When you’re line-editing Results, how far do you go in revising overstated significance without supplanting peer review? I routinely flag cases like p=0.051 reported as p<0.05 and ask for raw data or R scripts (statcheck often catches one or two inconsistencies per manuscript), but I want to keep the language precise and ethically neutral — what wording do you use?
I use a neutral template: “p=0.051; does not meet the prespecified α=0.05; report effect size and 95% CI,” and if they push for “trend” I mark it as descriptive only, no directional claims. Do you also ask them to cite the preregistered α or link to a justification, e.g., the ASA statement for context (https://www.amstat.org/asa/files/pdfs/p-valuestatement.pdf)? Close only counts in horseshoes.
Quick example: I add a parenthetical that anchors the claim to the analysis plan and multiplicity, e.g., “p=0.051 (two‑sided; unadjusted); exploratory per protocol,” and move any spin to Discussion with effect size/CI. If preregistered, I point to the record or ASA guidance (https://www.amstat.org/asa/files/pdfs/p-valuestatement.pdf) to keep the tone neutral. Would you standardize that parenthetical to note the correction method (e.g., Holm) when the manuscript hasn’t stated it?
But building on @twilli27: for near‑α results I add a one‑liner like “does not meet the prespecified threshold; estimate = X, 95% CI [A, B]; exploratory unless multiplicity‑adjusted,” and I request the exact test, tails, and software/version in a footnote — no “trend” fairy dust. Would you consider a brief house‑style note that prefers “compatible with no effect” for these cases?