Table 1 presents the results of the meta-analyses under both approaches. In both approaches the population effect size was estimated accurately as 0 (first row). However, it took on average much less published studies in publishing everything (3.84) than in selective publishing (8.36). Note that a vast number of on average 114 studies was needed to conclude that the effect is ‘smaller than small’ in selective publishing, which is about 30 times as many studies as in publishing everything. Hence this scenario also demonstrates that publishing everything is more effective than selective publishing. Another characteristic of selective publishing is again the heterogeneity in effect sizes (see the rows on ). The (absence of) heterogeneity is estimated correctly in publishing everything ( is 0 in 59.4% of the runs, and its average across runs is close to zero), but not in selective publishing ( is never zero, and its average corresponds to a considerably large standard deviation of effect size equal to .268). Note that the heterogeneity under selective publishing does not reflect heterogeneity of the underlying population effect size, but is an artifact of selective publishing.
