From: bart.debruyn@vlm.be
Date: 03 Oct 2001
Time: 08:45:31
Greetings Stan,
I just want some supplementary explanation on what you said:
> I ASKED -- "Does the software also contain decision rules taken into account the confidence limits ??? For example: Specification indicates that fraction f within lot: f > 97% result within sample: 96% +/- 3%" > YOU SAID -- "TP105 will calculate, by entering the sample results (n and x), one-sided or two-sided confidence limits for percent defective -- at any % confidence that you choose from 50% to 99.9999%. You can also enter n and (n-x) to calculate the confidence limits on percent good for that lot (or group of lots). If I understand the confidence limit requirement, a lot must be shown to be >97% good with 95 % confidence. That would mean that the lower 95% confidence limit on % good has to be equal to or higher than 97% good. In terms of %defective -- the upper 95% confidence limit would have to be equal to or less than 3% defective (100-97). Because of the relationship between confidence limits and decision rules, the RQL(LQ) should be 3%. That is, by definition, the upper confidence limit on % defective is identical to RQL. The % confidence is (1-beta) where beta is the consumers risk."
The lower 95% confidence limit on % good amounts in this example to 93% which is less than what has been specified (at least 97% good). So I understand that in this example there is no accordance with the specifications. I am surprised that one needs the upper confidence limit because this value depends on sample size. The controller can choose a small sample so that rejection of the lot follows.
example:
Our specifications indicate that a lot must be shown to be >97% good with 95 % confidence lot size: 1000 items
sample 1: 50 items with 49 good items (98%)
sample 2: 950 items with 931 good items (98%)
According to the ncs pearson calculator on the web the former confidence interval amounts to 13,5% and is of course much wider than the latter (0.7%).
Only if we decide to do about a 100% control, the lot shall be accepted.
In my opinion the "98% good" should be determinitave for the entire lot. Is it current to specify the minimum sample by a "maximum confidence interval" ? "The controller should take a sample that results in a confidence interval of maximum +- 5%.
bart de bruyn