Use of uncertainty in compliance

Significance of the topic

The correct use of measurement uncertainty when assessing compliance against specifications or regulatory limits is essential for reliable decisions in testing, quality assurance and enforcement. Accounting for uncertainty avoids incorrect acceptance or rejection of batches, products or samples and clarifies decisions when measured values are close to limit values. The Eurachem/CITAC guidance provides a principled decision framework based on quantified risk levels and explicit decision rules that introduce guard bands to separate acceptance and rejection zones.

Objectives and overview of the guidance

The guidance explains how to use uncertainty information to decide whether a measured value complies with an upper or lower specification limit, or both. It defines decision rules, acceptance and rejection zones, guard bands and decision limits. The aim is to provide a transparent method that links the laboratory measurement (including its uncertainty) to a target confidence about correct acceptance or correct rejection of a specification or regulatory limit.

Methodology and decision rules

Key concepts and steps:

• Define the measurand clearly and state the relevant specification limits (upper, lower or both).
• Specify a decision rule that quantifies acceptable probabilities for correct acceptance or correct rejection (for example, choose a confidence level or acceptable false acceptance/rejection rates).
• Estimate the measurement result and its uncertainty, in particular the uncertainty appropriate for a measured value at the specification limit(s).
• Calculate a guard band g that modifies the specification to create an acceptance zone and a rejection zone; the decision limit is the boundary between these zones.

Decision rule types:

• High confidence of correct acceptance: choose g so that if the measured value lies in the acceptance zone the probability that the true measurand lies within the specification is at or above a chosen confidence (e.g., 95%).
• High confidence of correct rejection: choose g so that if the measured value lies in the rejection zone the probability that the true measurand lies outside the specification is at or above a chosen confidence.
• Simple acceptance: set g = 0 (accept if measured value lies within the nominal specification limits); this is less conservative and may give higher risk of incorrect decisions near the limit.

Statistical foundations:

• Guard bands are derived from quantiles of the assumed distribution of the measurement error (normal, lognormal, etc.).
• For one-tailed 95% confidence with a normal approximation the z-quantile 1.64 is commonly used to compute g = z · u (one-sided), where u is the standard uncertainty relevant at the limit.
• The assumed distribution matters, especially for large relative uncertainties—lognormal assumptions may be appropriate for concentrations and must be considered explicitly.

Information required for compliance assessment

To perform an uncertainty-informed compliance decision you need:

• The measurand definition.
• Specification limits (upper and/or lower).
• The chosen decision rule and target confidence/probability levels (e.g., 95% for correct acceptance).
• The measured value.
• An estimate of measurement uncertainty appropriate for the value at the limit(s) (including sampling and analytical components if relevant).

Main results and discussion (illustrative examples)

Example 1 — focus on correct acceptance (nickel in steel):

• Measurand: mass fraction of Ni in a batch of steel; combined expanded uncertainty U = 0.2 % Ni with k = 2 (standard uncertainty u = 0.1 % Ni).
• Specification: 16.0 % to 18.0 % Ni.
• Decision rule: choose guard band for high confidence of correct acceptance at approximately 95% (one-tailed).
• Guard band: g = 1.64 · u ≈ 0.17 % Ni (rounded up to 0.17 % for safety).
• Acceptance zone: 16.2 % to 17.8 % Ni (rounded to one decimal place).
• Measured value 16.1 % Ni falls below the lower acceptance limit and lies in the rejection zone → conclude non-compliance under this decision rule. Under simple acceptance (g = 0) the same measured value would be compliant.

Example 2 — focus on correct rejection (banned substance):

• Measurand: concentration of a banned substance; relative standard uncertainty u_rel = 35% (high relative uncertainty).
• Upper limit: 2 ng/g.
• Decision rule: require that probability the true value exceeds the limit is ≥ 95% to declare non-compliance (high confidence of correct rejection).
• Guard band: using a lognormal model (appropriate when relative uncertainty is large) gives g ≈ 1.6 ng/g, so the acceptance limit becomes ~3.6 ng/g.
• Measured value 3.3 ng/g is below 3.6 ng/g → lies in the acceptance zone and the sample is compliant under the lognormal-based decision rule. If a normal model were used instead, the acceptance limit would be ~3.2 ng/g and the sample would be non-compliant; this highlights strong sensitivity to distributional assumptions.

Discussion points:

• Choice of decision rule changes conclusions near limits; laboratories and regulators must agree on acceptable risk levels (false acceptance/rejection probabilities) and the decision rule to be applied.
• Accurate estimation of uncertainty at the limit (including sampling) is critical; underestimating uncertainty can lead to excessive rejections, overestimating leads to excessive acceptances.
• Assumed error/distribution models (normal vs lognormal) can materially affect guard bands when uncertainty is large; justify model choice and consider transformations where appropriate.

Benefits and practical applications

Applying uncertainty-informed decision rules provides:

• Transparent, reproducible criteria for compliance decisions that explicitly balance risks of false acceptance and false rejection.
• Consistent handling of borderline cases where measured values overlap specification limits due to uncertainty.
• Guidance for setting operational policies (e.g., whether to use conservative guard bands or simple acceptance) for laboratories, producers, customers and regulators.

Practical uses include batch release testing, environmental monitoring, forensic thresholds, chemical contaminants enforcement and any QA/QC process requiring binary compliance decisions based on analytical results.

Future trends and potential applications

Expected developments and opportunities:

• Wider adoption of standardized decision-rule frameworks across sectors and regulatory bodies to harmonize compliance outcomes.
• Improved uncertainty models that integrate sampling variability, matrix effects and measurement method heteroscedasticity, often supported by Bayesian or mixed-effects approaches.
• Greater use of distribution-aware approaches (lognormal, beta, etc.) and simulation (Monte Carlo) to derive guard bands when analytical uncertainties are large or non-normal.
• Implementation of decision-rule logic in laboratory information management systems to automate compliance notifications and documentation while preserving auditability.
• Stakeholder-driven choices on acceptable risk levels and harmonized reporting templates that clearly state the decision rule, confidence levels and uncertainty estimates used for compliance calls.

Conclusion

Incorporating measurement uncertainty into compliance assessment is essential to make defensible, risk-aware decisions. The guard-band approach from the Eurachem/CITAC guidance provides a clear mechanism to separate acceptance and rejection zones based on chosen confidence levels and the uncertainty of measurement. Key practical requirements are an agreed decision rule, reliable uncertainty estimates at specification limits and an explicit statement of distributional assumptions. Implementing these principles improves transparency and consistency in compliance decisions across laboratories and regulatory contexts.

References

1. A. Williams and B. Magnusson (eds.) Eurachem/CITAC Guide: Use of uncertainty information in compliance assessment, 2nd edition, 2021.