Measurement Uncertainty

What Is Measurement Uncertainty?

Measurement uncertainty is the parameter that characterizes the range of values that can reasonably be attributed to a measurand, given incomplete knowledge of the measurement process and its influencing factors. It is not an error in the traditional sense but a statement of doubt: no measurement result is exact, and uncertainty quantifies how wide that doubt is. The concept is formalized by the Joint Committee for Guides in Metrology (JCGM) in the Guide to the Expression of Uncertainty in Measurement (GUM, JCGM 100:2008), which is the internationally accepted standard for uncertainty evaluation in calibration, testing, and research. Measurement uncertainty draws on probability theory, statistics, and physical modeling, and its proper evaluation is required for accreditation under ISO/IEC 17025.

Type A and Type B Evaluation

The GUM classifies uncertainty evaluation into two complementary methods. Type A evaluation applies statistical analysis to a series of repeated observations: the standard deviation of the measurement results estimates the random variability, and the standard uncertainty assigned to that component is the standard error of the mean. Type A methods are most informative when sufficient replicate measurements are available and when the measurement conditions can be held sufficiently stable. Type B evaluation covers all other ways of estimating uncertainty components, drawing on calibration certificates, manufacturer specifications, physical reasoning, and published data. A pressure transducer's drift specification from its calibration report, an ambient temperature effect derived from its datasheet, and a resolution limit read from a digital display all contribute Type B components. The NIST measurement uncertainty program provides guidance for applying both evaluation types and for combining them into a combined standard uncertainty through the law of propagation of uncertainty.

Uncertainty Propagation and Budgets

When a measurand is derived from several directly measured input quantities through a mathematical model, the uncertainties of the inputs propagate into the final result. The GUM's sensitivity coefficient approach uses partial derivatives of the model function to weight each input's uncertainty contribution, and the individual contributions are combined in quadrature under the assumption of statistical independence. An uncertainty budget documents each source of uncertainty, its standard value, its sensitivity coefficient, and its contribution to the combined standard uncertainty, giving practitioners a clear picture of which inputs dominate and where improvement effort should be directed. Expanded uncertainty, obtained by multiplying the combined standard uncertainty by a coverage factor k (typically 2, corresponding to approximately 95 percent confidence for normal distributions), is the form reported on calibration certificates and in test reports.

Measurement Uncertainty in Intelligent and Adaptive Systems

As machine learning and adaptive signal processing methods enter measurement systems, quantifying uncertainty becomes more complex. Intelligent sensors, adaptive filters, and data-driven models introduce both epistemic uncertainty (arising from limited training data or model misspecification) and aleatoric uncertainty (arising from irreducible sensor noise). Research surveyed in Machine Learning in Measurement Part 2: Uncertainty Quantification, published in IEEE Transactions on Instrumentation and Measurement, identifies Bayesian inference, Gaussian process regression, and conformal prediction as the main methods for attaching statistically rigorous uncertainty bounds to the outputs of learning-based measurement systems. Extending GUM-style uncertainty budgets to models with thousands of parameters remains an open area, and standards bodies are actively developing supplements to the GUM to address these cases.

Applications

Measurement uncertainty has applications in a wide range of disciplines, including: