Fig 4. Correlations between effect size, variation and statistical power.

(A) Correlation between normalized effect size and mean sample size. No correlation is found (r = 0.0007, p = 0.99; r = -0.26, p = 0.64 after adjustment), although sample size variation is limited. (B) Correlation between normalized effect size and coefficient of variation. Correlation of the whole sample of experiments yields r = 0.37, p<0.0001* (n = 336; r = 0.32, p<0.001 after adjustment for freezing levels). (C) Correlation between normalized effect size and statistical power based on upper-bound effect size of 45.6%. Correlation of the whole sample of experiments yields r = -0.12, p = 0.03 (r = 0.11, p = 0.84 after adjustment for freezing levels), but distribution is skewed due to a ceiling effect on power. (D) Correlation between normalized effect size and statistical power based on intermediate effect size of 37.2%; r = -0.16, p = 0.003* (r = -0.16, p = 0.48 after adjustment). (E) Correlation between normalized effect size and statistical power based on lower-bound effect size of 29.5%; r = -0.21, p<0.0001* (r = -0.1, p = 0.06 after adjustment). Asterisks indicate significant results according to Holm-Sidak correction for 28 experiment-level correlations.