Application of Poisson Underreporting Model to Examine Crash Frequencies at Signalized Three-Legged Intersections

Abstract

Before effective remedial treatments can be implemented at hazardous intersections, it is often necessary to identify the causal factors affecting accident frequency. However, a problem often encountered in safety studies is the underreporting of accidents. This biased reporting may affect the selection of the parsimonious model. This study analyzes the factors affecting road accident frequency at three-legged signalized intersection approaches in Singapore, with special emphasis to underreporting. The annual accidents at 104 three-legged signalized intersections are modeled as the sum of observed Poisson outcome of accident reports. The reporting mechanism is introduced as a probit equation. The model shows that several geometric, traffic, and traffic control factors significantly affected the accident occurrence. The total approach volumes, left-turn volumes, existence of unprotected left-turn slip roads, number of signal phases per cycle, use of permissive right-turning phase, and sites with sight distances less than 100 m or greater than 300 m appear to increase accident occurrence. However, the presence of right-turn channelization, provision of an acceleration section for left turning, existence of a surveillance camera, availability of median railings, and presence of an approach gradient greater than +5% may reduce the occurrence of intersection approach accidents. Moreover, the study shows that the reporting rate may drop because current law enforcement requires that only injury accidents and accidents at intersections in a residential area be reported.