Measurement Uncertainty:
Why Every Number Needs an Error Bar

The Number Is Not the Measurement

Picture two calibration laboratories, each measuring the same precision gauge block. Both report a measured length of 25.0000 mm. A superficial reading suggests identical results. But the first laboratory’s certificate reads 25.0000 mm ± 0.0002 mm (k = 2), while the second reads 25.0000 mm ± 0.0500 mm (k = 2). These two results are not the same — they are not even close to the same. The first tells you the true value almost certainly lies within a range of 0.4 micrometers. The second says it could be anywhere within a one-tenth-millimeter band. The difference between them is measurement uncertainty, and it changes everything about how those results can be used.

This example illustrates a principle that is easy to state and surprisingly hard to internalize: a number without its associated uncertainty is not a complete measurement result. It is a reading. It might be useful. It might not. You can’t tell without the uncertainty.

What Measurement Uncertainty Actually Means

The VIM 2.26 defines measurement uncertainty as a “non-negative parameter characterizing the dispersion of the quantity values being attributed to a measurand, based on the information used.” Translated into working language: uncertainty is a quantified range within which the true value of what you measured is expected to lie, at a stated level of confidence.

Uncertainty is not the same as error. Error is the difference between a measured value and the true value — and because we can never know the true value exactly, we can never know the error exactly either. Uncertainty is what we can know: a statistically or analytically derived statement of how much doubt surrounds our measurement result, based on everything we understand about the measurement process.

This distinction matters operationally. When an instrument reads 100.03 and the reference is 100.00, the error is +0.03. But the uncertainty of that comparison — which accounts for the reference standard’s own uncertainty, the instrument’s resolution, environmental conditions, and operator repeatability — might be ±0.05. If that’s the case, the apparent error of 0.03 is not even statistically distinguishable from zero. The measurement uncertainty shapes what conclusions you are scientifically permitted to draw.

The GUM Framework: A Universal Method

The internationally recognized methodology for evaluating and expressing measurement uncertainty is the NIST GUM — formally, JCGM 100:2008, Evaluation of measurement data — Guide to the expression of uncertainty in measurement, adopted in the United States through NIST Technical Note 1297. ISO/IEC 17025:2017 Clause 7.6 mandates its application in accredited calibration laboratories. The ASQ CCT Body of Knowledge tests it. Every serious metrology practitioner must understand it.

The GUM framework rests on a clear organizing principle: identify every source of uncertainty that contributes meaningfully to the total uncertainty of your measurement, evaluate each one, combine them mathematically, and express the result with a stated coverage factor and confidence level.

Type A and Type B: Two Paths to the Same Place

The GUM divides uncertainty evaluations into two categories based on the method of evaluation — not the source or importance of the uncertainty.

Both types carry equal mathematical weight in the uncertainty budget. The GUM does not privilege one over the other. A well-characterized Type B contribution from a NIST-traceable reference standard can be better known — and contribute less uncertainty — than a Type A evaluation based on only three repeat measurements taken under highly variable conditions.

Building the Uncertainty Budget

The uncertainty budget is the structured document — or calculation — that captures every significant uncertainty source, assigns a standard uncertainty to each, applies sensitivity coefficients where needed, and combines them into a single combined standard uncertainty. The process follows a defined mathematical procedure.

1. Define the measurement model. Identify the equation that relates your input quantities to the output measurand. For a simple direct comparison calibration, this may be straightforward. For a derived quantity like pressure calculated from force and area, the model includes multiple input variables.
2. Identify all significant uncertainty sources. Systematically consider: resolution of the instrument under test, uncertainty of the reference standard (from its calibration certificate), measurement repeatability, reproducibility across operators or environmental conditions, environmental factors (temperature coefficients, humidity effects, vibration), and equipment drift since last calibration.
3. Assign a probability distribution to each source. Repeated measurements follow a normal distribution. Resolution uncertainty follows a rectangular distribution (equal probability anywhere within the range). Systematic shifts often follow a rectangular or U-shaped distribution. The assumed distribution affects how you convert limits to standard uncertainties.
4. Calculate the standard uncertainty for each source. For Type A: standard deviation of the mean. For Type B normal distributions: divide the stated expanded uncertainty by k. For rectangular distributions: divide the half-width by √3.
5. Apply sensitivity coefficients. For each input quantity xᵢ, the sensitivity coefficient cᵢ = ∂y/∂xᵢ describes how strongly that quantity influences the output. For direct measurements, cᵢ = 1. For derived quantities, compute the partial derivatives of the measurement model.
6. Combine to get combined standard uncertainty. Use the law of propagation of uncertainty (assuming uncorrelated inputs).
7. Multiply by coverage factor k to get expanded uncertainty. Report U with the coverage factor and confidence level.

A Worked Example: Calibrating a Digital Micrometer

Consider calibrating a 0–25 mm digital micrometer using a NIST-traceable gauge block as the reference standard. The uncertainty budget might include the following contributions:

Combining these in quadrature (root sum of squares, with cᵢ = 1 for each source): u_c ≈ 0.000314 mm. Applying k = 2 yields an expanded uncertainty U ≈ 0.00063 mm, which would be rounded and reported as U = 0.00063 mm at k = 2 (approximately 95% confidence level). This is what appears on the calibration certificate — not the bare number, but the number with its full uncertainty context.

Coverage Factor, Confidence Level, and the Normal Distribution

The coverage factor k is the multiplier applied to the combined standard uncertainty to obtain the expanded uncertainty at a stated confidence level. For a normal (Gaussian) probability distribution:

A common error on calibration certificates is stating an expanded uncertainty with a coverage factor but not specifying the confidence level — or stating a confidence level without specifying the assumed probability distribution. ISO/IEC 17025:2017 requires all three: the uncertainty value, the coverage factor, and the confidence level.

Uncertainty and the Test Uncertainty Ratio (TUR)

Measurement uncertainty does not exist in isolation — it directly affects whether a calibration is technically valid. The Test Uncertainty Ratio (TUR) is the ratio of the tolerance of the item being calibrated to the expanded uncertainty of the calibration measurement process:

ANSI/NCSL Z540.3 requires a TUR of at least 4:1 unless an alternative decision rule with documented false accept risk is applied. When TUR falls below 4:1, guard banding — narrowing the acceptance zone by the amount of measurement uncertainty — is required to maintain the false accept risk at or below 2%. Stating “calibrated” on a certificate without knowing or documenting the TUR is a technical deficiency that many accreditation assessors specifically look for.

Reporting Uncertainty on Calibration Certificates

ISO/IEC 17025:2017 Clause 7.8.4.1 establishes specific requirements for what calibration certificates must contain. The measurement uncertainty requirements are unambiguous: calibration results must include “the measurement uncertainty of the measurement result in the same unit as the measurand or in a term relative to the measurand (e.g. percent)” and must identify the coverage factor and coverage probability.

Correct certificate language looks like this: “The expanded uncertainty of measurement is ± 0.00063 mm at a coverage factor of k = 2, corresponding to a coverage probability of approximately 95%, assuming a normal distribution.”

The most frequent deficiencies found in calibration certificate audits include: omitting the coverage factor, failing to state the confidence level, providing uncertainty only for the reference standard (not for the entire measurement process), and using inconsistent significant figures in the uncertainty statement relative to the measurement result.

Why This Matters Beyond Compliance

Measurement uncertainty is not a paperwork exercise. It is the quantitative answer to the question every measurement ultimately asks: how well do you actually know this?

Consider a pressure relief valve being calibrated with a set point tolerance of ±1.0 psi. If the calibration measurement uncertainty is ±0.8 psi (a TUR of only 1.25:1), a “pass” result near the tolerance edge could easily be a genuine failure masked by measurement noise — with potentially serious safety consequences. The uncertainty tells you how much trust to place in the measurement, and therefore, how much trust to place in the decision made from it.

This is why ISO/IEC 17025, ANSI/NCSL Z540.3, and A2LA policies all require uncertainty to be evaluated and reported. It is why the GUM was developed and why it has been adopted internationally. The number is not the measurement. The measurement is the number — and its error bar.