L'analyse log-linéaire de tableaux de contingence et le modèle logit I : principe, usage, intérêt et limites

This paper introduces the theoretical basis of the loglinear methods, as a substitute to analysis of variance when variables are categorical, and logit, as a substitute to regression analysis in the same context. Their specific advantage is to allow the modelling of categorical, thus qualitative data, often collected from surveys, while keeping their probabilistic nature. The stress is put on the concepts of interaction and odds ratio which allow to unveil complex phenomena with nonlinear relations