Goodness-of-fit measures for multiple correspondence analysis

This paper presents three results: (1) It is shown that the explained variance criterion is inappropriate for multiple correspondence analysis (MCA). (2) There are two strategies to interpret a MCA configuration: factor analytical interpretation and cluster analytical interpretation. Goodness-of-fit measures for both interpretations are constructed. (3) These measures give (a) a more adequate picture of the model fit and (b) allow to differentiate whether a factor analytical interpretation or a cluster analytical interpretation is more adequate.