Angular analysis (ANA) of correlation matrices is proposed in which variables are represented as vectors in a low-dimensional space and inter-variable correlation coefficients are approximated only by the cosines of the angles between variable vectors. ANA can be viewed as a method which approximates correlation coefficients with least parameters and as a constrained version of principal component analysis (PCA). Solutions of ANA are obtained with an alternate least squares procedure in which an algorithm for oblique Procrustes rotation is used. The performance of the procedure is assessed in a simulation study. ANA is compared theoretically and empirically with PCA.
