The objective of this report is to carry out a sensitivity analysis to assess the overall robustness of the road safety outcomes due to the uncertainty in various model input factors. Moreover, input factors that propagate most variance in the model output can be determined. In this research, the model output are the road safety outcomes, that is, the number of injury accidents saved while input factors are variables influencing the road safety outcomes. The sensitivity of the number of injury accidents saved is assessed using six factors (underreporting factor for injury accidents, modification factor for the autonomous risk, growth factor in traffic performance, the effectiveness of alcohol or drugs checks, the combined effect of automatic warnings of queues with variable signs and congestion warning signals (considered dependent or independent) and the combined effect of signs showing recommended speed in curves and new guardrails along embankments (considered dependent or independent)). By means of a computational model (Nambuusi et al., 2009 based on Reurings and Wijnen, 2008) the number of injury accidents saved in a fictitious case study is determined. Variance based sensitivity analysis methods (Saltelli (2002) and Saltelli et al. (2008)) are utilized to identify the factors that cause most uncertainty in the number of injury accidents saved. Such methods have advantages of being model-free, the ability to capture interaction effects in addition to the fractional contribution of an input factor to the variance of the model output and straightforward to interpret. The fictitious application (aimed at illustrating the sensitivity methodology) results in an average number of injury accidents saved by 2010 of 2,357 with a standard deviation of 159, a minimum of 1,774 and a maximum of 2,954. This wide range of the number of injury accidents saved implies a high uncertainty given the possible values of the input factors. The factors contributing to this uncertainty have been identified by means of a sensitivity analysis. The sensitivity analysis results have shown that considered singularly, the growth factor in traffic performance causes most uncertainty in the number of injury accidents saved. Moreover, the underreporting factor for injury accidents, the modification factor for the risk and the combined effect of two measures (new guard rails along embankments and signs showing recommended speed in curves) also have large first-order sensitivity indices. Further, the total-effect sensitivity indices (accounting for interaction effects) mainly indicate the effectiveness of alcohol or drug checks and the combined effect of two measures (automatic warnings of queues with variable signs and congestion warning signals; signs showing recommended speed in curves and new guardrails along embankments). In general, we conclude that the estimates of all factors are essential and require precise estimation if uncertainty in the number of injury accidents saved is to be reduced. However, the growth factor in traffic performance should be given priority. Future research in the development of the model should be devoted to improving certainty concerning all input factors. In addition, a more realistic case study for Flanders (using recent, detailed data of high quality) would provide interesting insights. (Author/publisher)
Abstract