We propose a test of dimension in multivariate regression. This test is in the spirit of tests on the rank of the coefficient matrix in a multivariate linear model, but it does not require a prespecified model. The test may be particularly useful at the outset of an analysis before a multivariate model is posited, because it can lead to low-dimensional summary plots that are inferred to contain all of the sample information on the multivariate mean function.
Copyright 2008 Elsevier B.V., All rights reserved.
- Central subspaces
- Dimension reduction
- Multivariate regression
- Regression graphics