Once you have worked out the uncertainty contribution from each source in a measurement, you still have to fold them into one number that describes the uncertainty of the final result. You cannot simply add them up, because errors that are independent do not all pull in the same direction at once. The combined standard uncertainty is that single merged figure, and the way it is calculated - in quadrature rather than by ordinary addition - reflects how independent errors actually accumulate. This guide explains what the combined standard uncertainty is, how the root-sum-of-squares rule works and why, and how correlation between inputs breaks the simple rule and forces a correction.
Combined Standard Uncertainty in one line: The combined standard uncertainty, usually written uc, is the single standard uncertainty of a measurement result formed by combining the standard uncertainty contributions of all its inputs. When the inputs are independent, the contributions are combined in quadrature, meaning uc is the square root of the sum of the squares of the individual contributions, a rule called root-sum-of-squares or RSS. If some inputs are correlated, the simple RSS rule no longer holds and correlation terms must be added, which can make uc larger or smaller than RSS alone would give.
The combined standard uncertainty is the step that turns a list of individual contributions into one number for the result. Each input has already been reduced to a contribution expressed in the units of the result, by taking its own standard uncertainty and weighting it by its sensitivity coefficient. The combined standard uncertainty answers the question, given all these contributions, how uncertain is the result as a whole. It is still a standard uncertainty, on the same one-standard-deviation footing as the pieces that went into it, which is what lets it be expanded later into a coverage interval.
The rule for combining independent contributions is root-sum-of-squares, also called combining in quadrature. You square each contribution, add the squares together, and take the square root of the total. The reason you do not simply add the contributions is that independent errors are not all at their maximum in the same direction at the same instant; sometimes one is high while another is low, and they partially offset. Squaring and rooting captures this statistical cancellation, so the combined uncertainty is larger than any single contribution but smaller than their straight sum.
A useful consequence of squaring is that the largest contributions dominate the result far out of proportion to their size, and small ones fade away. Because each contribution is squared before being added, a term that is, say, three times another contributes nine times as much to the sum of squares. This means that in a typical budget one or two large contributions set the combined standard uncertainty almost entirely, and shaving a small contributor makes almost no difference. Recognising which terms actually drive uc, through the quadrature sum, is the practical payoff of computing it this way rather than just adding.
The clean root-sum-of-squares rule rests on an assumption: that the input errors are independent of one another. Independence means that knowing one input happened to be high tells you nothing about whether another is high or low. When that holds, the partial cancellation captured by quadrature is exactly right. But it does not always hold, and when inputs are correlated the simple RSS rule is no longer correct on its own.
Correlation arises when two inputs share a common influence, so their errors tend to move together rather than independently. A frequent cause is a shared calibration or a shared reference: if two measurements are made with instruments calibrated against the same standard, an error in that standard pushes both readings the same way, so their errors are correlated. Another is a shared environmental influence, such as a temperature that affects several sensors at once. When errors move together, they no longer partially cancel; they can reinforce, and treating them as independent would understate the true combined uncertainty.
To handle correlation, the combination formula gains extra terms that account for how strongly, and in which direction, the correlated inputs move together. Positive correlation, where the errors tend to move the same way, adds to the combined uncertainty relative to RSS, because the errors reinforce instead of cancelling. Negative correlation, where they move oppositely, subtracts, because the errors offset more than independence would predict. The upshot is that with correlated inputs the combined standard uncertainty can be either larger or smaller than the naive root-sum-of-squares value, and getting it right requires knowing which inputs are linked and by how much rather than blindly applying RSS.
The combined standard uncertainty is one specific step in a longer sequence, and it helps to see where it sits. First each input is characterised by its own standard uncertainty. Then each is weighted by its sensitivity coefficient to give a contribution in the units of the result. Then those contributions are merged, in quadrature or with correlation terms, into the combined standard uncertainty uc. Finally uc is multiplied by a coverage factor to give an expanded uncertainty that states an interval at a chosen level of confidence. The combined standard uncertainty is the pivot between characterising the inputs and reporting a result: it is the last purely internal figure before the result is expressed in a form a user reads.
Understanding this placement clears up a common confusion between the combined standard uncertainty and the expanded uncertainty. The combined value uc is still at the one-standard-deviation level and is not yet the number you would typically quote as the measurement's uncertainty; expanding it with a coverage factor is what produces the interval most reports state. Keeping the two straight matters, because applying a coverage factor to the wrong figure, or reporting uc as if it were already expanded, misstates how confident the interval is.
In flow and process measurement this chain lives inside real systems. A flow computer or a SCADA-based measurement package effectively carries the ingredients of an uncertainty budget for a metering point, and the combined standard uncertainty is what characterises the quality of the reported flow before it is expanded for a statement of measurement uncertainty. When a cloud SCADA platform such as Merobix trends the inputs to a measurement over time, it lets an engineer see whether the contributions that drive uc are staying within expectation - for example whether a differential-pressure input is drifting toward a range where its contribution grows. In that sense the combined standard uncertainty is not just a paperwork figure but a description of measurement quality that can be watched and maintained through the same monitoring used to run the process.
For independent inputs it is calculated in quadrature, using the root-sum-of-squares rule: square each input's contribution to the result, add the squares, and take the square root of the total. This captures the fact that independent errors do not all reach their maximum in the same direction at once, so they partially cancel. The result, written uc, is larger than any single contribution but smaller than their straight sum.
The combined standard uncertainty uc merges all the input contributions into one figure that is still at the one-standard-deviation level. The expanded uncertainty is uc multiplied by a coverage factor, which widens it into an interval at a chosen level of confidence, and it is usually the number quoted in a measurement report. Combining comes first and expanding comes after, so uc is the step before expansion, not a substitute for it.
The simple root-sum-of-squares rule assumes the input errors are independent, so they partially cancel. When inputs are correlated, for example because they share a calibration reference or an environmental influence, their errors move together and no longer cancel in the same way. Positive correlation adds to the combined uncertainty and negative correlation subtracts, so correlated inputs require extra terms and can make the combined uncertainty either larger or smaller than plain RSS.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.