Prediction of Driver Action at Signalized Intersections by Using a Nested Logit Model

Abstract

The primary objective of this research was to identify factors that influenced whether a vehicle would stop or proceed through, either legally or illegally, during the yellow indication when the vehicle was approaching a signalized intersection. For this objective to be accomplished, naturalistic driver behavioral data were collected at 72 signalized intersection approaches selected from four regions of the United States. Data were obtained for 6,208 vehicles that were approaching a study intersection during the yellow interval. With these data, a nested logit model was developed to investigate the influence of various factors on the likelihood that a driver approaching a signalized intersection during the yellow interval would stop, would proceed through legally, or would commit red light running (RLR). The nested logit model represented an improvement over prior binary logistic regression models, because it allowed the simultaneous estimation of all three potential driver actions. With the use of this model, RLR was determined to be more likely to occur under the following conditions: (a) the duration of the yellow interval was equal to or less than 4.5 s, (b) the subject vehicle was part of a platoon, (c) the approach speed limit was less than or equal to 40 mph, (d) the subject vehicle was farther from the intersection at the onset of the yellow indication, or (e) the subject vehicle approached at a lower rate of speed. The results may be used to improve the accuracy of algorithms for real-time prediction of RLR, including those used in conjunction with traffic signal phase extension systems.