Ce rapport propose une méthodologie d'analyse des comportements des conducteurs routiers vis-a-vis du risque et de la vitesse en fonction du trafic et des caractéristiques de géométrie et de surface routières. Les auteurs y donnent une structure d'équations simultanées permettant de tenir compte des différents ajustements entre confort, vitesse et risque lies aux préférences des usagers. La formulation adoptée distingue parmi les composantes du risque observe ou objectif entre l'incertitude et la dangerosité et le risque calcule associe a la vitesse. On y utilise aussi pour la première fois une mesure empirique nouvelle du risque perçu qui permet d' isoler le rôle de ce facteur parmi les autres déterminants du choix de vitesse. L'estimation reposant sur un échantillon de plus d'un millier de sections de routes nationales. English abstract: This paper presents a method to analyse the user's attitude toward speed and road safety in relation to traffic, road design and surface characteristics. A simultaneous structure makes it possible to take into account the trade-offs among comfort, speed and uncertainty dependent upon user preferences. The authors' formulation notably distinguishes between two components of observed objective risk, namely calculated risk linked to speed, and uncertainty or "dangerousness", in the manner of Frank Knight. The authors also use for the first time a new empirical measure of "perceived risk" derived from random utility theory, with uncertainty at its core, facilitating the identification of the role of perceived risk among the determinants of speed choice. The estimation structure consists of three equation groups: the first two explain accident frequency and severity with discrete choice logit type models easily admitting of nonlinearity and choice-based sampling. The speed equations, explaining both the mean and the variance of speeds, also consist of non linear flexible-form models. The authors' choice-based sampling approach permits future completion of the data base at low cost, in order to correct potential biases due to missing variables that may not be orthogonal to those at the authors' disposal. (A)
Samenvatting