Re: Sample size selection for peel strength for variable data using a 95%/99% confide
I haven't seen that ISO standard, however, I can confirm that Steven is correct from a couple of sources (which may be cheaper to obtain than buying the ISO standard): Juran's Quality Control Handbook, 4th Edition, discussion on pp. 23.52-56 and table of K values on p. AII.36-37. Also see Experimental Statistics, by Mary Natrella, National Bureau of Standards Handbook 91, issued 1963, pp. 2-13 to 2-15, "Statistical Tolerance Limits" and tables A-6 and A-7. The handbook also includes a formula for calculating K values for confidence / reliability percentages which are not in the tables.
I'd like to ask a couple of follow-up questions of Statistical Steven (pardon my statistical ignorance – I’m kind of a newbie):
1. Is there a way to use Minitab to perform these calculations rather than look up k in a table? Does it calculate K if you input Xbar, s, upper and/or lower spec limits, confidence level, and proportion (reliability)? I found something close in the discussion about "Acceptance Sampling by Variables - Accept / Reject Lot", which mentions the Xbar +/- K*s formula, but does not give you the option of inputting the desired confidence / reliability factors.
2. There seems to be a danger in calculating K just based on my best guess of the standard deviation from limited data. If I underestimate the true standard deviation of the population, and I start making parts and running tests, then if my future sample standard deviations are unexpectedly large, I might have to reject the lot and fail the test. It occurred to me that one way to handle this is to calculate a confidence interval of the standard deviation, to the desired level of confidence, in this case 95%, and then use the upper bound of the interval for the standard deviation in my K value calculations. Is this correct? This way I build in a safety margin, so the chance of future lots failing is only 5%.
Any insight you have would be appreciated. Thanks.
Ari