ABSTRACT
Road accident risk assessment is a complex topic due to the large number of factors determining it and to the difficulties to collect data. In addition, most exposure factors influencing crash probability, such as environment and driver characteristics, are dependent on each other, so that it is not intuitive to devise a cause–effect scenario. The use of telematics devices, recently spreading among insurance and rental companies, provides new chances to collect exposure data, to define interpretive models of accident risk, and to explain variables relationships. Using global positioning system (GPS) data available through a long term rental company, the authors studied a sample of 900 vehicles. The authors aggregated raw data (e.g., road type covered, time, speed) in exposure metrics and organized them in a relational database. The authors built a number of multivariate logistic regression models, adopting a strategy to progressively refine them. The authors obtained a relatively high model fits (up to pseudo R2 0.301, Hosmer–Lemeshow p value 0.206) acquiring insights about the nonlinear association between explanatory variables and their outcomes. Interactions between variables were also examined. The results are, in general, in accordance with similar studies; regarding certain observed discrepancies, a discussion is provided to explain their origin, starting from the differences in associating predictors, outcome and interaction variables.
Acknowledgments
The authors are grateful to Arval Service Lease for providing data from a campaign of telematics devices installation (GPS-DR) on a subgroup of equipped vehicles. A special mention goes to Head of Business Transition & Systems of Arval Service Lease Dr. Sara Rossi.