Schemes for handling inconsistent matrices

Publisher Website
Google Scholar
Abstract
The increasing role of multivariate statistical techniques, a trend based on the growing availability of computers and on pressures for streamlined representation of multi‐dimensional arrays of data, carries with it the responsibility for using efficient techniques to manipulate matrices and arrays. It frequently happens that all records in a multivariate system (of flow records, quality measurements, or whatever) are not equally long, and that some subsets of these records do not present a consistent pattern of correlation when the entire bivariate correlation matrix is reconstructed. If one or more estimate is such that the entire correlation matrix is rendered inconsistent, repairs must be effected before continuing the multivariate analysis. Two algorithms for systematically finding a set of adjustments that satisfy the requirements of consistency are given. Of course, the results are not unique.

This publication has 0 references indexed in Scilit: