Using a regional case study, the paper investigates some methodological problems concerning the measure of the relationships between two sets of variables. Canonical analysis seems a good tool, but discussion of its advantages and drawbacks shows it is not fully satisfactory. Other methods are tested and their efficiency discussed: Regression, joint-set component analysis, canonical-subspace interpretation by communalities, component-based redundancy. In conclusion, simultaneous use of canonical analysis and other methods is recommended since their results appear complementary.
Testing Bivariate Normality Based on Nonlinear Canonical Analysis
