Manuscript Number: TE/2006/023538 MODELING THE PROBABILITY OF FREEWAY REAR-END CRASH OCCURRENCE by

A microscopic model of freeway rear-end crash risk is developed based on a modified negative binomial regression and estimated using Washington State data. Compared with most existing models, this model has two major advantages: 1) it directly considers a driver’s response time distribution; and 2) it applies a new dual-impact structure accounting for the probability of both, a vehicle becoming an obstacle ( o P ) and the following vehicle’s reaction failure ( f P ). The results show for example that truck percentage-mile-per-lane has a dual impact, it increases o P and decreases f P , yielding a net decrease in rear-end crash probabilities. Urban area, curvature, off-ramp and merge, shoulder width, and merge section are factors found to increase rear-end crash probabilities. Daily VMT per lane has a dual impact, it decreases o P and increases f P , yielding a net increase, indicating for example that focusing VMT related safety improvement efforts on reducing drivers’ failure to avoid crashes, such as crash-avoidance systems, is of key importance. Understanding such dual impacts is important for selecting and evaluating safety improvement plans for freeways.