1. Define the quantity being measured
The measurand is the quantity intended to be measured. Define it precisely: a component's length at a reference temperature may differ from its shop-floor length. Measurement location, contact force, orientation and procedure can also matter.
Measurement uncertainty is a non-negative parameter describing the dispersion of values attributed to the measurand, based on the information used. It is not merely the spread of readings. If the definition is ambiguous, an otherwise careful uncertainty calculation may describe a different quantity from the one your customer needs.
2. Separate uncertainty, error and tolerance
Measurement error is a measured value minus a reference value and can have either sign. Uncertainty does not identify this difference or its direction. Since reference values are not always known exactly, error is generally not fully known.
Tolerance defines the permitted range established by a product or process specification. It is not a property of the measurement method. Uncertainty describes the quality of measurement information; tolerance describes a requirement. Conformity assessment connects them through an agreed decision rule, rather than automatically comparing the reading alone with a limit.
3. Look beyond repeatability
Repeatability describes the spread of results under specified conditions. Identical readings do not establish zero uncertainty. Calibration uncertainty, finite resolution, temperature knowledge, component deformation or alignment may still contribute. Connect each proposed source to the actual measurement process rather than adding a generic list of instrument specifications.
Correct recognized significant systematic effects where practicable. The budget contains the uncertainty of a correction, not simply the correction's absolute value. Increasing the final uncertainty does not automatically justify ignoring a known correction. Check whether calibration, observed variation and other entries already include effects that you are about to count again.
4. Build a model and express standard uncertainties
A model relates the result y to input quantities, each assigned an estimate and standard uncertainty u, expressed as a standard deviation. Type A uses statistical analysis; Type B uses other information. Neither classification labels random or systematic effects.
The combined standard uncertainty u꜀ propagates input uncertainties through the model. Sensitivity coefficients describe how changes in each input affect the result. For independent inputs, add squared contributions rather than uncertainties themselves. Correlated inputs require covariance terms; using the same reference standard for several inputs can introduce such dependence.
5. Interpret expanded uncertainty carefully
Expanded uncertainty U is the combined standard uncertainty multiplied by a coverage factor k. The interval y ± U needs an intended coverage probability and an explanation of how k was obtained. An interval's width alone does not tell the reader how to interpret the measurement result.
k = 2 does not universally guarantee 95% coverage. It is a common approximation for sufficiently normal output distributions and reliably estimated standard uncertainty. Limited observations may require effective degrees of freedom; unusual distributions may need a different interval calculation.
6. Report a result that others can use
Report the result, its unit and uncertainty, identifying whether the uncertainty is standard or expanded. For U, include k and the intended coverage probability, with the basis for that interpretation. Describe the measurand and reference conditions, particularly when the result has been corrected to a reference temperature or another defined state.
Round at the end and align the result's decimal place with its uncertainty. Lower uncertainty helps near tolerance limits but does not prove conformity. Documenting the result enables the recipient to apply the appropriate decision rule.
Remember this
Uncertainty describes knowledge about a result, not its error or tolerance. Define the measurand, build the model, evaluate contributions and then report an interval with an explicit interpretation.
Sources and further reading
This guide explains fundamental GUM principles. Always adapt your model to the method and the information available about the measurement.
Practise with your own measurement.
Understanding comes with practice. Training or coaching connects theory with your own method and data.