Like many countries, France has an accident register, which contains informations that are provided by law enforcement officers. These registers have a number of features in common: they have several levels, contain a large number of variables and minor accidents are under-reported. Data processing for this type of register is often performed by using contingency tables without applying any statistical tests. This technique does not allow us to rank the different variables according to their contribution or reveal the multivariate nature of accident causation. A processing methodology is proposed that takes account of the points mentioned e and that is based on a division of accidents into three classes,using odds ratios and logistical regression. Factors increasing accident severity are particularly important when they involvge number of cases. This applies to drink-driving, young drivers, older pedestrians, a rural location, lateral obstacles and HGVs. On the basis of additional risk exposure considerations, motorcycles can be added to this list. These factors correspond to those, frequently mentioned in the context of road traffic accident studies (A). For the covering abstract of the conference see E217780.
Samenvatting