Every measurement error has two components. The systematic component repeats in the same way in each measurement, or changes according to a known law, so it shifts the result in one direction. The random component changes from one reading to the next with no predictable pattern, so it scatters results around a central value. The practical difference is that systematic error is corrected and random error is averaged out.
How they show up
| Aspect | Systematic | Random |
|---|---|---|
| Behavior | Constant or predictable | Unpredictable |
| Effect on the mean | Shifts it | Hardly affects it if there are many readings |
| Effect on dispersion | None | Increases it |
| How it is handled | Correction or adjustment | Repetition and averaging |
Typical examples
A badly set zero on a balance produces the same systematic error in every weighing. A reference temperature different from the calibration temperature introduces a predictable systematic effect. By contrast, electronic noise, vibration and small variations in how the part is placed produce random variation.
How they are detected
Random error becomes evident when the measurement is repeated: the dispersion of the readings estimates it. Systematic error does not appear on repetition, because all readings deviate equally; it is only discovered by comparing against a reference, that is, by calibrating, or by using another independent method. That is the very reason calibration exists.
How to interpret the result
Once the correction for a systematic effect is known, it is applied to the result. Even so, the correction has its own uncertainty, which must be included in the final uncertainty. An unknown systematic effect cannot be corrected and therefore becomes a risk: what looks like a precise measurement may be biased without anyone noticing.