Re: SPC question response

[GrantBlair@aol.com]

In a message dated 8/15/2001 2:50:16 AM Eastern Daylight Time, 
tpr42345@aol.com writes:

<< So how should a practitioner proceed? >>

I'll repeat again, fellas....it ain't that hard!!! All the process owner has 
to do is ask two questions:

   o How many DEFECTIVES do I have? 
   o How many DEFECTIVES do I not have?

If both questions can be answered: X oversize, undersize, cracked, etc. (N-X) 
right size, not cracked, etc.Then, you can express this as a percent, plot it 
as a p-chart,  safely assume the underlying distribution is binomial,and 
sigma is calculated by multiplying p(1-p).

However, it the question becomes:
 How many DEFECTS do I have, then you will have a difficult time answering 
the second question:
How many interruptions, scratches, etc. do I NOT have?
Now the underlying distribution becomes poisson, and the sigma calculation is 
simplified because p is so SMALL
that (1-p) is essentially the same as 1. (open a good Statistics book!...the 
derivation is pretty easy!!!).

Now, it has been correctly pointed out in another post that X-MR charts are 
an appropriate substitute for p-charts under certain circumstances, such as % 
efficiency/downtime, etc....this post also provided the statistical basis for 
this.

I'm also surprised anyone would claim that control charts with 3-sigma limits 
are only appropriate for time-ordered data. Chambers provided a number of 
examples where control charts with 3-sigma limits were used to distinguish 
between operators, etc. 


GrantBlair@aol.com
Ninety Six SC