Determination of number of linear latent variables in a set of experimental chemical data

Abstract

A criterion has been suggested for determination of number of linear latent variables in experimental chemical sets which is based on the principle of scatter analysis applied to the method of conjugated deviations. The criterion suggested has been tested on data with preliminarily given error, and correct numbers of latent variables were obtained until the error load 70%. The phenomenon of dominance of latent variables with the maximum variability has been analyzed and tested. The criterion suggested has been applied to three classes of data sets - homogeneous complete, homogeneous incomplete and nonhomogeneous incomplete. A good accordance with other criteria has been obtained first of all for the first data class, in most cases the found numbers of latent variables agreed with the experiment interpreted.