RE: The effect of special causes

[Steven_S_Prevette@rl.gov]

My question is this: Does special cause variation influence
the amount of common cause variation? In other words, could
a highly variable (noisy) process that is in "statistical
control" mask special causes because of its wide limits? 

++++

My response is:  I believe you will find that special cause variation is
independent of common cause variation.  The very nature of the definition
(common cause is continually present, special cause is only present at
certain discrete times) should imply they are independent.  But no, I do not
have a statistical proof.  The assumption made is that the amount of special
cause variation does not influence the amount of common cause variation. 

It is true that you may have a system where a high level of common cause
variation masks a source of special cause variation.  In application of Dr.
Shewhart's work, I believe this is not a major problem.  The model still is
to eliminate visible special cause variation first, then work on common
cause variation.  It is certainly possible that while investigating sources
of common cause variation you may find a special cause.  Or, more likely,
once you improve the process and common cause variation reduces, a
previously "hidden" special cause variation will become visible and then you
can attack it.

But you must be wary of taking the data at too high of a level, of including
"mixed outputs".  It is necessary to drill down into individual
contributors.

Let me propose a thought example is automotive crash deaths.  Perhaps the
number of deaths due to Firestone tire blowouts causing rollovers of Ford
Explorers was not visible in a chart of all world-wide highway deaths.  But
running some Pareto charts by model of automobile may have been useful,
running control charts for some of the leading models of automobile may have
also been useful.  A control chart analyzing death rates per miles driven
(or number of vehicles produced) may have helped to show a difference.  

As a personal example, I run control charts of injury rates at Hanford.
Overall, our injury rates are statistically significantly decreasing.  I
just rebaselined the OSHA recordable case rate to a lower average.  But I do
have one project within Hanford that is having a statistically significant
increase in its OSHA recordable case rate.  This piece of information both
shows up in a Pareto chart of injury rates, and in their individual control
chart of injury rates.  So it can be worthwhile to keep track of the major
parties within a larger system.

See also page 354 - 356 in OOC, where Dr. Deming points out that "study of a
mixture may obscure chance to improve" . . . "Even though the combined work
of a group may be in pretty good statistical control, control charts for
individuals may discover that one or two or more people are in need of more
training or transfer.  (See example of p. 384)"


Steve Prevette
QA Engineer, ESH Radiological Compliance
Fluor Hanford, A Fluor Global Services Company
ASQ Certified Quality Engineer
steven_s_prevette@rl.gov
509-373-9371