Circular dependence in a multiresponse mixed model under a split-plot design

Abstract

It is well known how to analyse correctly data arising from auniresponse fixed or mixed effects model under a split-plot design. Many results are available for both canonical dependence pattern (compound symmetry) and more general patterns. Conversely, for the multiresponse case, there are only few results concerning the repeated measurements context. Some references are here given for amultiresponse analysis under multivariate compound symmetry. The aim of this paper is to propose a simple methodology for MANOVA testing procedures if a more general dependence pattern,i.e., circular autoregressive multiresponse dependence pattern, is assumed. Conservative thresholds are suggested in order to perform a powerful test even if only vague information is available about parameters appearing in this dependence pattern.