Canonical correlation analysis is a statistical method proposed by Hotelling in 1936, best known for its theoretical qualities, since it encompasses many other statistical methods. It allows the linear relationships between two sets of variables measured on the same individuals to be described. Several generalizations of this method, grouped under the name Generalized Canonical Analysis, have been proposed in the literature to study jointly the linear relationships existing between several groups of variables measured on the same individuals. In this article, after describing linear canonical analysis and its generalizations, we compare three methods of generalized canonical analysis: the Sumcor method, the Ssqcor method and the Carroll method. To illustrate the application of these methods we start with a psychosocial questionnaire developed in the theory of social representations and which concerns the study of disobedience of traffic rules among young drivers. The results observed in this study show that the method of canonical correlation analysis and its generalizations are appropriate statistical tools for describing and analyzing the data collected in the field of social sciences. Furthermore in the study of social representations this approach allows us to illustrate through the “patterns” of answers, the internal structuring of the legitimate transgressions