Critical evaluation of a significance test for partial least squares regression

A significance test for determining the optimum model dimension for partial least squares regression is critically examined. It is derived that this s ignificance test is equivalent with a well-known t-test for principal compo nent regression. The results for a near-infrared data set suggest that the conventional methods of optimising the prediction error estimate obtained f rom a test set or cross-validation are to be preferred. (C) 2001 Elsevier S cience B.V. All rights reserved.