Egger Test Publication Bias: What It Can and Cannot Show

Egger test publication bias is a regression-based method for assessing whether study precision is associated with effect estimates in a meta-analysis. It is often used to examine funnel-plot asymmetry, but it does not establish that publication bias exists or identify why asymmetry arose.[1][2]

The result is best treated as one signal within a broader assessment of missing evidence and small-study effects. Interpretation depends on the effect measure, the number and spread of studies, and the context of the evidence; it should not replace comprehensive searching or investigation of potentially unreported results.[1]

What does Egger’s test actually test?

Egger’s test examines whether standardized effect estimates are associated with study precision; in its classic form, it tests whether the regression intercept differs from zero.[2]

The regression uses the effect estimate divided by its standard error as the outcome, and the reciprocal of the standard error as the predictor. The intercept is a test statistic for the association: it is not an estimate of how many studies are missing, nor is it a corrected pooled effect.[2]

This distinction matters because the method is often described informally as a “publication bias test.” More precisely, it tests for a pattern consistent with funnel-plot asymmetry or small-study effects. Even if that pattern is present, Egger’s test cannot determine whether selective non-reporting caused it.[1][2]

How should you interpret a significant or non-significant result?

A small p-value indicates evidence of an association consistent with asymmetry, not proof of publication bias.[1]

A non-significant p-value means the test did not detect asymmetry. It does not establish that the funnel plot is symmetrical or that missing evidence is absent.[1] The result should therefore be interpreted alongside the plot, the study characteristics, and the review’s search and reporting checks.

Avoid treating the p-value as a verdict on the validity of the meta-analysis. Instead, describe what the test detected and whether plausible features of the studies could explain the pattern. The estimate and uncertainty for the intercept can help readers assess the result; neither should be interpreted as a measure of missing studies.[1][2]

When is Egger’s test appropriate?

Cochrane advises using tests for funnel-plot asymmetry only when a meta-analysis includes at least 10 studies, as a rule of thumb, and not when study standard errors are similar.[1] This is a practical eligibility guide, not a guarantee that the test will be informative. Substantial heterogeneity may mean that more studies are needed.[1]

The original Egger test is also not suitable for every effect measure. Cochrane cautions that in some odds-ratio and standardized-mean-difference settings, effect estimates can be artefactually correlated with their standard errors.[1] Harbord or Peters tests are cited as alternatives for odds ratios, but the appropriate method depends on the outcome and analysis context.[1]

Before testing, check whether the studies have enough variation in precision for the regression to be meaningful, and whether the effect measure is suitable for the classic formulation. Prespecify the approach where possible, rather than trying several tests and selecting a significant result.[1]

Why can funnel-plot asymmetry occur?

Asymmetry can result from selective non-reporting, but it can also reflect differences in study methods, populations, or implementation; artefactual relationships between effects and standard errors; or chance.[1][2]

That range of explanations is why an asymmetry test cannot identify publication bias on its own. A plot may help reveal patterns that deserve investigation, but it does not establish their cause. Interpretation should consider whether smaller and larger studies differ in design or context, and whether the observed pattern is compatible with those differences.[1][2]

The direction of effect also matters when assessing small-study effects, so readers should not treat visual asymmetry as automatically indicating that smaller studies show larger benefits or harms.[2] Use the plot and test as prompts for contextual assessment, not as stand-alone evidence that studies are missing.

What should you report and investigate after the test?

Report the number of studies, effect measure, test method, intercept estimate and its uncertainty, and p-value.[1] Explain how the funnel plot, influential studies, and sensitivity analyses informed the interpretation.[1]

Also describe the limits of the test in the review’s context, including whether the study count and spread of standard errors made testing defensible. If multiple asymmetry tests were used, report them rather than presenting only the result that supports a preferred conclusion.[1]

Funnel plots and asymmetry tests are supplementary signals, not substitutes for a comprehensive search. Where records are available, compare study protocols or registry entries with publications to check for selectively unreported results.[1] A reproducible workflow can document the planned analysis, study count, method, and interpretation while keeping reviewers responsible for methodological judgement.

How is Egger’s test different from ROBINS-I?

Egger’s test assesses a pattern across studies, whereas ROBINS-I assesses risk of bias in a result from an individual non-randomized intervention study.[1][3] ROBINS-I can inform interpretation of small-study patterns, but it cannot determine whether studies or results are missing.[1]

Be clear about which version of ROBINS-I is being discussed. The familiar seven domains and low, moderate, serious, or critical judgements refer to the original tool; the official site describes ROBINS-I V2 as a draft subject to change, with six domains and revised guidance.[3][4][5] These assessments address different questions and should not be treated as interchangeable.

Key takeaways

• Egger’s test evaluates an association between study precision and effect estimates; it does not diagnose publication bias.[1][2]
• A small p-value is evidence consistent with asymmetry, while a non-significant result does not establish symmetry or absence of missing evidence.[1]
• Cochrane’s rule of thumb is to test only when a meta-analysis has at least 10 studies, and not when standard errors are similar.[1]
• Asymmetry may reflect selective non-reporting, study differences, artefacts, or chance, so interpretation needs context.[1][2]
• Report the method, study count, intercept and uncertainty, p-value, and how plots and sensitivity analyses informed interpretation.[1]

FAQ

What is Egger's test for publication bias in meta-analysis?

It is a regression test for an association between study precision and effect estimates, commonly used to assess funnel-plot asymmetry or small-study effects.[1][2] It does not establish that publication bias caused an observed pattern.

How does Egger's test relate to funnel plot asymmetry?

The test evaluates a statistical association that is commonly interpreted as evidence consistent with funnel-plot asymmetry.[1][2] A plot and test can flag a pattern for investigation, but neither identifies its cause.

What are the limitations of using Egger's test for publication bias?

The test cannot distinguish selective non-reporting from other explanations for asymmetry, and its suitability depends on the study data and effect measure.[1][2] A non-significant result also cannot rule out missing evidence.[1]

How should ROBINS-I results be used alongside Egger’s test?

ROBINS-I evaluates risk of bias in an individual non-randomized intervention study result, while Egger’s test examines a pattern across studies.[1][3] ROBINS-I may inform interpretation, but it does not determine whether evidence is missing.[1]