Fig. 1.

Statistical power, Type S and M errors as a function of the ‘true’ effect size (the alpha level is fixed at 0.05). The generic form of effect sizes (e.g. SMD, lnRR, Zr) are simulated from 0 to 1 with a fixed standard error (0.25). These panels (A–C) show that studies investigating larger true effects have higher power (A) and lower rates of Type M (B) and Type S (C) errors. If a study suffers from publication bias, the effect size is likely to be exaggerated, and consequently, the corresponding statistical power, Type M and S errors would be underestimated