Table 1.

Recommendations for Meta-Analysis and Application of p-Uniform and p-Curve

1. Check for evidence of p-hacking in the primary studies. In case of strong evidence or strong indications of p-hacking, be reluctant in interpreting estimates of traditional meta-analytic techniques and p-uniform and p-curve because their effect-size estimates may be biased in any direction depending on the type of p-hacking used. 2. Apply fixed-effect and random-effects meta-analysis, as well as p-uniform or p-curve, and report results conforming to the Meta-Analysis Reporting Standards (MARS; American Psychological Association, 2010, pp. 251–252.). 3. Check for direct or indirect evidence of publication bias. In case of evidence of publication bias, interpret results of p-uniform or p-curve rather than those of fixed-effect and random-effects meta-analysis; in the absence of such evidence, interpret results of fixed-effect and random-effects meta-analysis. 4. Set the effect-size estimate of p-uniform or p-curve equal to zero if the average p value of the statistically significant studies is larger than .025 5a. If the effect size is homogeneous or if the heterogeneity is small to moderate (I2 < 0.5), interpret the estimates of p-uniform and p-curve as estimates of the average population effect size; otherwise, these methods result in overestimates of average population effect size and should be interpreted as estimates of the average true effect size of only the set of statistically significant studies. 5b. In case of substantial heterogeneity (and if desired), create homogeneous subgroups of primary studies on the basis of theoretical or methodological considerations to estimate with p-uniform and p-curve the average population effect size underlying the studies in each subgroup