Skip to content
Incoming Inspection and Traceability | Part 2 of 6

Sampling Plans: Sample Size, Acceptance Number and What They Prove

How a sampling plan works in plain arithmetic, what zero defects in a sample really proves, and when to switch to tightened or reduced inspection.

Ask an inspector where the sample size of fifty comes from and you usually get one of two answers: it has always been fifty, or it comes from a table. Neither answer helps when a customer asks what the check proved.

A sampling plan is a bet with known odds. Three numbers make it, and the fourth number is the one you actually need.

The three numbers and the fourth

The lot size is the quantity the plan applies to. Use size bands, not one plan per delivery.

The sample size is how many parts are drawn and checked. It follows from the risk class and the switching state, not from the time available.

The acceptance number is how many defects the lot may contain in the sample and still pass. Zero for characteristics that matter, one or two for cosmetic ones.

The fourth number is the chance that a lot of a given quality is accepted. With an acceptance number of zero it is simple arithmetic: the share of good parts raised to the power of the sample size. A lot at 1 percent defective and a sample of 50 gives 0.99 to the power of 50, which is 61 percent. With an acceptance number above zero, add the chance of drawing exactly one defect, exactly two, and so on.

Plan Sample Accepts a 1 % lot Accepts a 5 % lot
Reduced 20 parts 82 % 36 %
Class B, normal 32 parts 73 % 19 %
Class A, normal 50 parts 61 % 8 %
Tightened 80 parts 45 % 2 %

Write those two percentages on the plan card. Everyone who uses the plan then knows what it does and does not promise.

What zero defects proves

The question an auditor asks is different: the sample was clean, so what may you say about the lot?

Turn the arithmetic around. With fifty parts and no defect found, the defect level at which the plan would have caught something nineteen times out of twenty is 5.8 percent. That is the whole claim. A sample of 32 proves better than 8.9 percent, a sample of 80 proves better than 3.7 percent.

Fifty clean parts does not mean the lot is clean. It means a lot worse than 5.8 percent defective would probably have been caught. Say that sentence out loud once in a meeting and the discussion about incoming inspection changes permanently.

Three limits follow from it. A sample cannot find a rare defect, so a part whose single failure causes a safety problem needs a full check, a detection device in the process or a supplier whose capability is proven. A sample drawn from the top layer of the first pallet is not a sample of the lot. And a plan written for a visible characteristic says nothing about a characteristic nobody checks.

The plan has to move

A plan that never changes is not a plan, it is a habit. Four switches, each triggered by the record and not by a discussion.

After one rejected lot, go to tightened: the sample doubles and the acceptance number is zero, until five lots in a row pass. After ten accepted lots with no complaint for six months, go to reduced: the sample halves, and one rejection ends it. After twenty accepted lots, an approved process on file and a signed release, go to skip lot: one lot in four is inspected. If a defect reaches the customer, every part is checked at the supplier cost until the fix is proven on three lots.

Named standards for attribute and variables sampling exist, and they are worth reading before you write your own tables. What matters more is that the rule is written down, that the log triggers it without anyone having to defend the decision, and that the supplier knows the rules before they apply.

Newsletter

One Lean idea.
Every two weeks.