Measurement Uncertainty Explained: What U=±0.5°C (k=2) Means and Why It Matters

Why uncertainty exists in every measurement

No measurement is perfect. Every time an instrument produces a number, that number is shaped by a web of influences: the instrument itself and its inherent limitations, the reference standard used to calibrate it, the ambient temperature and humidity in the room, the technique and judgement of the operator, the stability of the quantity being measured, and the resolution of the display. Each of these contributes a small — sometimes very small — amount of doubt to the final result.

The international framework for quantifying and communicating that doubt is the Guide to the Expression of Uncertainty in Measurement, known as the GUM (JCGM 100:2008), published by the Joint Committee for Guides in Metrology. The GUM is the foundation for how all ISO/IEC 17025 accredited laboratories calculate and report measurement uncertainty. It is not optional: accredited labs are assessed on whether their uncertainty budgets correctly implement the GUM methodology.

The key principle is this: because all measurements involve imperfect instruments, imperfect references, and real-world conditions, there is no such thing as a measurement result without some associated doubt. Uncertainty quantifies that doubt in a way that is consistent, comparable, and actionable.

Error versus uncertainty — a critical distinction

These two terms are frequently confused, and the confusion matters because they are fundamentally different concepts.

Error is the difference between the measured value and the true value. If a thermometer reads 100.3°C and the true temperature is 100.0°C, the error is +0.3°C. The problem with error is that it requires knowledge of the true value — which, in any real measurement situation, you do not have. The true value is a philosophical concept: you can approach it, but you can never know it exactly. This is why error cannot be reported on a calibration certificate as a standalone number without qualification.

Uncertainty sidesteps this problem. Instead of claiming to know the true value (which would require knowing the error), uncertainty estimates the range within which the true value is expected to lie, given everything the lab knows about the measurement process. It is a statement about what we do not know, expressed quantitatively.

The analogy is a tape measure reading 1.234 m. The error — the difference between 1.234 m and the actual length — might be +0.5 mm, but since you cannot see the true length directly, you estimate uncertainty as ±0.5 mm based on the tape’s accuracy and the resolution of its scale. The uncertainty tells you how much confidence to place in the reading; the error tells you how wrong the reading was, but only if you already know the answer.

Type A uncertainty — the statistical component

Type A uncertainty is evaluated by statistical analysis of a series of observations. The procedure is straightforward: take repeated measurements of the same quantity under the same conditions, calculate the standard deviation of those measurements, and divide by the square root of the number of readings to obtain the standard uncertainty.

If you take ten readings of a temperature bath with a reference thermometer and obtain a standard deviation s = 0.04°C, the Type A standard uncertainty is:

u_A = s / √n = 0.04 / √10 = 0.013°C

Increasing the number of measurements reduces the Type A contribution — taking 40 readings instead of 10 would halve it. Type A uncertainty catches the random effects that make measurements scatter: electrical noise in the measurement circuit, thermal fluctuations, vibration, operator-to-operator variation, and similar sources of variability. It is the component that would be reduced to zero in a hypothetical perfect measurement environment.

Type B uncertainty — everything else

Type B uncertainty covers all sources that are evaluated by means other than statistical analysis of repeated measurements. This includes values from calibration certificates of reference instruments, manufacturer specifications, published tables, theoretical calculations, and knowledge about the physical behaviour of the measurement system.

Type B is not a lesser or less rigorous category — it is simply a different method of evaluation. For many uncertainty sources, Type B is the only practical method: you cannot take repeated readings of the resolution of a digital display, for example, because it is a fixed property of the instrument.

Common Type B contributions and how they are calculated:

• Reference instrument calibration uncertainty: the calibration certificate for your reference thermometer states U = 0.10°C (k=2). The standard uncertainty is U/k = 0.10/2 = 0.05°C (assuming a normal distribution).
• Display resolution: a thermometer with a 0.01°C resolution display has a rectangular distribution over ±0.005°C. The standard uncertainty is 0.005/√3 = 0.003°C. The √3 divisor converts a rectangular distribution to a standard deviation.
• Bath non-uniformity: the manufacturer specifies the temperature bath is uniform to ±0.05°C across the working volume. Treating this as a rectangular distribution: 0.05/√3 = 0.029°C.
• Manufacturer specification (stated at 95%): divide by 2 (assumed normal distribution) to get the standard uncertainty contribution.

Combining uncertainties — root sum of squares

Once all individual uncertainty contributions are identified — both Type A and Type B — they are combined using the root sum of squares (RSS) method to obtain the combined standard uncertainty u_c:

u_c = √(u₁² + u₂² + u₃² + … + u_n²)

This is the GUM-recommended combination method when the uncertainty sources are independent of each other. Where sources are correlated, sensitivity coefficients and covariance terms are required — but for most calibration uncertainty budgets, the simple RSS formula is appropriate.

Working through the temperature calibration example in the table below numerically:

u_c = √(0.05² + 0.003² + 0.02² + 0.029² + 0.003² + 0.04²)
= √(0.0025 + 0.000009 + 0.0004 + 0.000841 + 0.000009 + 0.0016)
= √(0.005059)
= 0.071°C

The expanded uncertainty at k=2 is then U = 2 × 0.071 = ±0.14°C (rounded to 2 significant figures, which is standard practice for uncertainty reporting).

Coverage factor k and confidence levels

The coverage factor k determines how wide the uncertainty interval is, and therefore what confidence level the stated uncertainty represents. The relationship comes from the normal (Gaussian) probability distribution:

• k=1: approximately 68% confidence — the true value lies within ±u_c of the stated result about 68% of the time.
• k=2: approximately 95% confidence — the standard for ISO/IEC 17025 accredited calibration certificates.
• k=3: approximately 99.7% confidence — used for safety-critical applications where the risk of exceeding the interval must be minimised.

ISO/IEC 17025 calibration laboratories default to k=2 (95%) because it provides a practical balance between interval width and confidence. Reporting at k=1 would produce tighter-looking numbers but lower confidence; reporting at k=3 would produce more conservative intervals that could cause unnecessary instrument rejections in most applications.

When a calibration certificate is received without a stated k value, there is no way to know the confidence level of the reported uncertainty — and therefore no basis for using the uncertainty in conformity decisions. This is one reason ISO/IEC 17025:2017 clause 7.8.2 mandates that both U and k (or the coverage probability) must appear on the certificate.

Reading uncertainty on a calibration certificate

A typical calibration certificate for a temperature measurement device might include a results table with columns for nominal value, measured value, deviation, and expanded uncertainty. Understanding what each means in practice is essential for using the certificate correctly.

Consider a certificate entry that reads: Nominal 100.00°C | Measured 100.30°C | Deviation +0.30°C | U = ±0.50°C (k=2, 95%).

This tells you:

• The instrument reads 0.30°C higher than the reference standard at 100°C.
• The calibration lab is 95% confident that the true deviation lies somewhere between +0.30 − 0.50 = −0.20°C and +0.30 + 0.50 = +0.80°C.
• The instrument is not simply “wrong by 0.30°C” — it could be anywhere in the range −0.20°C to +0.80°C with 95% confidence, or slightly outside that range 5% of the time.

This interpretation is critical when the instrument is used to make decisions at a tolerance limit. If your product must be maintained at 100°C ± 1°C and the instrument reads 100.3°C at the nominal 100°C point, you cannot conclude that it reads “close enough” without considering the full uncertainty interval of −0.20°C to +0.80°C.

Uncertainty and conformity decisions — the 4:1 rule

The most practically important use of measurement uncertainty is in conformity decisions: is this product, process, or instrument within its specified tolerance? The measurement uncertainty directly affects the reliability of that decision.

The 4:1 rule — also called the test accuracy ratio (TAR) — states that the measurement uncertainty should be no more than 25% of the process tolerance for reliable conformity decisions. If your process tolerance is ±1°C, the 4:1 rule requires a calibration uncertainty U ≤ ±0.25°C.

When uncertainty exceeds 25% of tolerance, a significant “grey zone” exists around the tolerance limit — a range of instrument readings where the true value might be either inside or outside the tolerance. Within this grey zone, you cannot reliably determine conformance without additional information.

ISO 14253-1 addresses this formally through the concept of guardbanding. Rather than accepting or rejecting a measurement based solely on whether it falls within the nominal tolerance, a guardband shrinks the acceptance zone by the measurement uncertainty on each side. A measurement must fall within the tighter guardband to be confidently declared conforming — this ensures that even accounting for measurement uncertainty, genuine non-conforming values are not accepted.

Uncertainty in the traceability chain

Measurement uncertainty does not exist in isolation — it accumulates as you move down the traceability chain. Understanding this accumulation is essential for interpreting the uncertainty on your working instruments’ calibration certificates.

The chain begins at the national metrology institute. In Singapore, that is the National Metrology Centre (NMC), operated by A*STAR. NMC maintains Singapore’s primary standards — the physical realisations of the SI units — with the lowest achievable uncertainties. For temperature, the NMC primary standard uncertainty might be in the range of ±0.001–0.005°C.

An accredited calibration lab like Unitest holds reference instruments calibrated by NMC. These reference instruments carry the NMC’s uncertainty plus contributions from the calibration process, transport, and storage. The lab’s reference thermometer might carry an expanded uncertainty of ±0.05–0.10°C.

When the lab calibrates a working instrument, the lab’s reference uncertainty is a Type B input to the working instrument’s uncertainty budget. Combined with all the other contributions — bath stability, DUT repeatability, resolution, non-uniformity — the final expanded uncertainty on the working instrument’s certificate might be ±0.15–0.50°C depending on the parameter and measurement method.

By the time calibration reaches a working instrument, typical expanded uncertainties at k=2 are:

• Temperature: ±0.1°C to ±0.5°C depending on range and method
• Pressure: ±0.01% to ±0.1% of full scale
• Dimensional (length): ±0.002 mm to ±0.05 mm
• DC voltage: ±0.005% to ±0.05% of reading
• Humidity: ±0.5% to ±2% relative humidity

These numbers inform the 4:1 rule calculations for your specific instruments. A micrometer with a calibration uncertainty of ±0.003 mm is suitable for verifying a ±0.012 mm tolerance (4:1 ratio). A digital thermometer with ±0.3°C calibration uncertainty should not be used to verify a ±0.5°C process tolerance — the 4:1 rule fails at that combination.

Common questions from quality managers

My calibration certificate has no uncertainty stated — is that acceptable?

No. ISO/IEC 17025:2017 clause 7.8.2 explicitly requires that calibration certificates include the measurement uncertainty. A certificate without uncertainty is not compliant with the current standard. In addition, ISO 9001:2015 clause 7.1.5 requires measurement results with stated uncertainties for instruments contributing to quality decisions. If you receive a certificate without uncertainty from a lab claiming ISO/IEC 17025 accreditation, verify the claim at sac.gov.sg — the certificate may not actually be covered by the accreditation, or the specific measurement may fall outside the lab’s accredited scope.

The uncertainty on my certificate is larger than I need — what can I do?

There are two practical options. First, use a higher-accuracy instrument — one whose inherent accuracy and stability allow the calibrating lab to achieve a lower combined uncertainty. Second, use a higher-accuracy calibration laboratory with reference standards calibrated to lower uncertainties. Requesting a certificate with a smaller stated uncertainty from the same laboratory using the same instrument is not meaningful — the uncertainty budget reflects the actual physics of the measurement, not a reporting preference. If you need U ≤ ±0.05°C for a temperature measurement, you need an instrument and calibration lab both capable of achieving that level.

Can I use a 0.01 mm resolution caliper to verify a 0.05 mm tolerance?

Probably not reliably. A 0.01 mm resolution caliper calibrated at an accredited lab might carry a calibration uncertainty of U ≈ ±0.010–0.015 mm. Against a ±0.05 mm tolerance (half-tolerance = ±0.025 mm), the test accuracy ratio is approximately 1.7:1 — well below the recommended 4:1. The grey zone around the tolerance limit is large enough that conformance decisions become unreliable at the boundary. A calibrated micrometer (U ≈ ±0.003–0.005 mm) or a coordinate measuring machine would be appropriate tools for this application.