When to p and when not to p

Abstract

The p-value was proposed in the early 20th century as a potentially useful metric for statistical inference and was defined as “the probability of the observed result, plus more extreme results, if the null hypothesis were true,” in the context of a formal statistical testing process. A century later, the scientific community uses it extensively, mostly inappropriately, and often interprets it incorrectly. This editorial briefly reviews the history of the p-value, provides how to properly interpret it, and recommends when to use it and when not to use it.