Where does confidence collide?

A statement about the procedure that produced an interval, not about the one interval or the parameter inside it: if the estimation procedure is repeated on many samples, a stated fraction of the intervals it generates, the confidence level, will contain the fixed unknown parameter. Neyman set this out in his 1937 Outline of a theory of statistical estimation, defining confidence intervals and confidence coefficients as a frequency property of the method, and warning that the coefficient is not the probability that any single interval is correct. The referent is the long-run coverage of a rule, a number such as 0.95 attached to a procedure, and it says nothing about a person's feeling of certainty or their belief in themselves.[1]

Unit

The long-run coverage rate of an interval procedure, such as 0.95

Recovery means

Not applicable

Measured by

Derived analytically from the sampling distribution, or checked by simulation counting how often the interval covers the parameter

A person's graded belief that a specific judgment is correct, expressed as a subjective probability attached to an answer or a forecast. The central question is calibration: whether the answers a person calls 90 percent likely turn out right about 90 percent of the time. Fischhoff, Slovic, and Lichtenstein showed in 1977 that people are often more certain than they are accurate, assigning extreme confidence to answers that are frequently wrong, and Lichtenstein, Fischhoff, and Phillips reviewed the calibration literature in 1982 and documented systematic overconfidence across many tasks. The referent is the fit between a stated probability and an outcome rate for particular judgments, not the coverage of a statistical procedure, and not a stable belief in one's own worth.[2][3]

Unit

A subjective probability attached to one judgment, from chance to certainty

Recovery means

Not applicable

Measured by

Calibration curves and over/underconfidence scores relating stated confidence to the proportion of judgments that prove correct

A relatively stable belief in one's own worth and general ability to do well, treated as a personality characteristic rather than a judgment about a named task. Shrauger and Schohn built the Personal Evaluation Inventory in 1995 to measure self-confidence in college students across areas such as academics, social situations, and physical appearance, and separated a global sense of confidence from confidence in any one domain. This meaning sits close to self-esteem and overlaps in wording with general self-efficacy, but its referent is a trait-like feeling of assurance about oneself, not a graded belief that a particular action can be executed, which is self-efficacy, and not a probability placed on an answer.[4]

Unit

A person's stable sense of assurance about their own ability and worth

Recovery means

Not applicable

Measured by

Self-report inventories such as the Personal Evaluation Inventory, summed across life domains

The expectation, held without much active deliberation, that familiar people and systems will behave as they normally do, so that one proceeds without weighing the risk. Luhmann distinguished this in 1988 from trust proper: confidence is the taken-for-granted reliance on continuity, such as that the currency will hold or the bridge will stand, where a disappointment is blamed on the world; trust is chosen in awareness of an alternative and of possible betrayal, where a disappointment is blamed on one's own decision. The referent is a social and everyday attitude of assurance toward institutions and others, counted in surveys of confidence in government or the economy, and it is not a coverage rate, a calibration score, or a personality trait.[5]

Unit

An expressed reliance that a person, institution, or system will hold up

Recovery means

Not applicable

Measured by

Survey items on confidence in institutions, and behaviour that proceeds without hedging against betrayal