[ | Next | Previous | Up ]

Re: Validation Sampling Plans

From: Stan Hilliard
Date: 20 Oct 1999
Time: 23:09:07

Comments

Greetings,

You Said:

> What changes do you make to design a > sampling plan for an unvalidated process?

You do not have to make any changes to the sampling plan for unvalidated processes. Process validation is not an issue when applying acceptance sampling plans.

I will elaborate here on some related matters. Each type of plan has its own assumptions: Here are my thoughts about attribute sampling plans.

VALIDITY OF SAMPLING PLANS: The two things that make a sampling plan statistically valid are:

(1) that its designer/user knows and understands the probabilities/risks of the decision rule. The operating characteristic curve of the specific sampling plan describes those probabilities/risks as a function of the true lot fraction defective.

(2) that the oc-curve probabilities are true under the conditions of application of the decision rule.

POPULATION=LOT: These plans are lot-by-lot plans, as opposed to continuous sampling plans. It doesn't matter whether the items in the lot are aggregated together physically or whether they are a population by definition -- like all the items processed between 10:00am and 10:15am.

TEST METHOD: It is assumed that the test method can accurately evaluate the items in the sample.

RANDOM SAMPLING: By random sample, I mean that prior to selecting the sample, all items in the population have an equal probability of being selected.

STATISTICAL CONTROL: A sampling plan will work even when the process that generated the items is not in statistical control! In fact, an important value and purpose of a good sampling plan is to be able to detect changes in product quality when a process deteriorates. That is what justifies using the sampling plan.

EXAMPLE: Consider this extreme case. A warehouse contains items that are stacked on pallets. Select a sample randomly. Use an attribute sampling plan with a specified producer's point (AQL,alpha) and consumer's point (RQL,beta). See:

www.samplingplans.com/modern3.htm#EVALUATE

Suppose in that warehouse all the pallets were 100% good except one, and that one pallet contains 100% defective items. This aggregation of all the defective items onto one pallet is similar to an extremely out-of-control process.

(A) Consider each pallet to be a lot. The sampling plan will accept all the pallets except the one bad one, which it will reject. It picks out the "out of control" lot.

(B) Consider all items in the warehouse as the population/lot. This is similar to a process out of control within-lot. The oc-curve of the sampling plan still accurately describes the probabilities and risks of the decision rule. The thing that makes it work is the random selection of the sample -- the probability of selecting an item is independent of which pallet it was on.

The short answer to your question is: No change is necessary. An attribute sampling plan will work with product from an unvalidated process.

Sincerely, Stan Hilliard CQE,CRE,CQA,PE


Last changed: November 20, 2007