Type A and Type B Uncertainty

Estimation and combination of Type A and Type B experimental uncertainties.

Conceptual context

Every experimental result is an estimate rather than a perfect capture of reality. Repetition reveals fluctuations, while instrument resolution, calibration, and tolerance provide additional information. Type A uncertainty comes from statistical analysis of repetitions. Type B uncertainty comes from specifications, certificates, prior knowledge, or distribution models. They are methods of evaluation, not two absolute kinds of error. The goal is to obtain $x_bar$ as the central value, estimate $u_A$ and $u_B$, combine them into $u_c$, and, when required, communicate $U$ with a stated $k$.

🟢 Essential level

Three ideas must be separated. $x_bar$ summarizes the center of repeated observations. $s$ describes how widely individual measurements scatter. $u_A$ describes how much the mean would vary if the whole series were repeated. It therefore decreases as $n$ increases, although with diminishing returns. $u_B$ represents information not extracted from the current series, such as resolution, calibration, or tolerances. If both contributions are independent, they are not added directly; they are combined in quadrature to obtain $u_c$. More repetitions help against Type A uncertainty but do not remove a constant Type B contribution. Finally, $U$ scales the standard uncertainty into a coverage interval and must always be accompanied by $k$.

The correct strategy depends on identifying the dominant component and the experimental change that can actually reduce it. If the Type A component dominates, repeating the trial improves the result; if Type B dominates, no additional repetition will help, and it is better to change instrument, improve calibration, or reduce the stated tolerance.

🔵 Formal level

Basic formulas

The central value of repeated observations is obtained from:

Arithmetic mean

Scatter of individual observations is evaluated with:

Sample standard deviation

Using that scatter and $n$, standard uncertainty of the mean is:

Type A standard uncertainty

These three expressions are the foundation of any Type A evaluation: without $x_bar$, $s$, and its propagation into standard error, no later $u_c$ can be built.

Real student questions

Does Type A mean random and Type B systematic? Not exactly. They describe evaluation methods rather than absolute causes.

Why is $s$ different from $u_A$? One describes individual observations, the other uncertainty of the mean.

Does repetition remove all uncertainty? No. It reduces Type A under independence, but not a constant Type B contribution.

Why are uncertainties not added directly? Independent standard contributions combine in quadrature.

Which distribution should represent a tolerance? Rectangular is reasonable when all values inside the interval are equally plausible.

Can $k$ always be two? Only when the criterion is stated and the effective distribution supports that approximation.

Final synthesis

Type A uncertainty comes from repetition; Type B comes from additional metrological information. Both must be converted to compatible standard scales, combined under explicit assumptions, and communicated through a declared coverage interval. Repetition improves the mean, but only improvement of the whole budget improves the final result.