| Road surface conditions, particularly ice and snow, are not a permanent feature of the roadway. This lack of permanence in hazard formation, location, and duration makes effective signing difficult. The Washington State Department of Transportation (WSDOT) has questioned the sensibility of expending material and personnel resources to maintain ice warning signs when little proof exists of their effectiveness in improving public safety or protecting WSDOT against unnecessary liability.; A preliminary investigation was conducted to more fully understand the issues surrounding ice warning sign use and effectiveness. This preliminary investigation consisted of (1){A0}a national review, (2){A0}a review of current practices in Washington, (3){A0}a review of public attitudes and responsiveness, and (4){A0}a review of safety and liability issues. Lacking from this preliminary investigation was an adequate examination of the effects of ice warning signs on public safety. This research serves to prove or disprove the relationship between ice-related accident frequency/severity and ice warning signs using advanced econometric modeling methods.; On the basis of strong public support for ice warning signs, WSDOT's high ranking as a party at fault in the event of an ice-related accident, and the terms required to escape or minimize liability damages (i.e., proof of a reasonable and systematic process for placing signs), it is recommended that the use of warning signs be continued. However, given the lack of effect noted in this investigation of ice warning signs on either accident severity or frequency, efforts to improve (1){A0}motorist reaction to ice warning signs and (2){A0}sign placement should be pursued.; From a theoretical perspective, as confirmed through this investigation, a more suitable means of count data analysis is either Poisson or negative binomial regression when data are overdispersed. In instances where a preponderance of zeros exists in the data, dual-state models, such as the zero-inflated Poisson or the zero-inflated negative binomial models, show promise. Sufficient overall statistical fit (as noted by the log likelihood value at convergence) in combination with reasonable dispersion and zero-inflation test-statistics and coefficient estimates confirm this applicability. |
