Current crash prediction methods—such as those in the AASHTO Highway Safety Manual (HSM)—consist of safety performance functions (SPF), crash modification factors (CMF), and severity distribution functions (SDF). These tools use annual average daily traffic (AADT) data along with geometric and operational characteristics to predict annual average crash frequency of roadway sites. While these models have statistical merit, they do not allow users to accurately predict crashes for variable, short-term periods (e.g., peak periods, special events). This can be an issue for agencies wanting to assess the safety impacts of temporary works zones or time-of-day capacity changes (e.g., lane configuration or speed limit). The existing annual prediction convention also limits the models’ ability to quantify the effects of variables that fluctuate throughout the day (e.g., operating speeds, or operating speed variance). State departments of transportation require the ability to more accurately assess daily or hourly changes that could affect crash outcomes.
The objectives of this research were to: (1) develop short-term crash prediction models to estimate the safety performance of roadways. Consideration should be given to specific geometric, operational, and exposure characteristics (e.g., detours, variable speed limits), and routes that experience short-term capacity changes (e.g., hard shoulder running, reversible lanes). (2) Identify explanatory variables measured over short durations. Include more precise measures of exposure other than AADT (e.g., peak hour volume, special event volume), and factors such as speed (including speed variability) to help predict crashes for varied periods of time. (3) Develop an implementation tool suitable for practitioner use.
Status: The research is complete. Publication decision is pending.