Use of uncertainty in compliance

Significance of the topic

The correct incorporation of measurement uncertainty into compliance decisions is essential in analytical chemistry, regulatory testing and quality assurance. Decisions about whether a result meets a specification or legal limit affect product release, environmental safety, human health and legal liability. Clear, reproducible decision rules that explicitly account for uncertainty reduce the risk of false acceptance or false rejection and provide defensible, transparent outcomes for laboratories and stakeholders.

Objectives and overview of the guide

The Eurachem/CITAC guidance summarized here explains how to assess compliance with a specification or regulation when measurement uncertainty is non-negligible. The guide defines decision zones (acceptance, rejection) and introduces the concept of a guard band that shifts the acceptance boundary to control the probability of making incorrect decisions. The document provides practical recommendations, example calculations and emphasises the need to state the decision rule and the associated confidence levels.

Methodology and decision framework

Key concepts and required inputs:

• Measurand definition: the quantity being assessed must be clearly specified.
• Specification limits: upper, lower or both limits must be stated.
• Decision rule: the laboratory must choose a rule that defines acceptable probabilities of false acceptance and false rejection (e.g., high confidence of correct acceptance or correct rejection).
• Measured value and its measurement uncertainty at or near the specification limit: uncertainty must include relevant components (sampling plus analytical, if applicable).

Guard band and decision limit:

• A guard band (g) moves the effective acceptance boundary away from the specification limit to provide a predefined confidence of correct acceptance or rejection.
• Acceptance zone: measured values inside this zone lead to a conclusion of compliance with the chosen confidence level.
• Rejection zone: measured values in this zone lead to a conclusion of non-compliance.
• Decision limit: the boundary between acceptance and rejection zones; the overlap between uncertainty around a measured value and the specification is handled by the guard band.
• Simple acceptance corresponds to g = 0 (no guard band) and is acceptable if the decision rule permits it.

Statistical considerations:

• Choice of distribution (normal, lognormal, etc.) matters—especially for high relative uncertainties—and affects guard band calculation and the acceptance limit.
• Confidence levels and one-/two-tailed quantiles determine the multiplier used to compute the guard band from the standard uncertainty.

Examples and key results

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

• Measurand: mass fraction of Ni in a batch.
• Uncertainty: U = 0.2 % Ni (k = 2, ≈95 %), so u = 0.1 % Ni (includes sampling and analytical uncertainty).
• Specification: 16.0 % to 18.0 % Ni.
• Decision rule: require high confidence of correct acceptance (≈95 % one-sided).
• Guard band: multiplier 1.64 (one-tailed 95 % quantile) → g = 1.64·u ≈ 0.17 % (rounded up to 0.17 % for safety).
• Acceptance zone: 16.2 % to 17.8 % (rounded to one decimal place).
• Outcome: measured 16.1 % Ni lies below the lower acceptance limit (16.2 %) and is therefore in the rejection zone — batch deemed non-compliant under this decision rule. Under simple acceptance (g = 0) the batch would have been compliant.

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

• Measurand: concentration of a banned substance.
• Uncertainty: high relative standard uncertainty u_rel = 35 %.
• Specification: upper limit 2 ng/g.
• Decision rule: require 95 % probability that the measurand exceeds the limit to conclude non-compliance (high confidence of correct rejection).
• Guard band: for the assumed lognormal distribution g ≈ 1.6 ng/g, giving an acceptance limit of 3.6 ng/g.
• Outcome: measured value 3.3 ng/g is below the acceptance limit → sample compliant. Note that assuming a normal distribution would change the acceptance limit (≈3.2 ng/g) and could change the compliance decision; this highlights the sensitivity to distributional assumptions.

Discussion — practical implications

Practical points for laboratories and regulators:

• Decision rules must be predefined, documented and appropriate for the risk profile of the application (e.g., consumer safety vs. commercial release).
• Uncertainty budgets should include all relevant sources (sampling, preparation, instrument) when batch-level or lot-level decisions are made.
• The choice between prioritising correct acceptance or correct rejection depends on consequences: preventing release of non-compliant material versus avoiding unnecessary rejection of acceptable material.
• Distributional assumptions must be justified; for results with large relative uncertainty, lognormal models or other non-normal assumptions may be more appropriate.

Benefits and practical applications

Using uncertainty-informed decision rules offers:

• Transparent and reproducible compliance decisions that can be defended to regulators and customers.
• Reduced legal and commercial risk by explicitly managing probabilities of false acceptance/rejection.
• A framework adaptable to product specification testing, environmental monitoring, forensic thresholds and regulated contaminants.

Future trends and potential uses

Anticipated developments and opportunities include:

• Integration of guard-band decision rules into laboratory information management systems for automated compliance reporting.
• Harmonisation of decision-rule practices across regulatory frameworks and international standards.
• Broader adoption of Bayesian methods and probabilistic models to combine prior information with measurement results for more nuanced decision making.
• Improved uncertainty propagation for complex sampling and multi-stage analytical workflows, including Monte Carlo approaches.
• Development of sector-specific guidance (food, pharmaceuticals, environmental) that formalises acceptable confidence levels and decision priorities.

Conclusion

Incorporating measurement uncertainty into compliance assessment via guard bands and explicit decision rules provides a principled approach to balance the risks of false acceptance and false rejection. Laboratories should document their decision rules, include all relevant uncertainty components (notably sampling when applicable), and justify distributional assumptions. Consistent application enhances confidence in regulatory and commercial decisions and reduces downstream disputes.

References

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