From: Stan Hilliard
Date: 31 Oct 2002
Time: 00:04:16
The thing that makes an attribute sampling statistically valid is that the user knows the probabilities of accepting lots, if the lots are of various percent defective. Show them the oc curve of the plan that they are actually using, and for the normal plan.
The oc curve for an attribute plan depends only on n and AC. It doesn't matter whether they got the n and Ac from a standard or just made it up.
From the OC curves, show the user the RQLs of the plans for a consumer's risk of 5%. He/She might not be happy with the high RQL of the current plan.
The AQL based standards like ANSI Z1.4 are designed to allow users to set up sampling plans without considering the consumer's point of the oc curve. Since most organizations that use such plan have customers, they don't know their most important risk - the risk of passing off-grade (RQL) lots. In such cases the plans are invalid because the users do not know the consumers risk.
samplingplans.com/aqlprimer.htm samplingplans.com/usingoccurves.htm samplingplans.com/modern3.htm
Switching procedures give you different oc curves for each of the three plans. I think that a better way to adapt the sample size to the current quality is with sequential sampling. You would have smaller sample size when the quality is good, yet the plan has only one oc curve. Switching rules give you a less good oc curve for lots that are on reduced sampling - as indicated by the RQL.