Statistically Determining how many Tests to conduct.

[JLRothbauer]

Question: How to determine how many samples needed to ensure that you have statistical similarity, i.e. one more sample is just more numbers..

Background info....
I am conducting a series of tests on our newly complete Co-Extrusion Film line. We have produced a tremendous amount of samples and we are currently working our way through them.
Our mission is to measure the thickness of each individual layer in the 3 layer film using a microscope.
We have several hundred samples 82 inches wide that are tested every inch.
For each individual test there are three measurement made: Thickness Layer 1, Thickness Layer 1 + 2, Total Thickness.
So, in the end, you have a web profile indicating the film's thickness.
How do you determine how many samples are necessary. Each sample takes over 2 hours to do, so the difference between 5 and 10 is a huge amount of time.
When is, for example, 5 samples worth of data no better than 6 samples worth of data on a statistical level. How do I determine this?
Ugh, what an ugly post. I must apologize that I can't find the proper way to put my problem into words anyone can understand.

Well I thank you all in advance; any help would be greatly appreciated.

 

[Kevin Mader]

Do you have access to "Quality Planning and Analysis", Juran/Gryna?

You are testing a hypothesis and Chapter 10 in this book details what you need to know far better than I can put it to words here.

Regards,

Kevin

 

[JLRothbauer]

Thanks.

No, I don't actually have access to that book, but I do have others I can look into.

I have some statistical training under the belt, but my problem has always been application of the techniques when/where. I see that the isse is rearing its ugly head yet again.

I will go through my book on the topic of Testing an Hypothesis and relearn the technique.

Thanks for the advice.

 

[Kevin Mader]

When studying for the CQE, I found this book to be an invaluable source of information. Chapter 10 uses several types of tests (i.e. sample to a population, population to population, etc.). This should help you to determine what test to use, two or single tail tests in determining, or reference you to other reading that will help.

I'd get the book. I own two myself keeping one here at work, and the other at home.

Regards,

Kevin