Publication bias analyses

ProMeta includes a set of analyses to detect and adjust for publication bias and small-study effects, in the app and in the CLI. For when and why to use them, see Evaluating publication bias. No single test settles the question: use several and read them together.

Tests of small-study effects

• Egger's regression: classic, mixed-effects or cluster-robust (sandwich) variant, for dependent effects.
• Begg and Mazumdar's rank correlation test.
• Peters' test: funnel-plot asymmetry for log odds ratios, from the 2×2 tables.
• Time-lag bias (decline effect): regression of the effect size on the years since the first study.

Adjusted estimates and sensitivity

• PET-PEESE: effect adjusted for small-study effects.
• Trim-and-fill: imputes the studies missing from the funnel plot and gives the adjusted effect.
• Limit meta-analysis: treatment effect adjusted for small-study effects, with radial plots.
• Selection models and the Copas selection-model sensitivity analysis.
• Fail-safe N, including Orwin's criterion.

Plots

These analyses produce the funnel plot and the radial and scatter plots as editable SVG.

Methods assume independent effects unless stated; with several effects per study, ProMeta warns. Results must be checked independently (see the disclaimer).