What Is Calibration? Making an Instrument Tell the Truth

Calibration compares an instrument's output against a reference of known value and establishes the relationship between them. In the simplest case a single known value is measured and any offset is recorded or corrected. More usually several known values spanning the working range are measured, which reveals whether the response is linear and where it departs from linearity, and the resulting relationship is used to convert readings into results. The reference must itself be trustworthy, which raises the question of how its value was established, and answering that leads to a chain of comparisons reaching back to a primary standard. Calibration also establishes the uncertainty of a measurement, since the process reveals how much the readings scatter and how well the reference is known, and a result quoted without an uncertainty is incomplete.

The need is continuous rather than one-off, for reasons that are physical:

• •Mechanical wear and deformation in moving parts and in mounts
• •Electronic component ageing, which shifts the response of amplifiers and sensors over months and years
• •Temperature, which affects nearly every measurement and is compensated for imperfectly
• •Contamination of optical surfaces, electrodes and sensors
• •Mechanical shock, including transport, which can shift an instrument abruptly
• •Changes in the reference itself, if it is a material that degrades or a solution that evaporates

Traceability means that a measurement can be linked through a documented chain of calibrations to a primary standard, with the uncertainty at each step accounted for. That structure is what allows a measurement made in one laboratory to be compared with one made elsewhere, and it underpins international trade, medicine, engineering and law. The primary standards sit with national measurement institutes and, at the top, with an international system that has been progressively rebuilt so that units are defined by fixed values of physical constants rather than by objects. The last physical artefact, a metal cylinder that defined the kilogram, was retired in 2019 after it was found to have drifted relative to its copies by a small but significant amount over a century, which is exactly the problem that definitions based on constants avoid.

The word is used beyond instruments and the extensions are more than metaphor. A forecaster is well calibrated when events they assign a seventy percent chance to happen about seventy percent of the time, which is a measurable property assessed over many forecasts and is distinct from being accurate, since a forecaster can be calibrated and uninformative by always saying fifty percent. Research on expert judgement has found calibration to be generally poor, with confidence intervals far too narrow, and to improve substantially with feedback and practice, which is the basis of training programmes for forecasters. Statistical models are calibrated when their predicted probabilities match observed frequencies, and a model can discriminate well while being badly calibrated, which matters enormously when the output is used to make decisions rather than only to rank cases.

The practical failures are instructive because they are rarely about equipment. Instruments used outside their calibrated range give readings that look normal and are not valid, which is a common error in fields where a device is used for something it was not characterised for. Calibration performed under conditions unlike the working ones transfers poorly, particularly for temperature and humidity. Records that are not kept mean an instrument's history is unknown and results cannot be defended afterwards. Single-point calibration is treated as sufficient when the response is not actually linear. And the reference material is assumed to be right when it has expired, been contaminated or been prepared incorrectly, which is the failure that propagates furthest, since every result depending on it is wrong in the same direction and nothing internal to the measurements reveals it.

Comparing an instrument against known values establishes the relationship between reading and reality and reveals the uncertainty, without which a result is incomplete. Instruments drift through wear, component ageing, temperature and contamination, so the process repeats. Traceability links a measurement through documented steps to primary standards, which are now defined by fixed physical constants after the kilogram artefact was found to have drifted.