The traditional control charts for nonconformances (np and p) and defects (c and u) date back to the 1920s, and they rely on the normal approximation to the binomial and Poisson distributions, respectively. This approximation works best when the expected number of events is 4 to 6, or even greater. The control charts therefore perform better as quality becomes worse, which suggests there is something wrong with the whole picture.

P. J. Cooper and N. Demos introduced the idea of tracking multiple attributes (e.g., different defect types) on a single chart, in their article, “Losses Cut 76 Percent with Control Chart System” (Quality, 1991). The inspector enters the number of nonconformances or defects in a table, compares the entry to a tabulated control limit, and puts a red mark around the number if it exceeds the upper control limit. The tabulated control limit usually depended on the average sample size and historical nonconformance or defect rate, but it could conceivably be modified to reflect the actual sample size. The method still relied, however, on the normal approximation to the attribute distributions in question.

 …