Table 2. Problems in quantifying p-hacking and evidential value from a p-curve using text-mined data.

Cases where p-hacking not detected by binomial test	Cases where right skew not due to evidential value
P-values are reported as p < .05 and so excluded from analysisa	Where p-values used to confirm prior characteristics of groups being compareda,b
Limited power because few p-values between .04 and .05	Where p-values come from confirming well-known effects, e.g., demonstrating that a method behaves as expecteda,b
Where p-values ambiguous because rounded to two decimal placesa	Where ‘double-dipping’ used to find ‘best’ data to analyse
	P-values from model-fitting or testing of assumptions of statistical tests (where low p-value indicative of poor fit, or failure to meet assumptions)a,b

Notes.

^aProblems that can potentially be overcome by analysing data from meta-analyses.

^bProblems that are less likely to affect text-mined data from Abstracts.