Arriving at Sample sizes based on confidence levels

[Andrews]

One of the CSR's from customer asks us to use sample sizes consistent with 90% confidence if used for lot acceptance.What does it mean? How to derive a sample size based on the confidence level?

Kindly help.

 

[ralphsulser]

We had that from Visteon when they first issued their CSR, but fortunately they deleted the requirement for confidence level when they revised their CSRs in Feb. 05

 

[bpritts]

Andrews --

This type of requirement is missing some vital information.

There is an entire branch of quality engineering that will give you the
statistical confidence for your product based on inspection sampling.

In the USA these were popular in the 1950 to 1970 time frame. Our US
Army created a standard called Military Standard 105D to standardize acceptance sampling.

What this plan can do is tell you how many pieces you must sample once you know:

1. The total manufacturing lot size.
2. The percentage defective that you will accept.
3. The confidence level that you require (90%).

You can probably find data on MIL STD 105D on the web. I have it in a book titled Quality Control and Industrial Statistics - by Acheson Duncan. ASQC sells this book.

If you must do acceptance sampling, this method will give you the details.

The problem with this approach is that when you specify 0 as your acceptable percent defective, the sample size gets very large, perhaps as high as 100%.

Today, in North American automotive, we have pretty well stopped using these techniques. Instead, we use statistical process control to attempt to control the process, rather than accepting lots based on inspections. If the process is not capable to a very high probability, the customer will demand
that you inspect 100%.

Best regards,

Brad Pritts
Ann Arbor, Michigan

 

[Tim Folkerts]

Andrews,

I will echo most of what Brad said, except the MIL-STD-105 (also known as ASQ Z1.4) is not based strictly on any specific confidence level. Using the typical "Normal, Level II" sampling plans, the probabilty of accepting a lot where the defect rate is equal to the AQL is around 95%, but it can vary considerably (from around 90% to 99% depending on AQL & lot size).

So your confidence level of accepting good lots is around 95%. On the other hand, this tells you next to nothing about the probabilty of reject bad lots. Still, this is the industry standard, so if you tell your suppliers/customers that this is what you are doing, they will probably accept the sampling plan.

I posted a spread sheet in another recent thread (https://elsmar.com/elsmarqualityforum/threads/12836/) that allows you to input the confidence levels (actually alpha and beta values), along with what you would consider "good" and "bad" defect rates. Then you slide a little control until the sample size produces an acceptable sampling plan. I have made one refinement to the spread sheet (I suppose I ought to post the update, but I am at the wrong computer right now). Perhaps this can help you decide on a reasonable sampling plan.


Tim F