Check a reported p-value

How the check works

The p-value is a deterministic function of the test statistic and its degrees of freedom. Given t and df, there is exactly one correct p — no simulation, no lookup table, no rounding shortcut involved. This page recomputes that exact value from the underlying distribution (Student's t, F, or chi-square) and compares it to what you actually reported, the same logic behind statcheck, the R package that found reporting inconsistencies in roughly half of published psychology articles by doing exactly this at scale across the literature.

What counts as a mismatch

Your stated precision is respected, not silently overridden. If you report p = .03, any recomputed value that rounds or truncates to .03 is treated as consistent — both are legitimate ways a value could have been reported, and treating only one as correct would produce false flags on perfectly honest rounding. A flag is raised only when no reading of your reported value, at the precision you used, can produce the computed one.

Why a flag is not necessarily an error

Legitimate causes include one-tailed tests reported without saying so, Greenhouse–Geisser or Welch corrections that adjust degrees of freedom, multiple-comparison adjustments applied to p but not shown in the write-up, and statistics that were themselves rounded before p was computed from them. The check tells you a number needs a second look — it does not tell you the analysis is wrong, and it can't see your data or your design, only the numbers you typed in.

The distinction that matters

A rounding-level discrepancy is cosmetic — the kind of thing a careful copyeditor would catch. A decision error, where the reported and computed values fall on opposite sides of your alpha, is different in kind: it changes what the paper's conclusion actually is. Those are flagged separately and more prominently, because a reviewer or a reader will treat them very differently once they notice.