Multivariate Techniques, Robustness

In this article, three broad areas of robustness in the multivariate case are considered: inference, ordination and classification, and latent variable models. Also, the authors give an overview of the robustness of the main techniques encountered under each of them. In summary, across all the different areas the broad concerns are the effects that are departures from normality and/or equal dispersion matrices. Departures from normality are relatively unimportant, but lack of equality of dispersion matrices can have much more serious consequences. Keywords: robustness; multivariate analysis; inference; non-normality; skewness; kurtosis; dispersion; Hotelling's T2; likelihood ratio test; MANOVA; principal components analysis; discrimination; latent variable models; structural equation models; LISEREL; maximum likelihood; bias; consistency; efficiency; goodness of fit; level of a test; power