Re: Rule of thumb for SPC

[Rip Stauffer <ripstaur@vabch.com>]

Steve Prevette mentioned: "In the case of recursive
recalculations throwing more points outside the control limits, this
becomes a case where knowledge of the process is an absolute necessity"

Great point, Steve! This was the point Shewhart drove at in his 1931
book, and the whole point behind analytic studies. Knowledge of the
process is the MOST important ingredient in control charting. The math
is nothing--the math only provides the means to an answer--the GIGO (*)
principle still rules. You can get a computer to do the math for you,
and crank out the charts. But unless you understand the meaning of the
data and where they came from, you have no business plugging them into
the formulas. 

To the question of changing limits after removing points: Since all the
limits are dependent upon the calculated average range or standard
deviation, any extreme point used in those calculations may  inflate the
limits (especially true of points outside the limits in the dispersion
chart). This should only be true when calculating limits, however--once
limits are set, special causes should be investigated and their causes
eliminated if possible. Any change to the process brought about by the
elimination of part of the cause system might be a reason to recalculate
limits.   

One problem I have noted is that people plug data into a computer, then
allow the program to recalculate limits with each new point. This should
not be done. Once evidence of a reasonable degree of statistical control
is demonstrated, the limits should be set, and further points plotted
against those limits. Comparison of those further points against the
limits is what tells you when the state of control has shifted or been
compromised. If you allow the computer to update the limits with each
new point, gradual increases in dispersion (which otherwise might be
quickly detected in the range chart) could be masked.

Rip Stauffer
Naval Leader Training Unit
ripstaur@vabch.com

(*)  GIGO = garbage in, garbage out.
====================================================================