RE: Methods to Determine Special vs. Common Cause (fwd)

[Rip Stauffer <rstauffer@bluefirepartners.com>]

There has been a lot of good input on this thread. I think it's important to
step back from the question (and our assumptions) and examine it (and them).
I think this is a question that is at the core of the System of Profound
Knowledge, and our level of understanding of the question determines our
ability to put the SoPK to practical use.

Although a control chart provides us with a pretty good operational
definition, it is still only a conditional operational definition. Shewhart
didn't think that control charts could *determine* the difference between
special cause and common cause, he didn't believe that mathematical or
probability sturctures could do that in the real world. He recognized that
probability theory provided a convenient structure to build an indicator.
The control chart does not *isolate* special causes--in any deterministic
sense--it provides an economic set of rules for determining which signals
are probably worth investigating. Shewhart, Deming, Wheeler, and every other
serious practicioner I have met, have all pointed out the importance of two
things in studying processes: knowledge of statistical theory, and knowledge
of the process under study. My understanding of their writing is that both
are necessary; neither, by itself, is sufficient.

So if we look at a point outside the control limits as an indication of
special cause, what does that mean? One of the first definitions I learned
included "...not inherent in the system, comes from outside the system,
arises because of specific circumstances..." and some other stuff. That was
pretty good, but left me confused when I first saw Wheeler's "A Japanese
Control Chart." One of the special causes plaguing Tokai Rika is tool wear.
Isn't tool wear inherent in the system? Does it come from outside the
system? You could say that it "arises because of special circumstances," I
guess. I eventually had the opportunity to discuss it with Don, and I
learned that special or assignable cause indicates a shift or change in the
underlying system. This was a very profound moment, and signaled departure
from a learning plateau, for me anyway.

In the early '90s, in Costa Mesa, California, one of my Navy Colleagues
(Laurie) took part as a willing worker in a Red Bead Experiment. She drew a
20. (As I understand it, this was only the second Deming had seen to that
point...I think someone drew one after her, a couple of years later). She,
of course, was mortified, terrified, pale and shaking, as Deming said, "Let
me see that...let me see...WHAT HAPPENED?" This fear, despite her having
conducted the Red Bead in classes of her own, and having a real appreciation
for the dynamics of the exercise. 

In the discussion after the exercise, I gained a new insight into the
distinction. Deming had drawn the control chart, with Laurie's 20 glaringly
outside the limits. He asked the audience, "So is this a special cause? Was
Laurie's performance due to anything special?...Pretty good question, I
think...what do you think?" No answer came. He finally asked again, "Did
Laurie do anything special?" There were some heads shaking in the audience,
now, as understanding began to sink in. Finally he said, "No. Laurie did
nothing special. Her performance was determined entirely by the system. This
is a false signal." It doesn't matter that the probability structure worked,
and that you could "prove," mathematically, that there was a special cause.
Reality will out. I think that Deming was the only one in Costa Mesa (at
least at the beginning of that day) who had both the understanding of
statistics and the understanding of the process to make that determination. 

I believe that a special cause is any cause which creates a shift in the
performance of a system. There are some conditions in this definition, of
course. Operationally, you would be required to have established some
evidence of a reasonable degree of statistical control in the system under
study. With that evidence, you could say that you had only common or chance
causes present; that is, the interaction of forces within the system
produced some predictable amount of variation. If some change happens,
either through outside influence or through resonation within the common
cause system, the system performance will change. That definition works for
me, anyway. Operationally, I start looking for special causes when it's
indicated by signals on a control chart, but I don't let the chart make the
decision for me. Process knowledge will have to weigh in. 

The advantage for management is that you don't have to look at everything.
You have some rational basis for your decisions. You have a rational basis
for prediction. You can determine the practical limits of your knowledge.

As to whether there are other tools for determining special from common
cause, I don't know.  Myron once told me that the only two things that
really matter are "What is the question?" and "What method will provide the
best answer?" (If I got that wrong, Myron, please correct it for me). I
think our questions here are, "Do I have a predictable system?" and "When
has my system changed?" Because the systems or processes that we want to
study tend to be dynamic, the best method for answering the question tends
to be an analytic study. A control chart is really good; some things can be
suggested pretty strongly by patterns in run charts. t-tests, ANOVA, and
other hypothesis tests appropriate for determining differences between
populations in enumerative studies will not provide any answers to the first
question. They may provide some insight into the second question but will
probably never provide as much insight as control charts will. 

Hopefully, I haven't completely confused the issue. I have always felt that
it was a more complex question than it appears on the surface. 

Best regards to all,

Rip Stauffer, Senior Consultant
BlueFire Partners
1300 Fifth St. Towers, 150 So. Fifth St.
Minneapolis, MN 55402
612-344-1027
rstauffer@bluefirepartners.com
========================================================================