Why we need confidence intervals

Abstract

The estimation approach to statistical analysis aims to quantify the effect of interest as an "estimate" of a clinically relevant quantity and to quantify the uncertainty in this estimate by means of a confidence interval (CI). As such, results expressed in this form are much more informative than results presented just as p values. This article focuses on the principles rather than the mathematics of CIs and discusses interpretation of CIs and some common misuses. CIs can be constructed for almost all analyses. They are especially useful for avoiding misinterpretation of nonsignificant results of small studies. CIs should be provided routinely for the main results of trials and observational studies.