Confidence interval interpretation: Introducing the confidence interval function

Confidence intervals are often used incorrectly to perform hypothesis tests in the biomedical sciences. In this paper, the confidence interval function is introduced. It is demonstrated that its use helps to prevent errors in inference caused by a reliance on null hypothesis significance testing, and in particular, on the dichotomisation of results into “significant”, and “nonsignificant” categories. By way of three examples of confidence interval function use, it is shown that dependence on p values for inference can lead to a) unfounded claims of conflicting results, b) overlooking effects of practical importance, and c) incomplete appraisal of study results. Authors of epidemiologic studies are encouraged to routinely report confidence
intervals and use sketches of confidence interval functions to interpret them.