Two principle methodologies can be used for analyzing roadway safety:
Historic Performance Method - Uses crash history to identify or explore locations of concern.
Predictive Modeling Method - Calculates the predicted or expected crash frequency.
Most roadways suggested for analysis will include a combination of intersections and roadway segments. As such, intersection related crashes must be identified and removed from the roadway segments prior to any analysis that implements either method listed above. To ensure consistency, the following guidelines are recommended when separating intersection and roadway segment crashes:
Designate the intersection’s area of influence using the distances provided in the list below. These distances are based on a review of crash trends and typical designs used in Utah.
Crashes within the intersection’s area of influence that are flagged by the “intersection related” rollup field should be included in the intersection analysis.
All remaining crashes should be attributed to the roadway segment analysis.
Intersection Type and Area of Influence
Signal control 300 feet
Minor leg stop control 150 feet
All way stop control 100 feet
Yield control 100 feet
Uncontrolled 100 feet
Roundabout 300 feet
Offset Left-turn (CFI) 400 feet for primary intersection, 300 feet for upstream left-turn signals
Median Thru-U Turn 400 feet for primary intersection, 200 feet for downstream u-turn signals
R-Cut 400 feet for primary intersection, 300 feet for downstream left-turn signals
SPUI 500 feet
DDI 400 feet for each intersection
Active Transportation Only 100 feet
Railroad Crossing 100 feet
The Historic Performance method is used to identify locations where crash history suggests a problem may be present. This method is used for:
Corridor studies (simple studies only, where no change in roadway configuration is anticipated).
Evaluation of crash “hot spots”.
Providing guidance for project planning and scoping. A Project Safety Analysis (PSA) can also be used to complete a roadway segment or intersection analysis.
Short roadway segments and single rural intersections will often have only a few crashes to investigate. The predictive modeling method describe below in this article should be used in this situation when there is insufficient crash history to gain insight from the historic performance method.
The Historic Performance method addresses the following questions:
What roadway attributes increase crash risk?
What are the predominant crash types?
Are crash types disproportionate?
How might disproportionate crashes be mitigated?
Which mitigation efforts are viable?
Typically the Historic Performance Method uses a 5 year crash history. This crash history time period should be adjusted to reflect changes resulting from recent construction or significant changes in daily traffic volumes, which make this method non-representative. If changes to the segment or intersection have recently occurred, the predictive modeling method is preferred for analysis.
It is important to review the physical attributes of the roadway or intersection being analyzed. Roadway attributes help the analyst to understand crash risks and potential mitigations. Attributes of interest include the following:
Area type (the “Urban/Rural” rollup filter in AASHTOWare Safety software can be used to determine the area type).
AADT and projected AADT if available.
Number of lanes and presence of passing lanes or two-way-left-turn lanes.
Posted speed limits.
Shoulder width, both average width and the proportion of roadway with shoulders 2-feet wide or less.
Median treatments.
Driveway density.
Types of pedestrian/bicycle facilities.
Presence of roadway lighting.
Known safety elements, e.g. rumble strips, barrier, etc.
Intersection lane configuration, including turn lanes.
Intersection control type.
Attribute summaries are often presented as a proportion of the overall roadway segment. Where there are significant differences in a particular attribute, such as area type or number of lanes, the analyst should segment the analysis and report results separately.
The Historic Performance method includes a summary of the crashes observed, total crash counts, calculated crash rates, and/or crash proportions.
Predominate manners of collision are often evaluated more thoroughly by looking for similar crash attributes. The FHWA has developed a Crash Tree tool spreadsheet that can be used to filter crashes by attributes of interest. Generally the analyst should consider each of the following:
Severe crashes: Review crash narratives to ensure that crash attributes were correctly reported.
Lower severity crashes: Review attributes to find patterns that correlate with severe crashes. This helps explain and understand crash patterns.
Crashes can be disproportionate in both quantity (i.e. the proportion of a crash type is elevated), and in severity (i.e. the severity distribution of crashes is skewed). Consider the following items when identifying disproportionate crashes:
How does the crash severity distribution compare to Utah specific crash severity distributions?
How do the manner of collision proportions compare to typical proportions?
The list below shows crash types and attributes that should be reviewed when using the Historic Performance Method.
Roadway Segment Analysis
Pedestrian and bicycle
High risk drivers, including teens and seniors
Night/dark
Commercial motor vehicles
Driveway/access related
Roadway departure (including head-on and sideswipe opposite direction)
Speed related
Direction of travel
Wildlife related
Intersection Analysis
Pedestrian and bicycle
High risk drivers, including teens and seniors
Night/dark
Time of day
Red light running or disregard traffic control
Direction of travel
Vehicle maneuver
Tables 1-6 in the Google Sheet below show typical crash type proportions for various roadway and intersection types in Utah based on 2016-2020 crash data. These tables are updated regularly as new information becomes available. Proportions at a given site can vary dramatically based on the number of crashes that have occurred. Engineering judgment is required in determining whether a crash type is overrepresented.
The analyst should report all crash patterns that are deemed disproportionate. In some cases disproportionate crashes cannot be completely identified. The analyst should report the most prevalent crashes and also focus on ways to reduce their frequency.
Table 1: Crash Type Proportions by Roadway Types (2017-2021 crash data)
Table 2: Crash type proportions by intersection type (2016-2020 crash data, sample size shown in parenthesis)
Table 3: Average crash severity by roadway type (2017-2021 crash data, excludes intersection related crashes)
Table 4: Average crash severity by intersection type (2016-2020 crash data, sample size shown in parenthesis)
Table 5: Manner of collision proportions for roadway segments (2017-2021 crash data)
Table 6: Manner of collision proportions for intersections (2016-2020 crash data)
The predictive modeling method is the preferred analysis method outlined in the Highway Safety Manual (HSM). UDOT currently uses HSM algorithms coupled with UDOT specific calibration factors and distributions. Crash prediction can be used to do the following:
Quantify the intensity of a problem identified through a historic performance method analysis.
Compare future alternatives:
Corridor Studies (when complex situations exist or when comparing roadway type alternatives).
Interchange Access Change Request (IACR) safety analysis.
Environmental Impact Study (EIS) safety analysis.
Evaluating the safety impacts of design exceptions.
Comparing transportation planning alternatives.
Determine safety benefits of improvements where crash history is limited.
The Predictive Modeling method is dependent on safety performance functions (SPF), crash severity distributions, crash costs, and the alternatives being compared. If the alternatives being compared require a change in the SPF (e.g. changing an intersection from stop controlled to signal controlled), only the predictive values should be compared, no influence from current history can be applied. If the analysis focuses on changes to the same SPF (e.g. change in the CMFs being applied, or a change in AADT over time) then the expected crashes should be used in the analysis. Expected crashes are calculated using the weighting method outlined in the HSM.
The FHWA and AASHTO have developed spreadsheet tools and the Interactive Highway Safety Design Model (IHSDM) to calculate these values. The analyst should keep the HSM available when using these models. Some of the descriptions and limitations used as input are unique to the HSM and may conflict with the understanding of the analyst. UDOT currently uses HSM SPFs. Changes in facility types not found in the HSM are modeled with available CMFs (e.g. roundabouts, Hi-T, R-CUTs, etc.). Tables 7 and 8 include UDOT specific calibration factors that should be used with the SPFs available in the HSM.
Table 7: UDOT preliminary calibration factors for fatal and injury SPFs of the HSM (2010) for roadway segments
Table 8: UDOT preliminary calibration factors for SPFs of the HSM (2010) for intersection total crashes
Additional calibration factors will be provided in the future. SPFs not included in Tables 8 or 9 should assume a calibration factor of 1.0 unless more specific information is available.
When a severity distribution is available, the predicted or expected total crashes are used with unweighted crash costs. When severity distributions are not available, the predicted or expected fatal/injury (KABC) and the PDO crash frequencies are used with the weighted crash costs.
Use of the Predictive Modeling method to compare alternatives can become very complex. Analysts should keep in mind the following issues:
Not all design changes can be addressed through predictive modeling, only proposed changes that have quantifiable impacts or have viable CMFs can be calculated.
Roadway network impacts associated with an IACR or EIS can include multiple adjacent roadway segments and/or intersections. It is imperative that the design team agree on the analysis extents prior to beginning the analysis.
When using the same SPFs for various roadway or intersection configurations, the historic expected to predicted crash ratio is applied to the predicted future condition to calculate the expected future crashes.
There are several resources that provide AADT growth estimates, AADT estimates should be incorporated into the predictive method to account for future conditions.
Some roadway or intersection alternatives require creative solutions to obtain reasonable results for comparison purposes. All analysis processes and assumptions should be clearly documented.
As UDOT develops their own calibration factors, SPFs, CMFs, severity distributions, and manner of collision proportions, the predictive modeling method will become more accurate.