Lexicon · Tools and roadmap
Uncertainty
Fully specified name: Uncertainty (concept)
Limits on knowledge, measurement, estimation, or prediction.
- Scope note
- Do not collapse distinct uncertainties into an unlabeled confidence number.
- Synonyms
- None recorded as true synonyms in this context.
- Monograph
- Model-ready knowledge exchange (planned; brief only)
- External mappings
- None recorded. In W0 no mapping to SNOMED CT® or any external authority is asserted for this entry. A future mapping record will carry source system, identifier, version, relation type, evidence, author, confidence where meaningful, and review state.
- Formal status
- Editorial entry only. It is not part of any released ontology and a prose edit here changes no formal definition.
Draft. This entry is an unreviewed draft. Its sources have not been checked by a named person and no domain reviewer has approved it. Treat every claim as provisional.
Notes
Uncertainty is the acknowledged gap between what is stated and what is known. It takes several forms that are easy to run together. Measurement uncertainty is the spread in a value from the instrument and procedure. Sampling uncertainty is the spread from having observed a sample rather than a population. Model uncertainty is doubt about whether the equations or structure chosen are the right ones. Parameter uncertainty is doubt about the values fed into a model that is assumed correct. Prediction uncertainty combines these when a model is used to say what will happen. Epistemic uncertainty could in principle be reduced by more knowledge; aleatory uncertainty reflects variability that no further measurement removes.
Metrologists, statisticians, physicists, and clinicians each have conventions for one or two of these and often use the same words, such as "error" or "confidence," for different things. In dosimetry, for instance, the uncertainty in an estimated dose comes from counting statistics, time-point sampling, the geometric model, and the fit, and these do not add simply.
The scope instruction for this concept is a warning. A single unlabeled number offered as "confidence," whether from a statistical procedure or from a machine-learning system, hides which kind of uncertainty it represents and which it ignores. NucLex asks that uncertainty be recorded with its kind, its basis, and its source, attached to the parameter or claim it qualifies.