ABSTRACT
Extending simple correspondence analysis (CA) of a two-way table to many variables is not so straightforward. The usual strategy is to apply CA to the indicator or Burt matrices, but we have seen that the geometry is not so clear any more — for example, the total inertia and principal inertias change depending on the matrix analysed, and percentages of inertia explained are low. The Burt matrix version of multiple correspondence analysis (MCA) shows that the problem lies in trying to visualize the whole matrix, whereas we are really interested only in the off-diagonal contingency tables which cross-tabulate distinct pairs of variables. Joint correspondence analysis (JCA) concentrates on these tables, ignoring those on the diagonal, resulting in improved measures of total inertia and much better data representation in the maps.
