The National Academies

NCHRP 17-85 [Active]

Development and Application of Crash Severity Models for the Highway Safety Manual

  Project Data
Funds: $600,000
Staff Responsibility: David M. Jared
Research Agency: University of Connecticut
Principal Investigator: John Ivan
Effective Date: 11/15/2018
Completion Date: 11/15/2021


As the Highway Safety Manual (HSM) continues to evolve, determining the potential crash severity at a location becomes an increasingly vital component in predicting safety performance. For safety performance functions (SPFs) to be reliable, they must be consistent. Consistency in how states apply stand-alone severity or severity with frequency based tools and estimates are a fundamental requirement for the adoption and use of the HSM and its associated tools. Consistency is a function of several factors: (1) results that are in general agreement from a multitude of analytical techniques available to practitioner; (2) availability of data sources that allow for broad state and interstate analysis; and (3) interpretability of results for policy application at the national, state, and local agency levels.

Severity analysis tools, as currently available in the HSM, do not fully meet this definition of consistency–for various reasons, but primarily originating from the fact that the analytical techniques available in the published literature are not consistent in their estimation of crash severity probabilities and frequencies. The differences in severity estimations can be significant–from the methods currently in use that adopt the observed severity ratios to emerging methods which analyze the severity aspects at multiple scales–from spot location, to corridor and network levels.

Three factors are critical to adoption of safety analysis techniques–integrity with network screening methods (Part B of the HSM), data availability and requirements, and model predictive performance (Part C of the HSM). The HSM provides methods for model calibration and qualitative guidance on the model reliability, but lacks methods to identify which crash severity levels and crash types need to be modeled to ensure consistency. The default approach recommends using observed crash severity ratios for a facility type and using those factors to obtain severity specific expected values. Data quality, crash severity reporting, and how to model crash severity level combinations are significant challenges in terms of crash prediction reliability.

Crash prediction model results are currently used to make planning and project-level decisions without complete understanding of their reliability. Specific examples include: (1) lack of understanding of confounding factors due to inaccuracy of severity estimations; (2) poor understanding of models and their limitations; (3) use of models at or near their limits; and (4) inappropriate transfer of severity ratios and severity level definitions without consideration of location specific temporal and spatial factors. Continuing the use of models in this way may lead to suboptimal design of projects, degradation of model credibility, and open concerns of liability and public trust.


The objectives of this project are to: (1) assess the current HSM approaches to severity estimation and prediction using SPFs; (2) Identify gaps and opportunities in the current severity prediction/estimation procedures within the HSM; (3) Develop and validate new severity models to address the gaps and opportunities; and (4) develop a guidance document that includes protocols for the use and application of severity based models in a format suitable for possible adoption in the HSM.

The guidance and protocols should address:

  • Both the current HSM methods and any newly created models that consider, but are not limited to the following potential approaches:
    1. Number of crashes at each severity level by predicating total crash count and adjusting by assumed or observed crash severity distribution.
    2. Number of crashes at each severity level by predicting crashes for each of the individual severity categories.
    3. Estimated probability of crashes at each severity level.
    4. Number of crashes at each severity level by predicting both the total crashes and the crash severity likelihoods.
  • How crash frequency and severity models would be integrated.
  • New crash models shall include but are not limited to discrete choice models.
  • Statistical reliability in terms of, but not limited to, spatial, temporal, and statistical specification and fit.
  • Data quality and availability and how each affects the models application at both the national and local levels.
  • The criteria for model development and selection.
  • The use of the models for network screening (HSM Part B) and predictive methods (HSM Part C).
  • How project planning and design decisions are affected by different modeling approaches.
  • The performance of the models at varying AADTs, segment lengths, roadway functional classes, and contexts (e.g. roadside conditions, and speeds).

The guidance should include a number of case studies or illustrative examples that demonstrate the issues with model use and statistical reliability.

In addition, the guidelines should promote better informed planning decisions and the use of engineering judgment in the model application, greater implementation of crash prediction models, and better acceptance of model results leading to improved decision making.


The NCHRP is seeking the insights of proposers on how best to achieve the research objectives. Proposers are expected to describe research plans that can realistically be accomplished within the constraints of available funds and contract time. Proposals must present the proposers' current thinking in sufficient detail to demonstrate their understanding of the issues and the soundness of their approach to meeting the research objectives.

A kick-off web-conference call with the research team and NCHRP shall be scheduled within 1 month of the contract’s execution. The research plan proposed must be divided into tasks, with each task described in detail. The research plan must be divided into two phases.

Phase 1 will consist of information gathering and planning tasks culminating in the submittal of an interim report. The interim report will describe the work completed in the Phase 1 tasks and provide an updated research plan for the Phase 2 tasks as well as an outline of the framework and protocol at a minimum. The updated Phase 2 research plan should address the manner in which the proposer intends to use the information obtained in Phase 1 to satisfy the project objective. At the completion of Phase 1, a face-to-face meeting with NCHRP will be scheduled to discuss the interim report. Work on Phase 2 tasks will not begin until the updated research plan is approved by NCHRP. The project schedule shall include 1 month for NCHRP review and approval of the interim report.

The final deliverables will include, but not be limited to: (1) the guidance document in a format suitable for possible incorporation in a future edition of the AASHTO Highway Safety Manual; (2) a final report documenting the entire project and incorporating all other specified deliverable products of the research including the database used in the analyses and documentation sufficient for use by other research professionals; (3) an executive summary that outlines the research results; (4) a PowerPoint presentation with speaker notes describing the project background, objective, research methods, findings, and conclusions suitable for use in a webinar; and (5) a stand-alone memorandum titled “Implementation of Research Findings and Products” (see Special Note E for additional information).

The research plan may include additional deliverables as well as additional meetings with the NCHRP project panel via web-conferences. All web-conference calls will be hosted by NCHRP.

Status: (Assume) Research in progress.

To create a link to this page, use this URL: http://apps.trb.org/cmsfeed/TRBNetProjectDisplay.asp?ProjectID=4371