Prediction of responsibility for drivers and riders involved in injury road crashes. (September 2019)

• Record Type: Journal Article
• Authors: Garcia, Cédric; Viallon, Vivian; Bouaoun, Liacine; Martin, Jean-Louis
• Is Part Of: Journal of safety research. Volume 70(2019)
• Page Start: 159
• Page End: 167
• Publication Date: 2019-09
• Subjects: Road crash -- Responsibility analysis -- Scoring tool -- Statistical learning -- Crash risk factor; Industrial safety -- Periodicals; Accidents -- Prevention -- Periodicals; Safety -- Periodicals; Accidents, Occupational -- Periodicals; Sécurité du travail -- Périodiques; Accidents -- Prévention -- Périodiques; Accidents -- Prevention; Industrial safety; Periodicals; 363.1
• DOI: 10.1016/j.jsr.2019.07.001
• Languages: English
• ISSNs: 0022-4375

Abstract

Abstract: Introduction : Responsibility analysis allows the evaluation of crash risk factors from crash data only, but requires a reliable responsibility assessment. The aim of the present study is to predict expert responsibility attribution (considered as a gold-standard) from explanatory variables available in crash data routinely recorded by the police, according to a data-driven process with explicit rules. Method : Driver responsibility was assessed by experts using all information contained in police reports for a sample of about 5000 injury crashes that occurred in France in 2011. Three statistical methods were used to predict expert responsibility attribution: logistic regression with L1 penalty, random forests, and boosting. Potential predictors of expert attribution referred to inappropriate driver actions and to external conditions at the time of the crash. Logistic regression was chosen to construct a score to assess responsibility for drivers and riders in crashes involving one or more motor vehicles, or involving a cyclist or pedestrian. Results : Cross-validation showed that our tool can predict expert responsibility assessments on new data sets. In addition, responsibility analyses performed using either the expert responsibility or our predicted responsibility return similar odds ratios. Our scoring process can then be used to reliably assess responsibility based on national police report databases, provided that they include the information needed to construct the score. Highlights: Responsibility assessment quality is essential for valid responsibility analyses. We developed a data-driven scoring tool to predict responsibility. In our training set, responsibility of crash-involved drivers was rated by experts. Predictors mainly referred to inappropriate driver actions. Cross-validation showed that the scoring tool can be applied on new crash datasets.