Najprediktivniji i najpouzdaniji kriterijum

Abstract

A probably new method is proposed and a program is written for the analysis of relationships between two sets of quantitative variables. The proposed method is defined so that the reliability of a function of criterion variables is maximized, and at the same time is maximized the covariance between this function and a normalized function of predictor variables. It is shown that the function of criterion variables is simply the first principal component of criterion variables set, and that the function of predictor variables can be obtained by the normalized projection of predictor variables to the eigenvector of correlation matrix of criterion variables related to the maximal eigenvalue of the correlation matrix. An asymmetric measure of the relations between predictor and criterion variables is derived and a test of significance of this measure is proposed.