Some simple rules for interpreting outputs of principal components and correspondence analysis

A large literature has been devoted to the assessment of the right number of eigenvaluesin PCA and CA (two-way and multiple). Most of the publications are based ondistributional assuptions for the sample, or on bootstrap techniques. After havingrecalled some of the most important results, we present simple thresholds based on a« control chart » approach for eigenvalues as well as for contributions, distances