Modeling Fault in Fatal Pedestrian Crashes by Using Various Data Sources

Abstract

Binary logistic regression was used to model fault in 318 fatal pedestrian cases that occurred in Florida in the year 2000. The results were used to classify fault and identify factors that influenced fault. An expert fault assessment served as a control for predicting fault in each crash. The expert assessment team conducted a case review of each traffic crash by using additional data sources, such as traffic homicide reports, diagrams, photographs, accident reconstructions, and site visit notes. The logistic models correctly classified fault in anywhere from 84% to 97% of the cases. The existing Florida Department of Transportation algorithm correctly classified fault in only 56% to 58% of the same cases. Improvements in classification accuracy were shown to stem from two sources: the abundance of the data and the improved accuracy of the data. The mental state of the pedestrian and the driver were shown to be important in determining fault. Exhibiting a mental aberration, such as inattention, distraction, perception or decision error, or intoxication, increased the propensity for fault. Issues such as the number of lanes attempted in a crossing, the age of an individual, being a former vehicle occupant, having limited conspicuity, receiving a citation, and wet roads were also shown to be factors significant in determining fault.