Canonical Correlation is an advanced statistical technique
that can estimate the relationship between a set of dependent variables with a set of
predictor or independent variables. Multiple regression techniques, on the other hand, look
at the relationships between a single dependent variable and some predictor variables.
Canonical correlation places fewer restrictions on the modeling structure.
Consider an example in the credit industry. Suppose you wanted to predict both the credit usage
and the number of credit cards of individuals within a certain geographic market. You could
build separate regression models to accomplish this, or you could estimate a model
simulataneously to predict both number of credit cards as well as overall dollar usage. The
canonical correlation is a measure of strength of the overall relationships between the linear
composites of the predictor variables and the dependent variables. In general, it represents the
bivariate correlation between the two linear composites.