Research Papers (2009 – 2013)

Filename 4B-Heming-Jiang.pdf
Filesize 136.34 KB
Version 1
Date added May 7, 2014
Downloaded 8 times
Category 2011 CMRSC XXI Halifax
Tags Session 4B
Author/Auteur Heming Jiang

Abstract

1 Purpose. Over the past three years, the City of Edmonton has annually averaged 28,808 total collisions, resulting in an average of 4,725 injury collisions and more than $97,000,000 in property damage. On average, collisions at intersections contribute to 48% of total collisions, 65% of injury collisions, and 47% of fatal collisions. Red light cameras (RLC) are installed to deter drivers from running a red light, which can result in serious right angle vehicle collisions. RLCs automatically detect when a vehicle enters an intersection after the signal has turned red and takes a photograph of the violation and the vehicle’s license plate. The proposed model is different than the existing ones in the literature as it links the violations and issued tickets with the collisions. The purpose of the present study was to assess the “odds” of RLC locations to be high collision intersections (HCIs). The odds here represent the ratio of the probability of an intersection being a HCI over the probability of not being a HCI. The study assesses the “odds” of RLC locations to be high collision intersections (HCIs) and is able to rank the red light camera locations based on the probability of being a high collision intersection given either the violation rate or issued ticket rate. Method. Three years of City of Edmonton collision and RLC violations and issued tickets data from 2007-2009 were used for the analysis. The threshold for a high collision intersection was determined based on the median of red light running related collision rates of all RLC locations. An RLC location is considered as a HCI if the total collision rate exceeds the threshold. Binary logistic regression was used to assess the probability of HCIs given the violation rate. A similar analysis was also done to assess the probability of HCIs given the issued ticket rate. We also performed separate analysis on followed too closely collisions. Results. For any given RLC location in 3 years, with every one unit increase in violation rate, we expect to see about a statistically significant 7% decrease in the odds of being a HCI, about 3% decrease in the odds of being a high fatal and injury collision intersection, and about 4% decrease in the odds of being a high property damage only collision intersection. Furthermore, in 3 years, with every one unit increase in issued ticket rate, we expect to see about a statistically significant 9% decrease in the odds of being a HCI, about 4% decrease in the odds of being a high fatal and injury collision intersection, and about a 4% decrease in the odds of being a high property damage only collision intersection. These relationships suggest that the red light camera program was effective at reducing red light running related collisions. This program was slightly more effective in reducing collisions after the violators started receiving their tickets. This relationship is true regardless of the type of collisions (fatal and injury as well as property damage only collisions). We also computed rankings of the top 10 RLC sites in terms of probability of being high collision intersections. Conclusion. Overall, we see a statistically significant decrease in the odds of being a HCI with every one unit increase in violation/issued ticket rate. Also, the decrease in odds of being a high total collision location are higher than the decrease in odds of being a high injury/fatal or PDO collision location. When we change the predictor from violation rate to issued ticket rate, the effects do not seem to change. Total collision rates are better indicator response variables in predicting high collision intersections as the p-values show statistical significance at 5% level.

Heming Jiang