Correspondence analysis is a method for representing the association between the rows and columns of a contingency table in a
low-dimensional Euclidean space. Let 
be a nonzero nonnegative data matrix, let 
be the sum of all its entries, and put 
. After any rows or columns with zero marginal total are
removed, let 
and 
be the row and column marginal proportions and let 
and 
be the corresponding diagonal
matrices. Correspondence analysis forms the standardized residual matrix
 |
A singular value decomposition
 |
then gives coordinates for plotting the row and column profiles. The squared Frobenius norm of 
is the Pearson chi-squared statistic divided by 
, so the method decomposes the table's departure from row-column
independence.
