Collision prediction on combined horizontal and vertical alignments of two-lane rural highways

This study investigates the safety effects of combined horizontal and vertical alignments using accident occurrences on two-lane rural highways in Washington. Eight statistical models were developed to establish the relationships between vehicle accidents and their associated factors for eight combinations of alignments by the Poisson, negative binomial, zero-inflated Poisson, and zero-inflated negative binomial. Three selected models were validated. The findings show that degree of curvature is the most successful predictor for horizontal curves combined with vertical alignments. A minimum ratio of 25 of vertical curve radius to horizontal curve radius is recommended for a curve with radius of smaller than 6000 ft (or 1830 m). Vertical curves have relatively little influence on accident occurrences at horizontal tangents. The grade value and length of a grade increase accident occurrences when a horizontal curve or tangent is on a grade. A smaller curve should be avoided introducing at a steep grade.