Interactive demo · Statistical Methods for Modern Biology · Chapter 9

Correspondence analysis: the SVD of the chi-square residuals

For a table of counts, compare every cell with what independence predicts, eij = ni·n·j/n. The standardised residuals sij = (nij − eij)/√(n eij) have squared sum X²/n. Their singular value decomposition splits this total inertia into axes (λk = γk²) and gives coordinates for the rows and the columns on the same map.

Table

Biplot scaling

Axes

Show a row's profile

Scoreboard

Try these

Symmetric map: rows and columns both in principal coordinates (as fviz_ca_biplot). Row-principal: rows principal, columns standard, so every row point is the profile-weighted mean (barycentre) of the column points; the ring marks that barycentre. Column-principal: the reverse.

In the lecture notes