18.1 Acceptance Sampling & Sampling Types (incl. MSA)

18.1 Acceptance Sampling & Sampling Types (incl. MSA) (CQA BoK V.E.1–2 — Apply)

/practice/cqaPractice questions with detailed explanations

Sampling is how auditors and quality systems convert limited observation into defensible decisions. Domain V.E expects you to apply acceptance sampling concepts and sampling types—not design ANSI/ASQ Z1.4 tables from scratch on the exam, but recognize correct use, misuse, and audit implications. You will see sampling in supplier receiving, in-process checks, finished-goods release, environmental monitoring, document reviews, and audit field work itself.

Acceptance sampling — what it is and is not

Acceptance sampling is a statistical (or sometimes risk-justified) procedure that uses sample results to accept or reject a lot (batch, shipment, production period) against specified criteria. Classic elements:

What acceptance sampling is not:

• A guarantee that every accepted unit is conforming
• A replacement for process control, capability, or prevention
• Automatic proof of product safety when sample size is tiny relative to risk
• A license to skip inspection of critical characteristics when regulations or risk demand more

Auditors challenge plans that treat “we sample 5” as quality assurance without linking n, c, risk, and process knowledge.

Attributes vs variables acceptance sampling

Attributes sampling

Each unit is classified as conforming / nonconforming (or defects are counted). Plans are built around defectives or defects-per-unit.

Typical uses: visual defects, go/no-go gauges, functional pass/fail, labeling presence, attribute checklist items.

Strengths: simple data collection; clear lot rules; works for qualitative defects.

Weaknesses: needs larger samples than variables for comparable discrimination; loses information about how far a unit is from the limit.

Variables sampling

Units are measured on a continuous scale (dimension, weight, purity, tensile strength). Plans use sample mean and often standard deviation against specification limits.

Strengths: more statistical power per unit inspected; can detect process shift earlier; supports capability thinking.

Weaknesses: requires reliable measurement systems; assumptions about distribution (often normality); more calculation and training.

Worked scenario — attributes lot decision

A receiving plan for connector lots: n = 50, c = 1 (accept if ≤1 defective in the sample).

• Sample finds 0 defectives → accept lot under the plan.
• Sample finds 2 defectives → reject lot under the plan (even if 48 are good).

The auditor’s job is not only to recompute the count but to verify: Was the sample randomly selected from the lot? Was the inspection method defined? Was the plan authorized for that part criticality? Was the rejected lot contained and dispositioned?

Sampling types auditors must apply

Random sampling

Every unit in the defined population has a known, equal chance of selection (simple random sample). Use random numbers, systematic-with-random-start when justified, or software selection from a frame.

Audit use: selecting records from a closed period, units from a homogeneous lot, transactions from a log.

Failure mode: “grabbing the top of the pallet” or “asking the supervisor for typical lots” — convenience sampling dressed as random.

Stratified sampling

Divide the population into strata (shifts, lines, product families, suppliers, severity classes) and sample within each stratum. Improves representation when strata differ.

Audit use: CAPA sample by site; complaint sample by product family; internal audit samples balanced across departments.

Cluster sampling

Select clusters (cartons, days, machines, stores) then inspect all or many units inside selected clusters. Efficient when travel or access cost is high; higher design effect if clusters are homogeneous.

Audit use: sample three production days fully rather than one unit from every day across a quarter.

Nonstatistical sampling

Judgment, haphazard, or purposeful selection without probabilistic design. Allowed in many audit contexts when objectives are exploratory or risk-focused—if rationale is documented.

When it fits: following a known high-risk trail; investigating a specific complaint lot; testing a new control design.

When it fails: claiming statistical confidence or extrapolating “zero defects in 10 convenience samples” to the full population.

Risk-based sampling

Allocate more sample intensity to higher risk (patient safety, regulatory criticality, history of nonconformances, new suppliers, process changes). Lower risk areas get lighter sampling or skip-lot schemes when justified.

Apply-level auditor moves:

1. Map characteristic criticality (critical / major / minor).
2. Use history (supplier scorecards, process capability, previous findings).
3. Increase n or frequency after change, CAPA, or process instability.
4. Document why low-risk items were deprioritized.

Measurement system analysis (MSA) considerations

A sampling plan assumes the observed result is a trustworthy proxy for the true characteristic. MSA evaluates the measurement process: bias, linearity, stability, repeatability, reproducibility, and (for attributes) agreement.

Why CQA candidates care:

• Variables acceptance decisions collapse if gage R&R is a large fraction of tolerance.
• Attribute inspectors who disagree create inconsistent lot dispositions.
• Audit findings based on poorly calibrated instruments are invalid evidence.
• “We measured 30 units” is meaningless if the caliper is out of calibration or operators use different methods.

Auditor questions for MSA during sampling reviews

• Is the gage calibrated and within interval for the range used?
• Was R&R studied for this characteristic / fixture / method?
• Are operators trained and procedures controlled?
• For go/no-go: is there a master / attribute agreement study?
• Are environmental conditions (temperature, humidity) controlled when they matter?
• Do out-of-tolerance findings trigger impact assessments on prior measurements?

Scenario — sampling looks fine, MSA does not

A plant rejects lots on a critical dimension using n = 20 variables sampling. Capability charts look terrible. MSA later shows gage R&R ≈ 45% of tolerance with large operator-to-operator differences. The “process problem” is largely a measurement problem. Acceptance sampling amplified noise into costly false rejects—and may also have accepted bad product when bias ran the other way.

Audit application checklist (Apply-level)

1. Define the population (lot, period, scope) before sampling.
2. Classify characteristic (attribute vs variable; criticality).
3. Select sampling type matched to structure and risk; document if nonstatistical.
4. Confirm plan parameters (n, c or variables criteria) are authorized and current.
5. Verify selection method (random numbers, strata, risk rules)—not convenience.
6. Confirm MSA / calibration adequacy for the method.
7. Trace disposition of rejected lots and any “accepted under waiver” paths.
8. Link findings to procedure, standard, or customer requirement—not personal preference.

Common exam traps

• Confusing acceptance sampling with process control or capability
• Treating attributes and variables as interchangeable without noting sample-size and MSA differences
• Calling convenience sampling “random”
• Using stratified language without actually sampling within strata
• Claiming statistical confidence from pure judgment samples
• Ignoring MSA when measurement uncertainty dominates the tolerance
• Believing c = 0 with tiny n guarantees zero defects in the lot

Link forward

/study-guides/cqaFree exam prep with practice questions & AI tutor