ABSTRACT
In the previous chapters, we presented the 7 ^ -statistics and con trol charts for testing the hypotheses that a new multivariate observation (or group of observations) originates from the same distribution as that which generated the “target values.” When the Tj^-value signals that an observation is out of control, an important practical issue is to detect the variables which caused that signal. While it is common in industry to monitor individual process characteristics, the use of such procedures alone, is far from being satisfactory, since those procedures do not ac count for the correlations among the variables.
