A p value is not an independent number: it follows from the test statistic and its degrees of freedom. When it does not match, either a digit was mistyped or the analysis is not the one the text describes.
A p value is the area left in the tail of a distribution beyond your test statistic. Given the t and its degrees of freedom, the distribution is fully determined and so is the area. There is no room for interpretation: either the number is that one or it is not.
That turns every results line into an equation with three unknowns, all three of which get published: the statistic, the degrees of freedom and the p value. If the three do not fit together, there is enough information to know something is wrong, even without knowing which of the three is at fault.
It is the idea behind statcheck, used for years to audit the psychology literature at scale. This page does the same thing, for one line, in your browser.
Almost always a typo. A transposed digit copied from the software output, a lost minus sign, degrees of freedom left over from the previous model. Finding a mismatch accuses nobody of anything, and it is worth saying so if you are writing to an author.
Sometimes a different analysis from the one described. Degrees of freedom that do not match the reported N usually mean undeclared missing cases, or a one-tailed test written up as two-tailed.
And occasionally a misread rounding. If the text says p = .03 and the computation gives .0251, it matches: the author rounded. That is why the checker above tolerates rounding instead of demanding an exact hit.
p = .000. It does not exist: a p value is an area and cannot be zero. It is what SPSS prints when rounding to three decimals, and the correct form is p < .001. Worked through in how to report p values in APA 7.
Chi-square without its N. It is the only test carrying the sample size inside the parentheses, and without it the value cannot be interpreted: χ²(1, N = 200) = 8.45. See how to report chi-square.
Decimal degrees of freedom in a repeated-measures ANOVA. Not an error: they come from the Greenhouse-Geisser correction. See Mauchly’s test of sphericity.
No, and it is better not to pretend otherwise. The check says the three do not fit together, but any one of them could be the wrong one. If you are writing to an author, the honest move is to flag the inconsistency without assigning blame.
Almost never. Large-scale audits of the literature find inconsistencies in a high proportion of papers, and the majority explanation is transcription error. Fraud leaves different traces and is not diagnosed from a single line.
The usual causes are a one-tailed test not described as such, a correction (Bonferroni, Greenhouse-Geisser) applied to the p value but not to the published degrees of freedom, or a statistic from a model whose missing cases reduced the N.
No. The computation runs in your browser with the same library the Professional report uses: nothing is sent to any server and nothing you type is stored.
Yes, and it is a common use in journal clubs. Bear in mind that if you are reviewing an unpublished manuscript for a journal, that text is confidential: these fields send nothing anywhere, but you should not upload the full manuscript to any service.