Linking roadway geometrics and real-time traffic characteristics to model daytime freeway crashes - Generalized estimating equations for correlated data

    Authors

    M. Abdel-Aty; M. F. Abdalla;Trb

    Comments

    Authors: contact us about adding a copy of your work at STARS@ucf.edu

    Keywords

    LONGITUDINAL DATA; LINEAR-MODELS; ACCIDENTS; Engineering, Civil; Mathematics, Interdisciplinary Applications; Statistics & Probability; Transportation Science & Technology

    Abstract

    A model was developed to predict daytime crashes on a freeway when real-time traffic flow characteristics and roadway geometric features are known. The data used were for 337 daytime crashes that occurred on Interstate 4 in 1999 in Orlando, Florida (13.25 mi), where corresponding loop detector traffic data were available upstream and downstream of the location of every crash and for at least 30 min before the crash occurred. The traffic data included average speed, volume, and occupancy rate for every 30 s at 0.5-mi intervals. Traffic characteristics for noncrash cases at the same locations and at similar times of the crash cases were extracted and fed to the model. The generalized estimating equations technique with binomial probit link function was used to account for correlation between crashes that occurred at the same location. Three different correlation structures (independent, exchangeable, autoregressive) were used, discussed, and compared. The modeling results showed that the existence of an on-ramp increases the likelihood of a crash within 0.5 mi downstream of the crash location. High variability in speed for a period of 15 min in a certain location was shown to increase the likelihood of a crash 0.5 mi downstream. Unlike speed, low variability in volume over 15 min was shown to increase the likelihood of a crash at 1 mi downstream.

    Journal Title

    Statistical Methods and Safety Data Analysis and Evaluation

    Issue/Number

    1897

    Publication Date

    1-1-2004

    Document Type

    Article

    Language

    English

    First Page

    106

    Last Page

    115

    WOS Identifier

    WOS:000228263900014

    ISSN

    0361-1981; 0-309-09495-X

    Share

    COinS