Citation

Wang X, Abdel-Aty MA, Nevarez A, Santos J. Transp. Res. Rec. 2008; 2083: 86-95.

Copyright

(Copyright © 2008, Transportation Research Board, National Research Council, National Academy of Sciences USA, Publisher SAGE Publishing)

DOI

10.3141/2083-10

PMID

unavailable

Abstract

It is common in crash data recording and safety studies to designate crashes that occurred within a default radius from intersections as having occurred at or been influenced by intersections. The purpose of this study is to investigate thoroughly how intersection attributes affect this safety influence area and how the varied safety influence areas for intersection approaches improve safety analysis. First, a nationwide survey was conducted to review how police officers, crash records technicians, and state safety engineers identify intersection crashes. Data for a sample of 177 four-legged signalized intersections were collected from the state of Florida. On the basis of crash data and the approach features, the intersection's inside area (from the center to the stop bar, used to designate at-intersection crashes) and the safety influence area (measured upstream from the stop bar, used to designate intersection-related crashes) were explored by using classification trees. The research concluded that it is better to define at-intersection and intersection-related safety influence areas for each approach separately, and then to determine intersection influence area and size on the basis of approach features. Among the selected intersections, around 30% of the approaches had at least one intersection-related crash occurring beyond the physical boundary (left-turn lane), and these cases were examined. Test statistics showed that intersection-related crashes defined by a 250-ft boundary (the default used in many states) versus those defined by varied boundaries are significantly different. The approach-level intersection-related crashes were fitted by using the generalized estimating equations with a negative binomial. The significant factors identified were different for the models that used the fixed and the varied boundaries, and the model with varied influence areas had a better performance.