Results of negative binomial regressions of the number of reporting errors per paper.

Negative binomial regressions (N = 49; with a log link) of the number of misreporting errors per paper on the log of the number of test statistics, square root of the average p-value of genuinely significant results within the papers, and whether or not the data were shared for reanalysis. Analyses were estimated in SPSS 18.0 (The Zelig package in R gave similar results) with a robust variance estimator to deal with the possibility that errors were dependent within papers. Natural log and square root transformations were used to improve predictors' normality.