Audit

Why stats integrity matters

Nuijten et al. (2016) scanned 30,000+ psychology papers and found that ~50% contained at least one inconsistent reported p-value, with ~13% containing a decision-flipping inconsistency. A 2024 follow-up by Nuijten & Wicherts found that integrating statcheck-equivalent checks into peer review correlated with a 4.5× reduction in reporting errors at submission.

Reviewers and meta-analysts run these checks routinely. The point of this tool is to put the same checks in front of you before submission, with plain-English flag explanations, so a fixable arithmetic error never becomes a public correction notice.

Limitations

• Reported stats only. We can recompute p only when test statistic + df + reported p are all present in the paste. We can’t check raw data we don’t see.
• Corrections suspend recompute. If you tick “Bonferroni / FDR / Holm,” we suppress p-recompute (the math is family-specific) and run only descriptive checks (GRIM / GRIMMER / DEBIT).
• GRIM range. GRIM is informative for N ≤ 200 with integer-bounded scales (Likert, counts). Above that, the rounding band swallows the scale and we surface “skipped” rather than false-positive.
• GRIMMER upper-bound only. V1 flags SDs that exceed the theoretical maximum given the integer scale and N. The full Anaya 2017 algorithm additionally checks integer-partition consistency; that fuller enumeration ships in a future update. Flagged cases are correct; some genuinely-impossible sub-maximal SDs may currently pass.
• Not a substitute for statcheck-on-PDFs. For batch checks of full manuscripts (PDF / DOCX), use the official statcheck R package or the upcoming statcheck Word add-in. Audit is built for the “paste my Results paragraph and tell me what’s wrong” loop.