Measurement uncertainty is a non negative parameter characterising the dispersion of quantity values that could reasonably be attributed to the quantity being measured, based on available information. It is a central concept in metrology, the science of measurement, and is used because a measurement result is not an exact representation of an unknowable physical quantity. Uncertainty can arise from instrument calibration, resolution, repeatability, environmental conditions, sample variation, reference standards, data processing, model assumptions and other relevant influences.
The internationally influential terminology of measurement uncertainty treats the uncertainty as a quantified expression of incomplete knowledge about the measurand rather than simply as a mistake. A measurement result is therefore often properly reported as an estimated value together with an uncertainty interval or uncertainty statement. The uncertainty communicates how strongly the available evidence constrains the quantity's possible values.
The quantity being measured is called the measurand. Defining the measurand precisely is essential because uncertainty depends on what exactly is being measured and under which conditions. For example, the temperature of a liquid depends on location, time, equilibration conditions and the operational definition of the measurement. A length measurement may depend on temperature because materials expand or contract. A concentration measurement may depend on sample preparation and the chemical method used.
If the measurand is poorly defined, uncertainty analysis can become incomplete because important sources of variation have not been specified. Measurement uncertainty therefore begins with a measurement model describing how the reported quantity is related to measured inputs and relevant influencing quantities. The model provides the framework for identifying, evaluating and combining uncertainty contributions.
Uncertainty evaluation commonly distinguishes Type A and Type B evaluation. Type A evaluation uses statistical analysis of repeated observations or other data amenable to statistical treatment. Type B evaluation uses information from sources such as calibration certificates, instrument specifications, reference data, previous measurements, published information and expert technical knowledge. The distinction concerns the method used to evaluate uncertainty rather than a claim that one type is intrinsically more reliable.
Type A uncertainty might arise from the standard deviation of repeated measurements, while Type B uncertainty might arise from a calibration certificate specifying a bounded or expanded uncertainty. Multiple contributions are combined according to the measurement model, taking account of sensitivity coefficients and correlations where relevant. The resulting combined standard uncertainty can then be expressed in a form suitable for the intended measurement application.
Expanded uncertainty is often reported when communicating measurement results. It is obtained by multiplying a combined standard uncertainty by a coverage factor, commonly denoted by k, chosen according to the desired coverage probability and statistical assumptions. A result may therefore be expressed in the form measured value plus or minus expanded uncertainty, together with the coverage factor or other information necessary for interpretation.
The numerical interval should not automatically be interpreted as an absolute guarantee that the true quantity lies inside it. Coverage depends on the probability model, distributions and assumptions used in the uncertainty evaluation. Different reporting conventions are appropriate for different applications. Laboratory accreditation, calibration services, scientific experiments and industrial measurements may use formal uncertainty budgets and specified reporting procedures.
An uncertainty budget identifies significant contributions to the uncertainty of a measurement result. For a dimensional measurement, contributions might include calibration uncertainty, repeatability, resolution, temperature correction and alignment. For a mass measurement, balance calibration, buoyancy correction, environmental effects and repeatability may matter. In analytical chemistry, uncertainty may involve weighing, dilution, calibration curves, recovery, repeatability and sample preparation. The importance of each contribution depends on the sensitivity of the final result to that input. A source with a large numerical uncertainty may have little influence if the measurement is insensitive to it, while a seemingly small input uncertainty can dominate if the measurement model strongly amplifies it. Sensitivity coefficients quantify these relationships in formal uncertainty propagation.
Measurement uncertainty is related to but distinct from accuracy, precision and error. Precision describes the closeness of agreement among repeated measurements, while accuracy describes closeness to a reference or accepted value. Error is the difference between a measured value and a reference or true value when that difference can be defined. Uncertainty expresses incomplete knowledge about the measurand and does not simply mean that the instrument is inaccurate.
A measurement can have small uncertainty but substantial bias if an important systematic effect has been omitted. Conversely, a measurement can be unbiased on average while having substantial uncertainty because random variation is large. Calibration can provide information about systematic effects and uncertainty simultaneously, but calibration does not make an instrument automatically perfect.
Reducing measurement uncertainty requires understanding which contributions dominate. Increasing the number of repeated observations can reduce uncertainty associated with random variation in an estimated mean, but it will not generally reduce uncertainty from an unrecognised systematic bias. Better calibration standards, more stable environmental conditions, improved instruments, more precise reference materials or better measurement models can reduce other contributions. Greater instrument resolution does not necessarily improve total measurement uncertainty if another source dominates. Similarly, collecting huge quantities of data does not compensate for a poorly defined measurand. Efficient uncertainty reduction therefore focuses resources on the dominant uncertainty components rather than attempting to improve every component equally.
Measurement uncertainty is fundamental to scientific comparison and decision making because measured values are rarely exact. Two laboratories may report different numerical results while remaining statistically and metrologically consistent when their uncertainty intervals overlap appropriately. Conversely, two measurements with similar central values may have very different evidential strength if their uncertainties differ substantially. Uncertainty allows scientists to determine whether observed differences are compatible with measurement limitations or provide evidence for a real difference. It also supports calibration, conformity assessment, engineering tolerances, environmental monitoring, clinical measurement and international comparison. The central purpose is not to express doubt vaguely but to provide a quantitative, traceable description of the information available about a measurement.
