Re: SPC question response

["Statistical Process Controls, Inc." <fwheeler@spcpress.com>]

If you would like you may make the following part of my response.

Ever since E.S. Pearson misunderstood Shewhart in 1935 there have been
two divergent streams of thought about SPC.
In fact, Shewhart's 1939 book was actually a rebuttal to Pearson's 1935
book which became British Standard 600 (better known as BS 600).  The
difference between these two streams of thought was succinctly
characterized by Shewhart on pages 276-277 of his 1931 book when he
observed that if we knew the probability model, f(x), then we could find
two limits, A and B, such that the integral of the probability model
between A and B would be equal to some predetermined value P, where P is
close to 1.00.  (This is the traditional approach used in mainstream
statistics: fix the value for P and find the corresponding values for A
and B.)

However, Shewhart went on to observe that in practice we will never know
enough to fully specify f(x).  Therefore we cannot actually fix a value
for P and then find A and B.  If we insist upon a fixed value for P,
then we will have to CHOOSE a probability model, f(x), and then find the
corresponding values for A and B.  (This is the approach referred to in
the last paragraph of Ryan's comments.)  Of course, when we do this, we
will naturally be anxious about our assumption, hence all the noise
about normality and the need to check to see if the data are normally
distributed.  If our assumed probability model is wrong, then our
precious, predetermined value of P might not be exactly right!

Instead Shewhart chose to fix A and B—to place these points far enough
out (three sigma limits) so that no matter what probability model, f(x),
we might choose to use, the integral from A to B will always result in a
value for P that is reasonably close to 1.00.  Since the issue here is
filtration, rather than estimation or inference, exactness is not
needed.  Approximate solutions are sufficient when we are filtering our
the noise.

Therefore, Shewhart turned the statistical approach upside down.  Rather
than fixing the value for P and then letting A and B vary, he chose to
fix A and B and to let P vary.  And this is the difference between the
"statistical approach" and the "applied approach of Shewhart."  Many who
claim to have read Shewhart have failed to understand this difference.

--
Statistical Process Controls, Inc. & SPC Press
5908 Toole Drive, Suite C
Knoxville, Tennessee   37919   USA
Phone:  865-584-5005    Fax:  865-588-9440