When the tool becomes the tyrant.

This no-nonsense editorial asks just why exactly stats, such as p-values, lead scientists to deny differences that everyone else can easily see. This group of experts lists some suggestions on changing the paradigm of statistical interpretations, receiving hundreds of official endorsements within a matter of hours. First point of annoyance: claiming there is no difference between arms when the p-value is > 0.05. A more tolerable interpretation? The results failed to surpass a preordained degree of likelihood that a true difference exists (much less catchy, we know). Another point of contention: the extreme p-value dichotomy of > versus ≤ 0.05 that works to overemphasize “significant” results and de-value “non-significant” ones. In fact, they are so sick of over-confident conclusions, they propose re-coining "confidence" intervals as "compatibility" intervals. TBL: View the p-value as what it is, one of countless points along a continuum of certainty. | Amrhein, Nature 2019