Calibration of Highway Safety Manual Crash Prediction Models for Rural Intersections: A Case Study from Delaware

Abstract

This study presents a calibration of crash prediction models from the Highway Safety Manual (HSM) using localized data for rural intersections in the state of Delaware. The research evaluates six intersection site types under two crash assignment scenarios: (1) a 250-foot circular buffer around all intersections, and (2) a 528-foot buffer applied only to rural multilane divided highways, aligning with DelDOT’s methodology. Geographic Information System tools, satellite imagery, and Google Street View were used for data collection, classification, and validation. Calibration factors (CFs) were computed for each site type and assessed using modified R-squared, coefficient of variation, and cumulative residual (CURE) plots. Findings indicate that while the 528-foot scenario produced calibration factors 37% higher on average, it did not lead to improved model fit and may overestimate intersection-related crashes. Therefore, the 250-foot scenario is recommended for future applications. Among the site types analyzed, only RM4STM (rural, multilane, four-legged, stop-controlled) demonstrated an acceptable goodness-of-fit and is suitable for implementation. Delaware-specific crash distributions were also developed to replace default HSM values, including crash severity levels, collision types, and nighttime crash proportions. This study highlights the importance of periodic, data-driven calibration and offers a reproducible methodology for transportation agencies seeking to enhance the accuracy of safety performance models and prioritize roadway safety interventions more effectively.