A more recent concern is being out-of-trend (OOT). An OOT situation exists when one or more reportable results, collected
over a period of time, are compared with a historical data set or with a statistical model and determined to be different
from a practical significance standpoint.
The cause behind OOT occurrences can take several forms. The process could experience a sudden change in the average or have
an outlying data point. The standard deviation of the data could increase or decrease. The general direction of the data could
be gradually upward or downward, which is often called process drift. Finally, the process could cycle up and down. Cycling
can occur if the operators over control for the critical process parameters (also known as independent variables or factors)
and actually induce additional variability into the critical quality attributes (also known as dependent variables or responses).
Note that OOT is usually perceived as a negative event but it could be a positive one. If the variability decreases or if
the average moves towards the target, that would be a beneficial trend. If stability data does not change over the course
of the study, that would be a neutral trend. These changes need not reach the level of statistical significance to raise an
alert and initiate a response.
This concept is supported by the recent Supreme Court case for Matrixx (4, see page 118 for the full report). The court held
that a change or difference does not have to be statistically significant to warrant a preliminary examination of the data
or even an investigation. Thus, OOT is seen as a practical significance issue and not a statistical one.
The comparison of practical significance and statistical significance was discussed in the March 2010 Statistical Solutions
column on the role of statistical significance tests (5). The goal of a statistical significance test is to prevent us from
deciding something is important when it is just due to chance. In fact, if a difference or change is not of any practical
importance or significance, being statistically significant is frankly not of much interest.
Western Electric 8
The application of statistical concepts to manufacturing in the United States is often traced to 1924 when Dr. Walter Shewhart
wrote his famous memo describing for the first time the statistical control chart (or Shewhart chart) while he was working
at Western Electric in Cicero, Illinois.
At that time, the term "control" was innocuous enough and didn't raise questions when he and others spoke of control limits
and of being OOC. In the context of statistical control charts, if the pattern of values exceeds one or more the Western Electric
8 (WE8) rules, the data set is said to be "out-of-control" (6). (Note that most companies only use one or two, and no more
than three, of the eight rules to avoid over-control.) But we need an additional modifier here as well. It is clearer to say
the data set or process is "out-of-statistical-control" (OOSC).
This has been the case since 1924, and there is no reason to change such an old habit now. If someone speaks of being out-of-control
in the context of a statistical control chart we assume it is out-of-statistical-control. It should be pointed out that it
is possible to be out-of-statistical-control and still always meet the specifications.
Not all processes, for example biotechnology, will easily exist in a consistent state of statistical control. Such a situation
is often called being in a "state of engineering control." In the chemical and refining industries, engineering control is
more narrowly defined as the application of proportional, integral, or derivative tuners for process control.