11 Canonical Correlation Analysis

Every method so far has treated the \(p\) variables as one set. Suppose instead they divide naturally into two groups — physical measurements and performance measurements, say, or inputs and outputs — and the question is how strongly the two groups are related as groups.

Correlating every variable in one set with every variable in the other gives \(p_1p_2\) numbers and no summary. Canonical correlation analysis reduces this to a few, by asking for the linear combination of the first set most strongly correlated with a linear combination of the second.

Questions on this section

Stuck on something here? Ask below and it stays attached to this topic.