FFI - Fordonsstrategisk Forskning och Innovation

Real world data is important for safety development within the road transportation system. For car safety development in particular, methods to collect and analyse real world data on driver behaviour from normal driving, incidents and accidents are needed to address safety in driving. The main goal with this project is to explore and develop different analysis methods that can be applied to existing sources of real world data (e.g., accident mail surveys, naturalistic driving studies). The methods are evaluated based on their capabilities to: 1) Estimate how frequent different driver behaviours occur and/or 2) Contribute to the understanding of accident/incident mechanisms. The following studies were completed in this project during 1st of March 2009 – 31st Dec 2012: Study 1: Nonresponse analysis in an accident mail survey. Insurance data were retrieved as additional source of information to an accident mail survey for all cases where the mail survey was sent. These data were used to analyse and compensate for nonresponse by using logistic regression analysis and inverse propensity weights. Study 2: Accident case studies including the driver’s own descriptions in a mail survey and insurance documents In this study, mail survey variables were combined with accident descriptions provided by the involved road users in the mail survey questionnaire and insurance documents. These documents were then analysed for each case to establish if driver distraction or drowsiness were present at the time of the accident. This approach was also assessed on whether it contributed to an additional understanding of why the accident occurred compared to using mail survey variables alone. Study 3: Incident causation based on naturalistic driving data. This study was performed in collaboration with the DREAMi-project. Video-recordings of car-to-pedestrian incidents from a Naturalistic driving study performed in Japan were analysed. A method initially develop to classify accident causation was further developed for incident causation analysis and evaluated for the video-recorded events.