![]() | Only 14 pages are availabe for public view |
Abstract Procedures for determining the optimal number of principal components retained for further analysis is the main topic in this thesis. Most of the well known procedures given in the literature, in addition to a new proposed procedure called coefficient of multiple determination method, are described. We investigate the performance of our proposed method and we compare it with three other well known methods namely: eigenvalues greater than one method; the minimum average partial correlation method; and the parallel analysis method through a simulation study. Our method showed good performance as, and sometimes better than, the other comparative methods. Interpretation of the retained principal components is also considered. Some methods of rotation and simplification of the principal components presented in the literature, in addition to a new proposed method, are described and illustrated by different examples. Keywords: Principal component analysis, Reduction of dimensionality, Interpretability of components, Rotation of axes, Simplification of principal components. |