Computational aspects for perturbation problems in multivariate descriptive statistical methods

In the context of multivariate statistical techniques such as principal component analysis or correspondence analysis, a general computational approach is suggested in order to assess the sensitivity of the results to some perturbations of the basic parameters. In particular, approximations of the perturbed eigenvalues and eigenvectors are provided with a simple procedure permitting the evaluation of their accuracy without recalculating the analysis. A practical illustration is given by an example.