From significance to divergence: guiding statistical interpretation through language

Abstract

Purpose of review P values have long been central to medical research reporting, with the term ‘‘statistical significance’’ and a P value threshold of 0.05 being in common use since 1925. Despite a century of use, P values remain a topic of significant controversy and debate, particularly regarding their proper application and frequent misinterpretation. Much of this confusion stems from adoption of the everyday words ‘‘significance’’ and ‘‘confidence’’ as a label for the statistical concepts that are only loosely connected to their common meaning, subsequently exposing such misleading labels to a wide audience unaware of the disconnect. Recent findings To resolve this ambiguity, we take a look at the existing literature, conclude that this is a language issue and propose replacing ‘‘significance’’ with ‘‘divergence’’ to highlight the data’s divergence from the hypothesized null model. In addition, we propose renaming the ‘‘1-α% confidence interval’’ to ‘1-α% uncertainty interval’’ which would more accurately convey its role in representing uncertainty about the possible data-generating models for the observed data. Summary The revised terminology will help researchers and readers better understand P values and uncertainty intervals, aims to reduce reporting bias (especially for nondivergent results), and will temper unrealistic replicability expectations. It would also minimize misinterpretation and over-interpretation, promoting a clearer, more nuanced understanding of their use in statistical reporting while addressing ongoing misuse controversies.