Confidence intervals (CIs)

Lead Author(s): PeterBacchetti

These are used to indicate the uncertainty around an estimated or observed association, difference, or effect. Informally, a CI indicates the range of possible true values that are not too inconsistent with the observed data. Thus, we usually claim to have strong evidence against any value outside the CI, but not those inside it. In particular, negative results (claims for evidence against any substantial association, difference, or effect) should be supported by noting that one or both ends of the CI are too small to be of substantial biological, clinical, or scientific importance. See principles for interpreting results and P-value fallacy.

Although alternative levels are possible, 95% confidence intervals are most commonly used. These are (by definition) constructed by a method that will include the true value 95% percent of the time. Note, however, that the "95%" property pertains to the method and not to any specific interval that it produces. It is therefore incorrect to claim that there is a 95% chance that the true value falls within your CI. So use less specific, more informal phrasing when interpreting CIs.

More detailed technical explanation can be obtained at Wikipedia.