Bayes Success-Run on Attribute Data - Determining Sample Size

B

blazin912

I'm trying to recall the appropriate way to develop a sample size justification on Attribute Data. I believe I can use the Bayes Success-Run method, but would like some input/confirmation.

What we're trying to achieve.. perfection, :lol: but seriously, we need to test a limited number of samples (3-5) numerous times to build up our sample count to provide sufficient confidence that a failure will not occur. Failure count must be = 0, period end of story.

We're dealing with an issue that occurs at random currently due to EMI issues. While we're confident in our solution, we need statistical backing in the form of a go/no-go test sample size that will stand up to scrutiny.

As this will be a one time protocol run, running each unit for 100 activations ie 500 total samples is not out of the question, but I'd like to confirm the minimum number required.

Calculator I found came out with 457 with C= 99 and Rc = 99
 

Bev D

Heretical Statistician
Leader
Super Moderator
A simple formula for Bayes successful runs is:

worst case failure rate = 1/(n+2)

you must have zero failures in your sample and n is the number of sequential runs.

note - you can NEVER predict perfection.

So, if you take one unit and have 1000 successful sequential runs, your maximum expected defect rate (95% confidence level) is 1/1002 = .000998 = .1%

I don't remember where I got that formula from...
 
Top Bottom