I am working with our finance group on a testing plan for our financial controls (Sarbanes-Oxley related) and we are looking at using the C=0 Sampling Plans table to determine our sample size for these various tests. The question I have been asked is “what is the confidence level (power?) of these kinds of sampling plans”? I have searched the isixsigma site and have found many threads on this topic but have not found anything about relating them to a confidence level (power?). Can anyone provide help with answering this question? Thanks in advance
Of course. But there are a few questions that should be answered first.
Power of what? Confidence to do what?
Are you going to count defects or defectives? (i.e. poisson or binomial?)
C=0 alone does not define a sampling plan.Whatever you want to know about the performance of the sampling plan you will need the sample size too.
Typical questions that can be answered would be: “What is the risk to fail to reject a batch that contains more than 5% of defectives, which we would like to reject with a high probability?” “What is the risk to reject a batch with only 0.1% of defectives, which we would want to accept with a high probability?” “Which is the level of defects that has a 50/50 chance to be accepted or rejected?”And allways for a given control plan: n=…, c=…
I’ve never seen an answer to this question either. I’ve looked at C=0 inspection plans to use on incoming inspection of purchased parts. The typical plan comes straight out of the old MIL STD 105. For example at an AQL of 2.5 and a lot size of 40, those plans recommend a sample size of 5. Well, one defective in a lot of 40 would be 2.5% defective, and common sense says a sample size of 5 isn’t going to find a lot of 40 with 1 defective very often. Indeed, the hypergeometric function tells me you would find such a lot only 13% of the time. Not very good odds in my opinion! So perhaps someone else can explain where and why such plans are in use and promoted.
C=0 sampleing plans were meant to promote the idea that no level of defects is okay. They fail miserably in protection of the user. The OC curves are almost flat.
Read Deming for the best advice on sampling plans.
Thanks for the immediate reply! I posted a similar question on a sampling forum here:
Any response there would be appreciated (-:
I am forever, telling people that sampling to find small percentages requires a large number of data, and that checking out four or five to get that warm fuzzy feeling really doesn’t accomplish anything. Any support in that message is helpful to me. If you disagree, I’m open to that as well.
My own gut feeling on the matter is that the best sample size is zero, [You know they're good!] After that, a sample of one [an identity check] tells you the most, and after that? If it’s that important, you ought to consider 100% inspection. Anything in between, you’re just fooling yourself!
All good points so far. c=0 was meant to stop any serial type defects. If you found one defective part with any one defect on it, you have to go 100% if your check list means anything to the final application. For defects that were from random reasons, c=0 sampling does not work.
When considering sample size vs the lot size and the way the samples are to be pulled, one needs to include to cover all possible process streams – a machine, a batch of material, an operator/ a shift, a tool, etc.
Unfortunately, if you are talking to anybody who cares about cost (sometimes more than they care about quality), 100% inspection just isn’t practical.
The forum ‘General’ is closed to new topics and replies.