Let X be a random sample of size n from the population PX;θ with state space S, and gL,gU:Sn→R be statistics with gL(x)≤gU(x) for all x∈Sn. An interval estimator of θ is the random interval [gL(X),gU(X)], which is intended to contain the true parameter θ.
Let [gL(X),gU(X)] be an interval estimator of θ and x be an observed random sample. An interval estimate of θ is the realised interval [gL(x),gU(x)], which is equivalent to the statement gL(x)≤θ≤gU(x).
Let gL and gU be statistics. A one-sided interval estimate of θ is an interval estimate of the form (−∞,gU(x)] or [gL(x),∞), which is equivalent to the statement θ≤gU(x) or θ≥gL(x) respectively.
Let [gL(X),gU(X)] be an interval estimator of θ. The coverage probability of the interval estimator is the probability that θ lies within the random interval, which is the joint probability that the lower statistic is at most θ and the upper statistic is at least θ.
P(θ∈[gL(X),gU(X)])=P(gL(X)≤θ,gU(X)≥θ)
The coverage probability is a function of θ.
Let [gL(X),gU(X)] be an interval estimator of θ. The confidence coefficient of the interval estimator is the infimum of the coverage probability over all possible values of the true parameter.
θ∈ΘinfP(θ∈[gL(X),gU(X)])
The coverage probability is often constant as a function of θ, in which case the coverage probability and the confidence coefficient are equal.
Let [gL(X),gU(X)] be an interval estimator of θ and α∈(0,1). The interval estimator is a 100(1−α)% confidence interval for θ if and only if its confidence coefficient is at least 1−α.
P(θ∈[gL(X),gU(X)])≥1−α∀θ∈Θ
A confidence interval has probability at least 1−α of containing θ over repeated samples before the data are observed. A realised confidence interval either contains θ or does not, and the confidence coefficient is not a probability statement about θ given the observed data.