[Value and limits of a statistical aid in typological constructions]

Correspondence analysis has become the favoured method to describe qualitative data. Graphical representation allows to appreciate positions of different modalities between them and in relation with the two factorial design axis. Nevertheless, lecture and interpretation seem to depend on the researcher's assessment who doesn't know always how modalities are jointed. Tri-Deux statistical software directly makes this level readable. Developed by a sociologist engineer from the CNRS, it is free of charge and operates with PC or compatible material. Every time a modality is in attraction with an other (on the basis of independency deviations), one line links the two points: this tool permits to visualize how modalities of analysed variables gather or contrast together. All the connections are not equal, some of them are more important than others. This program allows to draw each connection with a decreasing order according to their weight and to distinguish the highest attractions from the lowest ones. An example with a study allows to understand interest of this method. The aim of this statistical way is researching a possible typology in the analysed corpus. It is necessary to outline, in this method, the significance of an empirical phase when deliverate research of disconnected graphical complexes (clinical entities) leads to rule out non discriminant variables. Analysis consists to study data according to their clinical relevance and to their dynamic cohesion. This time corresponds to a real 'come back to the clinical practice'.(ABSTRACT TRUNCATED AT 250 WORDS)