Estimation of carrier accident risk potential

Abstract

The cost of accidents on North America's highways is estimated to be in the billions of dollars annually. Truck safety represents a major safety issue and is of special interest to policy makers. Trucks are involved in a disproportionate number of accidents with respect to their population. Previous studies have analyzed truck safety from a number of perspectives including those of the highway, the driver, the vehicle and the environment. However, few studies have viewed this issue from a carrier perspective. In Ontario, 100% of truck accidents are attributed to 7% of carriers. If we can target carriers who are most at risk of accidents, we may be able to reduce a significant percentage of truck accidents. Carrier attributes and safety management practices can explain large truck accident involvements on a carrier specific basis. Authorities need to understand the accident risk potential of carriers and to identify those who are most at risk of accidents. Those carriers can then be selected for safety audits and other safety interventions.

Safety audits are expensive and, as a result, only a small percentage of carriers are audited in any given year. In Ontario, less than 1% of all carriers are normally audited. This statistic means that the probability of selecting high risk carriers is small. An efficient procedure needs to be established to identify for auditing those carriers with the greatest accident potential. The main objective of this research is to develop a methodology to identify those carriers that have the highest accident potential.

Analysis of Ontario accident data by carrier size underscored a difference between small and large carriers in terms of their respective accident rates per truck. Small carriers within the context of this study are those that have a fleet size of 10 trucks or less. Statistical models which are based on accident history (truck accident involvement in previous years), have proven to be a reasonably good predictor of future accident experience, for large carriers. Large carriers that have experienced a high number of accidents in previous years are likely to have a large number of accidents in the future and vice versa. For small carriers, however, accident history does not provide a good predictor for future accident behaviour. Other attributes of small carriers are needed to provide inference concerning future accident behaviour.

Small carriers comprise over 93% of all carriers registered to operate in Ontario and hence, identifying high risk carriers in the small carriers population is more difficult and more important. For the period of 1991 to 1995 in Ontario, small carriers represented 30% of all trucks registered in the province. These same carriers accounted for over 50% of all detentions and convictions for drivers and vehicles infractions. Thus, for small carriers, detentions and convictions could provide inference concerning future accident involvements.

In this thesis, an Empirical Bayes approach was adopted to identify small carriers with the highest risk of having accidents. The model was developed and tested based on the Carrier Vehicle Operator Record (CVOR) database of Ontario for the period of 1991 to 1995. This model makes use of statistical information on carrier convictions and detentions as prior and accident history as likelihood. The carrier attributes and accident data for the period of 1991 to 1994 were used to develop the model which was then tested and evaluated using the 1995 data.

Different sampling procedures were selected to be compared with the Bayes model. These included a) random selection based on the number of carriers in the population, b) random selection based on the number of trucks in the population, c) selection based on the previous one year of accident history, and d) selection based on the previous 2 years of accident history. Regardless of the number of carriers being sampled, the results of this analysis indicated that the Bayes approach performed significantly better than all other procedures in identifying carriers who experienced accidents in 1995. On average, the Bayes model identified between 3000% to 60% more carriers are compared to methods b) and d), respectively.

We found that for small carriers, detentions and conviction are significant factors in explaining accident involvements. While other variables might further explain accident involvements, many of these variables are unavailable in existing databases. The Bayes model makes an efficient use of the existing data collected in Ontario and does not require any additional data which are usually not available or are too expensive to collect.

This research offers a different approach to highway safety and the potential reduction of truck accidents by investigating the carriers' perspective. The research provides a methodology for identifying high risk carriers that should be targeted for safety interventions which would lead to the improvement of truck safety on highways.