Statistics Glossary → Egger's and Begg's tests
Egger's and Begg's testsconcept
Tests in meta-analysis of whether the funnel plot is asymmetric, i.e. whether small studies systematically report different effects from large ones.
When there is publication bias, small studies with non-significant results go unpublished; the small studies that remain tend to report large effects, and the funnel plot is skewed to one side. Egger's test checks this asymmetry by regression: the studies' effect sizes are regressed on their standard errors (equivalently, the standardized effect on precision); if the slope or intercept differs significantly from zero, a small-study effect is concluded. The Begg-Mazumdar test asks the same question nonparametrically, through Kendall's τ between the standardized effects and their variances.
Both tests have little power when there are few studies; they are usually recommended only with at least 10 studies, and Begg's test is even less powerful than Egger's. A significant result does not prove publication bias: asymmetry can also arise because small studies were done in genuinely different patient groups, from lower quality, from heterogeneity, or by chance. 'Small-study effect' is therefore the more accurate phrase. For measures such as the odds ratio, the effect and its standard error are mathematically related, so Egger's test can give false alarms.
An everyday example
There are 14 studies of a dietary supplement. Three large studies with over 500 participants report a small effect, while nearly all the studies with 20-30 people show large, positive results, and there are no small studies with negative findings. Egger's test gives p = 0.01; this strengthens the suspicion that small negative studies may have gone unpublished but does not prove it on its own.
In MerQur
Available in meta-analysis as 'Publication bias: Egger' (a yᵢ ~ SE regression using the model's own method, interpreted when k ≥ 10) and 'Publication bias: Begg' (Kendall's τ between standardized effects and variances).