Funnel Plot

A funnel plot is a simple scatter plot with a serious job: checking whether the studies pooled in a meta-analysis form a trustworthy body of evidence. Each dot is one study. The horizontal axis shows that study's estimated effect size — how big a difference it found. The vertical axis shows the study's precision, usually its standard error or sample size, with the most precise studies at the top. Because small, imprecise studies scatter widely while large, precise ones cluster near the true effect, a complete and unbiased set of studies should trace a symmetric shape that narrows toward the top — an inverted funnel. The name describes exactly what an honest evidence base ought to look like when you plot it. The funnel is a visual audit of the studies feeding a conclusion.

The reason to draw one is publication bias — the tendency for studies with striking, positive, statistically significant results to get published while null or unfavorable findings quietly disappear into file drawers. If the small studies that found nothing were never published, the bottom of the funnel loses its left or right side, and the plot tilts. That lopsided shape is the warning. Funnel-plot asymmetry can also flag small-study effects, where smaller trials systematically report larger effects than big ones for reasons beyond chance, such as lower quality or selective reporting. A funnel plot does not prove bias by itself, and asymmetry has innocent explanations too, but a badly skewed funnel is a red flag that the pooled estimate may overstate the truth. It turns an invisible gap in the literature into something you can see.

A funnel plot is easy to confuse with a forest plot, its constant companion in meta-analysis, but they answer different questions. A forest plot displays the actual result of the meta-analysis: each study appears as a point estimate with its confidence interval, stacked in a column, with a pooled summary — the diamond — at the bottom. It shows what the studies found and how they combine into an overall effect. A funnel plot, by contrast, is diagnostic. It does not report the summary effect; it interrogates whether the set of studies is complete and even-handed. The forest plot answers what the combined result is, while the funnel plot asks whether the studies behind it are a biased sample. You read the forest plot for the conclusion and the funnel plot for the caveat.

The two are complementary, and good meta-analyses show both. A forest plot with a tight, confident diamond looks reassuring, but if the accompanying funnel plot is badly asymmetric, that confidence may rest on a skewed literature that omits the studies which found nothing. Formal tests, such as Egger's regression test, put a number on the asymmetry a funnel plot shows by eye, and methods like trim-and-fill try to estimate what the pooled effect would be if the missing studies were restored. None of this replaces judgment: a funnel plot needs a reasonable number of studies to be readable, and with only a handful of trials its shape means little. Read the forest plot for the effect, the funnel plot for the integrity of the evidence, and the formal tests to sharpen what your eyes suspect.

Use a funnel plot whenever you pool studies and want to know if the evidence base is skewed before you trust the summary. Plot effect size against a sound measure of precision, include every study you can find rather than only the published, tidy ones, and look honestly at the shape — a clean inverted funnel is reassuring, a lopsided one is not. Back the visual read with a formal asymmetry test when you have enough studies, and treat trim-and-fill or sensitivity analyses as ways to ask how fragile the conclusion is. Remember that asymmetry has several possible causes — true small-study effects, differences in study quality, or genuine heterogeneity — so a tilted funnel is a prompt to investigate, not an automatic verdict of fraud. The point is to stress-test a conclusion, not to decorate it.

The traps are reading a funnel plot with too few studies and inventing patterns in noise, assuming a symmetric funnel proves there is no bias when it only fails to detect it, and treating asymmetry as automatic proof of publication bias when quality differences or real heterogeneity can produce the same shape. Analysts also err by choosing a precision measure that manufactures asymmetry, or by hiding the plot entirely so no one can check. The discipline is to draw the funnel plot on an adequate set of studies, pair the eyeball read with a formal test, consider innocent explanations for any tilt, and present it openly alongside the forest plot — so the pooled effect comes with an honest account of how complete and even-handed the studies behind it really are.

The name describes the inverted-funnel shape an unbiased set of studies traces when effect size is plotted against precision.

Etymology: source.

Search interest for this term over the last five years:

View interest-over-time on Google Trends →

Areas of marketing where funnel plot is a core concern: