RE: control charts and non-normality

[Steven_S_Prevette@RL.gov]

False alarms are always going to occur.  Cannot be avoided, unless you turn
off the alarm (try that on your smoke detector at home sometime).  Dr.
Shewhart's original work was based upon testing several non-normal
distributions, and through use of the Tchebychev Inequality.  I would not be
surprised (and I have heard this number before) that there is a 5% (average
run length of 20) false alarm rate for control charting that uses rules like
7 (or 8) points in a row the same side of the average, runs of 7, points
outside the control limits, etc.

I happily accept a false alarm rate of 1 in 20.  I say this in response to
the comment "I don't know any practitioner who would consider 50 to be an
acceptable value".  An average run length of 20 gives me in 19 results out
of 20 where I can tell the managers to stop trying to find a special cause
in every result, and focus on the systematic results.  By procedure, our
graphs have a minimum of 25 points on them, but few exceed 50 points.  That
means I could expect one or two false alarms per graph, possibly.

In performing the Red Bead Experiment, I have run into two false alarms out
of performing the experiment approximately 30 times.

Is the glass 5% empty or 95% full?

But Dr. Deming did admonish in Out of the Crisis to stop treating control
charting as an exercise in probability.  It is a prediction tool.  

There are statistical ways of increasing the sensitivity of control charts
without increasing false alarms.  Modifications to take into account
non-normality are perhaps one way.  EWMA and CUSUM control charting is
another.  But there is also another tradeoff that must be taken into
account:  that of complexity of analysis versus effectiveness.  I believe
(and I believe I have empirical evidence) that a non-statistician, with the
support of a statistician on call, can use control charts in a economic and
effective manner to manage a process.  

At some point, the benefits of the more advanced tools do not outweigh the
costs, both in culture and dollars.

Steve Prevette
ESH Planning and Performance
Fluor Hanford, A Fluor Global Services Company
ASQ Certified Quality Engineer
steven_s_prevette@rl.gov
509-373-9371
=====================================================================