ABSTRACT
A basic assumption in most multivariate control procedures is that the observation vectors are uncorrelated over time. When this assumption is true, the graph of any process variable against time should show only ran dom fluctuations. When the assumption is false, the patterns in such time plots are systematic and often indicate the existence of linear or quadratic trends. In these latter situations, incorrect signals can occur in the corre sponding multivariate control chart, and the effectiveness of the overall control procedure may be weakened [e.g., see Alt et al. (1977) or Montgomery and Mastrangelo (1991)].
