Re: den.list-d Digest V2001

[EJBeach@aol.com]

Comments on Assignable causes of Variation:

When Dr. Deming and Dr. Shewhart wrote about control charts, they were very 
specific about when to look and investigate for assignable or special causes 
of variation. They agreed that you should investigate your process for 
changes when a nonrandom pattern is found on the control chart. In addtion, 
you should investigate for the change when it is economical to do so. 
Although mathematical statistics is important to define the control limits 
for a process, the use of probability ends once the charts are constructed. 
You cannot possibly assign a probability to a particular pattern. There is no 
defined probability of an assignable cause based on the writings of Dr. 
Shewhart. Many authors based these probabilities on either a mathematical 
model - Normal, poisson, etc. and assume the process is perfectly located at 
the average line with one defined standard deviation. Of course, this is not 
possible. Processes waver. They are not static and particularly vary during 
the beginning of trying to stabilize the process. Although the Den has posted 
from many books  on the patterns to look for special causes, none have come 
from Shewhart.

It is understandable that if you are teaching SPC it is nice to give some 
idea of what a nonrandom pattern may look like on the chart. Dr. Shewhart was 
purposely vague - trend and cycles. How many is dependent on the economics of 
missing a shift and the cost of going to investigate the causes. In some 
processes it is a minor cost and can be quickly investigated. In other 
processes, such as large rolling millings, the cost of missing a change is 
prohibitively expenses - I would start looking for a change after 3 points 
trending. Better safe than sorry. The only answer is to understand the 
specifics of the process you are trying to control. In other words, learn 
what a change in your process would look like and define when to take action 
based on the cost of missing the change and the cost of investigating. It is 
purely an economical decision, not a probability calculation. 

I guess it takes 20 years of working with SPC to begin to understand the 
difference between the fallacy of using probability versus the economics for 
Dr. Shewhart's control limits for quality. Ed Baker is right, it will take 
another 40 years. How sad. 

Eileen J. Beachell
Quality Disciplines
ejbeach@aol.com