The Frequently Asked Questions are based on the knowledge and experience of the developers of the HSM as well as what we have learned so far about implementing the Program. If the information you need is not here, go to the User Discussion Forum to ask your question.

General Information

What Is the Highway Safety Manual, and Why Was It Developed?
The Highway Safety Manual (HSM) provides practitioners with information and tools to consider safety when making decisions related to design and operation of roadways. The HSM assists practitioners in selecting countermeasures and prioritizing projects, comparing alternatives, and quantifying and predicting the safety performance of roadway elements considered in planning, design, construction, maintenance, and operation. Prior to the HSM, there was no widely accepted tool available to quantitatively assess the impact of infrastructure decisions on safety.

Is the HSM Required?
Each state department of transportation can set its own policy related to use of the manual, if desired. The Federal Highway Administration does not require use of the Highway Safety Manual. The HSM is a tool to help practitioners perform data-driven safety analyses of roads, and is not a standard or a requirement.

How Is the Highway Safety Manual Related to the AASHTO Green Book and Roadside Design Guide?
The Highway Safety Manual provides information and tools for incorporating data-driven consideration of safety into the project planning and development process. The manual allows for determining the impacts of design and other decisions on the expected safety performance of a facility. The AASHTO Policy on Geometric Design of Highways and Streets (the "Green Book") and Roadside Design Guide are publications that present current information on design and operating practices that are in universal use in the United States. Where these publications present recommended ranges of values for given elements in the roadway or roadside environment, the HSM allows for determining the expected safety impact of using a specific value over another value.

Where Can I Find an Overview of the HSM?
The Highway Safety Manual website, www.highwaysafetymanual.org, has an Introduction or "primer" on the HSM, as well as a fact sheet and brochure.

Where Can I Find Information on Training?
The Training page of the HSM website has information on HSM courses currently available and under development. The National Highway Institute of FHWA has several courses related to the HSM. State departments of transportation can contact their FHWA Division Office for training assistance.

Can I Download the FHWA Webinar Recordings?
The recordings of the FHWA webinar series are posted on the HSM web site. If you're unable to view these recordings online, send a message to info@highwaysafetymanual.org for assistance.

Where Can I Find Other HSM Users?
There is an online User Discussion Forum that provides a place for HSM users to ask questions to other practitioners or to provide information on their experiences. This forum is monitored to ensure questions are answered if no other users are able to respond. You can view posts to this forum and responses without being a member. Additional information is on the User Discussion Forum page on the HSM website.

The Transportation Research Board Highway Safety Performance Committee deals with quantitative highway safety information to support inclusion of safety in decisions at all points in the project development process. This committee has a web site for HSM research and other information it develops and promotes. The committee meets during the annual TRB meeting in January and also holds a mid-year meeting; these meetings are open to all.

Where can I get help with the manual?
You can send questions on the Manual to info@highwaysafetymanual.org or you can post questions on the User Discussion Forum. Questions from both sources are passed along to technical experts for answers if needed. You can also browse the discussion forum to see if a similar question has been posted. As new technical resources are developed, they will be posted on the HSM website on the Implementation page.

Where do I find/send potential errata?
All Errata are posted on the HSM web site. Please submit information on potential errors in the HSM to info@highwaysafetymanual.org, or post the information on the User Discussion Forum.

How do I suggest additions to the HSM?
If you have suggestions on material to add to the HSM, please send a message to info@highwaysafetymanual.org, or post your suggestions on the User Discussion Forum.

When will the next edition of the manual be published?
It will be several years until publication of the second full edition of the HSM. The AASHTO Subcommittee on Safety Management Task Group on Technical Publications and the TRB Highway Safety Performance Committee have worked together closely on a workplan for the second edition, developed under an NCHRP project. This work plan helps to prioritize research needed for future editions.

NCHRP Project 17-45, Enhanced Safety Prediction Methodology and Analysis Tool for Freeways and Interchanges, developed predictive models for freeways and interchanges similar to those found in Part C of the HSM for other facility types. AASHTO is preparing these materials for publication as an interim volume of the HSM. A project (NCHRP 17-58) to develop predictive models for six-lane urban and suburban arterials and one-way arterials is underway and will be completed in 2016. Click here for more information.

FHWA's Crash Modification Factor Clearinghouse will periodically post new CMFs – new research is reviewed quarterly to identify CMFs for posting. You can also submit CMFs to be added to the database on the Clearinghouse website at www.cmfclearinghouse.org. The Clearinghouse contains information on the relationship of the CMFs on the website to those published in the HSM.

Can a State Department of Transportation Use Federal Highway Safety Improvement Program Funds to Purchase The HSM?
FHWA has provided this information related to use of HSIP funds for the HSM:

Since the publication of the Highway Safety Manual (HSM), the FHWA Office of Safety has received a few inquiries concerning the eligibility of Highway Safety Improvement Program (HSIP) funds to support HSM implementation activities. Questions such as this are handled on a case by case basis. While the law (23 U.S.C. 148) and regulation (23 CFR 924) governing the HSIP lists transportation safety planning and improvement in the collection and analysis of safety data as eligible highway safety improvement projects, these activities must directly support HSIP implementation efforts.

States may also leverage other federal-aid funds to support HSM implementation efforts. For example, state planning and research funds can be used to support data collection efforts. In addition, training is an eligible expense under core federal-aid programs. Improvements to the collection and analysis of safety data can also be funded by NHTSA Section 402 and 408 State Highway Safety Grant Programs. CFR Title 49 Part 350 Commercial Motor Carrier Safety Assistance Program also has limited applicability for commercial motor vehicle involved crashes.

Several States have used HSIP funds for HSM implementation and related projects. However, every situation is unique and you should direct questions regarding HSIP eligibility to your state's FHWA Division Office or Karen Scurry in the FHWA Office of Safety at 609-637-4207 or karen.scurry@dot.gov.

 

HSM Related Tools

This section provides an overview of the relationship of several tools to the Highway Safety Manual.

SafetyAnalyst

What is SafetyAnalyst?
SafetyAnalyst provides a set of software tools used by state and local highway agencies for highway safety management. It incorporates state-of-the-art safety management approaches into computerized analytical tools for guiding the decision-making process to identify safety improvement needs and develop a system-wide program of site-specific improvement projects.

How is SafetyAnalyst related to the HSM?
SafetyAnalyst is intended for network-wide application and includes procedures intended to implement the full scope of HSM Part B.

How can my agency obtain the SafetyAnalyst software?
The SafetyAnalyst software is available through AASHTO, and additional information can be found at www.safetyanalyst.org.

Interactive Highway Safety Design Model (IHSDM)

What is IHSDM?
A product of the Federal Highway Administration's (FHWA) Office of Safety Research and Development, IHSDM is a suite of software analysis tools for evaluating safety and operational effects of geometric design decisions on highways. IHSDM is a decision-support tool that provides estimates of existing or proposed highway designs expected safety and operational performance, and checks designs against relevant design policy values. Results of the IHSDM support decision-making in the highway design process. Intended users include highway project managers, designers, and traffic and safety reviewers in State and local highway agencies and in engineering consulting firms.

The IHSDM currently includes six evaluation modules (Crash Prediction, Design Consistency, Intersection Review, Policy Review, Traffic Analysis, and Driver/Vehicle). The IHSDM website (www.fhwa.dot.gov/research/tfhrc/projects/safety/comprehensive/ihsdm/) summarizes the capabilities and applications of the evaluation modules and provides a library of the research reports documenting their development. Information is available at the public software website, www.ihsdm.org, where users can register and download (free of charge) the latest release of IHSDM.

How is IHSDM related to the HSM?
The IHSDM Crash Prediction Module (CPM) is a faithful software implementation of the predictive methods documented in Part C of the HSM, which includes capabilities to evaluate rural two-lane, two-way roads, rural multilane highways, urban/ suburban arterials, and (using draft HSM Part C materials) freeways. The CPM estimates the frequency of crashes expected on a roadway based on its geometric design and traffic characteristics. The crash prediction algorithms consider the effect of a number of roadway segment and intersection variables.

IHSDM includes a CPM Calibration Utility to assist agencies in implementing the calibration procedures described in the Appendix to HSM Part C.

My agency wants to calibrate HSM Part C Predictive Models. How can IHSDM assist this effort?
A Calibration Utility is available in the IHSDM Administration Tool to assist agencies in implementing the calibration procedures described in the Appendix to HSM Part C.

The Crash Prediction panel of the Administration Tool contains three sections:

  • Calibration Data Sets;
  • Crash Distribution Data Sets; and
  • Model Data Sets.

The Calibration Data Sets interface provides a mechanism for users to enter, edit and organize the site data to be used to calculate the calibration factors for the various crash prediction models available in the IHSDM Crash Prediction Module (CPM) (and, thus, in HSM Part C). In addition to containing the user-entered site data, each Calibration Data Set is also linked to a Crash Distribution Data Set and a Model Data Set.

Within the Calibration Data Set interface, the user can choose to either "Calibrate Using Site Data" or "Manually Specify a Calibration Factor" for each of 26 crash prediction models (covering rural two-lane highways, rural multilane highways, urban/suburban arterials, and freeways).

When running a CPM evaluation, the user indicates which Calibration Data Set to use in that particular evaluation. The CPM then applies the appropriate calibration factors from the user-selected Calibration Data Set.

IHSDM Tutorial Lesson 11 (CPM Calibration) provides step-by-step instructions and hands-on exercises related to the calibration process.

If an agency has developed their own Safety Performance Functions (SPFs), can they be entered into IHSDM?
Yes. The IHSDM Administration Tool provides a mechanism for agencies to enter their own SPFs, as long as the SPFs follow the guidelines for development of jurisdiction-specific SPFs that are acceptable for use in HSM part C (HSM, Part C, Section A.1.2, p. A-9).

The CPM Dataset default configuration (HSM Configuration) file contains parameters and configuration data that define SPFs and Crash Modification Factors (CMFs) to be used in the IHSDM Crash Prediction Module. The default "HSM Configuration" may be copied and modified to reflect agency specific SPFs.

Can default crash severity and crash type distribution values be updated with agency-specific values in IHSDM?
Yes. The IHSDM Administration Tool provides a mechanism for agencies to modify the crash severity and crash type distribution values.

Crash Modification Factor (CMF) Clearinghouse

What is the CMF Clearinghouse?
The Crash Modification Factors Clearinghouse houses a web-based database of CMFs along with supporting documentation to help transportation engineers identify the most appropriate countermeasure for their safety needs. Using this site at www.cmfclearinghouse.org, users are able to search for existing CMFs or submit their own CMFs to be included in the clearinghouse.

What is the purpose of the CMF Clearinghouse?
The CMF Clearinghouse was established to provide transportation professionals:

  • A regularly updated, online repository of CMFs;
  • A mechanism for sharing newly developed CMFs; and
  • Educational information on the proper application of CMFs.

The purpose of the CMF Clearinghouse is to compile all documented CMFs in a central location. The CMF Clearinghouse provides a searchable database that can be easily queried to identify CMFs to meet user's needs.

The CMF Clearinghouse will be updated on a regular basis to add recently developed and documented CMFs. New CMFs will be identified via a periodic review of published literature. In addition, the CMF Clearinghouse provides a mechanism for transportation professionals to submit documentation of new CMFs to be considered for inclusion. Educational information on CMFs includes the "About CMFs" page, which summarizes useful information in the form of answers to frequently asked questions. The "Resources" page provides additional information on related trainings and publications.

The inclusion of all CMFs in the CMF Clearinghouse also serves an educational purpose. One important lesson is that reported CMFs have varying quality and applicability to a given user's needs.

The CMF Clearinghouse summarizes published information on each CMF, including how it was developed (e.g., study design, sample size, and source of data) and what are its statistical properties (e.g., standard error). Where available, a link is provided to the publication from which the CMF was extracted.

The CMF Clearinghouse reports this information in a standard format to enable users to make educated decisions about the most applicable CMF to their condition. To aid users in assessing the quality of the CMF presented, the CMF Clearinghouse reports a star quality rating. The star quality rating is assigned based upon the standard error of the CMF value, as well as the design, potential biases, data source, and sample size of the study that developed the CMF.

The CMF Clearinghouse also reports whether or not the CMF is included in the Highway Safety Manual (HSM). The HSM includes only a subset of CMFs that meet strict inclusion criteria. The CMF Clearinghouse provides the broader context of the larger population of CMFs, from which those included in the HSM were drawn.

The CMFs that are included in the HSM will typically have a higher star quality rating given the strict inclusion criteria. High quality CMFs do not exist for every countermeasure and, therefore, there are many countermeasures for which CMFs do not appear in the HSM. The CMF Clearinghouse includes any documented CMF; i.e., it includes CMFs that do not appear in the HSM either because they did not meet the HSM inclusion criteria or because they were documented after the Manual was completed. As a result, the Clearinghouse includes more CMFs for more countermeasures than the HSM.

Inclusion of a CMF in the CMF Clearinghouse does not constitute an endorsement of the CMF or support for its use. The burden is on the user to determine the most appropriate CMF for their analysis need. This determination should be made based upon the CMFs applicability to their condition (i.e. countermeasure being considered and conditions under which it is implemented) and the quality of the CMF.

What is a CMF?
A crash modification factor (CMF) is a multiplicative factor used to compute the expected number of crashes after implementing a given countermeasure at a specific site.

The CMF Clearinghouse presents both CMFs and Crash Reduction Factors (CRFs). What's the difference?
The main difference between CRF and CMF is that CRF provides an estimate of the percentage reduction in crashes, while CMF is a multiplicative factor used to compute the expected number of crashes after implementing a given improvement. Both terms are presented in the Clearinghouse because both are widely used in the field of traffic safety. Mathematically stated, CMF = 1 - (CRF/100). For example, if a particular countermeasure is expected to reduce the number of crashes by 23% (i.e., the CRF is 23), the CMF will be 1 - (23/100) = 0.77. On the other hand, if the treatment is expected to increase the number of crashes by 23% (i.e., the CRF is -23), the CMF will be = 1 - (-23/100) = 1.23.

These reduction estimates might also be expressed as a function. Crash reduction and crash modification factors are constants; crash modification functions allow the factor to vary for different scenarios, such as for different traffic volume scenarios.

I've seen the term "Accident Modification Factor" (AMF) before. Is that different than a Crash Modification Factor?
Aside from the name, AMF is the same thing as CMF (i.e., AMF of 0.80 = CMF of 0.80). Although the CMF Clearinghouse does not use the term "AMF", there are instances of its use in various areas of the safety field. For example, early drafts of the HSM used the term "AMF", but the decision was made to change the terminology to "CMF" for the final publication.

How can I apply multiple CMFs?
To begin, the CMFs being considered should all be applicable to the same "conditions" and "location." "Conditions" are defined by: setting (road type or intersection type), crash type (severity, manner of collision), time period (day/night), and traffic volume (if specified). All of these attributes are used to describe the CMFs in Part D. "Location" describes the road location to which the treatment associated with a CMF is being applied (e.g., the eastbound travel direction on a roadway, the curve at milepost 10.2, the northbound approach of an intersection).

Case 1: If the CMFs are all precisely defined for very specific conditions and the treatment is for the same location, then they are more likely to be highly related (i.e., targeting the same crashes). For example, two CMFs targeting rear-end crashes resulting in injury during daytime hours on the northbound approach at a stop-controlled rural intersection are likely to be highly related.

Case 2: If the CMFs are all defined for general conditions and nearby locations, then the degree to which they are related is unknown. Engineering judgment can be used to determine if each CMF is likely to affect specific conditions and whether these conditions overlap. Similarly, judgment can be used to determine if the associated treatments are applied at relatively distant locations (e.g., the other side of town, the other intersection approach). The CMFs are more likely to be related when they have more conditions in common, or the treatment influences the same drivers at the same (or a nearby) location.

If multiple countermeasures are implemented at one location, then common practice is to multiply the CMFs to estimate the combined effect of the countermeasures. In fact, there is limited research documenting the combined effect of multiple countermeasures. Although implementing several countermeasures might be more effective than just one, it is unlikely the full effect of each countermeasure would be realized when they are implemented concurrently, particularly if the countermeasures are targeting the same crash type.

For example, shoulder rumble strips and enhanced edgeline retroreflectivity would both target roadway departure crashes, so the CMFs for these treatments would be highly related. Other examples of related CMFs would be the use of increased lighting and installation of pavement reflectors, both of which would target nighttime crashes; and chevrons and advanced curve warning signs, both of which would target curve-related crashes.

Countermeasures that would be considered independent are those that target different crash types. For example, the installation of a pedestrian signal would be relatively independent of the installation of a left turn phase at an intersection, since the one addresses pedestrian-vehicle crashes while the other addresses left-turn opposite-direction crashes. Likewise, the conversion of a left turn phase from permissive to protected along with the installation of an exclusive right turn lane would be fairly independent in that they target different crash types.

Therefore, unless the countermeasures act completely independently, multiplying several CMFs is likely to overestimate the combined effect. The likelihood of overestimation increases with the number of CMFs that are multiplied. Therefore, much caution and engineering judgment should be exercised when estimating the combined effect of more than three countermeasures at a given location.

What does the star quality rating mean?
The star rating indicates the quality or confidence in the results of the study producing the CMF. The star rating is based on a scale (1 to 5), where a 5 indicates the highest or best rating. The review process to determine the star rating judges the accuracy and precision as well as the general applicability of the study results. Reviewers considered five categories for each study – study design, sample size, standard error, potential bias, and data source – and judged each CMF according to its performance in each category.

How is the star quality rating different from the notations (bold, italics, etc.) in the Highway Safety Manual (HSM)?
The star rating and the HSM notation are similar but different. Both indicate the same thing, which is a confidence in the CMF based on the quality of the study that produced it. In a rough sense, higher star ratings correspond to a bold face HSM notation and mid-range star ratings correspond to italics and asterisk HSM notations, but there is not a one-to-one comparison laid out between the two systems. The differences exist in the way the CMFs are reviewed to determine their quality.

The HSM review process applies an adjustment factor to the standard error from the study, and then assigns the bold and italic notations based on ranges of the adjusted standard error. The standard error is adjusted based mainly on the quality of the study design. The HSM assigns asterisk (*) or caret (^) notations based on the confidence interval of the CMF, which indicates the accuracy of the CMF estimate.

The CMF Clearinghouse review process rates the CMF according to five categories – study design, sample size, standard error, potential bias, and data source – and judged the CMF according to its performance in each category. It assigns a star rating based on the cumulative performance in the five categories. It differs from the HSM process in that it does not attempt to adjust the standard error as the HSM does, and it more explicitly considers criteria such as data source, which examines whether a study used data from just one locality or from multiple locations across the state or nation.

Are there available trainings related to the application of CMFs?
The National Highway Institute offers training resources on CMFs. Please visit the Resources Section at cmfclearinghouse.org to find out more on available trainings.

How does the CMF Clearinghouse relate to the Highway Safety Manual (HSM)?
The CMF Clearinghouse is just one of the tools and resources available to help transportation professionals make safety decisions. The first edition of the HSM, released in 2010, provides practitioners with the best factual information and tools to facilitate roadway design and operational decisions based on explicit consideration on their safety consequences.

The CMF Clearinghouse incorporates information relating to the HSM within its website. Users are able to view and search for CMFs included in the HSM. The CMF Clearinghouse includes all of the CMFs listed in the HSM.

That said, it should be understood that the CMF Clearinghouse only relates to the CMF portion of the HSM (Part D). The HSM also covers many other important topics for highway safety, including safety fundamentals, road safety management, and predictive methods.

How do you determine statistical significance?
A CMF is determined to be statistically significant if the specified confidence interval of the CMF does not include 1.0, since a value of 1.0 indicates no effect from the countermeasure. For a given CMF and standard error, the confidence interval will depend on the significance level that is used. The two most common significance levels are 0.05 (corresponds to 95% confidence interval) and 0.10 (corresponds to 90% confidence interval).

For the 95% confidence level, the confidence interval is equal to the CMF ± 1.96 * (standard error).
For the 90% confidence level, the confidence interval is equal to the CMF ± 1.64 * (standard error).

Example

The CMF for countermeasure A is 0.80 with a standard error of 0.15. The lower and upper limits of the 95% confidence interval are the following:
Lower limit: 0.80 – 1.96*0.15 = 0.80 – 0.294 = 0.506
Upper limit: 0.80 + 1.96*0.15 = 0.80 + 0.294 = 1.094

Since the 95% confidence interval (0.506, 1.094) includes 1.0, we cannot be sure that this CMF is statistically different from 1.0 (at the significance level 0.05, i.e., confidence level 0.95)

On the other hand, if the same CMF had a standard error or 0.09, then the lower and upper limits of the 95% confidence interval will be the following:
Lower limit: 0.80 – 1.96*0.09 = 0.80 – 0.1764 = 0.6236
Upper limit: 0.80 + 1.96*0.09 = 0.80 + 0.1764 = 0.9764

Since the 95% confidence interval (0.6236, 0.9764) does not include 1.0, this CMF is statistically different from 1.0 (at the significance level 0.05, i.e., confidence level 0.95).

Who uses CMFs and how are they used?
CMFs are used by several groups of transportation professionals for various reasons. The primary user groups include highway safety engineers, traffic engineers, highway designers, transportation planners, transportation researchers, and managers and administrators. As tools in the safety evaluation process, CMFs can be used to:

  • Capture the greatest safety gain with limited funds;
  • Compare safety consequences among various alternatives and locations;
  • Identify cost-effective strategies and locations;
  • Check reasonableness of evaluations (i.e., compare new analyses with existing CMFs); and
  • Check validity of assumptions in cost-benefit analyses.

Examples

  • A traffic engineer could use CMFs to evaluate the relative cost-effectiveness of several countermeasures for enhancing signal visibility. Countermeasures included increasing lens size, installing signal backplates, and installing dual red indicators in each signal head.
  • A highway designer could use CMFs to compare the cost and safety consequences between paved and unpaved shoulders.
  • A transportation planner could use CMFs to compare the long-term safety impacts of a series of roundabouts as opposed to a series of signalized and unsignalized intersections.

How are CMFs added to the Clearinghouse and what is the process for review?
CMFs for the Clearinghouse are obtained either through 1) a regular examination of presented or published material or 2) studies that are submitted by Clearinghouse users through the web site. All studies that are determined to be eligible for the Clearinghouse (i.e., studies that produce one or more CMFs) are submitted to a review process. Presented and published studies considered for the CMF Clearinghouse will be taken from the following sources:

  • Transportation Research Board Annual Meeting Compendium of Papers;
  • American Society of Civil Engineers Journal;
  • Institute of Transportation Engineers Journal;
  • Accident Analysis and Prevention;
  • Journal of Safety Research; and
  • Broad search of existing literature using the Transportation Research Information Services (TRIS).

The review process evaluates each study according to its study design, sample size, standard error, potential bias, and data source and gives a rating of excellent, fair, or poor to each category. These ratings correspond to point values in a scoring system that is used to determine the star quality rating.

How do I choose between CMFs in my search results that have the same star rating but different CMF values?
It's true that two or more CMFs for a particular countermeasure may have the same star rating but differing CMF values. It will be necessary for you to examine the information related to the applicability of the CMFs to determine how they differ. This could involve examining the brief data shown on the search results page (i.e., crash type, crash severity, roadway type, and area type) or looking at all the information about the CMFs by viewing the CMF details page for each one.

You should select the CMF that is most applicable to the situation in which you would like to apply the CMF (i.e., the characteristics associated with the CMF should closely match the characteristics of the scenario at hand). For example, CMFs often vary by crash type and crash severity. While it is useful to determine the change in crashes by type and severity, this should only be done when applicable CMFs are available for the specific crash type and severity of interest.

The figure below shows a snapshot of results for the countermeasure of "Installation of left-turn lane on single major road approach". You can see that the three CMFs listed in this figure all have 5-star ratings. However, the CMF values (0.65, 0.71, and 0.91) are all different.

 

From this initial view of the search results, it is relatively easy to tell the difference between the first CMF and the other two. Although all three are similar in crash type, crash severity, and roadway type, the first one (CMF of 0.65) is identified as being developed for a "Rural" area type, whereas the other two were developed for an "Urban" area type.

However, all information given on the search results page is identical for the second and third CMF. Therefore, it is necessary to examine the details of each CMF by clicking on the CMF value. When the details of each CMF are examined, it is apparent the CMF of 0.71 is intended for stop-controlled intersections, and the CMF of 0.91 is intended for signalized intersections.

It may be the case that two CMFs are exactly the same with respect to crash and roadway applicability. In these cases, it will be necessary to examine some of the other fields related to how and where the CMF was developed, such as:

  1. Score details. The reviewers who established the star quality rating did so by giving scores of excellent, fair, or poor to five categories: study design, sample size, standard error, potential bias, and data source. Many CMFs in the Clearinghouse are accompanied by details of the scores behind the star rating as shown in the image below.

     

    Clicking on the score details link will display a window showing the scores the CMF received in each category. Users of the Clearinghouse may examine the score details to compare two or more similar CMFs. For instance, although two CMFs may have received the same star rating, one may have a study design score of "Excellent" while the other is "Poor". It may be the case that a user may highly value study design and may use that category to decide between CMFs. Similarly, a user may prioritize some other category in their selection process and use that score to assist in selecting a CMF.

    It may also be useful to examine the fields in the CMF details pertaining to the scores, specifically sample size and standard error. It may be the case that two CMFs both received a score of "Excellent" for sample size, but one has a sample size of 1,000 while the other has a sample size of 3,000. Both of these sample sizes are large enough to qualify for an "Excellent" rating, however, all other factors being equal, the larger sample size would be preferred. Likewise, two CMFs may have both received a score of "Poor" for standard error, but one has a standard error of 0.75 while the other has a standard error of 0.90. In this case, the smaller standard error would be preferred.

  2. Similarity in locality of data used. The fields for "Municipality", "State", and "Country" indicate the area(s) from which data were used in developing the CMF. Many agencies prefer CMFs that were developed in locations that are similar or near their area for reasons of terrain, weather, and other characteristics. For example, a state department of transportation in a mid-western state may prefer using a CMF developed in Kansas over a CMF developed in West Virginia.
  3. Traffic volume range. The fields for "Major Road Traffic Volume" and "Minor Road Traffic Volume" indicate the range of traffic volumes that were used to develop the CMF. You should examine these fields to see which CMF has a traffic volume range that best fits your situation.
  4. Age of data. The field for "Date Range of Data Used" indicates the age of the data used in developing the CMF. Generally speaking, more recent data would be preferred (all other factors being equal). Studies conducted more recently typically use more advanced techniques, higher precision data, and have other advantages related to the progression of knowledge, data quality, and study methods that develop over time in the field of highway safety research. More recent data will also better reflect changes in vehicle fleet characteristics and technology.
  5. Original study report. In addition to providing the citation of the study, the CMF Clearinghouse provides a link, where possible, to the original study document for any CMF. This original document will typically be the final report or published article on the study that developed the CMF. A user of the Clearinghouse who is attempting to select between two similar CMFs may find it useful to refer to the original study report to understand the background of the CMF development. There may be details in the study report that would assist in the CMF selection process. Although the Clearinghouse contains extensive data for each CMF, it does not contain every detail from the study report. For example, the report may discuss details about the roadside character of the roads used in the CMF development. If the roadside character is significantly different from the roads in the user's jurisdiction, he or she may decide to select another CMF that was developed on roads with similar roadside characteristics.

How do the Clearinghouse crash severity terms Fatal, Serious Injury, and Minor Injury relate to the KABCO injury scale?
The initial idea was to use a standard KABCO scale for the Clearinghouse, but there is a problem encountered in standardizing or categorizing study details in the Clearinghouse database. The issue is that authors can and do report the details of their CMFs in many different ways. For crash severity, authors have been seen to report crash severity by KABCO, by MAIS, or simply by referring to "serious injury" and "minor injury". Thus, the Clearinghouse uses the lowest common denominator. There is no one-to-one comparison with KABCO, but the best comparison is that "Fatal" is always equivalent to K, "Serious Injury" would generally be A and B injuries, and "Minor Injury" would generally be C injuries.

How are states using the CMF Clearinghouse?
Many states refer to the CMF Clearinghouse for information on CMFs, traditional and new countermeasures, and other resources related to using CMFs. Although CMFs have been typically used for prioritizing safety projects and developing estimates in cost-benefit analysis, some states are beginning to use CMFs in other situations, such as design exceptions and alternatives analysis.

Oregon DOT has created a website for their staff that demonstrates how they use the Clearinghouse in their agency. Click here to visit their web site about the Clearinghouse.

PLANSAFE

Does PLANSAFE Implement the Highway Safety Manual (HSM)?
PLANSAFE – a software tool developed to complement NCHRP Report 546, Incorporating Safety into Long Range Transportation Planning – supports regional and statewide safety planning efforts. It does not implement portions of the HSM or overlap with the functionality of IHSDM or SafetyAnalyst. While PLANSAFE could be used for analysis of changes in individual locations, the level of detail needed to make decisions at this level is not supported by PLANSAFE, as the documentation for the tool cautions. For more information, refer to NCHRP Report 546 or the web page for NCHRP Project 8-44(02), under which PLANSAFE was developed.

 

Roadway Safety Management Process – HSM Part B

Minimum Segment Length (for Part B Analyses)

Should minimum segment lengths be established for use in HSM Part B analyses? SafetyAnalyst?
There are no explicitly prescribed procedures for HSM Part B analyses. Where SafetyAnalyst is applied for such analyses, roadway segments less than 0.1 mi in length are undesirable. Highway agency data bases with roadway characteristics often have many short segments because, whenever any of the many data elements in such data sets changes, a new roadway segment begins. SafetyAnalyst requires consideration of only a few variables in establishing roadway segments; therefore, longer segments can be used. The SafetyAnalyst data importing tool includes a capability to merge shorter segments from highway agency data bases to create longer segments.

Estimating Average Annual Daily Traffic (AADT) on Minor Cross Roads and Rural Low Volume Roads

When AADT values are estimated, results vary dramatically and data dispersion is extreme. Is using estimated AADT values (particularly on minor cross roads and rural, low volume roads) a good idea in SafetyAnalyst?
See answer to similar question in "Predictive Methods – HSM Part C" section.

HSM Part B/SafetyAnalyst Methods v. "Traditional" Procedures

If the underlying techniques in HSM Part C and SafetyAnalyst are so great, why do they give the same results as the older, apparently flawed procedures?
See answer to similar question in "Predictive Methods – HSM Part C" section.

 

Predictive Methods – HSM Part C

Minimum Segment Length (for Part C Analyses)

Should minimum segment lengths be established for use in HSM Part C analyses?
There is no minimum segment length necessary for use in HSM Part C analyses to estimate the predicted crash frequency (Np). The procedures have been developed so they can be applied to homogeneous segments as long or short as necessary. If a project being analyzed includes numerous segments shorter than 0.1 mi, consideration might be given to using the project-level Empirical Bayes (EB) procedure rather than the site-specific EB procedure to determine the expected crash frequency (Ne) because the locations of observed crashes may not be sufficiently accurate for application of the site-specific EB procedure. The site-specific and project level EB procedures are presented in HSM Part C Appendices A.2.4 and A.2.5, respectively.

Another interpretation of this question is, "Should analysts establish a minimum segment length when using Part C methods? If yes, what are the advantages?"

Part C provides the following guidance on this issue:

  • "When dividing roadway facilities into small homogenous roadway segments, limiting the segment length to a minimum of 0.10 miles will decrease data collection and management efforts." (p. 10-8)
  • "When dividing roadway facilities into small homogenous roadway segments, limiting the segment length to a minimum of 0.10 miles will minimize calculation efforts and not affect results." (p. 10-13)

Note: Software tools for implementing HSM Part C which automatically divide roadway facilities into homogeneous segments (e.g., FHWA's Interactive Highway Safety Design Model), effectively eliminating the need for analysts to establish segments "manually." See "HSM Part C and IHSDM" for more information.

HSM Part C and IHSDM

How is IHSDM related to HSM Part C?
The IHSDM Crash Prediction Module (CPM) is a faithful software implementation of the predictive methods documented in Part C of the HSM, which includes capabilities to evaluate rural two-lane, two-way roads, rural multilane highways, urban/ suburban arterials, and (using draft HSM Part C materials) freeways. IHSDM includes a CPM Calibration Utility to assist agencies in implementing the calibration procedures described in the Appendix to HSM Part C.

For more information on the relationship between IHSDM and HSM Part C, see "HSM-Related Tools," "Interactive Highway Safety Design Model (IHSDM)."

HSM Part C v. SafetyAnalyst

Why don't the HSM Part C and SafetyAnalyst require the same input information?
HSM Part C and SafetyAnalyst were developed for different purposes and, therefore, require different input data. The HSM Part C predictive methods were developed for application in the project development process, to quantify the safety performance of an existing facility and of proposed alternative improvements to the existing facility. For this reason, HSM Part C has extensive input data requirements; it is presumed that all of the HSM Part C input data should be available during the development or design of a particular project.

SafetyAnalyst is intended for network-wide application and includes procedures intended to implement the full scope of HSM Part B. SafetyAnalyst includes procedures to quantify the safety performance of an existing facility and of proposed alternative improvements to the existing facility, but these procedures are simpler and less sophisticated than the procedures used in HSM Part C. As a result, the required input data set for SafetyAnalyst is much less extensive than for HSM Part C, including only those items that are essential for analyses and which would be reasonable to expect could be assembled for the entire highway network.

Why don't HSM Part C and SafetyAnalyst give similar results?
Since HSM Part C and SafetyAnalyst were developed for different purposes and use different procedures and different data sets, it is naturally possible that that they will give different answers. Network screening procedures based on HSM Part C might theoretically provide more accurate results than the SafetyAnalyst procedures, but obtaining all of the required input data for HSM Part C for an entire highway network is likely to be a difficult challenge. Thus, SafetyAnalyst is generally best suited to network-wide analysis and to preliminary analyses of potential alternative improvements and HSM Part C is best suited to detailed analysis of potential alternative improvements as part of the project development process. Both SafetyAnalyst and HSM Part C use the Empirical Bayes (EB) method to compensate for potential bias due to regression to the mean.

Why are the SPF equations in HSM Part C different from those in SafetyAnalyst?
SafetyAnalyst was completed before the HSM and, therefore, the SPFs for SafetyAnalyst and the HSM were developed in separate efforts. However, SafetyAnalyst includes the capability for users to replace the default SPFs with either the SPFs from HSM Part C or their own agency-developed SPFs. HSM users are also free to replace the SPFs provided with HSM Part C with their own agency-developed SPFs (See HSM Part C Appendix, Section A.1.2).

Since the SafetyAnalyst and HSM Part C often give different answers, which one should be used?
The predictive method in HSM Part C would generally be preferred when all of the data needed to apply that method are available. However, it is unlikely that the full range of data needed to apply the HSM Part C method will be available for an entire highway network. It is assumed that many of the data elements needed for HSM Part C are likely to be available only when the development or design of a project is underway. SafetyAnalyst is intended to perform network-wide analyses with a data set that is more easily assembled for a highway network as a whole. For project-level analyses, HSM Part C methods should be used.

Estimating Average Annual Daily Traffic (AADT) on Minor Cross Roads and Rural Low Volume Roads

When AADT values are estimated, results vary dramatically and data dispersion is extreme. Is using estimated AADT values (particularly on minor cross roads and rural, low volume roads) a good idea in HSM Part C and SafetyAnalyst?
It is difficult to answer this question, in general, without knowing what AADT estimates were necessary and how they were made. The question does not make clear from what baseline the results were found to vary dramatically.

All AADT values are, to some extent, estimates unless a permanent count station is located on the site in question. Highway agencies generally have reasonable AADT estimates for roadway segments on the state highway system. AADT values are often unavailable for local roads, including minor-road legs of intersections with state highways, and estimates need to be made to provide exposure data for crash analysis tools. Clearly, the better the estimates made, the better the results.

Highway agencies need better procedures for estimating AADTs where data are not available. One State (Kentucky) is currently beginning research to develop better methods to estimate AADTs for local road networks considering development type and density and the character of intersecting roads.

The Empirical-Bayes (EB) Procedure

Under what conditions is it appropriate to apply the EB procedure?
As noted in HSM Volume 2, Section A.2.1 (Determine Whether the EB Method is Applicable):

"If a future project is being planned, then the nature of that future project should be considered in deciding whether to apply the EB Method.

The EB Method should be applied for the analyses involving the following future project types:

  • Sites at which the roadway geometrics and traffic control are not being changed (e.g., the "do-nothing" alternative);
  • Projects in which the roadway cross section is modified but the basic number of through lanes remains the same (This would include, for example, projects for which lanes or shoulders were widened or the roadside was improved, but the roadway remained a rural two-lane highway);
  • Projects in which minor changes in alignment are made, such as flattening individual horizontal curves while leaving most of the alignment intact;
  • Projects in which a passing lane or a short four-lane section is added to a rural two-lane, two-way road to increase passing opportunities; and
  • Any combination of the above improvements.

The EB Method is not applicable to the following types of improvements:

  • Projects in which a new alignment is developed for a substantial proportion of the project length; and
  • Intersections at which the basic number of intersection legs or type of traffic control is changed as part of a project."

If the EB procedure is applicable for only some sites (homogeneous segments or intersections) within a project, is it appropriate to apply the EB to those sites and not others – or should EB only be used if it can be applied to all sites in a project?
As noted in HSM Volume 2, Section A.2.1 (Determine Whether the EB Method is Applicable):

"If the EB Method is applied to individual roadway segments and intersections, and some roadway segments and intersections within the project limits will not be affected by the major geometric improvement, it is acceptable to apply the EB Method to those unaffected segments and intersections."

The intent of the above paragraph might be restated as follows: "If the EB Method is applied to individual sites and some sites within the project limits will not undergo a major geometric improvement, it is acceptable to apply the EB Method to these sites. In other words, the site-specific EB Method can be applied to some sites within the project limits and not applied to other sites."

If multiple alternative designs for a project are to be evaluated using HSM Part C methods and the EB procedure is applicable to one or more – but not all – alternative designs, should the EB procedure be used where applicable? Or, should EB only be used if it can be applied to all alternative designs in the project? In other words, is it appropriate to compare results (expected crashes) for alternative designs using EB to alternatives not using EB?
Although this issue is not directly addressed in the HSM, the authors of draft HSM Part C Chapters 18 and 19 (Predictive Methods for Freeways and Ramps), wrote the following in a proposed Appendix B to HSM Part C for NCHRP Project 17-45 (Enhanced Safety Prediction Methodology and Analysis Tool for Freeways and Interchanges):

"If alternative improvements are being evaluated for a given project and the EB Method is being considered, then the EB Method will need to be consistently applied to all alternatives being evaluated. If the EB Method cannot be consistently applied to all alternatives, then it should not be used for any alternatives (i.e., the predictive method should be used without EB adjustment). This approach recognizes that there is typically a small difference in the results obtained from the predictive method when it is used with and without the EB Method. If the EB Method is not applied consistently, such differences will likely introduce a small bias in the comparison of expected crash frequency among alternatives."

Calibration Factors – Use of Factors from SafetyAnalyst in HSM Part C (and vice versa)

Can calibration factors from SafetyAnalyst be used in HSM calculations?
See "Calibration" section.

HSM Part C Methods v. "Traditional" Procedures

If the underlying techniques in HSM Part C and SafetyAnalyst are so great, why do they give the same results as the older, apparently flawed procedures?
The premise of this question that the traditional procedures and the new procedures always give the identical or nearly identical answers does not fit actual experience. In some cases, HSM Part C and SafetyAnalyst may give answers that are about the same as the traditional procedure, but in many other cases they will not. It is evident from Equations (A-4) and (A-5) on page A-19 of HSM Part C that, if the predicted crash frequency (Np) and the observed crash frequency (No) happen to be approximately equal or if the predicted crash frequency (Np) and the overdispersion parameter (k) are both large, then the expected crash frequency (Ne) determined with the EB method may be about equal to the traditional answer (No). However, in many cases, the expected crash frequency (Ne) will differ substantially from the traditional answer (No). And, there is no easy way to know in advance, without applying the HSM Part C or SafetyAnalyst procedures, whether the traditional answer is accurate or not. Given a choice between a traditional procedure that may be accurate in some cases, but not in all cases, and new procedure that should be accurate in all cases, it makes most sense to apply the new procedures. Software can make the new procedures relatively easy to apply.

Use of CMFs from outside HSM Part C with Part C models

Is it appropriate to use CMFs from outside HSM Part C with the Part C models? For example, if a project includes all-way STOP-controlled intersections (which have no CMFs in HSM Part C), is it acceptable to apply CMFs for this type of intersection from HSM Part D, or other sources (e.g., the CMF Clearinghouse) to the Part C models?
HSM Part C, Section C.7 (Methods for Estimating the Safety Effectiveness of a Proposed Project) describes four methods for estimating the change in expected crash frequency of a proposed project or design alternative, in order of predictive reliability:

"The Part C predictive method provides a structured methodology to estimate the expected average crash frequency where geometric design and traffic control features are specified. There are four methods for estimating the change in expected average crash frequency of a proposed project or project design alternative (i.e., the effectiveness of the proposed changes in terms of crash reduction). In order of predictive reliability (high to low) these are:

  • Method 1 – Apply the Part C predictive method to estimate the expected average crash frequency of both the existing and proposed conditions.
  • Method 2 – Apply the Part C predictive method to estimate the expected average crash frequency of the existing condition and apply an appropriate project CMF from Part D (i.e., a CMF that represents a project which changes the character of a site) to estimate the safety performance of the proposed condition.
  • Method 3 – If the Part C predictive method is not available, but a Safety Performance Function (SPF) applicable to the existing roadway condition is available (i.e., an SPF developed for a facility type that is not included in Part C of the HSM), use that SPF to estimate the expected average crash frequency of the existing condition. Apply an appropriate project CMF from Part D to estimate the expected average crash frequency of the proposed condition. A locally-derived project CMF can also be used in Method 3.
  • Method 4 – Use observed crash frequency to estimate the expected average crash frequency of the existing condition and apply an appropriate project CMF from Part D to the estimated expected average crash frequency of the existing condition to obtain the estimated expected average crash frequency for the proposed condition.

In all four of the above methods, the difference in estimated expected average crash frequency between the existing and proposed conditions/projects is used as the project effectiveness estimate."

So, the most desirable method (Method 1) is to apply Part C methods. However, since there are no Part C models for all-way stop controlled intersections, the HSM states that Method 2 is the best alternative for this case (a minor road stop controlled intersection could serve as the "existing condition" and the all-way stop as the "proposed condition"). Note that the CMF in Part D for converting minor-road stop control to all-way stop control (Table 14- 5, p. 14-12) only applies when Model Uniform Traffic Control Devices (MUTCD) warrants are met.

A caveat: In the particular example used for illustration here, the above guidance (that Method 2 is the best alternative) only applies to the case where the intersection exists and it is two-way-stop-controlled (TWSC). The analyst should not use the Part C predictive method for TWSC intersections + the Part D CMF for converting a TWSC to all-way stop-controlled (AWSC) to develop an equivalent Part C AWSC predictive method. The inputs to the Part C TWSC predictive method are based on (and calibrated to) data from TWSC intersections. A true AWSC predictive method would be based on inputs from AWSC intersections.

If the existing intersection is AWSC, or if an AWSC alternative is being considered for new alignment, then Method 3 (i.e., develop an AWSC SPF) should be used for the evaluation. The Part D CMF for converting a TWSC intersection to AWSC would not be applicable in this case, but other Part D CMFs could be used with the new SPF.

The Introduction to Part C also provides a discussion on applying multiple CMFs and the assumption of independent effects, which should be kept in mind:

"The CMFs are multiplicative because the most reasonable assumption based on current knowledge is to assume independence of the effects of the features they represent. Little research exists regarding the independence of these effects. The use of observed crash data in the EB Method (see Section C.6.6 and Appendix A to Part C) can help to compensate for any bias which may be caused by lack of independence of the CMFs." (p. C-16)

"Where multiple treatments or countermeasures will be applied concurrently and are presumed to have independent effects, the CMFs for the combined treatments are multiplicative. As discussed above, limited research exists regarding the independence of the effects of individual treatments from one another. However, in the case of proposed treatments that have not yet been implemented, there are no observed crash data for the future condition to provide any compensation for overestimating forecast effectiveness of multiple treatments. Thus, engineering judgment is required to assess the interrelationships and independence for multiple treatments at a site.

The limited understanding of interrelationships among various treatments requires consideration, especially when several CMFs are being multiplied. It is possible to overestimate the combined effect of multiple treatments when it is expected that more than one of the treatments may affect the same type of crash. The implementation of wider lanes and shoulders along a corridor is an example of a combined treatment where the independence of the individual treatments is unclear because both treatments are expected to reduce the same crash types. When implementing potentially interdependent treatments, users should exercise engineering judgment to assess the interrelationship and/or independence of individual elements or treatments being considered for implementation within the same project." (p. C-16)

Is there any guidance for applying multiple CMFs within a project? (i.e., is there a limit to the number of CMFs that should be multiplied together?)
As noted in the previous question, the Introduction to Part C provides a discussion on applying multiple CMFs and the assumption of independent effects:

"The CMFs are multiplicative because the most reasonable assumption based on current knowledge is to assume independence of the effects of the features they represent. Little research exists regarding the independence of these effects. The use of observed crash data in the EB Method (see Section C.6.6 and Appendix A to Part C) can help to compensate for any bias which may be caused by lack of independence of the CMFs." (p. C-16)

"Where multiple treatments or countermeasures will be applied concurrently and are presumed to have independent effects, the CMFs for the combined treatments are multiplicative. As discussed above, limited research exists regarding the independence of the effects of individual treatments from one another. However, in the case of proposed treatments that have not yet been implemented, there are no observed crash data for the future condition to provide any compensation for overestimating forecast effectiveness of multiple treatments. Thus, engineering judgment is required to assess the interrelationships and independence for multiple treatments at a site.

The limited understanding of interrelationships among various treatments requires consideration, especially when several CMFs are being multiplied. It is possible to overestimate the combined effect of multiple treatments when it is expected that more than one of the treatments may affect the same type of crash. The implementation of wider lanes and shoulders along a corridor is an example of a combined treatment where the independence of the individual treatments is unclear because both treatments are expected to reduce the same crash types. When implementing potentially interdependent treatments, users should exercise engineering judgment to assess the interrelationship and/or independence of individual elements or treatments being considered for implementation within the same project." (p. C-16)

For further discussion, see "How can I apply multiple CMFs?" in the "Crash Modification Factor (CMF) Clearinghouse" section of this document.

Unpaved Roads

Does Chapter 10 – Predictive Method for Rural Two-Lane, Two-Way Roads apply to unpaved roads?
No data from unpaved roads were used to develop any of the HSM predictive methods. So, Part C is not intended for evaluation of unpaved roads.

Urban v. Rural Classification

The HSM states that "the definition of urban and rural areas is based on Federal Highway Administration (FHWA) guidelines which classify urban areas as places inside urban boundaries where the population is greater than 5,000 persons." However, agencies may have facilities that exhibit a "rural" cross section (i.e. paved shoulders and roadside ditches) within urban areas as defined above. They may also have a number of facilities that exhibit an "urban" typical section (i.e. curb & gutter, closed drainage system, etc.) outside of urban areas as defined above. In conducting HSM analyses, should the agency assign the facility type based on its geographic location as defined above, or based on its geometric features (presence of ditches vs. curb & gutter)?
The HSM seems to allow for user discretion in this case, though it is not entirely clear. On p. 12-2, it states, "Classifying an area as urban, suburban, or rural is subject to the roadway characteristics, surrounding population and land uses and is at the user's discretion." However, it then continues, "In the HSM, the definition of 'urban' and 'rural' areas is based on FHWA guidelines which classify 'urban' areas as places inside urban boundaries where the population is greater than 5,000 persons. 'Rural' areas are defined as places outside urban areas which have a population less than 5,000 persons. The HSM uses the term 'suburban' to refer to outlying portions of an urban area; the predictive method does not distinguish between urban and suburban portions of a developed area. The term 'arterial' refers to facilities that meet the FHWA definition of 'roads serving major traffic movements (high-speed, high volume) for travel between major points.'"

One State DOT was trying to calibrate the prediction models and had a number of roadway sections in what FHWA would classify as rural areas, but whose characteristics (sidewalks, curb & gutter, parking, etc.) and surrounding land use (small town) suggested that they functioned more like urban roads. The recommendation given to the State was to use their judgment to include such roads in one category or another (i.e., rural or urban/suburban).

So, as a minimum, it is certainly acceptable to assign the facility type based on a roadway's geometric features and surrounding land use. An agency should not feel constrained by the FHWA guidelines if other indicators strongly suggest otherwise.

Predictive models were developed using data from the states (typically via the Highway Safety Information System, or HSIS) wherein the definition of urban and rural are likely to follow the definitions offered in the HSM. More specifically, the state infrastructure databases are derived from Highway Performance Monitoring System (HPMS) data, which follow the FHWA definitions offered in the HSM. In short, the predictive models are calibrated based on the definitions provided in the HSM. Substantial deviation from these definitions is likely to compromise the accuracy of the prediction.

 

Calibration

IHSDM Calibration Utility

My agency wants to calibrate HSM Part C Predictive Models. How can IHSDM assist this effort?
A Calibration Utility is available in the IHSDM Administration Tool to assist agencies in implementing the calibration procedures described in the Appendix to HSM Part C. The Crash Prediction panel of the Administration Tool contains three sections:

  • Calibration Data Sets ;
  • Crash Distribution Data Sets ; and
  • Model Data Sets.

The Calibration Data Sets interface provides a mechanism for users to enter, edit and organize the site data to be used to calculate the calibration factors for the various crash prediction models available in the IHSDM Crash Prediction Module (and, thus, in HSM Part C). In addition to containing the user-entered site data, each Calibration Data Set is also linked to a Crash Distribution Data Set and a Model Data Set.

Within the Calibration Data Set interface, the user can choose to either "Calibrate Using Site Data" or "Manually Specify a Calibration Factor" for each of 26 crash prediction models (covering rural two-lane highways, rural multilane highways, urban/suburban arterials, and freeways).

When running a Crash Prediction Module (CPM) evaluation, the user indicates which Calibration Data Set to use in that particular evaluation. The CPM then applies the appropriate calibration factors from the user-selected Calibration Data Set.

IHSDM Tutorial Lesson 11 (CPM Calibration) provides step-by-step instructions and hands-on exercises related to the calibration process.

Calibration vs. Safety Performance Function (SPF) Development

How much is gained in accuracy by using an agency-developed safety performance function (SPF) rather than a calibrated SPF? Under what circumstances are calibrated SPFs satisfactory and under what circumstances is there a clear advantage for agencies that develop their own SPFs?
The SPFs presented in HSM Part C, when calibrated to local conditions, should provide acceptable level of accuracy for application of the HSM Part C procedures. The HSM does not require that each agency develop their own SPFs, because a requirement for SPF development might have become an impediment to highway agency implementation of the HSM. However, there is every reason to believe that agency-developed SPFs should be even more accurate than calibrated SPFs from the HSM. As long as local SPFs are developed with properly applied statistical techniques, it is only natural that statistical models developed with local data should be more accurate than models developed with data from elsewhere and calibrated to local conditions. Guidance for the development of SPFs with highway agency data has been provided in HSM Part C Appendix A.1.2, and more detailed guidance is available for SafetyAnalyst (see www.safetyanalyst.org). An expanded guide on SPF development is currently being prepared in an FHWA project. In summary, use of SPFs presented in HSM Part C and calibrated to local conditions is acceptable; use of SPFs developed from an agency's own data using proper statistical techniques is acceptable.

There can be no general quantitative answer on how much better an agency-developed SPF will be in comparison to a calibrated SPF. Clearly, this will vary on a case-by-case basis. The response to the next question describes how the usefulness of agency-developed SPFs and calibrated SPFs can be compared.

How should the usefulness of agency-developed SPFs be assessed to determine whether they are preferable to calibrated SPFs? (goodness-of-fit measures? over dispersion parameter values? direct application of both SPFs to test data sets other than the data used in their development?)
In comparing SPFs, those models with better goodness-of-fit measures and smaller overdispersion parameters are generally preferable. However, it should be kept in mind that the goodness-of-fit measure and overdispersion parameter for any model is determined with respect to the data set used to develop that model. Thus, the goodness-of-fit and overdispersion parameters for calibrated SPFs from HSM Part C and agency-developed SPFs are not necessarily comparable. The best method for comparing calibrated SPFs from HSM Part C to agency-developed SPFs would be apply both models to sites from the road network of the highway agency of interest and compare the observed and predicted crash frequencies. Such comparisons should be made using sites other than the sites used in fitting the agency-developed SPF and sites used in calibrating the HSM SPF. A portion of the available data set can be held aside from model development and calibration for this purpose.

Calibration Factors – Use of Factors from SafetyAnalyst in HSM Part C (and vice versa)

Can calibration factors from SafetyAnalyst be used in HSM calculations?
Calibration factors from SafetyAnalyst cannot be used in applying the HSM Part C procedures, and calibration factors from HSM Part C procedures cannot be used in applying SafetyAnalyst, because calibration is performed differently in SafetyAnalyst and HSM Part C. In SafetyAnalyst, the calibration procedure addresses the calibration of SPFs by themselves. In the HSM Part C procedures, the entire predictive method, including both SPFs and crash modification factors (CMFs), is calibrated.

Is it appropriate to develop separate calibration factors for each crash type and severity level? (For example, the Chapter 12 predictive method includes the prediction of multi-vehicle and single-vehicle crashes separately by severity. Should separate calibration factors be developed for each crash type and severity?)
There is no problem with calculating separate calibration factors for each crash type and severity level, as long as the agency has enough observed crashes in their calibration data set to support that type of breakdown. As noted in section A.1.1.2 of the Appendix to HSM Part C, the desirable minimum sample size for the calibration data set is 30 to 50 sites, and the entire group of calibration sites should represent a total of at least 100 crashes per year. So, if an agency wishes to derive separate calibration factors by crash type and severity level, then each of those factors would require a minimum of 30 to 50 sites and 100 observed crashes per year.

Calibration when observed crashes per year for all sites combined and/or the number of sites are less than the recommended minimum

What should be done in the case of insufficient data for a certain facility type? For example, an agency wants to calibrate the model for 3-leg signalized intersections on rural 2-lane roadways, but all sites in the agency database combined have less than the recommended 100 crashes per year. Can a valid calibration factor still be determined?
HSM Part C section A.1.1.2 notes, "For each facility type, the desirable minimum sample size for the calibration data set is 30 to 50 sites, with each site long enough to adequately represent physical and safety conditions for the facility." Where practical, calibration sites should be randomly selected from a larger set of candidate sites. "Following site selection, the entire group of calibration sites should represent a total of at least 100 crashes per year." However, in practice, some agencies have found that – for certain facility types – all sites combined do not total more than 100 crashes per year and/or there are fewer than 30 sites (e.g., intersections) in the available database. In that case, it is recommended that all available data be used to determine the calibration factor.

 

Safety Performance Function (SPF) Development

Calibration vs. Safety Performance Function (SPF) Development

See "Calibration"

Data Quality for SPF Development

What data quality is needed for agency datasets used to develop SPFs?
Naturally, any SPF developed with highway agency data will be only as good as the data from which the SPF is developed. And, unfortunately, most existing safety data bases are far from perfect, but agencies have no choice but to work with the data they have available. The following guidelines describe the desirable characteristics of data bases for SPF development.

The development of SPFs from data for a specific highway agency requires data on roadway segment or intersection characteristics and data on crash history that can be linked together by location. This linkage is necessary so that each crash can be attributed to a particular roadway segment or intersection. Crashes that are classified as occurring at an intersection or are classified as intersection-related and occur within 250 ft of an intersection should be attributed to that intersection. (At particular intersections where traffic queues are known to extend beyond 250 ft from the intersection on a daily basis, a distance greater than 250 ft for intersection-related crashes may be considered.) All other crashes should be attributed to the roadway segment on which they occur. Available data bases for roadway segment or intersection characteristics should contain all the data needed to identify sites of the specific type under consideration (e.g., rural undivided two-lane, two-way roadway segments or three-leg unsignalized intersections with STOP-control on the minor road), as well as traffic volume and any other variables to be considered as independent variables. In addition to the crash location guidelines presented above, crash data bases should include, at a minimum, crash severity and crash type.

The database needs to include all of the attributes identified as "base conditions" in the predictive method. The SPFs must be calibrated to respect all of the base values specified in the appropriate Part C chapter.

Agency-Specific SPFs v. HSM/SafetyAnalyst SPFs

If a highway agency develops agency-specific SPFs with its own data, and the agency-developed SPFs have a different shape from those found in HSM Part C or SafetyAnalyst, is there something wrong? Which SPF should be used?
Research has not established any single best shape for particular SPFs. The functional forms used for SPFs in HSM Part C and SafetyAnalyst will accommodate a variety of shapes depending on the values of the coefficients in the fitted model. Thus, if an agency-developed SPF that was developed with properly applied statistical techniques appears to have a different shape than the calibrated SPF from HSM Part C, the agency-developed SPF may still be preferable to the calibrated SPF. For guidance on comparison of agency-developed and calibrated SPFs, see "Calibration," Calibration vs. Safety Performance Function (SPF) Development.

Shapes of SPFs – variations from one agency to another

Why do the shapes of SPF curves for a given facility type often vary markedly from one agency to another or from one region to another within a state? Are such variations real or are they artifacts of sample size or data quality issues?
Not enough is known about SPFs to provide a general answer to this question. But, there is no doubt that researchers developing SPFs have encountered variations of this kind. There is no indication that such variations are the results of sample size limitations, crash reporting thresholds, or data quality issues, but the explanation of these variations is uncertain. Research is needed on this issue.

Minimum Segment Length (for SPF Development)

What should an agency do with all of the very short segments that do not meet minimum suggested threshold of 0.1-mi length?
For SPF development, a minimum roadway segment length of 0.1 mi is desirable. Shorter roadway segments are undesirable because the segment characteristics may not be in place for sufficient length to truly affect crash risk and because data on crash locations may not be accurate enough to assign each crash to the appropriate road segment. Thus, segments shorter than 0.1 mi should generally not be included in data bases for SPF development.

IHSDM and Agency-Specific SPFs

If an agency has developed their own SPFs, can they be entered into IHSDM?
Yes. The IHSDM Administration Tool provides a mechanism for agencies to enter their own SPFs, as long as the SPFs follow the guidelines for development of jurisdiction-specific SPFs that are acceptable for use in HSM part C (HSM, Part C, Section A.1.2, p. A-9).

The crash prediction Model Dataset default configuration (HSM Configuration) file contains parameters and configuration data that define SPFs and CMFs to be used in the IHSDM Crash Prediction Module. The default "HSM Configuration" may be copied and modified to reflect agency specific SPFs.

 

Crash Modification Factors (CMFs) – HSM Part D

What is a CMF?
A Crash Modification Factor (CMF) is a value that quantifies the expected change in crash frequency at a site as a result of implementing a specific countermeasure. Countermeasures can also be called "treatments" or "safety treatments". A CMF can estimate the expected change in crash frequency for total crashes, a particular crash type, or a particular severity. A CMF is expressed as:

 

A CMF can also be a crash modification function, which is a formula used to compute the CMF for a specific site based on its characteristics. Crash modification functions are useful because it is not always reasonable to assume that a treatment will have the same safety effect at sites with different characteristics (e.g., safety benefits may be greater for sites with higher traffic volumes). A crash modification function allows the CMF to change over the range of a variable or combination of variables.

What is meant by the "base condition"?
The values of CMFs in the HSM are determined for a specified set of base conditions. These base conditions represent the site conditions before implementation of a treatment. This allows comparison of treatment options against a specified reference condition. For example, CMF values for the effect of lane width changes are determined in comparison to a base condition of 12-ft lane width. Under the base conditions (i.e., with no change in the conditions), the value of a CMF is 1.00.

What is the numerical value of a CMF?
A countermeasure with a CMF greater than 1.0 is expected to increase crashes at the site following installation, while a countermeasure with a CMF less than 1.0 is expected to decrease crashes. A countermeasure with a CMF equal to 1.0 is not expected to have an effect on crash frequency.

How do I apply a CMF?
A CMF is a multiplicative factor applied to an estimate of the expected crash frequency. The estimate can represent a particular crash type or particular crash severity as specified in the HSM or in the underlying study associated with the treatment. There are examples at the end of this section that show sample calculations.

In the HSM, what does the CMF standard error mean?
The standard error (SE) indicates the anticipated variation in the results of the CMF. A smaller SE indicates more certainty in the results. The CMFs with a SE less than 0.1 are identified in the HSM through the use of bold text. Table 1 summarizes the formatting used in the HSM to indicate the SE of a CMF. The SE is used to calculate a confidence interval for the CMF value. The equation for the 95th percentile confidence interval is CMF ± (1.96*SE). In the HSM, this equation is rounded to CMF ± (2*SE). The examples at the end of this document demonstrate application of the confidence interval.

Table 1: Formatting used in the HSM to indicate the Standard Error of a CMF

Standard Error Font
<0.10 Bold
0.10 < standard error < 0.20 Normal
0.20 < standard error < 0.30 Italic

How do I select a Treatment/CMF?
When selecting treatments and CMFs, it is important to make sure the treatment is applicable to the site of interest. For example, the same countermeasure used on different road types may have different effects. Therefore, applying a CMF at a location that does not correspond to the setting (i.e. base condition) identified in the study may provide an erroneous estimate of the expected change in crash frequency. This could result in infrastructure investments that may not be as beneficial as expected.

Other factors that influence the applicability of the treatment to the situation under consideration are: area type (rural vs. urban), study location (differences in driver characteristics), traffic volumes, speed limit, and traffic control. To the extent possible this information is described in the HSM, or presented in the CMF table.

Can a CMF have a different effect on different crash types?
A CMF may have a different effect on different crash types or severities. As an example, consider Table 2 below. Notice the CMF for all crash types and all severities is equal to 0.86 plus or minus a standard error of 0.05, but for head-on and opposing direction injury crashes, the CMF is equal to 0.75 plus or minus a standard error of 0.2.

Table 2: from HSM Table 13-46 – Effect of Installing Centerline Rumble Strips

Road Type

AADT

Crash Type (Severity)

CMF

Std. Error

Rural
(two-lane)

5,000 to 22,000

All types (all severities)

0.86

0.05

All types (injury)

0.85

0.08

Head-on and opposing-direction sideswipe (all severities)

0.79

0.1

Head-on and opposing-direction sideswipe (injury)

0.75

0.2

Base condition: Absence of centerline rumble strips.

In some cases, a treatment may increase certain crash types (i.e., CMF > 1.0) while reducing others (i.e., CMF < 1.0). For example, Table 3 presents CMFs from Table 14-7 of the HSM for installing a traffic signal at a rural stop-controlled intersection. Notice that the installation of a traffic signal is expected to reduce angle and turning crashes, and increase rear-end crashes; the net benefit is hopefully a reduction in the overall severity of crashes. The potential for differential crash effects underscores the importance of properly applying CMFs—only apply CMFs to the applicable crash types and severities.


Table 3: from HSM Table 14-7 – Potential Crash Effect of Converting from Stop to Signal Control

Road Type

AADT

Crash Type (Severity)

CMF

Std. Error

Rural
(3-leg and 4-leg)

Major road
3,261 to 29,926

Minor road
101 to 10,300

All types (all severities)

0.56

0.03

Right Angle (all severities)

0.23

0.02

Left Turn (all severities)

0.40

0.06

Rear-end (all severities)

1.58

0.2

Base condition: Absence of centerline rumble strips.

How do I calculate the effect of multiple treatments at one site?
The process for estimating the combined effects of multiple countermeasures is to multiply the CMFs together and then apply this factor to the number of crashes. The CMFs should only be multiplied when the CMFs apply to the same set or subset of crashes at the site. This requires the crash data to first be segmented (e.g., using severity or collision type distribution data), and then apply the applicable CMF for each treatment to its respective subset of crash data. It is important to note this approach basically assumes that CMFs function independently of each other and the magnitude of the expected crash reduction of implementing each of the countermeasures is the same as if implemented individually. This assumption needs more research; therefore, the HSM advises that not more than three CMFs be multiplied together.

For further discussion of this topic, see HSM-Related Tools > Crash Modification Factor (CMF) Clearinghouse > "How can I apply multiple CMFs"; and Predictive Methods – HSM Part C > Use of CMFs from outside HSM Part C with Part C models > "Is there any guidance for applying multiple CMFs within a project (i.e., is there a limit to the number of CMFs that should be multiplied together?)"

What is the difference between CMFs in the HSM and CMFs in the CMF Clearinghouse?
Currently, the two main resources for CMFs are the HSM and the FHWA CMF Clearinghouse (www.cmfclearinghouse.com). A significant difference between these two resources is in how the CMF values are presented. For each treatment in the HSM, one CMF is presented for a given crash type or severity based on the best available research. The CMF may be based on a single study or may represent an aggregate value based on multiple studies. The CMF Clearinghouse is a comprehensive database of all the CMFs available for a given treatment. The quality of the CMF is rated on a one to five star basis. All of the treatments and CMFs in the HSM are in the CMF Clearinghouse. The CMF Clearinghouse is updated regularly, with new CMFs from researchers and state agencies.

For further discussion of this topic, see HSM-Related Tools > Crash Modification Factor (CMF) Clearinghouse > "How does the CMF Clearinghouse relate to the Highway Safety Manual?"

What is the difference between the CMFs in Part D and Part C of the HSM?
As noted in the HSM, "The information presented in Part D is based on an extensive literature review of published transportation safety research, mostly dated from the 1960s to June 2008. The CMFs from the literature review process were evaluated during an 'Inclusion Process,' based on their standard errors, to determine whether or not they are sufficiently reliable and stable to be presented in the HSM. A standard error of 0.10 or less indicates a CMF value that is sufficiently accurate, precise and stable. Part D includes all CMFs assessed with the literature review and inclusion process, including measures of their reliability and stability. These CMFs are applicable to a broad range of roadway segment and intersection facility types, not just those facility types addressed in the Part C predictive methods. Some Part D CMFs are included in Part C and for use with specific SPFs. Other Part D CMFs are not presented in Part C but can be used in the methods to estimate change in crash frequency described in Part C." (HSM Part D, p. D-7 and D-8)

Each CMF underwent a rigorous evaluation, however, there are two key differences between the CMFs in Part C and Part D of the HSM:

  • The CMFs in Part C have been formally integrated into a safety–prediction methodology. Part D provides a catalog of CMFs for a variety of facility types; these CMFs have not been formally integrated into a safety-prediction methodology.
  • Some of the CMFs in Part C have been reviewed and approved for inclusion in the HSM by an expert panel, but have not necessarily had the "inclusion process" applied to them. All of the CMFs that passed the "inclusion process" are included in HSM Part D.

What does a CMF calculation look like?
The following examples illustrate the use of CMFs and associated standard errors to estimate the expected effect of a given strategy. The first example shows how the standard error is incorporated to estimate the likely range of effect. The second example shows how CMFs are identified and compared to determine the relative effects of different design values.

Example Calculation 1

Using the treatment in Table 2, the CMF and 95 percent confidence interval for installing centerline rumble strip on a two-lane rural highway will be:

CMF All Types, All Severities = (0.86+2(0.05)) = 0.96,OR (0.86-2(0.05)) = 0.76

The CMF for adding centerline rumble strips on rural two-lane highways could be between 0.76 and 0.96 for all crash types and all severities.

CMF Head-On, Opposing Direction, Injury = (0.75+2(0.20)) = 1.15,OR (0.75-2(0.20)) = 0.35

For head-on, opposing-direction or sideswipe injury crashes, the CMF could be between 0.35 and 1.15. Note the impact of the standard error to the estimated range of benefit to head-on, opposing-direction or sideswipe injury crashes.

Every location is different; variations in the actual performance of a treatment can be expected if implemented at several different sites.

For further discussion, see HSM-Related Tools > Crash Modification Factor (CMF) Clearinghouse > "How do you determine statistical significance?"

Example Calculation 2

Given a two-lane rural highway segment with 12-foot travel lanes, 4-foot paved shoulder, and Average Annual Daily Traffic (AADT) of 8,000 vehicles per day; evaluate the expected difference in crash frequency if the roadway cross-section is altered to an 11-foot lane width with 5-foot paved shoulders. The average total crash frequency on the segment is equal to 20 crashes per year. The average crash frequency of run-off-the road, head-on and side-swipe crashes is equal to 11 crashes a year, which comprises 55 percent of the total observed crashes.

Solution: The first step is to identify the appropriate CMFs for lane width. Table 4 (from Table 13-2 of the HSM) provides CMFs for lane width on rural two-lane roadway segments. Based on the AADT, the CMF corresponding to the existing condition is 1.00 (base condition) and the proposed condition is 1.05 based on an AADT of 8,000 vehicles per day. Note that these CMFs only apply to single-vehicle run-off-road and multiple-vehicle head-on, opposite-direction side-swipe, and same-direction side-swipe crashes.

Table 4: from HSM Table 13-2 - CMF for Lane Width on Rural Two-Lane Roadway Segments

Lane Width

Average Annual Daily Traffic (veh/day)

< 400

400 to 2,000

> 2,000

< 9 feet

1.05

1.05 + 2.81 x 10-4 x (AADT – 400)

1.50

10 feet

1.02

1.02 + 1.75 x 10-4 x (AADT – 400)

1.30

11 feet

1.01

1.01 + 2.5 x 10-5 x (AADT – 400)

1.05

> 12 feet

1.00

1.00

1.00

NOTE: The collision types related to lane width to which these CMFs apply are single-vehicle run-off-road and multiple-vehicle head-on, opposite-direction side-swipe, and same-direction side-swipe crashes.

The next step is to identify the appropriate CMF for shoulder width using Table 13-7 of the HSM (Table 5). The CMF for the existing conditions with a 4-foot paved shoulder is equal to 1.15, and the CMF for the proposed 5-foot paved shoulder is interpolated as approximately 1.075. Once again, these CMFs only apply to single-vehicle run-off-road and multiple-vehicle head-on, opposite-direction side-swipe, and same-direction side-swipe crashes.

Table 5: from HSM Table 13-7 – CMF for Shoulder Width on Rural Two-Lane Roadway Segments

Shoulder Width

Average Annual Daily Traffic (veh/day)

< 400

400 to 2,000

> 2,000

0 feet

1.10

1.10 + 2.5 x 10-4 x (AADT – 400)

1.50

2 feet

1.07

1.07 + 1.43 x 10-4 x (AADT – 400)

1.30

4 feet

1.02

1.02 + 8.125 x 10-5 x (AADT – 400)

1.15

6 feet

1.00

1.00

1.00

> 8 feet

0.98

0.98 – 6.875 x 10-5 x (AADT – 400)

0.87

NOTE: The collision types related to shoulder width to which this CMF applies include single-vehicle run-off-road and multiple-vehicle head-on, opposite-direction side-swipe, and same-direction side-swipe crashes.

Since the CMF values identified for lane width and shoulder width only apply to a particular subset of crash types, these CMFs must be converted to total crashes, using HSM Equation 13-3:

CMF = [(CMFra-1.0)×pra]+1.0

Where,

CMFra = crash modification for related crashes

Pra = related crashes expressed as a proportion of total crashes

No adjustment is necessary for the 12-foot lane width. The CMF for the 11-foot lane width is calculated as follows:

CMF=[(1.05-1.0)×0.55]+1.0=1.0275

The same process is used to convert the CMF for shoulder width to total crashes.

4-foot shoulder:

CMF = [(1.15-1.0)×0.55]+1.0 = 1.0825

5-foot shoulder:

CMF = [(1.075-1.0)×0.55]+1.0 = 1.0413

Finally, the combined influence of the lane and shoulder width is determined by multiplying the respective CMFs.

Existing condition (12-foot lane, 4-foot shoulder):

CMF = 1.0×1.0825 = 1.0825

Proposed condition (11-foot lane, 5-foot shoulder):

CMF = 1.0275×1.0413 = 1.0699

Based on the results, the proposed condition is expected to perform slightly better than the current condition with approximately 1.26 percent fewer total crashes (1.0825-1.0699). It would be necessary to perform a benefit cost analysis to determine if the required investment for the proposed configuration provides enough benefit to justify the change.

Crash Modification Factor Clearinghouse

See "HSM-Related Tools"

CMFs in the Roadway Safety Management Process

Information in this section was extracted from Crash Modification Factors in Practice: Quantifying Safety in the Roadway Safety Management Process. [1]

How can CMFs be applied in the Roadway Safety Management Process?
CMFs can be applied in the roadway safety management process to help select countermeasures and prioritize projects through an economic evaluation (e.g., benefit-cost analysis). The roadway safety management process is a six-step process as shown in Figure 1 and outlined in the HSM.

Figure 1: HSM Six-Step Roadway Safety Management Process

 

The Highway Safety Improvement Program (HSIP) Manual [2] identifies this process as planning, implementation, and evaluation, where planning covers problem identification, countermeasure identification, and project prioritization. In either case, CMFs can play a role in the countermeasure selection and project prioritization components of the roadway safety management process. While not directly applicable to the application of CMFs, one can develop new CMFs in the safety effectiveness evaluation component of the process.

How can CMFs be used/applied in the Countermeasure Selection process? (What is the role of CMFs in the Countermeasure Selection process?)
In this step, potential countermeasures are developed to address the contributing factors identified in the safety diagnosis. Physical, financial, and political constraints need to be taken into consideration during this task as well as the potential impacts on safety, mobility, and the environment. CMFs can provide valuable information to assist in the countermeasure selection process, particularly the quantification of safety impacts.

With respect to countermeasure selection, CMFs can play a valuable role by indicating which candidate treatments are associated with the greatest expected reductions in crashes. From the diagnosis step, a list of contributing factors is generated. The first step in the countermeasure selection process is to identify a list of potential countermeasures to address the specific contributing factors. Contributing factors and related treatments are identified in the NCHRP Report 500 Series [3] for several specific topics.

CMFs can help to reduce the list of potential treatments to more manageable levels by grading the treatments in terms of expected safety effectiveness. For example, those treatments with CMFs less than 1.0 could be carried forward for further evaluation, while treatments with CMFs greater than or equal to 1.0 may be eliminated from further consideration as they are likely to result in an increase in crashes. Of course, there may also be physical, financial, and political constraints, but CMFs are a useful tool for sifting through the initial list of potential treatments.

The CMF alone is not always enough information to immediately include or discount a treatment from further consideration. CMFs are developed using various study designs, sample sizes, and study periods. As such, there is a wide range in the quality and reliability of CMFs. The standard error of a CMF should be considered as it indicates the potential variability in the estimate. The standard error can be used to define a confidence interval which indicates the range of values that contain the true treatment effect with a given level of confidence. A CMF confidence interval which includes 1.0 suggests that a treatment is not highly effective and may be completely ineffective. Consequently, it would be reasonable to give less consideration to treatments for which the associated CMF has a confidence interval that includes 1.0. Furthermore, it may be prudent in some situations to give greater consideration to treatments with smaller confidence intervals because of the greater level of certainty in the results.

How can CMFs be used/applied in the Economic Appraisal process? (What is the role of CMFs in the Economic Appraisal process?)
The economic appraisal step of the highway safety management process seeks to compare the benefits of safety improvements to the costs of implementing those improvements. There are two main types of economic appraisals: benefit-cost analysis and cost-effectiveness analysis. In benefit-cost analyses, the safety benefits of potential treatments are translated into monetary values and then compared to treatment costs. In contrast, a cost-effectiveness analysis does not convert safety benefits into monetary terms. Instead, the cumulative treatment costs are divided by the estimated number of reduced crashes to approximate the cost per crash reduced. CMFs may be utilized in either type of analysis to estimate the reduction in crashes.

With respect to the economic appraisal, the main function of CMFs is to help estimate the benefits of proposed treatments as part of benefit-cost or cost-effectiveness analyses. Depending on which type of economic appraisal is conducted, benefits may be quantified in different forms. In a benefit-cost analysis, benefits are measured in terms of monetary values. Specifically, estimated crash reductions are converted to monetary values using average crash costs. In a cost-effectiveness analysis, benefits are quantified simply as the estimated reduction in crashes. In either case, CMFs are used to estimate the change in crash frequency associated with proposed treatments.

What is the role of CMFs in Safety Effectiveness Evaluation?
The safety effectiveness evaluation step of the roadway safety management process assesses how an implemented safety treatment or set of safety treatments affected the frequency and severity of crashes. During this step, evaluations of individual treatments or combinations of treatments can be carried out based on various performance measures. It is often possible to develop CMFs in this step of the process. If the goal is to develop CMFs, there are numerous study designs that can be utilized which have varying levels of complexity and quality. More information about the various approaches to develop CMFs can be found in A Guide to Developing Quality Crash Modification Factors [4] and Recommended Protocols for Developing Crash Modification Factors [5]. This step is intended to provide quantitative indicators of effectiveness in order to guide future highway safety decision-making and policy development.

What are potential benefits associated with applications of CMFs in the safety management process?
There are several potential benefits associated with the application of CMFs in the safety management process. Specifically, CMFs provide a means to quantify the safety impacts of decisions and help to raise awareness of safety. The application of CMFs also helps to prioritize potential treatments and provides decision-makers with the information needed to identify cost-effective strategies.

Using CMFs as part of a benefit-cost analysis is not only beneficial to prioritizing the suggested countermeasures for a particular site, but also helps in the management of a safety program.

What are potential challenges to applying CMFs in the safety management process (and opportunities to overcome these challenges)?
Potential challenges may arise when applying CMFs in the roadway safety management process. Many are directly related to limitations in the progress of CMF research, while others apply to the lack of understanding of CMFs. Despite decades of advancement in CMF research, there are still knowledge gaps that present obstacles for practitioners seeking to apply CMFs in the roadway safety management process. The Introduction to Crash Modification Factors [6] provides general guidance related to the application of CMFs. The following are general challenges associated with the application of CMFs and opportunities to overcome challenges. The discussion includes specific concerns and lessons learned based on actual experiences with the application of CMFs in roadway safety management efforts.

Availability of CMFs

A notable potential challenge is the availability of CMFs for specific countermeasures. The CMF Clearinghouse contains over 3,000 CMFs for a wide range of safety countermeasures under a variety of conditions. However, CMFs are still lacking for a large number of treatments, especially combination treatments and those that are innovative and experimental in nature. Furthermore, CMFs may not be available for certain crash types and severities.

Applicability of CMFs

CMFs are developed based on a sample of sites with specific conditions. While a CMF may be available for a given treatment, it may not be appropriate for the scenario under consideration. For example, there may be significant differences between the characteristics of a proposed treatment site and the sites used to develop the CMF (e.g., different area type, number of lanes, or traffic volume). The CMF Clearinghouse and HSM provide information to help users identify the applicability of CMFs.

A related challenge may be that multiple CMFs exist for the same treatment and conditions. This is particularly challenging when multiple studies have estimated CMFs for the same countermeasure and combination of crash type and severity level, but yielded dissimilar results.

If the CMFs also apply to the same roadway characteristics, then the selection can become even more difficult. A star quality rating – which appraises the overall perceived reliability of a CMF using a range of one to five stars – is provided by the CMF Clearinghouse and may be helpful in these circumstances to identify the most suitable CMF. However, the ratings of the different CMFs may be similar as well. If the various CMFs have a fairly small range of values, then this situation may not be of great concern. Yet, it is possible for the CMFs to vary significantly and even have contradictory expected outcomes (i.e., some CMFs greater than 1.0 and others less than 1.0). In such cases, this potential situation would be highly challenging to overcome. Additional guidance on how to select the most applicable CMF is posted on the CMF Clearinghouse under FAQs.

Estimating the Effects of Multiple Treatments

The current practice for many agencies is to assume that CMFs are multiplicative; this is the current method presented in the HSM and posted on the CMF Clearinghouse. There are relatively few studies that estimate CMFs for combinations of countermeasures. It is far more common for studies to estimate CMFs for individual treatments. Consequently, it is difficult to accurately estimate the effects of combinations of treatments. In brief, the recommended approach (and many of the alternatives) is problematic in the sense that applying the combined CMF may overestimate or underestimate the true crash effects, particularly if the treatments target similar crash types. More information regarding the application of multiple CMFs is available in recent articles [7],[8].

Insufficient Expertise

A specific challenge could be that there is insufficient expertise within an agency to apply CMFs. While CMFs are not a new tool, they have only recently gained popularity among safety professionals. There are a number of opportunities to apply CMFs in aspects of transportation engineering (e.g., roadway safety management process), but it may be necessary to solicit input or assistance from those who are more familiar with the selection and application of CMFs. If an agency does not have the needed expertise related to CMFs, then they can solicit outside expertise from the State Safety Engineer, FHWA Division Office, or consultants for further guidance and assistance with the selection and/or application of CMFs and interpretation of results. The National Highway Institute also offers several courses related to the quantification of safety using CMFs, including the Application of CMFs (#380093) and Science of CMFs (#380094).

Inconsistency across decentralized States

Where multiple districts/divisions/regions exist within a State, there is the potential for inconsistency with respect to the selection and application of CMFs. This can result from a number of factors, including available resources and range in expertise. There is need to encourage the consistent selection and application of CMFs in the roadway safety management process within a State, particularly if the districts/divisions/regions are competing for the same pool of funding.

Estimating Annual Crashes without Treatment

To quantify the expected safety performance for a given alternative, it is necessary to estimate the annual crashes without treatment. The applicable CMFs are then applied to the annual crashes without treatment to estimate the annual crashes with treatment. The annual crashes without treatment can be estimated using several methods, with each bringing certain strengths and weaknesses. The most basic approach is to use the observed crash history of the site of interest (i.e., short-term or long-term average) to estimate annual crashes without treatment. This method is relatively simple but is highly susceptible to regression-to-the-mean bias (i.e., random fluctuation in crashes over time) and could overestimate or underestimate the annual crashes without treatment. Another option to estimate annual crashes without treatment is to employ SPFs, which provide the predicted number of crashes. SPFs help to account for the random nature of crashes at a single site by incorporating data from other similar sites. The drawback to using SPFs is that, unless they are developed using local data, they may not accurately reflect local conditions and again could overestimate or underestimate the annual crashes without treatment. The HSM presents the Empirical Bayes method as yet another option, which combines both the observed crash history of a site and the predicted crashes from the SPF to compute the expected crashes.

The prior discussion assumes that the crash history is available and applicable for a given site. In some cases, the crash history may not be available (e.g., new construction); in others, the crash history may not be applicable (e.g., significant changes in the alignment). For both scenarios, it may be necessary to rely on SPF predictions, but it is suggested that the SPFs be calibrated to local conditions before applying them, whenever possible. The Introduction to Safety Performance Functions [9] provides general guidance related to the selection, calibration, and application of SPFs.

 

 

Specific Technical Issues

How should default values in the HSM be handled?
The HSM provides default values for items such as costs of injuries, severity distributions, and percentage of animal crashes (refer to Section A.1.3 of the Appendix to HSM Part C for guidance on developing default values). These defaults are the most appropriate values for inclusion in a manual to be used by a variety of agencies and organizations across the country, but if a user has reliable local values, the local values should be used instead of defaults. This will provide results more applicable to the specific situation for which the HSM is being used.

Does the HSM Cover…

Guidelines for crash testing safety hardware?
No. Resources for this include the Manual for Assessing Safety Hardware and the earlier NCHRP Report 350: Recommended Procedures for the Safety Performance Evaluation of Highway Features.

One-way streets?
The HSM briefly discusses the expected impact of removing an unwarranted signal on a one-way street (Part D, Chapter 14) and converting one-way streets to two-way (Part D, Chapter 17). One-way arterials will be included in upcoming research sponsored by the National Cooperative Highway Research Program (project 17-58).

Pedestrians? Bikes?
HSM information on non-motorized road users is spread throughout the manual. Chapter 13, on CMFs for roadway segments contains information on the expected impact of treatments related to pedestrians and bicyclists. There is not enough information available to develop CMFs, but this chapter does contain information on trends. Chapter 12 includes a pedestrian crash prediction method for signalized intersections, including SPFs and CMFs; pedestrian crash adjustment factors for stop-controlled intersections and segments; and bicycle crash adjustment factors for segments and intersections.

Parking?
While expected impacts of treatments related to on-street parking are discussed in the HSM (Part D, Chapter 13), many issues related to parking have not undergone the type of study necessary for inclusion in the HSM. An on-street parking CMF is part of the crash prediction method for urban/suburban arterials in Chapter 12.

Animal crashes?
The HSM provides default percentages of animal crashes for use in Part C predictive method. For rural two-lane roads, this information is in Chapter 10. Chapter 12 contains the information for urban and suburban arterials. For rural multi-lane roads (Chapter 11), this information is not available. If percentages of animal crashes are available for a state or region for which an analysis is being performed, these values can be used instead of those provided by the HSM.

At-grade rail crossings?
The HSM will provide some information on crash effects of treatments related to highway-rail grade crossing traffic control and operational elements (Part D, Chapter 16). There are crash modification factors for signs and markings, signals and gate (active and passive), and illumination. There are a few other treatments for which trends are discussed, but for which enough information was not available to provide a CMF. These treatments are strobes, four-quadrant gates, pre-signals, and constant warning time devices.

 

References

  1. Crash Modification Factors in Practice: Quantifying Safety in the Roadway Safety Management Process. Publication FHWA-SA-13-010, Federal Highway Administration, Washington, D.C., 2013. Available online at: http://safety.fhwa.dot.gov/tools/crf/resources/cmfs/docs/management.pdf
  2. Highway Safety Improvement Program (HSIP) Manual. Publication FHWA-SA-09-029, Federal Highway Administration, Washington, DC, 2010.
  3. National Cooperative Highway research Program (NCHRP). NCHRP Report500 Series: Guidance for Implementation of the AASHTO Strategic Highway Safety Plan, Volumes 1-20. Transportation Research Board, Washington, DC, 2004. Available online at: http://safety.transportation.org/guides.aspx
  4. Gross, F., Persaud, B., and C. Lyon. A Guide to Developing Quality Crash Modification Factors. Publication FHWA-SA-10-032, Federal Highway Administration, Washington, D.C., 2010. Available online at: http://safety.fhwa.dot.gov/tools/crf/resources/fhwasa10032/
  5. Carter, D., R. Srinivasan, F. Gross, and F. Council. Recommended Protocols for Developing Crash Modification Factors. National Cooperative Highway Research Program, Transportation Research Board, February, 2012.
  6. Introduction to Crash Modification Factors. Publication FHWA-SA-13-015, Federal Highway Administration, Washington, D.C., 2013. Available online at: http://safety.fhwa.dot.gov/tools/crf/resources/cmfs/docs/intro.pdf
  7. Gross, F. and Yunk, K. "Crash Modification Factors: An Overview of Its Applications." Public Roads. Federal Highway Administration, Washington, D.C., 2011.
  8. Gross, F., Hamidi, A., and Yunk, K. Investigation of Existing and Alternative Methods for Combining Multiple CMFs. Federal Highway Administration, Washington, D.C., 2011.
  9. Introduction to Safety Performance Functions. Publication FHWA-SA-13-016, Federal Highway Administration, Washington, D.C., 2013. Available online at: http://safety.fhwa.dot.gov/tools/crf/resources/cmfs/docs/safety_performance_funtions.pdf
​​