From STA238 Probability Statistics and Data Analysis II

Confidence intervals

Judging an estimator told you how your estimator sits relative to the parameter: where it centres, and how far it spreads. Your estimator came from a random sample, so it has a distribution and would have come out differently on a different sample.

Now use that to say how far the parameter is from what you got.

θ^±cse(θ^)\hat\theta \pm c \cdot \mathrm{se}(\hat\theta)

  • Estimate — what you computed
  • Standard error — the spread, from page 02, with sample quantities substituted for the unknown population ones
  • cc — a percentile, looked up in whatever distribution the estimator follows

What you write: “We are 95% confident that the true value of θ\theta is between \ell and uu.”

Not “there is a 95% probability that θ\theta lies in (,u)(\ell, u).” θ\theta is fixed, so it has no probability. The 95% is a property of the procedure: 95% of all possible samples of size nn yield an interval containing θ\theta.

Finding cc for the mean

DataConditionUse
Normalσ\sigma knownzz
Normalσ\sigma unknownt(n1)t(n-1)
Unknownnn largezz, by the CLT
Unknownnn smallbootstrap

Finding cc for other parameters

ParameterDistribution
Meanzz or t(n1)t(n-1), per above
Variance or σ\sigmaχ2(n1)\chi^2(n-1), not symmetric, so two separate lookups
Proportionzz, by the CLT