Meta-analysts are limited to the consiseration of those experiments they can find in the published research literature. Unfortunately, not all experiments are published, either because the experimenters do not submit them for publication or because the journal editors turn them down. This is a serious problem, because a disproportionate number of the unpublished studies are those that fail to confirm the experimental hypothesis, those with a p-value greather than .05. This means that the sample of studies considered by the meta-analyst tends to be biased in favor of the experimental hypothesis. To solve this problem, meta-analysts have developed what they call a file-drawer statistic. This statistic estimates the number of unpublished studies (presumably locked away in experimenters' file drawers), each with a p-value of .50 (pure chance), that would be needed to raise the p-value of the enlarged composite to .05. When this number of experiments is absurdly large, as it often is, the meta-analyst concludes that the file drawer effect is not a serious problem after all and the conclusion about the reality of the experimental hypothesis can be sustained.
Tillbakahållande av experiment med negativa resultat
Om experiment med positiva resultat rapporteras oftare än de med negativa resultat blir det sammanslagna resultatet högre
Alias: file drawer effect, file-drawer statistics, supression of null results och tillbakahållande av experiment med negativa resultat
normal