RE: Statistical Control and Purpose

["Mowery, R. Neal (RNM) " <RNM@y12.doe.gov>]

When limited amounts of data are used to establish some indication of
statistical comtrol, there may be no evidence of special causes, even when
the processes are not managed.  I know that Steve often has to use limited
amounts of data on these new processes, and this may account for his
observation.  My experience in a business environment similar (but not
identical) to Steve's is that if you haven't seen evidence of a special
cause in an unmanaged process, you just haven't collected enough data.  

Other factors, however, may contribute to this phenomenon.  A process may
have elements so bad that their variation overwhelms the signals of special
cause due to other elements.  For example, if a measuring system is very
poor relative to the parameter under study, (i.e., a large percentage of the
total variation is caused by the measuring system), then random variation in
the measuring measuring would mask special causes inside the producing
process.  Thus an unmanaged process might appear to be stable simply because
the lens we view the process through is so bad.  Service systems are
notorious for having poor measuring systems.

Similarly, if we cannot plot data in the order in which the quality
characteristic was produced, we are likely not to see special causes in the
process, as some powerful tests are lost along with the order information.
But is that a control chart?

I'll acknowledge that charts of defects (p-charts) of unmanaged processes do
tend to converge to a stable, near-100% defect rate.  But that is a trivial
case. 

I interpret Deming use of "common cause" and "special cause" as evidence
that he believed that assignable sources of variation exist in all
processes, but the control chart tells you when it is economically feasable
to search for them.  I disagree with Steve's use of the phrase, "Achieved a
state of statistical control."  I don't see how an unmanaged process can
ever claim to be steady state simply because the data we have collected *so
far* have failed to show any signals of special cause.  Doesn't the word
"unmanaged" imply that we cannot assume stability?  Instead, perhaps we
would simply conclude that control charts of unmanaged processes may not
always show signals of special cause immediately.

Just a thought.

Neal



*********************************************
Neal Mowery 
Statistician - Lockheed Martin Energy Systems
Oak Ridge, Tennessee
rnm @ y12.doe.gov		Voice (865) 574-0796
*********************************************
======================================================================