MerQuR+Statistics Glossary

Statistics Glossary → Publication bias

Publication biasconcept

The tendency of the published literature to overstate the true effect because studies with significant or expected results are more likely to be published.

Significant findings find their way into journals more easily, whereas non-significant results often stay in the file drawer. Because a meta-analysis can only combine the studies it can find, the pooled effect size can be systematically inflated. The traces show up especially in small studies: a small study needs to find a large effect to get published.

The main tools are the funnel plot (effect size against precision), Egger's regression test and Begg's rank correlation test (which test funnel asymmetry), the trim-and-fill method (which gives a corrected estimate by imputing studies thought to be missing) and the fail-safe N (how many unpublished null studies would be needed to make the result non-significant). Funnel asymmetry is not proof of publication bias: genuine differences between small and large studies, heterogeneity or chance can also create it. These tests have low power when there are few studies (e.g. fewer than 10).

An everyday example

Twelve small studies showing that a food supplement improves memory have been published, while most studies that found no effect were never published. Someone looking only at the published studies would conclude the supplement works quite well, whereas if all the studies ever done could be seen, the average effect might be much smaller.

In MerQur

The Meta-Analysis analysis offers publication bias assessments (Egger, Begg, trim-and-fill, fail-safe N).

Analyses where this concept appears

Meta-Analysis

Related concepts

Funnel plotEgger's and Begg's testsTrim-and-fill