A 2013 'coin-size estimation' study was retracted after Simonsohn's forensics found its per-condition values too similar across experiments to be random samples (p<.000025)

research-integrityfraud-detectionstatisticsdata-coladareplicationretraction

In a September 2013 post on Data Colada, Uri Simonsohn described a third case caught by the same "excessive similarity" forensic signature — the method behind his earlier Sanna and Smeesters catches: a coin-size–estimation study whose corresponding author was Wen-Bin Chiou. The distributions of reported per-condition values across two separate experiments were, on Simonsohn's inspection, "almost identical" — a match that honest independent sampling from two experiments should not produce. Simonsohn ran a bootstrap simulation directly against the reported value lists and stated the result plainly: "The hypothesis that these data came from random samples is rejected, p<.000025." The paper was subsequently retracted.

The author's response is part of what makes the case a clean instance. Chiou attributed the pattern to a research assistant having sorted the data; Simonsohn rebutted this in the same post, noting that "the similarity of data is based on the frequency of values across samples, not their order, so sorting does not explain that the data are incompatible with random sampling" — i.e. the diagnostic is a property of which values appear and how often, not of their arrangement, so re-ordering cannot manufacture or dispel it.

This is a fourth concrete "too clean to be real" case for the vault's research-integrity cluster, independent of Sanna/Smeesters and of the superconductivity case mapped in moc-too-clean-to-be-real. It is the same inference Fisher applied to Mendel's peas and that Talmudic law encodes for judicial unanimity — real independent processes carry noise, so its absence is a signature of a defect in provenance. See observation-suspicious-perfection-independence-absence-signals-defect for the general law.