Corrections of Self-Selection Bias in Crash Causality Study: An Application on All-Red Signal Control
Abstract
All-red (AR) interval is designed as a method of clearance interval to safely clear vehicles that enter the signalized intersection. The provision of AR is generally expected to reduce the occurrence of crashes, though there are situations that AR is not proved to be effective because it is used at intersections with a higher potential for crashes. This controversial result however, does not indicate that the AR interval is a contributing cause of crashes. Therefore, the self-selection bias of signal designs needs to be corrected when estimating their effect in improving safety. To address the selection-bias problem at signalized intersections, a Heckman two-stage approach is adapted. First, a probit model is developed to explain the interrelationship between the AR interval and highway geometry, traffic volume and environmental variables. Second, the selection bias term (or Heckman correction) is included in the second stage to build two negative binomial models for locations with and without an AR interval. Three-year crash data on urban signalized intersections in the Detroit metro area is used to validate the proposed models. The results show that the impact of AR intervals is 51.7% reduction in total crashes on the treated intersections, and 42.5% on a random intersection.
Publication Title
Journal of Transportation Safety and Security
Recommended Citation
Mishra, S., & Zhu, X. (2015). Corrections of Self-Selection Bias in Crash Causality Study: An Application on All-Red Signal Control. Journal of Transportation Safety and Security, 7 (2), 107-123. https://doi.org/10.1080/19439962.2014.929603
