Reading measurement uncertainty on a calibration certificate — coverage factor, decision rule and why 'within tolerance' isn't always self-evident

The guide on calibration of measuring equipment covers why traceable calibration at an ISO/IEC 17025-accredited laboratory is required, and at what interval it takes place. This article goes one level deeper into one specific part of the calibration certificate itself: the stated measurement uncertainty, the coverage factor used to calculate it, and the decision rule the laboratory applies to determine whether an instrument is declared "within tolerance".

Why every measurement carries an uncertainty

No measurement — not even one performed by a calibration laboratory itself — produces an exact value. Every reported measured value comes with a measurement uncertainty: a range around the measured value within which the true value lies with a certain degree of probability. This uncertainty is built up from several contributions (the accuracy of the reference standard, environmental influences, resolution of the measuring system, repeatability of the measurement) and is summarised by the laboratory into a single reported figure: the expanded measurement uncertainty.

The coverage factor: from standard uncertainty to a reported interval

The expanded measurement uncertainty stated on a certificate is obtained by multiplying the (statistical) standard uncertainty by a coverage factor (k). In calibration practice, k = 2 is very commonly used, which for a normal distribution corresponds to a confidence interval of roughly 95% (more precisely, about 95.45%). A certificate that reports a measurement uncertainty without stating the coverage factor used and the associated confidence level therefore gives an incomplete picture of what that number actually means.

Note: with a small number of measurements (a low number of degrees of freedom), it may be more correct to use a factor from the t-distribution instead of k=2 — a detail usually not visible to the reader of a certificate, but which the laboratory itself must weigh in accordance with the applicable metrological guidelines. For the exact calculation rules, refer to the standard and international metrological guidelines (VIM/GUM), not to this guide article.

The decision rule: how uncertainty factors into "pass/fail"

Since ISO/IEC 17025:2017, a laboratory issuing a statement of conformity (for example "within tolerance" or "meets specification") is required to apply a decision rule and state it on the certificate (§7.8.6.1 and §7.8.6.2). That rule defines how the reported measurement uncertainty is factored into determining whether a result is conforming: is a measured value already rejected once the tolerance limit plus the uncertainty is exceeded ("guard banding", a stricter assessment), or is only the bare measured value tested against the tolerance limit without accounting for the uncertainty (a simpler, less strict rule)? Both choices are legitimate, but they produce a different risk of an erroneous "pass" verdict, and must therefore be explicitly stated on the certificate rather than assumed implicitly.

Why this matters in practice for a value close to the limit

For the user of a calibrated instrument (for example an insulation resistance tester deployed during a NEN 3140 inspection), this is more than a theoretical matter: a measured value that falls just within the tolerance limit, but whose uncertainty band extends beyond that limit, carries a different level of confidence than a measured value that sits well within tolerance with a narrow uncertainty band. A certificate that only states "within tolerance: yes" without the underlying measured value, uncertainty and applied decision rule leaves this nuance invisible to whoever evaluates the certificate.

Practical relevance

When assessing a calibration certificate for measuring equipment deployed for NEN 3140 inspections (see also the guide on measuring instruments and CAT classification), it is worthwhile to look not only at the "pass/fail" conclusion, but also at the stated measured value, the reported measurement uncertainty with its coverage factor, and the applied decision rule — particularly when a measured value sits close to a tolerance limit and the outcome of an inspection depends on it.

Common mistakes

Reading only the "within tolerance" conclusion without examining the underlying measured value and measurement uncertainty, causing a borderline case to be treated as fully certain.
Ignoring the coverage factor and confidence level when interpreting a reported measurement uncertainty, while these determine how "wide" the uncertainty interval actually is.
Assuming every laboratory applies the same decision rule, while ISO/IEC 17025 allows laboratories room to choose their own, explicitly documented rule — comparing certificates from different laboratories without checking that rule can be misleading.
Confusing measurement uncertainty with the instrument's own tolerance: the tolerance (for example from NEN-EN-IEC 61557) is a requirement on the instrument, while the measurement uncertainty is a property of the calibration measurement itself — two different, but interacting, quantities.