Fig 1. Frequency distribution of numerical P-values (P-curve).

Two hypothetical examples of P-curves show the expected distribution if (a) selection bias is present or (b) certain types of reverse P-hacking occur (see [12]). Data collected from articles testing the effect of confounding variables between treatment(s) and control groups are shown in (c). In (c), darker red points represent a higher frequency of P-values/bin among the randomised datasets. The horizontal red line denotes the theoretical even distribution across 20 bins (188 articles in 20 bins = 9.4 articles/bin). The grey rectangles emphasize intervals of interest in the two hypothetical scenarios. We only used P-values presented to at least two decimal places. The full data file of the 1,805 articles consulted and the 250 from which data were extracted and the R scripts used for the analysis and to generate the figures are available at https://figshare.com/s/f3bb7dfefdaa8976d3a1 or https://osf.io/au7yc/.