Re: Statistics and Action (2)

[GrantBlair@aol.com]

Quoting David Kerridhge:

>>>The confusion has arisen, I believe, in the minds of those
who treat control charts as if they are significance tests.
The only purpose, from this point of view, is to "prove" that
a change has taken place - and as soon as possible.
>From Shewhart's operational point of view, the purpose is
to decide on economically worthwhile action - and the problem
dissolves at once.
Cumulative sum charts can be a useful graphical technique.
But I cannot think of any circumstance in which it would
be my main, let alone only, way to study a process.<<<
[end quote]


David, I think you do a disservice to the quality gurus you reference to 
suggest that a control chart's existence is to collect data to 'study a 
process' in order to determine 'economically worthwhile action'.
The purpose of a control chart is to separate common cause and special cause 
variation. If this is not done, then ACTIONS taken to IMPROVE the process 
will only result in making the process WORSE
.
You point out, correctly, that there are two types of improvement actions:
1. Tracing and eliminating the cause of changes of mean.
2. Adjust to center the mean.
However, your conclusion that it is NOT 'desirable to take immediate action'' 
but rather to 'have plenty of observations' is not supported from practical 
experience in either of the cases. 
It has been shown in numerous cases that additional sampling following an 
indication of special cause will result  in the APPEARANCE of statistical 
control, even when special cause exists (3 cases out of 4). I have seen this 
firsthand many times...the process owner learns very quickly that taking 
another sample (and not plotting the current one, if he can get away with 
it!!) results in a point inside control limits, and the problem "has gone 
away".
 More importantly, by not taking a look at the process, the opportunity has 
been lost to determine root cause. Then, when special cause returns, the end 
result will be identification of an incorrect cause as something which 
happened on "someone else's watch." In many cases, this can be addressed by 
plotting the data as a CUSUM, in which the change in slope will indicate WHEN 
the change occurred, far more accurately than a traditional Shewhart chart 
(See the red bead example in Chapter XXII of Duncan's "Quality Control and 
Industrial Statistics 3rd Edition")

Historically, CUSUM charts were developed in the U.K. by Imperial Chemical 
Industries for control of chemical processes. This theory led to EWMA charts, 
which have the advantage of estimating an value which can used to adjust the 
process. EWMA charts are extensively used in the chemical process industry 
because chemical processes fit the A model  so well. Let's consider a typical 
chemical process, in which one or more raw materials is reacted to produce a 
finished product. Typically, raw material will be fed to a number of reaction 
vessels, which are Heated/Pressurized/Agitated to facilitate the reaction. 
Raw material variation is addressed by averaging results across these vessels 
and plotting this average as an EWMA, then adjusting the process when special 
cause is indicated. If you consider the model, nothing is gained from taking 
additional measurements, since you cannot "unmix" a liquid, powder, or chip, 
once the raw material been loaded into the supply tank...you must 
continuously 'trim' the results to minimize variation as a raw materials 
shift works its way through the process.Now, changes in the individual 
vessels are best addressed as difference charts. CUSUM techniques are often 
useful to identify root causes (e.g. temperature, pressure, sensors, 
controllers, etc.).

 Personally, it is interesting to me that studies to "debunk" these charts 
invariably use processes which DO NOT FIT this model (Wheeler uses camshaft 
diameters!!!). 


GrantBlair@aol.com
Ninety Six SC