Abstract
The Hotelling's T 2control chart is typically recommended for the preliminary analysis of multivariate observations in quality monitoring applications. This chart detects many types of out-of-control patterns, but it is not sensitive to small shifts in the mean vector. We recommend that when used in the preliminary analysis of multivariate data, multivariate CUSUM charts should have a zero shrinking parameter for greatest sensitivity to small shifts in the mean vector. When the multivariate EWMA chart is used for preliminary analysis, we recommend that it be applied to the data in reverse time order as well as in time order, to avoid asymmetrical performance in the detection of shifts. For all three charts, simulation is necessary to determine accurate control limits, because of the use of estimated parameters.