Dear Forum Members:
The prototype web site mentioned in my last e-mail has
been restructured.
The indirect contact form is now located at the bottom of
the page attached to the next-to-the-last button.
In addition, you may still contact me directly through
drking at gti.net.
Also, I would like to thank Dr. Su of Wyeth for pointing
out one of the key error sources that some overlook.
Another significant error source that some overlook is
that process capability approaches should only be applied
when a BATCH-REPRESENTATIVE SET of data values clearly
has a near normal (Gaussion) distribution .
If the distribution of data is NOT Gaussion, then one
should use an appropriate distribution-free approach
(which typically, requires testing several times as
many representative samples units [2 - 10 times] as is
the case when the raw data values can validly be treated
as being from a normally distributed population).
Hopefully, this additional information will be of help
to those attempting to use any process capability or other
approach based on the representative set of values being
members of a normal or near-normal (pseudo-Gaussion)
distribution.
Finally, I would caution all who are considering the use
of any statistical test to first understand the underlying
principles upon which that test is based.
After all, anyone can enter data into a "statistical program"
and initiate a test.
However, if you do not really understand exactly how the
test functions and the presumptions upon which it is made,
then the results from the test may totally mislead you.
As many have said, a little knowledge is a dangerous thing
-- knowledge w/o understanding is even more dangerous.
Respectfully,
Paul G. King
drking@gti.net
|