Introduction to Acceptance Sampling

Acceptance sampling is a statistical quality control technique used to decide whether a lot of materials or products should be accepted or rejected based on the inspection of a sample. It is a fundamental tool in manufacturing, supply chain management, and regulatory compliance, particularly when testing is destructive, the lot is large, or 100% inspection is impractical. The two primary categories of acceptance sampling are attribute sampling and variable sampling. Each method offers distinct advantages and trade-offs in terms of information gathered, sample size required, cost, and risk discrimination. Understanding these differences is essential for quality engineers and managers to design effective inspection plans that balance risk, cost, and quality assurance.

What is Attribute Sampling?

Attribute sampling classifies each inspected unit as either conforming (acceptable) or non-conforming (defective) based on a binary criterion. The quality characteristic is not measured; only its presence or absence of a defined defect is recorded. For example, a visual inspection for scratches, a functional test (pass/fail), or a go/no-go gauge check are typical attribute inspections.

In attribute sampling, the number of defective items in the sample is counted. The lot is accepted if the number of defects is less than or equal to a specified acceptance number (c); otherwise, it is rejected. This method is straightforward to implement and requires minimal training for inspectors. It is widely used in industries such as electronics, textiles, and consumer goods where quick disposition decisions are needed.

Key Concepts in Attribute Sampling

  • Acceptable Quality Level (AQL): The maximum defect rate that is considered acceptable for the process. Typically, a lot with a defect rate at or below the AQL will be accepted with a high probability (e.g., 95%).
  • Lot Tolerance Percent Defective (LTPD) or Rejectable Quality Level (RQL): The defect rate that the consumer considers unacceptable. A lot at the LTPD will be rejected with a high probability (e.g., 90%).
  • Operating Characteristic (OC) Curve: A graph showing the probability of accepting the lot as a function of the true defect rate. The shape of the OC curve determines the discrimination power of the sampling plan.
  • Producer’s Risk (α): The probability of rejecting a good lot (defect rate at AQL). Typically set at 0.05 or 0.01.
  • Consumer’s Risk (β): The probability of accepting a bad lot (defect rate at LTPD). Typically set at 0.10 or 0.05.

For example, a simple single sampling plan for attributes specifies a sample size n and an acceptance number c. If a lot contains 5,000 items, a plan such as n=200, c=3 means: inspect 200 items; if 3 or fewer are defective, accept the lot; otherwise reject. The OC curve for this plan can be computed using the hypergeometric or binomial distribution.

What is Variable Sampling?

Variable sampling (also called variables sampling) involves measuring a continuous quality characteristic (e.g., length, weight, voltage, hardness) on each sampled unit. The measurements are then used to estimate the lot mean and standard deviation, and to calculate the percentage of non-conforming units based on specification limits. This method provides more information per unit inspected, allowing for smaller sample sizes than attribute sampling for the same level of discrimination.

Types of Variable Sampling Plans

Variable sampling plans are often categorized based on whether the process standard deviation is known or unknown, and whether single or double specification limits apply.

  • Known Sigma (σ known) plans: Used when the standard deviation is stable and well-estimated from prior data. Sample sizes can be smaller.
  • Unknown Sigma (σ estimated) plans: More common in practice; the sample standard deviation (s) is used. Requires a slightly larger sample to account for estimation error.
  • Plans with one or two specification limits: For example, a plan ensuring that no more than 1% of items exceed an upper limit (e.g., maximum voltage).

Form of the Decision Rule

In variable sampling, the lot is accepted if a computed statistic (such as the sample mean plus a factor times the standard deviation) is within a critical region. For a plan with an upper specification limit (USL) and known sigma, the acceptance criterion might be: x̄ + kσ ≤ USL, where k is a constant determined from the AQL and the sample size. The k method is one common approach. Another is the M method, which estimates the fraction nonconforming and compares it to a maximum allowable fraction.

Key Differences Between Attribute and Variable Sampling

The following table summarizes the major distinctions between the two approaches:

Factor Attribute Sampling Variable Sampling
Data Type Qualitative (pass/fail, go/no-go) Quantitative (measurements on a continuous scale)
Information per unit Low – only binary classification High – measures the degree of conformance
Sample size required Generally larger to achieve same discrimination Smaller (often 30–60% of attribute sample size)
Cost per inspection Lower (simple tests, minimal equipment) Higher (requires measurement devices, trained personnel)
Risk discrimination Steeper OC curves possible with large n Steeper OC curves achievable with smaller n
Applicability When quality is binary (e.g., leak/no leak) or multiple attributes When key characteristics are measurable (dimensions, purity, strength)
Destructive testing Works, but larger sample size increases cost Good because smaller sample size reduces product loss
Traceability to process Limited – only yields count of defects Excellent – measurements help identify shifts or trends
Statistical complexity Low – binomial or hypergeometric calculations Moderate to high – assumes normal distribution (or known sigma)

Operating Characteristic Curves

The OC curve is a powerful tool for comparing sampling plans. For attribute plans, the OC curve typically follows a binomial distribution shape. For variable plans, the OC curve is based on the normal distribution and is generally steeper for a given sample size, meaning that variable sampling can better discriminate between good and bad lots with fewer samples. For example, a variable plan with n=30 can achieve the same discrimination as an attribute plan with n=100, provided the normality assumption holds.

Statistical Background: OC Curves, AQL, and LTPD

The design of any acceptance sampling plan involves balancing the producer’s and consumer’s risks. The AQL and LTPD are two points on the OC curve. The plan is chosen such that:

  • At the AQL, the probability of acceptance is at least (1 – α), typically 0.95 or 0.99.
  • At the LTPD (or RQL), the probability of acceptance is no more than β, typically 0.10 or 0.05.

For attribute sampling, the sample size n and acceptance number c are determined from these two points using binomial tables or software. For variable sampling, the sample size n and the k-factor (or M-limit) are determined from the normal distribution parameters. A key advantage of variable sampling is that the same AQL/LTPD protection can be achieved with a sample size roughly 1/3 to 1/2 of that required for attribute sampling.

Industry Standards and Applications

Attribute Sampling Standards

The most widely used standard for attribute sampling is ANSI/ASQ Z1.4 (formerly MIL-STD-105E). This standard provides tables for sample size code letters and acceptance numbers for different inspection levels (normal, tightened, reduced). It is applicable for continuous production and batches. Other standards include ISO 2859-1, which is harmonized with ANSI/ASQ Z1.4.

Variable Sampling Standards

For variable sampling, ANSI/ASQ Z1.9 (formerly MIL-STD-414) is the prevalent standard. It provides plans for both known sigma and unknown sigma cases, and for one or two specification limits. The standard includes tables for sample size and k-factors. ISO 3951-1 is the international equivalent.

Many industries also develop their own internal standards or use modified versions of these plans. For example, the automotive industry (AIAG) recommends specific sampling procedures in the Production Part Approval Process (PPAP).

When to Use Attribute Sampling vs. Variable Sampling

Attribute Sampling is Preferred When:

  • The quality characteristic is binary (e.g., presence of a leak, color match, functional check).
  • Measurements are not available or are too costly.
  • Multiple different defect criteria are checked on the same sample (e.g., visual inspection for scratches, dents, and discoloration).
  • The process distribution is not normal or cannot be normalized.
  • Speed and simplicity are paramount, and trained measurement operators are scarce.

Variable Sampling is Preferred When:

  • The quality characteristic is measurable on a continuous scale (e.g., diameter, tensile strength, viscosity).
  • Sample size must be minimized (e.g., destructive testing, high-value items).
  • Better risk discrimination is needed; variable sampling provides a steeper OC curve and lower risk of misclassification.
  • Process control feedback is desired; measurements indicate not just acceptance but also process shifts.
  • Regulatory or customer requirements mandate variable sampling (e.g., pharmaceutical potency testing).

Practical Considerations and Limitations

While variable sampling offers efficiencies, it relies on the assumption that the quality characteristic follows a normal distribution (or can be transformed). If the distribution is heavily skewed or multimodal, the calculated fraction nonconforming may be inaccurate. In such cases, attribute sampling is more robust or nonparametric alternatives can be used.

Another consideration is the cost of measurement. If a variable test is expensive, the savings from a smaller sample may be offset. However, if the product cost is high, the smaller sample size can still result in overall cost savings from reduced testing.

For more information on acceptance sampling theory and standards, refer to the following external resources:

Conclusion

Attribute and variable acceptance sampling are complementary tools in the quality engineer’s arsenal. Attribute sampling is simple, widely applicable, and does not require numerical measurements. Variable sampling provides greater information per sample, allows for smaller sample sizes, and offers superior risk discrimination when the measured characteristic is normally distributed. The choice between them should be guided by the nature of the product, the cost and feasibility of measurements, the required statistical protection, and the need for process feedback. By understanding the strengths and limitations of each method, organizations can design cost-effective inspection plans that maintain high quality standards while minimizing waste and risk.