Improved Prediction Models for Crash Types and Crash Severities (2021)

• May 4th, 2021

This report describes efforts to develop improved crash prediction methods for crash type and severity for the three facility types covered in the 2010 Highway Safety Manual (HSM)—specifically, two‐lane rural highways, multilane rural highways, and urban/suburban arterials. For each, models were estimated for undivided and divided (multilane rural and urban/suburban only) segments and three‐ and four‐leg stop‐controlled intersections and four‐leg signal‐controlled intersections (also three‐leg signal‐controlled intersections for urban/suburban arterials). The models use data for segments and intersections with “base conditions” that are defined specifically for each facility type. Only the observations that satisfy the defined base conditions were used for estimating these models. For urban/suburban arterial segments, because no sites met all base conditions for roadside fixed objects and median width, these variables were included in the models only if considered appropriate for that crash type and if the variable was statistically significant in the model and with the expected direction of effect. For some crash types, the number of driveways was also directly included in the models where warranted. These base condition models provide predictions that can be adjusted for actual conditions at a place of prediction, such as lane and shoulder width, the presence of lighting, and other pertinent factors. Content describing these models and instructions for applying them has been prepared for inclusion in the second edition of the HSM. A revisit of the HSM’s procedure for calibrating prediction models for transfer to other jurisdictions is also described and recommendations for updating that procedure offered. Average condition models were also estimated using all available valid data points available from the state data used for each facility type.

Read the full report here.