How to Build an ISO/IEC 17025 Measurement Uncertainty Budget

When an assessor asks for the budget

Accreditation assessments often raise a simple question: how did you arrive at the uncertainty attached to this result? A laboratory may have valid equipment, current methods and a traceable calibration chain, yet still struggle to show the arithmetic behind its reported uncertainty. This guide walks through preparing an ISO/IEC 17025 measurement uncertainty budget from the first definition of the measurand to the final report entry.

The approach follows the logic of the Guide to the Expression of Uncertainty in Measurement (GUM), which ISO/IEC 17025:2017 identifies in clause 7.6. You do not need a full metrology research program. You need a clear model, defensible input data and calculations that another technical reviewer can reproduce.

1. Define the measurand and the decision

Start by writing the measurand as a quantity, with its unit and the conditions under which it is obtained. “Customer temperature” is too broad. “Steel plate temperature at 20 °C, measured with a thermocouple at the designated monitoring point” gives the analyst something to model.

Next, record how the result will be used. A pass/fail specification check, a contractual tolerance or a screening decision changes the acceptable level of uncertainty and may require guard bands. ISO/IEC 17025 clause 7.8.6 requires the laboratory to report uncertainty when a decision rule is needed for conformity assessment. Clause 7.8.3 also addresses reporting results with measurement uncertainty.

Prepare three documents before calculating anything:

• A written measurand and decision-rule statement.
• The applicable test or calibration method, including its stated measurement uncertainty.
• A list of every input quantity, its unit, source and permitted range.

2. Gather input data and assign distributions

For each input, collect the best available estimate and the evidence that supports it. Type B evaluations use scientific judgment and prior data; Type A evaluations use repeated observations. Laboratory experience, manufacturer specifications, calibration certificates, environmental records and instrument resolution all qualify as legitimate evidence.

Assign a probability distribution that matches the information you actually have. Use a normal distribution for many repeated measurements, a rectangular distribution when only a bounded range is known, and a triangular distribution when a central estimate is more likely than either limit. Document the reasoning rather than copying a distribution from a template.

Do not confuse a calibration certificate’s stated uncertainty with the entire measurement uncertainty. The certificate covers the calibration laboratory’s contribution. Your method, sample handling, environmental effects and repeatability still need to be represented.

3. Build the model and quantify each contribution

Write the measurement model as an equation. For a straightforward calibration, the estimated value may be the indication plus a correction factor, with corrections for reference value, sensitivity and environmental effects. For a complex test method, the model may come from the method’s measurand relationship or a validated measurement model.

For each input, convert its uncertainty into a standard uncertainty. For a Type A evaluation, calculate the sample standard deviation of the observations and divide it by the square root of the number of observations. For a rectangular input with half-width (a), divide (a) by √3. For a normal input, use the standard uncertainty directly.

Sensitivity coefficients connect each input to the output. A dimensionless coefficient of 1 means a one-unit change in that input shifts the result by one unit. A coefficient of 0.01 means a 100-unit change produces a one-unit change in the result. Calculate each contribution as the product of the sensitivity coefficient and the standard uncertainty, then express it in the output unit.

Keep a table with columns for the input symbol, estimate, unit, uncertainty source, distribution, divisor, standard uncertainty, sensitivity coefficient and final contribution. This table is usually the most useful working record during an assessment.

4. Combine contributions and check the dominant terms

Add the individual variance contributions to obtain the combined standard uncertainty (u_c). The root-sum-of-squares calculation is appropriate when the inputs are treated as independent. Use a covariance term when a genuine, documented correlation exists, such as the same reference thermometer contributing to two inputs.

Review the largest contributors. If one term accounts for most of the budget, check whether its input data can be improved. A smaller calibration uncertainty may justify better temperature control, more representative sampling or additional repeats. Reallocating effort to a minor term rarely reduces the total result.

Keep enough significant figures in the working. A combined standard uncertainty of 0.8 kPa becomes 0.9 kPa after rounding, so an intermediate value of 0.79 kPa is not meaningful. Follow the rounding rules in your method or in ISO/IEC Guide 98-3, where the final reported uncertainty is normally rounded to two significant digits.

5. Apply the coverage factor and assign a statement of conformity

Multiply the combined standard uncertainty by a coverage factor (k) to obtain the expanded uncertainty (U). For many laboratory applications, (k = 2) corresponds to approximately 95 % coverage, provided the effective degrees of freedom are adequate. Use (k = 3) for approximately 99.7 % coverage when that level is required, or justify another factor.

State the result clearly. A typical entry reads: “(U = 0.42\ \text{kPa}), (k = 2), approximately 95 % level of confidence.” The wording should follow the terminology used in the applicable ILAC-G8 guideline and the laboratory’s own reporting procedure.

Where the result is compared against a specification, document the decision rule. ISO/IEC 17025 clause 7.8.6.1 requires the laboratory to document the decision rule applied when interpreting the result in relation to a specification limit, taking into account the level of risk. Guard banding, acceptance intervals and direct comparison each carry different assumptions, so choose one deliberately and apply it consistently.

6. Validate, approve and retain the record

Before releasing the budget, confirm that the calculation reproduces the reported uncertainty, that input sources are traceable to current records, and that the stated coverage factor is defensible. A second technical reviewer should be able to follow the logic from the input table to the final number.

Tie the budget to the method validation and calibration records. Clause 7.6.1 requires laboratories to identify and quantify the contributions to uncertainty using appropriate analysis. Clause 7.6.2 addresses the need to estimate uncertainty contributions from all significant sources, including those arising from the measurement process itself. Clause 7.6.3 requires the laboratory to ensure the uncertainty estimate is valid and up to date, and clause 7.6.4 requires it to be communicated in the required form.

Update the budget when the method changes, when new data alter a dominant input, or when proficiency testing shows a systematic offset. Record the trigger for each review so an assessor can see that the estimate is controlled rather than occasional.

Takeaway

An ISO/IEC 17025 measurement uncertainty budget is a working record, not a final spreadsheet attachment. Define the measurand, justify each input, combine contributions, select a coverage factor and document the decision rule. When an assessor asks how you know the reported uncertainty is right, the answer should sit in the budget, one reviewed line at a time.