A forest plot and a funnel plot are the two figures most readers expect in a meta-analysis, and they answer entirely different questions. The forest plot reports the result: it lays out every study's effect estimate and confidence interval alongside the pooled effect, so you can see how large the combined effect is and how consistent the studies are. The funnel plot interrogates the evidence base itself: it plots each study's effect size against its precision to ask whether small, unflattering studies are likely missing from the literature. One tells you what the answer is; the other tells you whether you should trust that the answer is complete.
What a forest plot is built to answer
The forest plot exists to communicate the combined effect and its consistency in a single column. Each study contributes a row, with a square marking its point estimate and a horizontal line marking its confidence interval. A summary diamond at the foot of the figure carries the weighted average across all studies, and its width shows how precise that average is. When you scan a forest plot, you are reading three things at once: the direction and size of the effect, how much each study weighs, and whether the studies agree well enough to be pooled in the first place.
Crucially, the forest plot says nothing about whether the studies in front of you are the studies that should exist. It faithfully summarises whatever you fed it. If half the small null trials on this topic were never published, the forest plot will still draw a tidy diamond, and that diamond may be biased. If you want a full walkthrough of every mark on the figure, our guide on reading a forest plot element by element covers the squares, the line of no effect, and the diamond in order.
What a funnel plot is built to detect
The funnel plot is a scatter plot, not a stacked column. It places each study's effect size on the horizontal axis and a measure of that study's precision, almost always its standard error, on the vertical axis. By convention the vertical axis is inverted, so the most precise large studies sit at the top and the imprecise small studies spread out across the bottom. The whole point is to surface small-study effects and publication bias: the systematic tendency for small studies with disappointing or null results to go unpublished, which quietly inflates a pooled estimate.
Because precision improves with sample size, the large studies cluster tightly around the true effect while the small studies scatter widely. In an unbiased evidence base, that scatter is symmetric, and the cloud of points forms the inverted triangle that gives the plot its name.
How to read a funnel plot
Start with the overall shape. A complete, unbiased set of studies produces a roughly symmetric funnel: precise studies at the narrow top, imprecise studies fanning out evenly to the left and right at the wide base. The features worth checking, in order, are these.
- Symmetry around the pooled estimate: draw a vertical line at the pooled effect and ask whether the small studies are balanced on both sides of it. A gap on one side is the classic fingerprint of missing studies.
- The inverted funnel shape: confirm that spread genuinely widens as precision falls. If the small studies do not scatter more than the large ones, the plot is not behaving as a funnel and asymmetry judgements become unreliable.
- Egger's test: this regression-based test puts a number on the visual impression by testing whether small studies report systematically different effects. A small p-value flags funnel plot asymmetry that is unlikely to be chance.
- Contour-enhanced funnel plots: these overlay shaded bands of statistical significance onto the funnel. If the gap of missing studies falls in a region of non-significance, publication bias is the likely cause; if missing studies fall in significant regions, the asymmetry probably comes from something other than suppression.
The contour-enhanced version is the single most useful upgrade to a plain funnel, because it separates "studies are missing because they were null" from "studies are missing for some other reason," and those two stories have very different implications for your conclusion.
Where funnel plots mislead, and their limits
A funnel plot is a screening device, not a verdict, and it is easy to over-read. Its most important limitation is the number of studies. With fewer than roughly ten studies, the plot has too few points to judge symmetry reliably, and Egger's test loses power, so any claim of bias from a sparse plot is speculation dressed as evidence. Most guidance recommends ten or more studies before you formally test for asymmetry at all.
The deeper trap is interpreting asymmetry as proof of publication bias when funnel plot asymmetry has several causes. Genuine heterogeneity between studies can bend the cloud out of symmetry, as can true small-study effects driven by differences in trial conduct, dose, or population rather than by suppression. Poor methodological quality concentrated in the smaller studies can produce the same pattern. This is exactly why a high I-squared value should change how you read a funnel: if the studies are very heterogeneous, asymmetry may reflect real between-study variation, not a missing literature. Our explainer on interpreting the I-squared heterogeneity statistic shows how to weigh that variation before you blame the funnel.
Treat the funnel plot, then, as one line of evidence about completeness, to be read together with the registered protocol, the search strategy, and the clinical plausibility of a missing-studies story. A single asymmetric plot is a prompt to investigate, never a conclusion on its own.
How the two plots work together
In a complete meta-analysis the two figures are partners, and the order of reading matters. First you build and read the forest plot to establish the result: the direction and size of the pooled effect and how consistent the studies are. Then you build the funnel plot to stress-test that result, asking whether the evidence base behind the diamond is likely to be complete or whether small negative studies appear to be missing. A clean, symmetric funnel lets you report the forest plot diamond with confidence; an asymmetric funnel means you should qualify the same diamond, or run a sensitivity analysis, before you quote it.
This is why the strongest results sections present both. The forest plot answers "what is the combined effect, and how consistent is it," and the funnel plot answers "is the evidence base likely missing small negative studies." Reporting only the first leaves a reviewer wondering whether the second was checked. If you need the underlying combined estimate first, our pooled effect size calculator will produce the weighted average and confidence interval that both figures depend on.
Build both for your own data
The practical takeaway is simple: do not stop at the result figure. Once you have extracted your study estimates, you can generate a publication-ready forest plot to read the combined effect, then produce the matching funnel plot to assess publication bias and check whether your evidence base looks complete. The two figures together turn a pooled number into a defensible claim, which is precisely what a well-conducted systematic review and meta-analysis is expected to deliver.