The underlying idea
A Type B uncertainty component usually comes from information that is not statistical: a manufacturer’s specification, a tolerance, the resolution of a display. Often that information says only that the true value lies between −a and +a. If nothing else is known, the most honest assumption is that any value in the interval is equally likely. That is a rectangular (uniform) distribution.
How it is converted to standard uncertainty
For an interval of half-width a, the standard deviation of a rectangular distribution is a divided by the square root of three. That is the divisor applied. The component is then expressed as a standard uncertainty and can be combined with the others as a root sum of squares.
Typical situations
- Resolution of a digital indicator: the real value can be anywhere within half a division of the least significant digit, so the half-width is half the resolution.
- Specification limits: when a manufacturer states an accuracy as a maximum limit without indicating a distribution or confidence level.
- Hysteresis or mechanical play: the maximum range of variation is known, but not how it is distributed.
- Temperature effects when all that is known is that the environment stayed within a band.
- Certificates that give an error limit rather than an expanded uncertainty with a coverage factor.
When it is not appropriate
If the certificate of a standard gives an expanded uncertainty with its coverage factor, the rectangular distribution is not used: the value is divided by that factor. If there are physical reasons for values near the center to be more likely, a triangular distribution is more realistic; its divisor is the square root of six and it produces a smaller uncertainty. If repeated data are available, the component is Type A and is calculated from the standard deviation of the data.
What an engineer sees in practice
In an uncertainty budget, the rectangular distribution appears as a row with its half-width, the root-three divisor and the sensitivity coefficient. Using it is a conservative decision: among reasonable distributions with the same limits, the rectangular one gives the greatest spread unless the extremes are assumed to be more likely than the center. That is why it is chosen when information is scarce. The judgment behind it must be documented, because it is a decision by the metrologist and not a measurement result.