Use of uncertainty in compliance

Importance of the topic

The correct use of measurement uncertainty in compliance assessment is fundamental for reliable decision-making in regulated and quality-controlled environments. Accounting for uncertainty prevents both false acceptance and false rejection of batches, products or materials, and ensures that decisions reflect the true level of risk acceptable to stakeholders. The Eurachem/CITAC approach provides a structured, risk-informed framework to convert measured values and their uncertainties into operational acceptance and rejection rules.

Objectives and overview of the guide

The guide aims to provide practical guidance on how to judge compliance with specification limits or regulatory thresholds when measurement uncertainty is non-negligible. It defines acceptance and rejection zones separated by a decision limit, introduces the concept of a guard band to control risk, and explains how to select decision rules that meet specified confidence (probability) criteria for correct acceptance or correct rejection. Two worked examples illustrate differing emphases: one on achieving high confidence of correct acceptance and another on high confidence of correct rejection.

Methodology

The method rests on these core elements:

• Define the measurand and the applicable upper and/or lower specification limits.
• Select a decision rule that states the desired confidence levels (probabilities) for correct acceptance and/or correct rejection.
• Estimate the measurement uncertainty for values at or near the limit(s). This uncertainty should include sampling and analytical components where relevant.
• Calculate a guard band g that adjusts the nominal specification limit to create an acceptance zone and a rejection zone. The guard band is chosen so that a measured value inside the acceptance zone implies a probability of correct acceptance greater than or equal to the chosen confidence level (e.g., 95%).
• Apply the decision rule: if the measured value falls inside the acceptance zone the item is accepted; if it falls in the rejection zone the item is rejected. Overlap between zones is resolved by the decision limit between them.

Key practical points:

• A guard band g = 0 corresponds to simple acceptance (no adjustment for uncertainty).
• The statistical model and distributional assumptions (normal, lognormal, etc.) used to derive g are critical and can change the compliance outcome.
• The chosen confidence level (often expressed as 1-α) must reflect stakeholder risk tolerance for false acceptance or false rejection.

Main results and discussion

The guide distinguishes clear-cut cases where the uncertainty interval is entirely on one side of the limit from ambiguous cases where the interval overlaps the limit. For ambiguous cases it prescribes a decision-rule approach using guard bands. Two illustrative examples demonstrate application:

• Nickel in steel (focus on correct acceptance): Measurand = mass fraction of Ni in a batch. Combined standard uncertainty u = 0.1 % Ni (k = 2 for 95 %). Specification 16.0–18.0 % Ni. Choosing high confidence of correct acceptance (one-tailed 95 %) leads to a guard band g ≈ 1.64 u ≈ 0.17 % Ni. Rounded acceptance zone becomes 16.2–17.8 % Ni. A measured value of 16.1 % Ni falls in the rejection zone → batch declared non-compliant. If simple acceptance were chosen (g = 0), the same measurement would be compliant.
• Banned substance (focus on correct rejection): Measurand = concentration of a banned substance with large relative uncertainty (u_rel = 35 %). Upper regulatory limit 2 ng/g. To achieve 95 % confidence of correct rejection, a guard band was calculated (assuming a lognormal distribution) giving an acceptance limit ≈ 3.6 ng/g; a measured value of 3.3 ng/g then lies in the acceptance zone and the sample is compliant. If a normal model had been assumed instead, the acceptance limit would be lower (≈ 3.2 ng/g) and the same measurement could be non-compliant. This underscores the importance of correct distributional assumptions when uncertainty is large.

Discussion highlights:

• Guard bands translate probabilistic requirements into operational limits and make trade-offs between false acceptance and false rejection explicit.
• The choice of confidence level and distribution model materially affects the guard band and the compliance outcome; justification should be documented.
• Uncertainty budgets must include sampling uncertainty when the item under test is a lot or batch, otherwise the guard band may be underestimated.

Benefits and practical applications

The approach provides laboratories, regulators and quality managers with a transparent, reproducible method to:

• Make defensible pass/fail decisions when measurements are close to specification limits.
• Align decision thresholds with acceptable legal or commercial risk levels.
• Document decision rules and uncertainty assumptions for audits and disputes.

Practical applications include industrial QC of raw materials and products, environmental and food contaminant monitoring, forensic and regulatory testing, and any context where non-compliance has significant consequences.

Future trends and possibilities for use

Expected developments and opportunities include:

• Wider adoption of harmonized decision-rule templates by regulators to reduce ambiguity across jurisdictions.
• Integration of Monte Carlo and Bayesian approaches to better handle complex uncertainty distributions and correlated components.
• Automation of guard-band calculation in laboratory information management systems to ensure consistent application of decision rules.
• Enhanced guidance on sampling uncertainty and on the choice of statistical distributions for non-ideal data (e.g., censored, highly skewed measurements).

Conclusion

Incorporating measurement uncertainty through explicit decision rules and guard bands yields more robust and transparent compliance decisions than simple acceptance alone. The Eurachem/CITAC guidance gives a practical framework for translating uncertainty into operational acceptance and rejection zones, but successful implementation requires careful uncertainty estimation, clear specification of confidence levels, and justified statistical assumptions.

Reference

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