ABSTRACT
It is often desirable to restrict attention to part of a data matrix, leaving out either some rows or some columns or both. For example, the columns might subdivide naturally into groups and it would be interesting to analyse each group separately. Or there might be categories corresponding to missing values and one would like to exclude these from the analysis. The most obvious approach would be simply to apply CA to the submatrix of interest — however, one or both of the margins of the submatrix would differ from those of the original data matrix, and so the profiles, masses and distances would change accordingly. The approach presented in this chapter, called subset correspondence analysis, fixes the original margins of the whole matrix, using these to determine the masses and χ2-distance in the analysis of any submatrix. Subset CA has many advantages; for example, the total inertia of the original data matrix is decomposed amongst the subsets, hence the information in a data matrix can be partitioned and investigated separately.
