RE: Semantics and 6 sigma

[Rip Stauffer <rstauffer@bluefirepartners.com>]

Perhaps I can shed a bit of light on the so-called "Six Sigma metric." I am
not here to praise it or bury it, just to try to explain a few things I have
learned about it. There are some problems with it, as most statisticians in
the Six Sigma business will tell you. The problems are not as much a matter
of statistical theory as they are definitions of contributing terms,
however. 

It helped me to stop thinking of it as any rigorous statistical measure.
It's a comparative metric, a convention. Although it is based on the idea of
sigma, the real underlying concept is that if you can predictably produce
stuff that's twice as good as what the customer is asking for, you are
probably doing pretty well.

So you do what you can to make things better. In some cases, you have SPC
charts and capability studies that estimate how well you are producing
measurable things, and you get some rough estimate from that as to how many
defects you might make per million opportunities (Before everyone gets their
statistical hackles up, however, let me point out that I have not met a
serious statistician yet--even in the Six Sigma world--who actually thinks
you can convert a Cpk into anything resembling a real fraction defective).
You also have to count some things...how many errors are there? How many
mistakes in assembly, etc.? (Of course, FMEA and Poka Yoke are incorporated
to cut these errors as much as possible, but these are mistakes, and have to
be measured instead of counting). This is all aggregated into a measure
called Defects Per Million Opportunities or DPMO. The DPMO for the process
or step under study is compared to a table, and a corresponding "process
sigma" value is selected. 

There are other, more mathematical ways to calculate it. Sometimes the
inverse of yield is used as a Z-score and sigma is calculated from that. It
doesn't make any difference, however. You are not calculating a number that
anyone thinks is real. It's just a gage of process performance, used to
track whether or not the process is improving, and for comparing processes
to one another.

I guess I should point out that, as in any capability study, it only applies
to one point in time. You have to have a process which is stable over time
in order for it to have any meaning, but it is based on comparing a snapshot
from the stable system with specs.

The 1.5 sigma shift built into the table is an unfortunate thing. It
originated from some studies at Motorola that suggested that a process
average might shift as much as 1.5 sigma units without detection on a
control chart. You can dispute that 'till you're blue in the face (I do). In
fact, a friend of mine who has seen that study, said that had the fine folks
at Motorola used the Western Electric zone tests, the shift would have been
readily discovered, and we wouldn't have this part of the debate. However,
they didn't know that, so they put that shift in the table, and it has
become the convention. As long as you and I both use the same flawed table,
we have a basis for comparison. 

As in Gresham's law, it's apparently harder to change the tables than to
continue using them. It's sort of an American tradition, like using inches
and yards when centimeters and meters are actually easier and more
universally recognized.

As to whether Six Sigma represents an improvement over Shewhart...to me it's
not even a relevant question. I have met a lot of statisticians, both in and
out of the Six Sigma world, who have never even read Shewhart. Most of the
ones I work with have, and see no conflict. Six Sigma is like TQM in a lot
of respects. There are many definitions of it, and many applications. Some
are useful, some are successful, some are disastrous. The best applications
I have seen came from people who had strong Deming backgrounds. Some of the
most successful Six Sigma organizations have learned many of the 14 points
over several years of continual improvement. Some haven't learned much. All
seems to depend on whether they have constancy of purpose, whether they
deploy the new philosophy, break down barriers, eliminate fear, etc..

I really don't want an extended argument on the merits/flaws in Six Sigma.
We can continue to pick Six Sigma apart, or we can try to get in and improve
it. I used to do the former, but have recently decided to try the latter. I
recently held some "Black Belt" courses in which I introduced the concept of
variation via the Red Bead and the Funnel. I doubt that the "Six Sigma
Academy" would approve, but it helped my students a great deal.

Best regards to all, Rip

Rip Stauffer, Senior Consultant
BlueFire Partners
1300 Fifth St. Towers, 150 So. Fifth St.
Minneapolis, MN 55402
612-344-1027
rstauffer@bluefirepartners.com