Measurement Uncertainty and Calibration: What It Is and How to Calculate It

Measurement uncertainty — a quantitative assessment of the doubt in a measurement result — is the foundation of reliable quality control. In simple terms, it indicates the range within which the true value of the quantity is expected to fall with a certain probability.

For example, if a result is recorded as 1.750 ± 0.001 kg at a 95% confidence level, this means the actual value is expected to be between 1.749 and 1.751 kg at the specified level of confidence.

This approach is important because a "bare" number without uncertainty is often misleading. Two values may appear equally precise yet have very different reliability.

One of the most common mistakes in quality practice is treating a measurement as absolutely precise. In real conditions, the result is always affected by variable factors:

• the condition and accuracy class of the instrument;
• how a specific operator performs the operation;
• temperature, humidity, vibrations, and other environmental conditions;
• the measurement method and its reproducibility;
• the preparation of the sample or control object.

Therefore, the question is not whether uncertainty exists, but whether it is correctly determined, controlled, and accounted for in decision-making.

In the calibration process, uncertainty arises from several groups of sources. The most common are:

• metrological characteristics of the reference standard;
• resolution (discrimination) of the instrument;
• repeatability of measurements;
• equipment stability over time;
• impact of environmental conditions;
• uncertainty stated in higher-level traceability certificates.

This is precisely why calibration cannot be reduced to a "one-time check." It is a systematic procedure that confirms whether an instrument is fit for a specific application.

These terms are often confused, although they describe different things:

• error is the deviation between the instrument reading and the reference value;
• uncertainty is the confidence interval for the result that accounts for the cumulative effect of various factors.

An error can be corrected or compensated for within the method. Uncertainty must always be declared to show the quality of the result and the level of confidence in it.

In metrological practice, two types of evaluation are typically applied.

Evaluated statistically based on a series of repeated measurements. Typically uses the mean value and standard deviation to quantify the spread of results.

This approach works well when there is a sufficient number of repeats and a stable measurement process. The recommended minimum number of repeats is usually at least 10 observations, although critical measurements may require more. The larger the sample, the more reliable the estimate of standard uncertainty. It is also important to verify that no systematic trends exist within the series of repeats, as a trend may indicate instrument drift or changing conditions.

Evaluated from other information sources when the statistical Type A method is unavailable or impractical:

• calibration certificates;
• manufacturer data;
• reference values;
• prior instrument operating experience.

Type B is often applied where performing a large series of repeated measurements is impractical or infeasible. To use this approach correctly, it is essential to identify the probability distribution for each source — rectangular, triangular, or normal — and calculate the standard uncertainty accordingly. The correct choice of distribution significantly affects the final combined uncertainty result.

Uncertainty is not a "scientific formalism" but a management tool. It directly affects process quality and economics.

Practical benefits for a company:

• increased confidence in control results;
• better comparability between batches, shifts, and sites;
• reduced risk of incorrect accept/reject decisions;
• greater resilience to claims and disputes with customers;
• easier compliance audits and conformity assessments.

For enterprises operating under ISO 9001, ISO/IEC 17025, IATF 16949, or sector-specific standards, uncertainty control is part of mature metrological management.

The source material outlines the basic calculation logic. Below is an adapted practical algorithm that can be applied in corporate procedures.

Compile a complete list of factors affecting the measurement:

• repeatability;
• instrument resolution;
• environmental instability;
• measuring instrument drift;
• reference standard errors, etc.

The more complete the list at this stage, the more reliable the final result. It is recommended to use a cause-and-effect diagram (Ishikawa diagram) or a factor table to systematically cover all potential sources. Missing even a single significant factor can lead to an underestimated uncertainty and incorrect conclusions about the suitability of the result.

For each source, determine the evaluation method:

• through statistical analysis of repeats (Type A);
• through external data or documentation (Type B).

All values must be converted to a consistent standard uncertainty format to correctly perform the combination. For Type B, this means identifying the probability distribution type and applying the appropriate divisor. For example, for a rectangular distribution, the standard uncertainty equals the half-width of the interval divided by √3. Thorough documentation of each component greatly simplifies verification during audit.

Once the components are defined, calculate the combined standard uncertainty using the root sum of squares rule:

u_c = sqrt(u1^2 + u2^2 + ... + un^2)

where u1...un are the standard uncertainties of individual sources.

This approach accounts for the contribution of each factor and produces an integral estimate. The formula assumes that the uncertainty components are uncorrelated. If correlation exists between sources, covariance terms must also be included. In practice, for most calibration procedures the correlation is negligible, but this assumption should be documented and justified.

Next, multiply the combined uncertainty by the coverage factor k:

U = k * u_c

The most commonly used values are:

• k ≈ 2 for a confidence level of approximately 95%;
• k ≈ 3 for a level of approximately 99%.

The specific value of k is selected according to the methodology and industry requirements. For a more precise determination of k with a small number of degrees of freedom, the Student's t-distribution may be applied. The chosen value of k and the justification for the corresponding confidence level must always be recorded in the measurement protocol.

The final result is recorded in the format:

X = x ± U, with the confidence level indicated.

For example: 1.750 ± 0.001 kg (95%).

Without this notation, the user of the result cannot adequately assess its suitability for a specific decision. It is also important to follow rounding rules: the expanded uncertainty is typically rounded up to two significant figures, and the measurement result is rounded to the same decimal place as the uncertainty. This approach aligns with GUM recommendations and ISO/IEC 17025 requirements.

Imagine you are controlling the weight of a critical part. The following data is obtained:

• mean value: 1.750 kg;
• combined uncertainty: 0.0005 kg;
• coverage factor k = 2.

Then:

• expanded uncertainty U = 2 * 0.0005 = 0.001 kg;
• final result: 1.750 ± 0.001 kg (95%).

If the customer tolerance is tight, it is precisely this interval that determines whether the result is reliable enough for product acceptance or whether additional control is needed.

For the system to work reliably, it is important not only to know how to calculate but also to organize the process.

Recommended minimum:

• approve methodologies with clear evaluation rules;
• maintain a register of measuring instruments and a calibration schedule;
• centralize certificates and change history;
• train personnel to interpret results in x ± U format;
• regularly review sources of uncertainty;
• conduct internal traceability checks.

This approach reduces operational risk and simplifies preparation for external audits.

Before an audit, verify that the following basic requirements are met:

• all critical instruments have current calibration status;
• certificates are stored in a single controlled environment;
• methodologies include uncertainty evaluation rules;
• personnel understand the difference between error and uncertainty;
• measurement records contain the full result format;
• traceability to reference standards is documented;
• corrective actions for deviations are initiated on time.

If even one of these items is weak, the risk of audit nonconformity increases, along with the risk of flawed production decisions.

The most common causes of a weak metrological system:

• using non-validated methodologies;
• formal calibration without fitness-for-purpose analysis;
• absence of Type B evaluation from documentary sources;
• incorrect rounding and result formatting;
• "manual" certificate storage without version control;
• interpreting a number without considering the uncertainty interval.

All of these mistakes share a common consequence: management makes decisions based on data whose reliability has not been confirmed.

Measurement uncertainty and calibration are essential elements of a mature quality system, not an optional add-on. They show how much a result can be trusted and help make sound management and technical decisions.

The approach of "we measured and wrote down a number" no longer works in modern conditions. It is necessary to declare the result together with measurement uncertainty, control sources of variation, and maintain traceability. The methodological foundation for measurement uncertainty evaluation is in the GUM guide published by BIPM.

For companies operating in competitive and regulated segments, a systematic approach to measurement uncertainty is directly linked to quality, productivity, and market trust.