Measurement Uncertainty is a non-negative parameter characterising the dispersion of quantity values that could reasonably be attributed to a measurement result. It expresses the limited knowledge about the value of the measurand and acknowledges that a measurement result is not an exact representation of an underlying quantity. Uncertainty is a fundamental concept in modern metrology and is distinct from the older, broader notion of "error".
Error is the difference between a measured quantity value and a reference quantity value, whereas uncertainty describes incomplete knowledge about the quantity value and the range of values reasonably associated with the measurement result. Uncertainty may be expressed as a standard uncertainty, combined standard uncertainty, expanded uncertainty, or another appropriate measure depending on the measurement model and reporting convention.
Measurement uncertainty arises from many sources. These can include instrument resolution, calibration uncertainty, environmental conditions, repeatability, reproducibility, sample variability, operator effects, reference standards, incomplete knowledge of physical constants, approximations in the measurement model, and uncertainty in input quantities. Some components can be evaluated statistically from repeated observations, while others are evaluated using information such as calibration certificates, previous measurements, manufacturer specifications, scientific literature, or expert judgement.
The traditional distinction between Type A and Type B evaluation describes whether a component is evaluated by statistical analysis of repeated observations or by other information. Both types are legitimate components of uncertainty evaluation, and the distinction concerns the method of evaluation rather than whether the source is intrinsically random or systematic.
A measurement result is often represented as an estimate together with an uncertainty, for example x ± U, where x is the measured value and U is an uncertainty quantity under a stated coverage convention. A standard uncertainty is commonly associated with a standard deviation-like quantity, while expanded uncertainty is obtained by multiplying a combined standard uncertainty by a coverage factor.
A coverage factor of approximately 2 is frequently used to obtain an interval corresponding to approximately 95% coverage under suitable assumptions, but the relationship is not universally exact and depends on the probability distribution, degrees of freedom, and interpretation of the coverage interval. Uncertainty should therefore not be described casually as a guaranteed error bound. It quantifies a stated level of uncertainty under a specified evaluation framework.
The evaluation of measurement uncertainty often begins by defining a measurement model that relates the measurand to input quantities. If a measurand y is calculated from input quantities x₁, x₂, and other variables, uncertainty in those inputs propagates through the model. For approximately linear models with suitably characterised uncertainties, the law of propagation of uncertainty combines the variances and covariance terms associated with the input quantities. Sensitivity coefficients describe how strongly the output responds to changes in each input. When the relationship is strongly nonlinear, probability distributions are non-normal, or correlations are important, more advanced approaches such as Monte Carlo methods may be appropriate. The uncertainty of a final result therefore depends not only on instrument specifications but on the entire measurement model.
Uncertainty differs from accuracy and precision. Accuracy is a qualitative concept concerning closeness of agreement with a reference value and is not normally expressed as a numerical quantity in the same way as uncertainty. Precision describes the closeness of agreement between repeated measurements under specified conditions. Uncertainty quantifies the dispersion associated with the measurement result and can include contributions from both repeated-observation variability and other sources of incomplete knowledge.
A highly precise measurement can have substantial uncertainty if systematic components have not been well characterised, while a measurement with relatively broad repeatability can still have a well-evaluated uncertainty budget. These distinctions are essential because reducing random variation alone does not necessarily reduce the total measurement uncertainty.
An uncertainty budget is a structured accounting of the uncertainty contributions associated with a measurement. It commonly lists input quantities, their estimated values, probability distributions or evaluation methods, standard uncertainties, sensitivity coefficients, and resulting contributions to the combined uncertainty. Correlations between inputs may need to be included because treating correlated quantities as independent can produce an incorrect result.
The largest contribution is not necessarily the easiest one to reduce, so uncertainty budgets can help laboratories identify which aspects of a measurement procedure warrant improvement. Calibration, environmental control, improved instrumentation, better sampling, refined models, or additional observations may reduce particular components. The budget also provides a transparent basis for communicating how the reported uncertainty was obtained.
Measurement uncertainty is essential for comparing measurements. If two laboratories report different numerical values, the difference cannot be interpreted properly without considering their associated uncertainties. Overlapping uncertainty intervals do not automatically prove that measurements agree, and non-overlapping intervals do not automatically prove incompatibility, because formal comparison depends on the measurement model and uncertainty evaluation. In metrology, concepts such as measurement compatibility, reference values, and degrees of equivalence provide more rigorous frameworks for comparison. Uncertainty is also important in calibration, conformity assessment, manufacturing tolerances, environmental monitoring, clinical measurements, and scientific experiments because decisions often depend on whether a measured value is sufficiently close to a specification or reference.
Measurement uncertainty does not mean that scientific measurements are unreliable or that nothing can be known precisely. It quantifies the limitations of knowledge at a particular level of measurement capability. Unit definitions may be exact within the SI framework, while realisations and practical measurements have uncertainty. Reducing uncertainty requires understanding which sources dominate and improving the corresponding measurement process; simply reporting more decimal places does not reduce uncertainty.
Proper reporting normally rounds uncertainty appropriately and reports the measured value to a compatible decimal position. Measurement uncertainty is therefore a central part of quantitative scientific communication: it allows numerical results to express both an estimate of the measurand and the degree of knowledge supported by the measurement process.
