Abstract

Under the assumption of multivariate normality, we study linear classification rules for multi-level multivariate data where more than one response variable is measured on each experimental unit on more than one site at several time points. It is very common in clinical trial study to collect measurements on more than one variable at different body positions (sites) repeatedly over time. The classification rules, with and without time effect, and with certain structured and unstructured mean vectors and covariance structures, are very efficient in small sample scenario, when the number of observations is not adequate to estimate the unknown variance-covariance matrix. We introduce a parametrically parsimonious model for the classification rules by using a doubly exchangeable covariance structure. The classification rules are applied to a real dataset as well as to simulated datasets. We compare our findings with other linear classification rules for three-level multivariate data as well as with the traditional linear classification rule.