Definition
Bias is the systematic error of a measurement, estimated as the difference between the average of the indications obtained by repeatedly measuring the same thing and an accepted reference value. It measures how far, on average, the instrument or method is displaced from the correct value.
Bias versus dispersion
Every measurement error has two components: the random part, which changes from one reading to the next, and the systematic part, which repeats. Averaging many readings reduces the random part but does not touch the bias. An instrument can have very good repeatability, with readings very close to one another, and still be biased.
| Characteristic | Dispersion | Bias |
|---|---|---|
| What it describes | Variation between readings | Average displacement |
| Reduced by averaging | Yes | No |
| How it is detected | By repeating the measurement | By comparing with a reference |
How it is estimated
- Measure a standard or reference material with a known, traceable value several times.
- Calculate the average of the indications.
- Subtract the reference value. The result is the estimated bias, with its own uncertainty.
Because bias is estimated from readings that have dispersion, you can only claim it exists if it is large relative to the uncertainty of its estimate.
What to do with it
If the bias is known and stable, it is corrected: a correction is applied to the result or the instrument is adjusted. That is, in essence, what calibration does. The correction has its own uncertainty, so the bias does not disappear completely. If the bias is not known, its effect must be counted as an uncertainty component.
What the engineer sees
It shows up as an instrument that always reads slightly high or low compared with other equipment, or a method that systematically gives a result different from another laboratory’s. It can vary with the level of the quantity, so it is advisable to estimate it at several points in the range.
When it changes
Bias can evolve with wear, contamination or the environment. A bias estimated today is not a permanent guarantee; that is why it is monitored with periodic verifications.