Crash Estimation at Signalized Intersections along Corridors

Abstract

Intersections could be considered as isolated when the distance between them is long because, the influence between them is negligible. Signalized intersections, especially closer ones along a certain corridor, are spatially correlated and will influence each other in many respects. Use of the basic negative binomial regression for correlated crash frequency data leads to invalid statistical inference due to incorrect test statistics and standard errors that are based on the misspecified variance. Generalized estimating equations (GEE) provide an extension of generalized linear models to the analysis of correlated data and can account for the spatial correlation among signalized intersections. In this study, 476 signalized intersections from 41 corridors are selected in Orange, Brevard, and Miami–Dade Counties in Florida. Because the distance between some intersections along some corridors is very long, the intersections along the 41 corridors were divided into 116 clusters. The spatially correlated crash frequency data were fitted through use of GEE models with a negative binomial link function for three correlation structures. Subsequent relative effect analysis identified the relative significance for the variables in the models. Intersections with a large total number of lanes, heavy traffic, short signal spacing, high speed limits along corridors, and a large number of phases per cycle were correlated with high crash frequencies. Intersections having three legs, with exclusive right-turn lanes on both roadways, having a protected phase for left-turning traffic from a corridor, and located in open county or primarily residential areas had lower crash frequencies.