Egger’s Test and Funnel Plot Asymmetry, Explained

A funnel plot shows publication bias visually, as an asymmetric scatter of studies around the pooled estimate. Egger's test formalizes this into a specific statistical test, and understanding exactly what it measures, and where it breaks down, prevents both over-reliance on a nonsignificant result and over-interpretation of a significant one.

What the test actually does

Egger's test runs a regression of each study's standardized effect estimate against its precision, typically the inverse of its standard error, and tests whether the resulting regression line's intercept differs significantly from zero. In a perfectly symmetric funnel plot, this intercept should be close to zero; a significant deviation suggests asymmetry beyond what chance alone would produce.

Why this specific test, rather than just eyeballing the plot

Visual inspection of a funnel plot is genuinely useful as a first pass, but it's also subjective, and different reviewers can reasonably disagree about whether a given plot looks "asymmetric enough" to be concerning. Egger's test provides a formal, reproducible statistical answer to the same underlying question, giving readers a specific p-value rather than relying on a subjective visual judgment call.

The study-count limitation, stated plainly

Egger's test performs poorly with fewer than ten included studies, producing unreliable results in either direction -- it can fail to detect real asymmetry when studies are few, and it can also produce a spuriously significant result from a small number of studies that happen to scatter unevenly by chance. Running Egger's test on eight or nine studies and reporting the resulting p-value without this caveat is a common but avoidable overstatement of what the test can actually tell you at that sample size.

Asymmetry is not proof of publication bias specifically

A statistically significant Egger's test tells you the funnel plot is asymmetric; it does not, by itself, prove the cause is publication bias specifically. Genuine heterogeneity in the true effect across different study contexts, small-study effects unrelated to selective publication -- such as smaller studies genuinely using less rigorous methods for reasons other than selective reporting -- and simple chance can all produce a similar asymmetric pattern. Egger's test result should be interpreted as evidence of asymmetry requiring further explanation, not as a direct, confirmed measurement of publication bias.

Alternatives and complements to Egger's test

Begg's rank correlation test is an older, generally less statistically powerful alternative, still occasionally reported alongside Egger's test in older literature. Trim-and-fill provides a complementary sensitivity analysis, estimating how many studies might be "missing" to restore funnel plot symmetry and recalculating the pooled estimate as if they were present -- useful as a sensitivity check on how much your conclusion might shift under a specific assumption about missingness, though it should not be presented as a corrected, superior estimate replacing your original pooled result.

Egger's test with binary outcome data

The standard formulation of Egger's test was developed primarily with continuous outcome effect sizes in mind, and applying it directly to odds ratios or risk ratios from binary outcome data has some known statistical complications, since the standard error of a log odds ratio is mathematically related to the effect size itself in a way that can artificially induce apparent asymmetry. Modified versions of the test, such as the Harbord test, were developed specifically to address this issue for binary outcomes, and using an appropriately modified test for binary data, rather than the standard continuous-outcome version, is worth checking for in whatever statistical software you're using.

Reporting Egger's test transparently

State your Egger's test result alongside the actual funnel plot, not as a replacement for showing it, since the visual plot and the formal test together give a reader more complete information than either alone. If your study count falls below the roughly ten-study threshold where the test becomes unreliable, say so explicitly rather than reporting a p-value without this important caveat attached.

A practical checklist before reporting

Confirm you have at least roughly ten included studies before treating Egger's test result as informative on its own. Confirm you're using an outcome-type-appropriate version of the test if your data is binary rather than continuous. Present the funnel plot itself alongside the test result, not instead of it. And discuss what the asymmetry, if found, might actually reflect in your specific review's context, rather than defaulting to "publication bias" as the automatic explanation without considering the genuine alternatives.

Egger's test in software practice

Most standard meta-analysis packages implement Egger's test as a built-in function, typically alongside the funnel plot itself, making it straightforward to generate both together as a single analytical step rather than as separate, potentially inconsistent add-on calculations. Running both from the same software call, on the same underlying dataset, also reduces the risk of a data-handling inconsistency between the visual plot and the reported statistic. Given how frequently reviewers specifically check whether a reported Egger's test result actually matches the accompanying funnel plot's visual appearance, running both together from a single, consistent analysis step is a small efficiency that also meaningfully reduces the risk of an easily avoidable internal inconsistency in your results section. Building this pairing into your standard analytical workflow, rather than treating the funnel plot and the formal test as separate, occasionally mismatched outputs generated at different points in your process, is a small procedural habit with a genuinely outsized payoff in manuscript consistency. It also, as a secondary benefit, saves genuine time during your own final pre-submission review, since you are checking one consistent output rather than reconciling two separately generated pieces of analysis after the fact. In the end, the goal of any publication bias assessment, Egger's test included, is giving your reader an honest, well-supported sense of how much confidence to place in your pooled estimate, not simply producing a p-value to include as a checklist item. Approached this way, the test becomes a genuine analytical aid rather than a procedural formality completed simply because a reviewer might otherwise ask for it. That distinction, small as it sounds, genuinely changes how the whole exercise gets approached and reported.