Paste the results section of a paper. We find t(df)=…,
F(df1,df2)=…, χ²(df)=… and r(df)=…
tests, recompute p from the statistic, and flag the lines where the
reported p does not match. Everything runs in your browser — nothing is
uploaded.
Paste the results
Results
Reported
Test
Reported p
Computed p
Verdict
«Grossly inconsistent» — the computed p falls on the other side of
α = 0.05 from what the paper reports (Nuijten et al. 2016). «Minor
inconsistent» — the significance decision stands, but the printed
p-value disagrees by more than rounding.
What this is
Reporting errors in psychology papers are common: the Nuijten et al.
2016 scan of 30 000 papers found that one in eight reported p-values
implied a different significance decision from the one printed. The
authors released an R package, statcheck, that does the
calculation; this page is a small, browser-only version so you can run
it on your own manuscript (or a submitted paper) without installing R.
Only the four classical tests (t, F, χ², r) with their degrees of
freedom spelled out in parentheses — the same shape statcheck
parses.
The computation uses an incomplete-beta and incomplete-γ
implementation (Numerical Recipes 6.2, 6.4). Our test suite compares
against R's pt, pf and pchisq
to within 10⁻⁴.
«p < .05» and the reported statistic are both read verbatim.
One-tailed tests, corrected p-values (Bonferroni, Holm, BH) and
tests reported without df are out of scope — the page silently
skips them.