2.3. Analysis

A score that brings together all the violations that are frequently or fairly frequently committed was calculated, with two classes (0-6 and greater than 6); the score of 6 corresponding to the last quartile in the entire sample.

The upper bound for the number of brothers and sisters was fixed at 3, which corresponds to the last quartile in the control group.

We did not use the Eysenck’s classification of EPI because we had no clinical objective. We present the scores distribution in both samples and compare them for the three dimensions measured by the EPI test (Neuroticism, Extroversion, Lying score). We were particularly interested in the extremes of distribution, very high scores which correspond to the two last of the 10 classes of Eysenck.

An initial descriptive analysis was conducted to characterize the two samples with regard to sociodemographic, transport, lifestyle and personality variables. The comparisons were made by tests that compared averages (Student’s t tests) and the Khi2 test for qualitative variables.

Logistical regression analysis was used to determine which variables were linked to course-taking for each gender, all other things being equal. For this, all the variables that were significantly linked (with a threshold of 10%) to course-taking during the univariate analysis were input to the multivariate model . We also kept the violation score in the last model as a variable of interest although it was not significant (forced variable), in accordance with authors who explained that it is the principal variable associated in a crash (Malaterre 2000).The variables which became non-significant at the 5% level were removed one by one, applying the stepwise method starting with the least significant value (the highest p-value). The odds-ratios (OR) for each factor were thus calculated, with their 95% confidence intervals (CI).

Processing was conducted using the SAS 9.1 software.