SIX SIGMA

[Terry Peterson <gtp@excels.demon.co.uk>]

One of my clients supplies the automotive industry.  I have been helping
them implement QS_9000 with Deming / Shewhart techniques, so far with
good results.
However, they are now starting to become interested in 'six-sigma',
because there are indications that Ford may require this of their
suppliers, presumably as yet another stick to beat them with to get
costs down.
I have tried to get info on six sigma, but meet with a barrier to
understanding; even buying a book from ASQ didn't help.  It seems to
have an attractive message for CEO's - save millions on the bottom line;
and the best way to cement it in the organisation is to have it
implemented and policed by a priestly caste - black belts.  But, to get
the keys to further understanding, you need to attend an 'academy' at
high cost.
Does anyone know if six sigma is any more than process management and
continual improvement in a different guise?
I am tempted to believe that it is based on a spurious piece of
statistical mythology.  For example, can anyone explain the following
extract from Mikel Harry's book, in Shewhart/Deming terms; which
attempts to explain that 6 sigma can in fact be regarded as 4.5 sigma.
 
 - quote -
Long- Term vs. Short-term Capability 
One strength of the Six Sigma approach is that it recognizes and
accounts for the fact that processes vary over time.  Inevitably, when
data about a process are gathered over a period of time, observers will
see that the process does not always perform on target, or within
specified limits or boundaries. 
When discussing a critical-to-quality characteristic (CTQ) such as the
length of time it takes to respond to a customer's phone inquiry, the
maximum acceptable time to answer an inquiry becomes the upper
specification limit (USL).  For some CTQs, there might also be a minimum
response time, or lower specification limit (LSL).  In the case of the
telephone inquiry, an LSL may be specified to ensure that adequate
research time is given to "getting the facts straight." In addition,
there is usually a target time (T), a limit usually centered between the
USL and LSL.  The USL and LSL create a bandwidth of acceptable
performance. 
Owing to the natural sources of variation, we would expect a
distribution within the design bandwidth of the length of time it takes
to handle various calls.  By comparing the actual process distribution
spread (or process bandwidth) with the limits allowed by the USL and LSL
(the design bandwidth), companies can quantitatively see how capable
their processes are.  As the process bandwidth narrows in relation to
the design bandwidth, process capability increases.  The reverse also
holds true. 
Over time, the process bandwidth will increase in size due to process
centering errors.  In other words, over an extended period of time, the
process center (average) may be on target.  However, during any given
interval of time, the process average will be off target for one reason
or another.  The aggregate effect of shifts in the process center will
widen the process bandwidth.  When this happens, the probability of
seeing a defect jumps radically.  In a nutshell, process centering
errors over time degrade capability, which, in turn, increases the
likelihood of defects. Consequently, yield drops and costs go up. 
The average time-to-time centering error for a "typical" process will
average about 1.5 sigma.  In other words, a four sigma process will
normally shift and drift from its design target value (T) by about 1.5
sigma over an extended period of time.  For a typical process, this
means that factoring in the shift and drift factor of 1.5 sigma, divided
by the existing sigma level-say, 4.5 sigma-will equal .375.  In other
words, 38 percent of the inherent short-term capability is "lost" due to
normal process centering errors that occur over time.  This means that a
process that measures at four sigma short-term capability actually runs
at about 2.5 sigma over repeated cycles of the process.  Years of
theoretical and empirical research on this subject have proven this to
be true. This amount of shift and drift is inevitable, and has to be
accounted for during the design cycle of the process, product, or
service.  So when companies claim that their processes are at six sigma,
what they are really saying is that the short-term capability of their
processes is six sigma; the long-term performance, however, is 4.5 sigma
because of process centering errors. 
To cope with the phenomenon of shifts and drifts, any given design
(industrial or commercial) should be what we call "robust" to at least a
1.5 sigma shift in centering of all critical-to-quality characteristics
so as not to show any practical change in performance or yield.  The
methods used to accomplish this are incorporated in Design for Six Sigma
(DFSS) practices. 
The Six Sigma Breakthrough Strategy recognizes the importance of "shift
and drift" and makes it possible to account for them when assessing
process capability. 
In order to precisely represent the unknown distribution of a process
and its output data, the total process variation is subdivided into
short-term and long-term components.  Sigma short-term is the metric for
the effect of the trivial, while sigma long-term measures the effect of
both the vital few and the trivial many.  We also refer to this as
discrete and continuous data. 
Using 1.5 sigma as a standard deviation gives us a strong advantage in
improving quality not only in industrial processes and designs, but in
commercial processes as well.  It also allows us to design products and
services that are relatively impervious, or "robust," to natural,
unavoidable sources of variation in processes, components, and
materials.  The following table shows what the sigma shift means with
regard to the number of nonconformities or defects per million
opportunities.

SIGMA QUALITY LEVELS BEFORE AND AFTER A SHIFT IN THE AVERAGE DPMO* 

SIGMA LEVELS    WITHOUT SHIFT   WITH SHIFT

1               317400          97700
2               45400           308537
3               700             66807
4               63              6210
5               0.57            233
6               .002            3.4
To compensate for the inevitable consequences associated with the
centering errors, the distribution mean is offset by 1.5 standard
deviations.  The adjustment provides a more realistic idea of what the
process capability will be over repeated cycles.
* Defects per million opportunities

- end quote -

Can anyone direct me to the "Years of theoretical and empirical research
on this subject, that have proven this to be true"?
 I really do need some guidance !!
Regards

-- 
Terry Peterson
===========================================================================