When you build an uncertainty budget, every line on it has to get a number, and there are really only two ways to arrive at that number: you either measure the scatter yourself by repeating the observation, or you take it from information someone else has already provided. That split is exactly what the Type A and Type B classification captures. It is not about which sources are bigger or more important; it is about how each contribution was evaluated. This guide explains what Type A and Type B evaluation mean, how to tell which category each line of a budget belongs to, and why the distinction is about method rather than the nature of the error itself.
Type A vs Type B Uncertainty in one line: Type A and Type B are the two methods for evaluating an uncertainty contribution. A Type A evaluation is done statistically from a series of repeated observations, typically by taking the standard deviation of the data you gathered. A Type B evaluation is done by any other means, drawing on information such as a calibration certificate, a manufacturer's specification, published data, or reasoned judgment, rather than from repeated measurements you performed. The distinction describes how the uncertainty was estimated, not how large it is or how it behaves, and both types are combined the same way once expressed as standard uncertainties.
A Type A evaluation is what most people picture when they think of measuring uncertainty: you make the measurement several times under the same conditions, look at how much the results scatter, and quantify that scatter statistically. The usual tool is the standard deviation of the repeated readings, which describes how much an individual observation typically departs from the mean. From that you derive the standard uncertainty of the result, often the standard deviation of the mean when the reported value is an average of the repeats.
The defining feature of a Type A evaluation is that it comes from data you actually collected in this measurement or a directly relevant repeatability study. Because it is statistical, it captures the real, observed variability of the process as it ran, including random effects you might not have anticipated or itemised. This is its great strength: it does not rely on assumptions about how a component behaves, only on what the data showed. Its limitation is that it only sees the sources of variation that were free to vary during the repeats; anything held fixed, such as a systematic offset in a reference, will not show up in the scatter and has to be accounted for another way.
Type A evaluation is therefore the natural choice for the repeatability and random components of a budget - the run-to-run scatter of a reading, the noise on a signal, the variability of a process being sampled. Where you have, or can obtain, a set of repeated observations, evaluating their uncertainty statistically is straightforward and well grounded. The number of repeats matters, because a standard deviation estimated from few observations is itself uncertain, which is why Type A evaluations from small samples are treated with appropriate caution.
A Type B evaluation covers every way of estimating an uncertainty that is not the statistics of your own repeated observations. Instead of gathering data, you draw on information that already exists. The classic source is a calibration certificate, which states the uncertainty of the reference or instrument you used. Others include a manufacturer's accuracy specification, published physical data, the resolution of a display, the tolerance of a component, and reasoned engineering judgment about an effect you cannot easily measure directly. In each case you are importing an uncertainty rather than deriving it from fresh measurements.
Because Type B information often arrives in a form that is not already a standard uncertainty, part of a Type B evaluation is converting it into one. A specification quoted as a plus-or-minus limit, for instance, has to be turned into a standard uncertainty by assuming a distribution for how the true value sits within those limits and applying the appropriate divisor. A certificate that quotes an expanded uncertainty with a coverage factor has to be divided back down to the standard level. This conversion step is characteristic of Type B work and is where judgment about the assumed distribution enters.
Type B evaluation is indispensable precisely because so many important uncertainty sources cannot be captured by repeating a measurement. A systematic offset in a reference standard does not vary from run to run, so no amount of repetition will reveal it; it has to come from the certificate. The same is true of an instrument's specified accuracy across its range, or the uncertainty of a physical constant used in a calculation. Far from being a lesser method, Type B evaluation is how the fixed, systematic, and externally-characterised parts of a budget get their numbers, and in many practical measurements the Type B contributions are the dominant ones.
Classifying a budget line comes down to one question: was this number obtained from the statistics of repeated observations I made, or from some other source of information? If it came from the scatter of your own repeats, it is Type A; if it came from a certificate, a spec sheet, published data, or judgment, it is Type B. A common source of confusion is to assume Type A means random and Type B means systematic, but that is not the definition. Type A and Type B describe the method of evaluation, while random and systematic describe the nature of the effect, and the two classifications are independent of each other.
That independence is worth dwelling on, because it is the point most often missed. A random effect could in principle be evaluated either way, and a Type B evaluation can perfectly well describe a random component if you are taking its magnitude from a specification rather than measuring the scatter yourself. Likewise, once each contribution has been expressed as a standard uncertainty, the way it is combined with the others does not depend at all on whether it was Type A or Type B. The two types are combined together on equal footing into the combined standard uncertainty, using the same quadrature rules. So the classification affects how you get the number, not how you use it afterward.
In practical measurement and SCADA-based metering, most of the budget for a given point tends to be Type B, because it is built from instrument specifications and calibration certificates rather than from repeated field trials, while a Type A repeatability component is added where a repeatability study or run-to-run data is available. Knowing which lines are which helps an engineer maintain the budget: the Type B lines are refreshed when instruments are recalibrated and new certificates issued, while the Type A lines can be revisited when fresh repeatability data is gathered from the operating system. A cloud SCADA platform that trends a measurement over time can even feed the Type A side, since the observed run-to-run variability of a logged reading is exactly the kind of data a statistical, Type A evaluation uses. Keeping the classification clear is what lets each part of the budget be updated from the right source.
Type A evaluation estimates an uncertainty statistically from a series of repeated observations you made, typically using the standard deviation of the data. Type B evaluation estimates it by any other means, such as a calibration certificate, a manufacturer's specification, published data, or reasoned judgment. The distinction is about the method used to get the number, not about how large the uncertainty is or how it behaves.
No, that is a common misconception. Type A and Type B describe how an uncertainty was evaluated, while random and systematic describe the nature of the underlying effect, and the two classifications are independent. A Type B evaluation can describe a random component if you take its size from a specification, and once expressed as standard uncertainties, both types are combined together in exactly the same way.
An uncertainty taken from a calibration certificate is a Type B evaluation, because it comes from prior information rather than from the statistics of repeated observations you performed. Since certificates often quote an expanded uncertainty with a coverage factor, part of the Type B work is dividing it back down to a standard uncertainty before it can be combined with the other contributions. It is then combined with any Type A components on equal footing.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.