Statistical Control

["John McConnell" <wysowl@msn.com.au>]

Some observations regarding statistical control follow:

1.    GENERALLY (not always) manufacturing processes tend towards =
instability and non-manufacturing processes tend towards stability.  =
Usually, much hard work is required to stabilise a manufacturing process =
at the outset.  This is counter-intuitive for many who expect processes =
dominated by machines to exhibit greater stability than those that are =
filled with people but with relatively few machines (except perhaps =
computers). =20

2.    Non-manufacturing processes may exhibit much variation, but they =
still tend towards stability; they tend to be robust.  (Remember, =
"stable" means neither "good" nor "low in variation".)  Any manufacturer =
can attest to the fact that the process can be humming along nicely in =
one moment, only to suddenly become chaotic or to rapidly alter its =
level of performance.  Very small alterations to inputs or to operating =
conditions can easily destabilise most manufacturing processes. =20

3.    The comment that processes tend to find a level of equilibrium, or =
to stabilise themselves, may have an element of truth; but even if it =
were absolutely correct it would be of little use to most of us.  This =
is because the statement assumes that no outside forces are acting on =
the process.....something I have never witnessed in business.....maybe =
others have.  Besides, are we not supposed to be on the path of =
continuous improvement?  Does this not imply changes to the process?

4.    Even in manufacturing, high level measures sometimes exhibit =
pretty good control.  My current project is a cement works.  Overall =
plant stoppages look pretty stable when plotted monthly.  Weekly plots =
reveal specials and shifts in the average.  As we break the plot down to =
categories, sub-categories (and so on, and so forth, per tedium ad =
nauseam) we find that not one reliability measure is even close to =
stable.  However, when accident data and service type processes are =
examined, all reveal an underlying stable system.  (We have a stable =
system for hurting people.)  The current estimate is that this plant =
would not only improve quality, but also would increase output by about =
15% by doing nothing other than stabilising the existing processes =
involved.  Then the serious fun starts as improvements are pursued.

One central reason to be concerned about statistical control is that =
until stability is established it is difficult, and sometimes nigh on =
impossible,  to properly establish causal relationships.  Often, this =
explains why a bunch of hard working people find that many of their =
efforts amount to naught.  The lesson is clear.  First stabilise the =
process, then set about establishing causal relationships and making =
improvements.

Hopefully, someone will find these ramblings useful. =20

Remember the 15th Point!

John McConnell
Wysowl Pty Ltd
Brisbane, Australia
wysowl@msn.com.au
=========================================================================