Comparing IT-based driver assessment results against self-reported and actual crash outcomes in a large motor vehicle fleet
This paper reports on three recent studies undertaken by Napier University s Centre for Mathematics and Statistics on behalf of Interactive Driving Systems. The studies were based on a large group of British Telecom (BT) van and company car drivers, who had undertaken an on-line RoadRISK assessment of their attitude, hazard perception, behaviour, knowledge and personal exposure.
The study had the following four aims.
1. Identify the most at-risk drivers.
2. Compare driver assessment scores against crash outcomes. 3. Evaluate the capabilities of the RoadRISK driver assessment program to predict likely crash outcomes.
4. Compare the outcomes from self reported crash data against insurance claims.
To achieve these aims, the following three studies were undertaken using cross tabulation and logistic regression.
1. In March 2003 the overall assessment scores of just over 8,000 drivers were compared against their self reported crashes.
2. In October 2003 the individual section scores of just over 16,000 drivers were compared against their self reported crashes.
3. In March 2004 the individual section scores of just over 16,000 drivers were compared against their actual crashes (based on insurance claims data).
The main results from these studies were as follows.
1. A relatively small group of drivers appeared to be involved in a disproportionate number of crashes.
2. There was a positive relationship between driver assessment scores and crash outcomes for both the self-reported and actual crash data.
3. The system has some capability to be used as a predictor of likely crash outcomes for fleet drivers, but it should not be seen as a substitute for other best practice fleet risk management processes.
4. Recommendations were made on how the assessment tool could be improved.
Finally the paper discusses some of the limitations of the project. Many of these were data related, such as the problems with self-reported crash data, the processes involved in obtaining accurate claims information in large vehicle fleets and the dilemma of evaluating event-led driver training programs in an objective way.