method from Figure B-5. The result is an expected crash frequency of 17.93 crashes over the 5-year period, or 3.59 crashes per year. ( )= â + â â =Expected Crashes 0.17 12.7 1 0.17 19 17.93 crashes Figure B-7. Crash modification function for changing signal spacing. Where: X = signal spacing (in thousands of feet) for the base condition, and Y = signal spacing (in thousands of feet) for the condition of interest. The next two subsections provide details on how to select and apply CMFs. Selecting CMFs Identifying the most appropriate CMF requires the analyst to consider several factors related to the applicability, quality, and availability of CMFs. CMF Applicability. CMF applicability is the first and most important factor considered in the process. The applicability of a CMF is based on factors that describe the facility type, crash type, and crash severity for which it was developed. The user should examine the differences between the characteristics of the sites used to develop the CMF and the subject site. In an ideal situation, the CMFs would have been developed using data from sites that are identical to the subject site. In reality, the CMFs identified for the proposed strategy will often differ from the
Overview of Quantitative Safety Analysis 129  subject site on at least a few, if not many, characteristics. To reasonably assume the CMF value is applicable to the subject site, it should match the subject site on site characteristics that are known to have a statistically significant influence on the value of the CMF such as traffic volume, number of intersection approaches, and number of lanes. Factors that describe CMF applicability are often found in the CMF Clearinghouse (www. cmfclearinghouse.org) and the original research report. Sometimes this information is clear (e.g., it is within the name of the strategy). For example, if the name of the strategy is âinstall rumble strips on horizontal curve,â then it is apparent the CMF applies to horizontal curves and not all roadway sections. Sometimes it is necessary to review the details from the CMF Clearing- house or the original research report to understand the CMF applicability. CMF Quality. CMF quality can be used as a secondary consideration when multiple appli- cable CMFs exist for the strategy of interest, and agencies may choose to consider only the higher- quality CMFs. CMFs are characterized by quality in terms of the data and statistical methods used to develop the CMF. In general, higher-quality CMFs control for statistical biases, have larger sample sizes, and have smaller standard errors. Using higher-quality CMFs will result in better decision-making. Lower-quality CMFs may suffer from potential issues such as regression-to- the-mean, misleading representations of strategy effectiveness, or large confidence intervals, all of which may lead to overestimating the strategy effect, which can affect related decisions. The CMF Clearinghouse provides a star quality rating to indicate the relative quality of CMFs. The CMF is rated on a scale of one to five stars, with five stars indicating the highest-quality CMFs. CMF Availability. The primary source of CMFs is the CMF Clearinghouse, which serves as a web-based repository of CMFs compiled from research studies. Analysts may use the CMF Clearinghouse and apply the previous guidance to search for suitable CMFs. The CMF Clearing- house provides tutorials and instructional videos on how to search for CMFs using basic and advanced search functions. Another Source of CMFs. Another source of CMFs is the Highway Safety Manual (1st Edi- tion); however, all of these CMFs are also included in the CMF Clearinghouse. Note there are two types of CMFs in the Highway Safety Manual (1st Edition): CMFs for the Part C Predictive Method and CMFs for individual strategies in Part D. The Highway Safety Manual (2nd Editionâ forthcoming) does not provide any CMFs. Instead, it provides adjustment factors (not CMFs) in the Part C Predictive Method and information on identifying, selecting, and applying CMFs in Part D. Further, the Highway Safety Manual (2nd Editionâforthcoming) establishes a meth- odology for assessing and approving CMFs. CMFs that do not meet the Highway Safety Manual inclusion criteria may still be appropriate for use in practice; however, users should understand the potential limitations and biases associated with the CMF. Applying CMFs CMFs are applied as multiplicative factors. To estimate the crashes for the condition with the strategy, multiply the estimated crashes for the base condition by the CMF, as shown in Figure B-8. The observed, predicted, or expected crash measures could be used to estimate the crashes for the base condition as described previously. The next section will explain when each method is appropriate. These methods are applicable when estimating the safety impacts of individual access management strategies (see Chapter 3) as well as when estimating the safety performance of segments or intersections (see Chapter 4). Figure B-8. Applying CMF to estimate change in crashes.
130 Application of Crash Modification Factors for Access Management CMFs should only be applied within the context for which they were developed. For example, Table B-1 presents CMFs related to corner clearance where the base condition is no driveways present within 50Â feet of the downstream corner. Specifically, the CMFs represent the effect of the presence of a driveway within 50Â feet of the downstream corner of a four- legged signalized intersection compared to no driveways present. The context of the roadway is a four-legged, signalized intersection within the specified range of traffic volumes. If the base condition for the scenario of interest differs from these conditions, then the analyst should consider using a CMF with a base condition that better matches the conditions of interest. For example, if the intersection of interest is a three-legged signalized intersection, then these CMFs may not be the most appropriate. CMFs also apply to specific crash types and severities. As shown in Table B-1, one CMF applies to total crashes (all crash types and all crash severities) while the others apply to specific crash types and severities. The key takeaway is that CMFs should only be applied to the applicable crash types and severities. For example, it would be appropriate to use the CMF for rear-end crashes to estimate the change in rear-end crashes; however, CMFs can change by crash type and severity. As such, it would not be appropriate to use the CMF for all crashes to estimate the change in an individual crash type or crash severity. Example: An agency is considering the installation of left-turn lanes on both major road approaches of a four-legged, signalized intersection on an urban arterial with a major road traffic volume of 10,000 vehicles per day and a minor road traffic volume of 8,000Â vehicles per day. Assume the expected crash frequency is 3.59 crashes per year under base conditions. From the Highway Safety Manual (1st Edition), the CMF for installing left-turn lanes on two approaches is 0.81 (AASHTO 2010). The computation to estimate the crash frequency with the left- turn lanes is based on Figure B-8. The result is an expected crash frequency of 2.91 crashes per year with the left-turn lanes, a reduction of 0.68 crashes per year compared to the base condition of no left-turn lanes. = â =Estimated Crashes with Strategy 0.81 3.59 crashes year 2.91crashes year Intersection Type (Applicability) AADT Crash Type (Severity) CMF CMF Clearinghouse ID (Le et al. 2018) 4-legged signalized (presence of driveway on 1 downstream corner) Major road: 10,406 to 93,000 Minor road: 500 to 48,000 All (All) 1.33 9736 4-legged signalized (presence of driveway on 1 downstream corner) Major road: 10,406 to 93,000 Minor road: 500 to 48,000 All (KABC) 1.29 9740 4-legged signalized (presence of driveway on 1 downstream corner) Major road: 10,406 to 93,000 Minor road: 500 to 48,000 Rear-end (All) 1.36 9744 4-legged signalized (presence of driveway on 1 downstream corner) Major road: 10,406 to 93,000 Minor road: 500 to 48,000 Left-turn, Right-turn (All) 1.22 9756 Base condition: 4-legged signalized intersection with no driveways present within 50 feet of the downstream corner. Table B-1. CMFs for driveway presence on downstream corner within 50Â feet of signalized intersection.
Overview of Quantitative Safety Analysis 131  In some cases, multiple strategies are applied to a single location, such as signalizing a stop- controlled intersection, adding turn lanes, and relocating or consolidating driveways. Each of these strategies may have a separate CMF, or a single CMF may represent the safety impact of the combined strategies. The preferred approach is to estimate the combined safety effect of multiple strategies with a single CMF that represents the combined effect. In the absence of such a CMF, practitioners may need to apply separate CMFs representing the individual strategy effects to estimate the combined safety effect. A single method for estimating the combined effect of multiple strategies is not appropriate in all scenarios. Refer to the following FHWA training videos for guidance on estimating the combined effects of multiple strategies: 1. Selecting a Method to Analyze Multiple CMFs (https://www.youtube.com/watch?v= OPvAjUpT6Dg) 2. Applying a Method to Analyze Multiple CMFs (https://www.youtube.com/watch?v= 48M7TBKTCM0) In general, the following abbreviated guidance applies when the CMFs for two strategies apply to the same crash types and severities: 1. When one or both CMFs are greater than 1.0, the multiplicative method is appropriate. 2. When both CMFs are less than 1.0, consider the specific case representing the potential for overlap: a. For zero overlap in strategy effects, the additive method is appropriate, which assumes the full effect of both strategies with a maximum reduction of 100 percent (or a CMF of 0). b. For some overlap in strategy effects, the dominant effect or dominant common residuals method may be appropriate, whichever produces the largest combined reduction (or smallest CMF). c. For complete overlap in strategy effects, the dominant effect method is appropriate because it only considers the most effective strategy. There is an opportunity to extend these methods to estimate the joint safety effect of more than two strategies; however, there is a lack of research to verify the accuracy of combining more than two individual CMFs. Selecting an Appropriate Approach to Quantify Safety Performance This section provides guidance to help decide when it is appropriate to use observed, predicted, and expected crashes as the basis for analysis; however, it is also important to apply judgment and select the approach that is most appropriate for the scenario at hand. Each approach has specific strengths and limitations, and these approaches differ in terms of effort, reliability, and data requirements. In some cases, data availability, decision complexity, and time constraints may limit the available options. Observed Crashes Table B-2 provides a summary of the strengths and limitations of using observed crash fre- quency to estimate long-term safety performance. The observed crash frequency approach is relatively simple, requires less data than predicted and expected crash frequency, and considers the historical crash data for the location of interest. The reliability of this approach improves with a smaller degree of fluctuation in annual observed crashes. One way to reduce the vari- ability in annual observed crashes is to increase the sample size (i.e., use a larger sample of observed crashes). The primary limitation is that it does not account for potential changes that affect safety performance at the location of interest from year to year. For example, it does not account for changes in traffic, weather, crash reporting, road conditions, land use, vehicle fleet,
132 Application of Crash Modification Factors for Access Management or driver behavior, all of which could impact the safety performance of the facility. It solely uses past site crash history as the basis for estimating future safety performance. The observed crash frequency approach assumes the historical observed crashes represent the future safety performance in the absence of any changes to the site. As such, it may not be reliable when the site characteristics change drastically from the past condition to the present or future condition of interest. However, if historical crash trends are relatively stable (minimal annual fluctuations in crashes) or if analysis skill or time is limited, then it might be appropriate to use observed crashes for quantifying safety performance. One way to assess the stability of annual crashes is to plot the crashes over time and visually inspect for trends and fluctuations, as shown in the following example. For a more formal statistical test, refer to Hauerâs work, which suggests a process to determine whether there is a change in safety performance over time at a particular site (Hauer 1996). Strengths Limitations Relatively simple Requires less data than predicted and expected crash frequency Considers historical crash data for location of interest Does not account for annual fluctuations in observed crashes Does not account for changes in ⢠Traffic ⢠Weather ⢠Crash reporting ⢠Road conditions ⢠Land use ⢠Vehicle fleet ⢠Driver behavior Table B-2. Strengths and limitations of observed crashes. Example: Considering the observed crash history shown in Figure B-9, what is a reasonable estimate of the crash frequency for the present or future year? a. 3 crashes per year (5-year average) b. 4 crashes per year (3-year average) c. More than 5 crashes per year d. None of the above Figure B-9. Observed crash history for an example scenario at a specific site.
Overview of Quantitative Safety Analysis 133  Predicted Crashes Table B-3 provides a summary of the strengths and limitations of using predicted crash fre- quency to estimate long-term safety performance. This approach assumes the SPF prediction is representative of future safety performance. While it accounts for annual fluctuations in observed crashes as well as changes in traffic volume, roadway characteristics, and general time trends (e.g., changes in weather, crash reporting, land use, vehicle fleet, and driver behavior), it does not consider historical crash data for the location of interest. Therefore, this approach is appropriate when the observed crash history is not informative of future conditions, which gen- erally occurs when there are major changes to the land use, traffic operations, or facility type. For example, if the existing condition is a two-lane road and the future condition of interest is a four- lane road, then the historical observed crashes under the two-lane condition probably do not reflect the safety performance under the future four-lane condition. In this case, the predicted crash frequency is generally appropriate. Using predicted crashes to quantify safety performance could also be appropriate when comparing design alternatives; however, this approach requires calibration of the SPFs to local conditions. Expected Crashes Table B-4 provides a summary of the strengths and limitations of using expected crash fre- quency to estimate long-term safety performance. The expected crash frequency combines the observed and predicted crash frequency using the EB method. This approach accounts for annual fluctuations in observed crashes as well as changes in traffic volume, roadway geometrics, and general time trends (e.g., changes in weather, crash reporting, land use, vehicle fleet, and driver behavior). The expected crash frequency approach assumes the combined information from observed and predicted crashes is representative of future safety performance. As such, it may not be reliable when the site characteristics change drastically from the past condition to the present or future condition of interest. It also requires a calibrated SPF and expertise in applying the EB method. A reasonable estimate for the present or future crash frequency is likely âcâ (more than 5 crashes per year) because there is a clear trend showing an increase in observed crashes. This underlying trend could be due to several factors, but the important point is that the average observed crash frequency, using either 3 or 5 years of observed crash history, does not capture this trend. This shows one limitation of only using average observed crashes to quantify the safety performance at a location. Strengths Limitations Accounts for annual fluctuations in observed crashes Accounts for changes in ⢠Traffic volume ⢠Roadway characteristics ⢠General time trends (e.g., changes in weather, crash reporting, land use, vehicle fleet, and driver behavior) Does not require historical crash data for location of interest Requires calibrated SPF Does not consider historical crash data for location of interest Table B-3. Strengths and limitations of predicted crashes.
134 Application of Crash Modification Factors for Access Management Decision Process It is important to understand the strengths and limitations of these approaches because the reliability of the approach will impact the reliability of analysis results and decision-making. In selecting an appropriate analysis approach to estimate safety performance, there is a need to consider the level of rigor, scope of the project, and availability of data. Other considerations in selecting an analysis method include 1. Annual fluctuation in observed crashes; 2. Changes in geometry, traffic control, or traffic volume; and 3. SPF availability and reliability. Table B-5 provides an overview of the characteristics associated with observed, predicted, and expected crashes to help determine when each is (or is not) appropriate. As indicated in Table B-5, if there is a high degree of annual fluctuation in the observed crashes or changes in traffic volume, then it may not be appropriate to use observed crash fre- quency as this can lead to inaccurate estimates of long-term safety performance. If there is a need to account for annual fluctuation in observed crashes or changes in traffic volume, then the predicted or expected crashes are more appropriate. The predicted and expected crashes can help to normalize annual fluctuations in crashes and account for changes in traffic volume when a calibrated SPF is available. The following list provides general guidance when selecting an approach to quantify safety performance: 1. Expected crashes are typically more reliable than observed or predicted crashes when site char- acteristics remain relatively constant from the past condition to the present or future condition of interest. Strengths Limitations Accounts for annual fluctuations in observed crashes Accounts for changes in ⢠Traffic volume ⢠Roadway characteristics ⢠General time trends (e.g., changes in weather, crash reporting, land use, vehicle fleet, and driver behavior) Considers historical crash data for location of interest Requires calibrated SPF Does not account for major changes from the base condition to the condition with the strategy of interest Requires expertise to use EB method Table B-4. Strengths and limitations of expected crashes. Characteristics Observed Crash Frequency Predicted Crash Frequency Expected Crash Frequency Appropriate with sample of observed crashes, small degree of fluctuation in annual observed crashes, and consistent site characteristics over time â â â Appropriate to normalize annual crash fluctuations -- â â Applicable when facility type changes -- â -- Requires calibrated SPF -- â â Accounts for traffic volume changes -- â â Considers historical crash data for location of interest â -- â Note: ⢠indicates potentially appropriate; -- indicates not appropriate Table B-5. Summary of approaches for estimating safety performance.
Overview of Quantitative Safety Analysis 135  2. Predicted crashes are typically more reliable than observed or expected crashes if the site characteristics change drastically from the past condition to the present or future condition of interest. The predicted crashes may also be preferred when there are only 1 or 2 years of observed crashes, which can lead to a higher degree of annual fluctuation in observed crash frequency. 3. Observed crashes may provide a reasonable estimate when at least 5 years of historical crash data exist, and year-to-year fluctuations in crashes are limited. In addition to the general guidance provided, Table B-6 offers additional scenarios to help select an appropriate analysis method. Scenario Question Answer Action 1: Estimate safety performance of existing roadway and traffic conditions 1.1: Is a crash history available to estimate future crashes without treatment? Yes Go to Question 1.2 No Use predicted crashes from calibrated SPF and apply adjustment factors (CMFs) as necessary; otherwise, it is not possible to estimate safety performance 1.2: Is a calibrated SPF available to estimate predicted crashes for the facility type of interest? Yes Go to Question 1.3 No Use observed crashes; otherwise, it is not possible to estimate safety performance 1.3: Does the SPF reflect existing design operations and access conditions? Yes Use expected crashes No Use expected crashes and apply adjustment factors (CMFs) as necessary 2: Estimate safety performance of existing roadway conditions under future traffic conditions 2.1: Is a crash history available to estimate future crashes without treatment? Yes Go to Question 2.2 No Use predicted crashes from calibrated SPF and apply adjustment factors (CMFs) as necessary; otherwise, it is not possible to estimate safety performance 2.2: Is a calibrated SPF available to estimate predicted crashes for the facility type of interest? Yes Go to Question 2.3 No Use observed crashes and apply ratio of future traffic to current traffic to adjust for expected changes in traffic; otherwise, it is not possible to estimate safety performance 2.3: Does the SPF reflect existing design, operations, and access conditions? Yes Use expected crashes and apply ratio of predicted crashes for future traffic to predicted crashes for current traffic to adjust for expected changes in traffic No Use expected crashes, apply ratio of predicted crashes for future traffic to predicted crashes for current traffic to adjust for expected changes in traffic, and apply adjustment factors (CMFs) as necessary 3: Estimate safety performance of proposed alternative that incorporates some changes to the roadway geometrics and/or operations 3.1: Is a crash history available and applicable to estimate future crashes without treatment (i.e., no major changes to facility type)? Yes Go to Question 3.2 No Use predicted crashes from calibrated SPF and apply adjustment factors (CMFs) as necessary; otherwise, it is not possible to estimate safety performance 3.2: Is a calibrated SPF available to estimate predicted crashes for the proposed alternatives associated with the facility type of interest? Yes Use expected crashes and apply adjustment factors (CMFs) to adjust safety performance estimate to reflect proposed alternative No If crash history is available and applicable to estimate future crashes without treatment (i.e., no major changes to facility type), then use observed crashes, apply ratio of future traffic to current traffic to adjust for expected changes in traffic, and apply CMFs to adjust for proposed alternative; otherwise, it is not possible to estimate safety performance Table B-6. Decision process to select appropriate method.
136 Application of Crash Modification Factors for Access Management Summary Quantitative safety performance can help inform decision-making processes. Safety perfor- mance is defined by the estimated frequency and severity of crashes. While multiple methods exist to estimate safety performance, there is a need to consider the strengths and limitations in selecting an appropriate method. Method selection is contingent on the availability and quality of historical crash data, the expected change in site characteristics, expertise, and time con- straints. The three methods for quantifying safety performance are listed below with a summary of each: ⢠Observed Crashes: Historical crash data (observed crashes) may provide a reasonable esti- mate of safety performance when past conditions are representative of present or future conditions, at least 5 years of crash data exist, and year-to-year fluctuation in crashes is limited. However, the use of observed crashes does not account for annual crash fluctuations or changes in traffic, weather, road conditions, land use, vehicle fleet, and driver behavior. ⢠Predicted Crashes: Predicted crashes from SPFs provide a reasonable estimate of safety per- formance when the site characteristics are expected to change drastically from the past condi- tion to the present or future condition of interest. Predicted crashes can account for annual fluctuations in observed crashes and changes in traffic volume, roadway characteristics, and general time trends. While the use of predicted crashes does not require historical crash data, it does require a calibrated SPF. Example: Consider a scenario where there is no observed crash history for the location of interest, there are minor changes in geometric and traffic control characteristics, and there is a reliable, calibrated SPF for the facility type of interest. What is an appropriate method to estimate the long-term safety performance of the site? (a) Observed crash frequency (b) Predicted crash frequency (c) Expected crash frequency In this case, the appropriate answer is âb,â predicted crash frequency, because a reliable SPF is available and both the observed and expected crash frequency require historical crash data. Example: Consider the same scenario, but at least 5 years of historical crash data exist, and fluctuation in crashes from year to year is limited. What is an appropriate method to estimate the long-term safety performance of the site? (a) Observed crash frequency (b) Predicted crash frequency (c) Expected crash frequency In this case, the most appropriate answer is âc,â expected crash frequency. While the expected crash frequency would provide a more reliable estimate, any of the methods would provide a reasonable estimate. The tradeoff between the reliability of the approach and the resources needed to conduct the analysis should be considered. The following scenarios provide examples of selecting a method to estimate the long-term safety performance of a site.
Overview of Quantitative Safety Analysis 137  ⢠Expected Crashes: Expected crashes are based on a weighted average of the observed and pre- dicted crashes through use of the EB method. It accounts for annual fluctuations in observed crashes and changes in traffic volume, roadway characteristics, and general time trends, while also considering historical crash data. While the use of expected crashes can account for changes over time, it may not be appropriate if site characteristics are expected to change significantly. This method also requires historical crash data and a calibrated SPF. CMFs can be applied in any of the three methods for quantifying the safety performance of specific conditions with varying degrees of reliability. Specifically, the observed, predicted, or expected crashes would be used to estimate the safety performance of the base condition. The estimate of crashes for the base condition is then multiplied by the CMF to estimate the expected change in safety performance associated with changing the geometric or operational conditions of the roadway. While CMFs can be applied to any of the three methods, the expected or pre- dicted crashes are typically more reliable because the use of observed crashes does not account for annual crash fluctuations or changes in traffic, weather, road conditions, land use, vehicle fleet, and driver behavior. CMFs also apply to specific conditions, crash types, and severities. As such, CMFs should only be applied within the context for which they were developed and to the applicable crash types and severities.
