SAL1 key terms

The complete set of items, people, transactions, or observations that a study or analysis is intended to describe. Conclusions drawn from a sample are generalised to this whole group.

A subset of a population selected for measurement or review, used to draw conclusions about the whole. Its usefulness depends on how representative the selection process is.

A selection method in which every member of the population has a known, non-zero chance of being chosen, reducing the risk that the selection process itself distorts results.

A systematic tendency for the chosen sample to differ from the population in a way that matters, so results are consistently too high or too low rather than merely imprecise.

An observation lying far from the general pattern of the other data. It may be a recording error, a genuine rare event, or a sign that the assumed model is wrong.

Values absent from a dataset that were expected to be recorded. If the absence is linked to the value itself, ignoring it can bias any conclusion drawn.

A single figure summarising the middle or typical value of a set of numbers, most often the mean, median, or mode. Each behaves differently when the data are skewed or contain outliers.

A measure of how widely values are spread around their mean, expressed in the same units as the data. A large value indicates results that vary substantially from observation to observation.

A number, usually between minus one and plus one, describing how strongly two quantities move together in a straight-line fashion. It says nothing about which one influences the other.

A third factor that influences both of two apparently related quantities, creating a link between them that disappears once the third factor is taken into account.

A technique for estimating how a quantity of interest changes as one or more explanatory factors change, usually by fitting a mathematical relationship to observed data.

Applying a relationship fitted to observed data to situations outside the range covered by that data. Results become progressively less reliable the further the new situation lies beyond the evidence.

Building a model so closely tailored to the particular data used that it captures random noise as if it were real pattern, and therefore performs poorly on new data.

A judgement that an observed result is unlikely to have arisen by chance alone under a stated assumption. It indicates detectability, not importance or size of effect.

A range of values, calculated from data, that is likely to contain the true unknown quantity, together with a stated level of confidence in that claim.

A formal procedure comparing observed data with what would be expected if a stated proposition were true, producing a decision rule about whether to reject that proposition.

A representation that explicitly allows for randomness, so that results are described by a range or distribution of possible outcomes rather than a single fixed answer.

A statement accepted as true for the purpose of an analysis without proof. Assumptions should be identified, justified, and tested for the effect they have on conclusions.

Repeated recalculation of a result while varying one input or assumption at a time, to show which factors drive the answer and how much the answer could move.

Examining outcomes under a small number of coherent, internally consistent alternative futures, chosen to illustrate plausible extremes rather than to attach probabilities to them.

Deliberately applying severe but plausible adverse conditions to see whether an entity or model can withstand them, and where the breaking points lie.

Independent checking that a model is fit for its intended use, covering the logic, the data, the assumptions, and the reasonableness of outputs against other evidence.

The property of data being complete, accurate, consistent, and protected from unauthorised change throughout their life cycle, from capture through storage to use.

The amount and type of exposure an organisation is willing to accept in pursuit of its objectives, stated in advance so that decisions can be judged against it.

The threshold at which an omission or misstatement is large enough, in amount or in nature, that it could reasonably influence the decisions of those relying on the information.

The policies, procedures, and checks built into a process to prevent, detect, and correct errors or irregularities, and to keep activity within agreed limits.

The recorded, chronological evidence of who did what, when, and why, sufficient for an independent reviewer to reconstruct how a figure or decision was reached.

Examination of work by someone with comparable expertise who was not involved in producing it, intended to catch errors and challenge judgements before the work is relied upon.

The ability of another competent person, working from the recorded inputs, methods, and documentation, to arrive at the same result independently.

A situation where a personal, financial, or other interest could improperly influence professional judgement. It should be disclosed and managed, even where no improper act has occurred.

A known weakness or boundary of a model, such as simplified assumptions, restricted data, or behaviour outside the calibration range, which users must understand before relying on outputs.

The state of not knowing which outcome will occur, whether because outcomes are random or because knowledge is incomplete. It differs from risk when possible outcomes cannot be listed or quantified.