OEM / private label · Multi-factory sourcing · Export project coordination
sales@moxasourcing.comCall UsChat on WhatsApp

Acceptance Sampling: Plans, Risks and Inspection Work

A single attributes acceptance plan needs two numbers, not a sample quantity alone.

Acceptance Sampling: Plans, Risks and Inspection Work: Sample size n, Acceptance count c, Decision risks, Inspection work
Original educational diagram by Moxa Sourcing.
Quick comparison
Evidence or termWhat it means here
Sample size nThe number of units inspected in the specified stage.
Acceptance number cThe permitted count used by the specified acceptance rule.
Operating-characteristic curveAcceptance probability across assumed quality levels; not a guarantee for each shipment.

What does the evidence explain?

How do sample size and acceptance number define a single plan?

A single attributes acceptance plan needs two numbers, not a sample quantity alone. In the handbook notation, n is the number of randomly selected items inspected and c is the acceptance number. The decision rule in NIST sections 6.2.3 and 6.2.3.2 accepts when the observed number of defective items d is at most c, and rejects when d exceeds c. Thus the boundary includes equality: a result of exactly c belongs on the acceptance side of this stated rule. The inspected characteristic and definition of a defective item must already be clear before the count can be interpreted.

The distinction matters when comparing inspection proposals. Two proposals can examine the same number of units but make different decisions because their acceptance numbers differ. Conversely, matching acceptance numbers does not make two plans equivalent when their sample sizes differ. Record the pair and the rule together. A purchase order that records only “inspect n pieces” leaves the disposition rule unresolved. If the parties adopt a published standard, record its actual edition and selected procedure as well; this explanation does not supply protected standard tables or select an order-specific plan.

What changes when an inspection uses double or multiple sampling?

In double sampling, the first stage has three possible outcomes: acceptance, rejection, or a requirement for a second sample. When the third outcome occurs, the final decision uses the combined results from both samples. The second sample is therefore part of an agreed procedure. It is not a fresh opportunity to ignore an inconvenient first result. The inspection file needs to preserve the stage-one count, the instruction to continue, the stage-two count and the cumulative basis for the final decision.

Multiple sampling extends the staged procedure beyond two samples. NIST describes smaller sampling requirements as its advantage, while also identifying the complexity of implementing several stages as a practical tradeoff. “Multiple” does not mean repeating an inspection indefinitely until an acceptable result appears. Before work starts, the chosen plan must tell the inspector when to accept, reject or continue. A short final report should still allow the reader to reconstruct which path was followed. Without that history, a final count may conceal that the parties used different stopping rules.

Are sequential sampling and skip-lot sampling the same shortcut?

They change different parts of the inspection design. Sequential sampling examines items one at a time. After each examined item, the procedure calls for acceptance, rejection or another item. The changing quantity concerns how many units must be examined before deciding about the lot. Skip-lot sampling instead concerns which submitted lots receive inspection: only a fraction are inspected. The two names should not be used interchangeably on an order or inspection instruction.

A record of a sequential inspection should retain the observation sequence and stopping decision. A record that a lot was skipped should identify that it was not inspected under the skip-lot arrangement, rather than presenting an empty inspection form as a passing result. These are record-design applications of the two definitions. NIST’s introductory classification does not itself establish eligibility, a switching rule or permission to skip a particular mugwort shipment. Those details require the actual adopted procedure and agreement. An earlier accepted shipment does not, by the definitions alone, authorize omitting inspection of the next one.

Why do producer risk and consumer risk need two quality points?

The two risks concern different mistakes at different assumed defect levels. In the handbook model, producer risk is the probability of rejecting a lot when its defect level equals the stated AQL. Consumer risk is the probability of accepting a lot when its defect level equals the stated LTPD, a designated poor quality level. Reducing these ideas to “the inspection has a five percent risk” removes the condition that gives the number its meaning. A risk statement should identify which event, which quality point and which plan it describes.

AQL and LTPD are not the observed defective count in today’s sample. They specify quality points used to investigate or design the plan’s acceptance behavior. For the two-point design, the desired acceptance probability is one minus producer risk at the first point and consumer risk at the second. The source does not make any illustrative risk level compulsory for an order. Choosing commercially acceptable protection requires agreement on the characteristic and consequences; quoting an AQL label alone does not describe the protection against accepting a much poorer lot.

What does an operating-characteristic curve say about an inspection?

An operating-characteristic, or OC, curve places an assumed lot defect fraction on the horizontal axis and the plan’s probability of acceptance on the vertical axis. A point on this curve answers a conditional question: if lots have that defect fraction and the model applies, how often would this plan accept? It does not state that a particular accepted shipment has that quality, or that the acceptance probability is a confidence level for the shipment’s actual defect fraction.

The curve also explains why pass/fail labels alone cannot compare inspection plans. Two plans may both have accepted the few lots observed so far but have different probabilities of accepting poorer quality. Conversely, a plan can reject a comparatively good lot with a nonzero probability. Reviewing the whole relevant part of the curve makes that tradeoff visible. Keep the assumptions next to the curve, especially the sampling model and the quality levels being compared. A curve calculated for one pair of sample size and acceptance number does not describe another pair merely because the inspection service uses the same checklist.

When does the handbook use a binomial sampling calculation?

NIST’s worked single-plan calculation assumes the lot is very large relative to the sample. Removing sampled items then does not significantly change the remainder, allowing the count of defective items in a random sample to be modeled approximately with a binomial distribution. The calculation requires n, the sample size, and p, the assumed fraction defective. Acceptance probability is obtained by adding the probabilities for counts from zero through c, including c. This is a model calculation with explicit conditions, not a measurement of the unknown p in a shipment.

The source illustrates n = 52 and c = 3. Under its binomial model, acceptance probability is approximately 0.998 at p = 0.01, approximately 0.930 at p = 0.03, and approximately 0.223 at p = 0.10. These are contrasting points for the same plan, not three sample outcomes and not recommended limits for mugwort products. A randomly selected sample and the large-lot condition belong with the interpretation. If an inspection proposal supplies a probability without identifying its model or input values, ask for those missing elements before comparing the claimed protection with another proposal.

Why does average outgoing quality depend on rectifying rejected lots?

The handbook’s average outgoing quality, AOQ, model combines sampling with a specific rectification process. Rejected lots are screened completely and defective items are replaced with good units. Accepted lots and rectified rejected lots then contribute differently to the outgoing stream. The model’s long-run result cannot be attributed to sampling alone: it depends on the stated complete screening and replacement of rejected lots. A supplier who merely holds, returns or partially sorts a rejected lot has not thereby demonstrated the modeled rectification process.

Using N for lot size, n for sample size, p for incoming fraction defective and Pa for acceptance probability, the stated model gives AOQ = Pa × p × (N − n) / N. Its meaning is an average over the combined program, under the assumed incoming quality and rectification conditions. It is not a guarantee on an individual accepted lot. The practical review question is therefore whether the proposal actually includes the rectification route assumed by its calculation, and what record identifies the affected lots and replacement work. Do not turn a theoretical AOQ figure into an observed supplier result.

Is an outgoing-quality limit a maximum for every shipped lot?

AOQL describes the maximum of the average-outgoing-quality curve for the stated rectifying inspection program. It is a limit on the modeled long-run average as incoming quality varies. It is not an upper bound on the fraction defective in every individual lot that passes sampling. This difference follows from the object being summarized: an outgoing mixture over repeated lots, including screened rejected lots, rather than a complete count of each accepted shipment.

At very good incoming quality, the modeled outgoing fraction defective is small. At the other extreme, very poor lots tend to be rejected and completely rectified under the model, so their outgoing contribution is also improved. The curve rises between these extremes before reaching its maximum and falling again. That behavior depends on rectification. If the inspection program does not perform the assumed screening and replacement, a statement about AOQL from this model no longer describes that program. Record whether a quoted “quality limit” is an acceptance specification, an OC design point or an AOQL calculation; the three labels answer different questions.

Why can total inspection work greatly exceed the sample size?

For a program that completely screens rejected lots, inspecting the initial sample is only part of the work. NIST expresses average total inspection as ATI = n + (1 − Pa) × (N − n). The first term is the initial sample. The second adds the remaining items in lots that are rejected and therefore receive complete screening. As the acceptance probability changes with incoming quality, the expected inspection workload changes even when the original sample size stays fixed.

The model places the average workload between the sample size and the lot size. When no lots are rejected, the initial sample accounts for the work; when every lot is rejected, complete screening dominates. This explains why comparing service proposals only by their initial sample quantities can miss a substantial operational difference. Ask whether screening of rejected goods is included, who performs it and how its scope is recorded. Those commercial arrangements are separate from the statistical formula. An ATI calculation does not demonstrate that screening has been purchased, performed or successfully completed for the current order.

How does average sample number help compare inspection plans?

Average sample number, ASN, concerns the number of units examined through the sampling stages. For a fixed single-sample plan, each lot has the specified sample size. For staged plans, the amount examined depends on when the procedure reaches an acceptance or rejection decision. NIST describes an ASN curve against incoming defect level as a way of examining a plan’s sampling efficiency. It is a different workload measure from ATI, which includes full screening of rejected lots under the rectifying program.

A plan with a lower average sampling requirement can still be more difficult to schedule and execute because its final workload is not known in advance for each lot. NIST frames selection between suitable single and multiple plans as a tradeoff between average savings, added complexity and uncertainty in day-to-day inspection work. Compare protection first, then examine the operational tradeoff among plans that meet the agreed properties. A smaller average sample is not, by itself, evidence of either weaker protection or a better choice. The stopping instructions and the team’s ability to follow and record them matter to implementing the selected method.

FAQ

How do sample size and acceptance number define a single plan?

A single attributes acceptance plan needs two numbers, not a sample quantity alone. In the handbook notation, n is the number of randomly selected items inspected and c is the acceptance number. The decision rule in NIST sections 6.2.3 and 6.2.3.2 accepts when the observed number of defective items d is at most c, and rejects when d exceeds c. Thus the boundary includes equality: a result of exactly c belongs on the acceptance side of this stated rule. The inspected characteristic and definition of a defective item must already be clear before the count can be interpreted.

How does average sample number help compare inspection plans?

Average sample number, ASN, concerns the number of units examined through the sampling stages. For a fixed single-sample plan, each lot has the specified sample size. For staged plans, the amount examined depends on when the procedure reaches an acceptance or rejection decision. NIST describes an ASN curve against incoming defect level as a way of examining a plan’s sampling efficiency. It is a different workload measure from ATI, which includes full screening of rejected lots under the rectifying program.

Sources and scope

Sources and scope: NIST/SEMATECH handbook sections 6.2.2, 6.2.3 and 6.2.3.2. Definitions and stated models, with editorial record applications; numerical single-plan example is illustrative, not an order recommendation. Protected ISO tables and source figures are not reproduced. NIST 6.2.2 NIST 6.2.3 NIST 6.2.3.2

Chat on WhatsApp+852 5725 4301