Article
French
ID: <
10.4000/msh.2789>
·
DOI: <
10.4000/msh.2789>
Abstract
In this paper, we propose a method of specific Correspondence Analysis which allows to treat questionnaires when some responses are missing, and thus to free oneself from the yoke of complete disjunctive encoding. The method of specific analysis is presented within the general framework of Geometric Data Analysis for a Euclidean cloud, then particularized to multinumerical protocols and to questionnaires. We show that, in this approach, beweighted Principal Component Analysis (PCA) is privileged and that Multiples Correspondence Analysis (MCA) is equivalent to a biweighted PCA on indicator variables. Finally, we compare the specific analysis to the conventional one by writing inequalities between eigenvalues and studying the rotation of principal subspaces when one goes from the global analysis to the specific one.