Development of Tools for Short-term Monitoring of Road Accident Trends
V Gitelman, A-S Hakkert, E Doveh, A Cohen, Transportation Research Institute, IL
Development of a tool for monitoring accident trends to assist in decision making, similar to the mechanism of control charts used in industry.
Road accident trends are under permanent focus of road safety authorities. However, these trends are subject to long-term changes and temporal fluctuations. Thus, in current road safety work the frequently asked questions are: do the changes observed in accident/ injury/ fatality numbers over the last (1-2) year indicate a significant increase/ decrease in those numbers? If significant changes in road accident trends took place, which road safety areas were involved? Answers on such questions have obvious practical implications as they provide a basis for carrying out necessary changes in the current plans of road safety interventions. Moreover, for the monitoring of accident trends it would be useful to have decision-making tools enabling to judge monthly changes in road accident figures, similarly to the mechanism of control charts in industry. In Israeli practice, such questions were raised by the National Road Safety Authority when a relative increase was observed in the national fatality numbers in 2001-2002. (Presently, these questions are under examination again as a 10% decrease was observed in the annual fatality numbers in 2005-2006).
The purpose of this paper is to present the analysis method and statistical tools which were developed for monitoring the short-term national accident trends. The study considered the changes in fatalities and severe injuries in 2001-2002 in comparison with previous years, 1995-2000. The research questions were: (a) was the change in the accident trend over the last years significant? (b) Did a significant change occur in the number of injuries in the last years as opposed to the previous years? (c) Had the trend of the previous years continued in 2001-2002, would the injury numbers be different from those actually observed? (d) Did the process in 2001-2002 remain within the boundaries of statistical quality control? The analysis considered both general accident injury numbers and their sub-divisions according to the main representative areas (which may be helpful for further planning of interventions) such as: urban versus rural roads, heavily-travelled rural roads, accidents with selected vehicle types (private and commercial light vehicles, trucks, buses and taxis).
The analysis was based on fitting explanatory models to the monthly frequencies. The model applied a Poisson distribution and accounted for seasonal effects (monthly and yearly where available). Depending on the research questions considered, the models were fitted to the previous years in order to predict the injury numbers for the years examined (2001-2002) and compare the numbers predicted with actually observed. One model was fitted for the whole period (1995-2002) in order to examine "breaking points" for the trends at the end of 2000 and of 2001. The control charts were fitted to the series of residuals of accident injuries, where the "mean line" and the boundaries were assigned for monthly figures of 2001-2002 based on the numbers observed in 1995-2000. For the control charts, types of events indicating a deviation from the "normal" process were defined for the application.
The findings were that in 2001-2002 a significant increase was observed in fatality numbers on rural roads, including heavily-traveled roads and accidents with private cars; at the same time, a significant decrease was observed in fatalities in urban areas. The conclusions were consistent through all the research questions considered. The application of control charts enabled both an earlier identification of changes and recognition of "corrected" trends in some processes.
Association for European Transport