Measurement uncertainty is a quantified estimate of the doubt about a measured value. It is the interval, written as value ± U, within which the true value is expected to lie at a stated confidence level. Every measurement has one, and a result without it is only half a result.
It matters because a number on its own hides how much you can trust it. A bore that reads 25.00 mm from a gauge with an uncertainty of ±0.01 mm is a different fact than the same reading from a gauge with ±0.05 mm of doubt, one comfortably clears a ±0.03 mm tolerance and the other cannot honestly say whether the part conforms at all. This guide covers what uncertainty is, why it is unavoidable, how Type A and Type B components combine, and how the coverage factor turns them into the figure you actually report.
What is measurement uncertainty?
Measurement uncertainty is a parameter that characterizes the spread of values that could reasonably be attributed to the thing being measured. It is not a mistake and it is not the reading being wrong, it is the honest width of the band around the reading. The governing reference is the Guide to the Expression of Uncertainty in Measurement, known as the GUM and published as JCGM 100:2008, which every accredited calibration laboratory in the world works to.
The key idea is that you report a value and an interval together. A well-formed result looks like 25.002 mm ± 0.008 mm (k = 2): the best estimate, the expanded uncertainty, and the coverage factor that tells the reader what confidence the interval carries. Drop any of the three and the reader cannot use the number to make a conforming-or-not decision.
Why does every measurement have uncertainty?
Because no instrument, operator, or environment is perfect, and the imperfections stack up. The gauge has finite resolution and its own calibration doubt. The part is not perfectly clean or perfectly at 20°C. The operator applies slightly different pressure each time. None of these are blunders, they are the residual, irreducible scatter that remains after you have done everything right. Uncertainty is the discipline of naming those contributions and adding them honestly instead of pretending they are zero.
This is also why uncertainty is different from a single instrument's stated accuracy. The manufacturer's accuracy spec is one input to the budget, not the whole answer. The reading also inherits doubt from the reference standard it was calibrated against, from temperature, from the fixture, and from repeatability on the day. A real uncertainty statement rolls all of them together, which is why two shops using the same model of gauge can honestly report different uncertainties.
What is the difference between Type A and Type B uncertainty?
The GUM sorts every contribution into one of two buckets by how you evaluated it not by what it is. Type A components are evaluated by the statistical analysis of repeated observations, you take a series of readings and compute the standard deviation, then divide by the square root of the number of readings to get the standard uncertainty of the mean. Type B components are evaluated from any other information: a calibration certificate, a manufacturer's specification, a handbook value, the gauge's resolution, or engineering judgment about the environment.
The split is about method, not importance. A resolution limit and a temperature effect are both Type B because you did not get them from repeated readings; the run-to-run scatter you measured this morning is Type A. Both end up as standard uncertainties in the same units, and once converted they are treated identically. Getting the Type B conversions right, turning a ± spec into a standard uncertainty by dividing by the right distribution factor, is where most budgets go wrong.
How do you combine uncertainties into expanded uncertainty?
You convert every source to a standard uncertainty, add them in root-sum-square, then scale by a coverage factor. The GUM lays out a repeatable procedure, and following it in order is what keeps a budget defensible:
- Define the measurand. State exactly what you are measuring and under what conditions, the bore diameter at 20°C, for instance. A fuzzy measurand makes every later step ambiguous.
- List the sources. Identify every contribution: repeatability, the reference standard's calibration, resolution, temperature, operator, fixture. Missing a source is the most common way a budget lies.
- Quantify each one. Get a magnitude for each source, Type A from the standard deviation of repeated readings, Type B from the certificate, spec, or resolution.
- Convert to standard uncertainties. Put each in the same units and reduce it to one standard deviation. A ± limit with a rectangular distribution is divided by √3; a certificate quoted at k = 2 is divided by 2.
- Combine in quadrature. Add the standard uncertainties in root-sum-square, the square root of the sum of the squares, to get the combined standard uncertainty, uc.
- Apply the coverage factor. Multiply uc by a coverage factor k to get the expanded uncertainty U = k × uc. Use k = 2 for roughly 95% confidence in most industrial work.
- Report the result. State the value, ± U, the coverage factor, and the confidence, for example 25.002 mm ± 0.008 mm (k = 2, ~95%).
| Source of uncertainty | Type | How it is evaluated |
|---|---|---|
| Repeatability of readings | A | Standard deviation of repeated measurements ÷ √n |
| Reference standard calibration | B | Certificate value ÷ its coverage factor |
| Gauge resolution | B | Half the resolution, rectangular, ÷ √3 |
| Temperature / thermal expansion | B | Estimated ± limit, rectangular, ÷ √3 |
| Combined standard uncertainty | Root-sum-square of the above | |
| Expanded uncertainty U | k × combined standard uncertainty |
By the numbers. The framework and its terms come from the GUM, published as JCGM 100:2008 by the Joint Committee for Guides in Metrology (BIPM/JCGM publications). Its rule for the coverage factor is that a factor of k = 2 gives an expanded uncertainty with a level of confidence of approximately 95% for a normal distribution, and in general k falls in the range 2 to 3 (GUM section 6, determining expanded uncertainty). k = 1 corresponds to about 68% and k = 3 to about 99.7%. NIST summarizes the same guidance for U.S. practice in its technical note on evaluating and expressing uncertainty (NIST Technical Note 1297).
How is uncertainty different from error and tolerance?
Error is the difference between a measured value and the true value, a single, usually unknowable, number. Uncertainty is the width of the doubt around your result, and it is what you can actually state. If you knew the error exactly you would just correct for it; you keep uncertainty precisely because you do not. Bias is the known, correctable part; uncertainty is what remains after correction.
Tolerance is different again: it is the allowed range for the product feature, set by design, not by the measurement. The two meet in the conformance decision. A useful rule of thumb is the test uncertainty ratio, the tolerance band should be several times wider than the measurement uncertainty, or you cannot reliably tell conforming parts from non-conforming ones. When uncertainty eats a large share of the tolerance, you are guessing, which is exactly the situation a measurement system analysis is designed to expose before it costs you a bad shipment.
Why does measurement uncertainty matter on the plant floor?
Because every conformance call, capability index, and control-chart point rests on it. If you accept a part that reads just inside tolerance but the uncertainty band straddles the limit, you may be shipping a defect; reject the mirror image and you may be scrapping a good part. Uncertainty is what tells you how much guard-band to leave. It is also inseparable from metrological traceability the GUM definition of traceability explicitly requires that each calibration in the chain contribute to the stated uncertainty, so a traceable result and an uncertainty statement are two views of the same thing.
Uncertainty also feeds directly into the studies a quality team runs. A gage R&R quantifies the repeatability and reproducibility slices of the budget, and a process capability study is only honest if the measurement doubt is small relative to the spread it is judging. The practical move is to keep the budget alive: capture repeatability, out-of-tolerance events, and re-measures at the point of inspection instead of on a clipboard, so a gauge whose uncertainty is quietly growing shows up as rising scatter during the shift. That live feedback is what Harmony gives a plant, and it is the difference the team at CLS saw when measurement data moved from next-morning paperwork to something visible while the line was still running. A well-run calibration program keeps the reference end of the budget solid; live floor data keeps the working end honest.