ISO 17025 measurement uncertainty: how to build a budget you can defend

What does ISO/IEC 17025 ask about measurement uncertainty?

ISO 17025 measurement uncertainty work starts with one question: how far from the true value could this reported result reasonably be? The standard answers the "whether" in clause 7.6, Evaluation of measurement uncertainty, which sits among the process requirements next to method selection, technical records and ensuring the validity of results. It leaves the "how" to the laboratory.

Uncertainty also depends on other clauses, and an assessor will follow those threads:

• 6.5 Metrological traceability. The uncertainty stated on your reference standards' calibration certificates is usually the first line of a calibration budget.
• 7.2 Selection, verification and validation of methods. Verification and validation data supply repeatability, reproducibility and bias figures.
• 7.7 Ensuring the validity of results. Control charts and proficiency testing show whether the claimed uncertainty holds up in routine work.
• 7.8 Reporting of results. Test reports (7.8.3), calibration certificates (7.8.4) and statements of conformity (7.8.6) are where uncertainty reaches the customer.

The edition in force is ISO/IEC 17025:2017, confirmed in 2023 with no amendment. Method validation itself is covered in our method verification and validation guide; this page stays with the uncertainty that follows from it.

How to build an uncertainty budget, step by step

An uncertainty budget is a table that turns a list of doubts into one number with a stated confidence.

1. Define the measurand

State exactly what is being measured, under which conditions and by which method. "Temperature indicated by a digital thermometer at 100 °C in a stirred liquid bath" is a measurand; "thermometer accuracy" is not.

2. Write the measurement model

Express the result as an equation of its inputs, even a simple one such as error = indicated value minus reference value plus corrections. The model shows which inputs carry uncertainty and how strongly each one moves the result.

3. List the sources

Work through reference standards, their drift, the environment, the item's resolution, the operator, sampling and sample preparation, and repeatability. A cause-and-effect sketch drawn with the analysts helps.

4. Quantify each source

Some contributions are evaluated statistically from repeated observations (often called Type A); others come from certificates, specifications, manufacturer data or experience (Type B). Each is converted to a standard uncertainty by dividing by the factor that suits its distribution.

5. Combine and expand

Independent standard uncertainties, each multiplied by its sensitivity coefficient, are combined as the square root of the sum of squares. The combined figure is then multiplied by a coverage factor to give the expanded uncertainty that appears on the report.

6. Challenge the result

Compare the expanded uncertainty with the spread of your control chart and your proficiency testing performance. A budget smaller than the scatter you actually see has missed something.

A worked example: calibrating a digital thermometer (illustrative)

The figures below are illustrative only: a budget for comparing a digital thermometer with a reference thermometer in a liquid bath at one point, 100 °C. Your own values must come from your own evidence.

The bath and the resolution outweigh the reference thermometer, so a better reference would barely change the result. The repeatability term is small only because ten readings were averaged; if routine calibrations use three readings, the budget must say three.

Uncertainty in testing laboratories

Chemical, microbiological and materials testing rarely suits the neat component-by-component model above. Sample heterogeneity, matrix effects and analyst technique interact. Many testing laboratories therefore work "top down", building the estimate from data they already hold:

• within-laboratory reproducibility from long-term quality control results, covering different days, analysts and equipment;
• bias, estimated from certified reference materials, spiked samples or proficiency testing results;
• any significant source those data do not capture, such as sampling in the field or a sub-sampling step, added separately.

The approach is only as strong as the data behind it. With only a few months of control results, say so and revisit the estimate as data accumulate. Proficiency testing and interlaboratory comparison, which feed this evidence, have their own guide.

Uncertainty, decision rules and statements of conformity

When a measured value sits close to a limit, its uncertainty interval can straddle that limit, and the answer "pass" may be wrong. The 2017 edition dealt with this by introducing the decision rule: the laboratory agrees with the customer, during the review of requests, tenders and contracts (7.1), how uncertainty will be accounted for, and then reports statements of conformity (7.8.6) in line with that rule.

The usual choice is simple acceptance, comparing the value directly with the limit, or guarded acceptance, where a guard band derived from the uncertainty narrows the limit so a "pass" is less likely to be false. Published guidance on decision rules, such as ILAC G8, is widely used; now that ILAC's work has moved to Global ACI, confirm with your accreditation body which guidance document and edition it expects you to refer to.

Whatever rule is chosen, three things must be traceable in the record: the rule agreed, the customer's agreement, and the uncertainty used when the statement was made.

What the uncertainty record should contain

Assessors check whether the number on the report can be reconstructed and defended. A complete record for each method and range usually holds:

• the measurand, the method reference and the range the budget covers;
• the measurement model and the list of sources considered, including those judged negligible and why;
• the value, distribution, divisor and origin of each contribution, with the certificate, characterisation or data set it came from;
• the combined and expanded uncertainty, the coverage factor and the rounding applied when reporting;
• the author, the reviewer, the date, and the triggers for revision: a new reference standard, a changed method, new equipment, a move of premises or a poor proficiency testing result.

If the budget lives in a spreadsheet, treat that spreadsheet as part of your controlled information: locked formulas, a check calculation and a version history.

Mistakes that weaken an uncertainty budget

• Copying the certificate uncertainty as the whole answer. The reference standard's uncertainty is one input; the laboratory's own process adds to it.
• Double counting or dividing twice. An expanded uncertainty taken from a certificate must be divided by its own coverage factor before it enters the budget.
• One budget for every range. Resolution, drift and environment behave differently at the ends of a range.
• A budget nobody revisits. A change of reference, method or location that leaves the budget untouched is an easy finding.
• An uncertainty smaller than the evidence allows. A claim tighter than the control chart or proficiency testing results invites a technical assessor's questions at once.

For the wider list of what goes wrong at assessment, see our guide to the ten common mistakes.

ISO 17025 measurement uncertainty: questions buyers ask