Multivariate least squares and its relation to other multivariate techniques

From MaRDI portal





The purpose of this paper is to show how a general multivariate least squares method (LS) can be used for estimation of the connection between two, possibly very large, data sets. It turns out that the LS approach provides a convenient unified framework for general description and comparison of various multivariate methods used for the same purpose, facilitates their understanding and helps to identify their usefulness for various real-world applications. Besides this, the LS technique does not require a specialized software so it can be implemented even in Excel. The authors consider canonical correlation analysis (CCA) and principal components (PCA) analysis as the LS approximations and compare them with partial least-squares (PLS) methods. They demonstrate that the robust CCA approach corresponds to the variant of PLS family methods and yields the most convenient results for practical applications in the analysis of large and complex sets: its loadings in contrast to CCA are not prone to multicollinearity and in contrast to PCA are not too intensive or evenly distributed. A numerical example of the management problem is also used to compare various methods.











This page was built for publication: Multivariate least squares and its relation to other multivariate techniques

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4455492)