Definition
A confidence interval is a random interval procedure designed to cover a fixed parameter at a stated long-run frequency under specified sampling assumptions. A 95 percent confidence interval is not, in the frequentist definition, a 95 percent probability statement about the already fixed parameter after the interval is observed. Method, dependence, selection and target parameter determine its meaning.
How it is calculated
A symmetric normal-approximation confidence interval at level 1 minus alpha is estimate theta-hat plus or minus the normal critical value z times its standard error.
| Symbol | Meaning | Unit |
|---|---|---|
| CI | confidence interval | theta units |
| alpha | tail probability | probability |
| theta | target parameter | parameter units |
| z | standard normal critical value | dimensionless |
| SE | standard error | theta units |
Interpretation
Imagine repeating the whole study and building an interval each time. The method is designed so a chosen fraction of those intervals covers the fixed target. Once one interval is observed, that repeated-sampling coverage property does not become a literal probability assigned to the target.
Assumptions and conventions
-
The interval procedure and coverage target are stated.
-
The standard-error and critical-value approximation is appropriate for the sampling process.
Same units as theta; alpha and z are dimensionless.
Limitations and common misreadings
-
A confidence interval is not automatically a credible interval.
-
Selection after looking at results can invalidate nominal coverage.
References
- Larry Wasserman. (2004). All of Statistics: A Concise Course in Statistical Inference. Springer Texts in Statistics. Chapter 9, Parametric Inference. doi:10.1007/978-0-387-21736-9. Source

