Watch on YouTube

5:18

Effect sizes, confidence intervals, and how results get misread

A confidence interval does everything a p-value does and more. It gives the range of effects compatible with the data, so it carries magnitude and precision at once, while a p-value says only how surprising the data would be if there were no effect. That difference explains the most damaging misreading in the literature. When an interval includes no effect, the study could not separate benefit from harm, which is a different statement from showing the intervention does not work. We judge importance, and not only significance.

Also covered: visual inspection with acetic acid · heterogeneity across studies · risk ratio · clinically meaningful effect · 0.05 significance threshold