Sensitivity analysis of a model can help us determine relative effects of model parameters on model results. In this study, the sensitivity of the accident prediction model proposed by Zegeer et al. [Zegeer, C.V., Reinfurt, D., Hummer, J., Herf, L., Hunter, W., 1987. Safety Effect of Cross-section Design for Two-lane Roads, vols. 1–2. Report FHWA-RD-87/008 and 009 Federal Highway Administration, Department of Transportation, USA] to its parameters was investigated by the fractional factorial analysis method. The reason for selecting this particular model is that it incorporates both traffic and road geometry parameters besides terrain characteristics. The evaluation of sensitivity analysis indicated that average daily traffic (ADT), lane width (W), width of paved shoulder (PA), median (H) and their interactions (i.e., ADT–W, ADT–PA and ADT–H) have significant effects on number of accidents. Based on the absolute value of parameter effects at the three- and two-standard deviation thresholds ADT was found to be of primary importance, while the remaining identified parameters seemed to be of secondary importance. This agrees with the fact that ADT is among the most effective parameters to determine road geometry and therefore, it is directly related to number of accidents. Overall, the fractional factorial method was found to be an efficient tool to examine the relative importance of the selected accident prediction model parameters.
Rapidly growing population and urbanization result in a significant increase in number of vehicles and traffic accidents as well. Unfortunately, nearly one million people lose their lives and about ten millions face different injuries worldwide in various traffic accidents every year. Additionally, millions of dollars are lost due to accidents and a significant amount of money is spent for health treatment and rehabilitation. All these have made transportation planners and transportation engineers consider road safety more seriously. Designing safer roads and/or rehabilitation of current road geometries are among important factors in reducing traffic accidents.
Traffic accidents take place due to a number of factors such as road, environment, driver and vehicle, and their interrelationship. Therefore, traffic accident prediction models are developed including certain parameters to represent the effect of these factors. Since average daily traffic is considered the most effective parameter many researchers have taken it as the only parameter in their models (McDonald, 1953, Webb, 1955, Pickering et al., 1986, Hauer et al., 1988 and Bonneson and McCoy, 1993 among others). In order to obtain more reasonable accident predictions, road geometry elements were also included in some accident prediction models (e.g., Belmont, 1954, Zegeer et al., 1987, Zegeer and Deacon, 1987 and Li et al., 1994).
Sensitivity analysis of a model is very useful to determine relative effects of model parameters on model results. The most commonly used sensitivity method is the traditional “change one-factor-at-a-time” approach. The major weakness of this method is its inability of identifying multiple factor interactions among the model parameters. As discussed in the following section, several studies have been presented in the literature to investigate the sensitivity of accident prediction models to traffic and road geometry parameters by the standard “change one-factor-at-a-time” approach. However, in addition to main parameter effects, multiple factor interactions should also be identified to better understand internal model dynamics of accident prediction models.
As an alternative approach, the factorial design method developed by Box et al. (1978) has been employed in various sensitivity analyses of environmental models (e.g., Henderson-Sellers, 1992, Henderson-Sellers, 1993, Liang, 1994, Liong et al., 1995, Barros, 1996, Henderson-Sellers and Henderson-Sellers, 1996, Yildiz, 2001 and Gebremichael and Barros, 2005). Unlike the traditional approach, this method has the advantage of testing the sensitivity of model results: (a) to changes in main parameters and (b) to interactions between groups of parameters.
The objective of this study is to utilize the factorial design method in the sensitivity analysis of a widely used accident prediction model proposed by Zegeer et al. (1987) which includes both traffic and road geometry parameters besides terrain characteristics.
The sensitivity of a selected accident prediction model to its model parameters was analyzed by the fractional factorial analysis method. In the study, the accident prediction model proposed by Zegeer et al. (1987) was used since it incorporates both traffic and road geometry parameters in addition to terrain characteristics. The evaluation of sensitivity analysis revealed that main parameters ADT, W, PA and H, and two-parameter interactions ADT–W, ADT–PA and ADT–H have significant effects on number of accidents. Based on the absolute value of parameter effects at the 3σ and 2σ thresholds ADT was found to be of primary importance, while the remaining identified parameters and parameter interactions seemed to be of secondary importance. This is in agreement with the fact that ADT is an effective parameter in determining road geometry and hence is directly related to number of accidents. In conclusion, the sensitivity analysis performed in this study shows that the fractional factorial design method is an efficient way of examining the relative importance of the selected accident prediction model parameters.
