Forest plot

A forest plot is the standard way a meta-analysis shows its work. Instead of handing you one pooled number, it shows every study that went into that number, laid out so you can see whether they actually agree with each other or merely average out.

Reading the plot

Each row is one study. A square marks that study's own estimate, and a horizontal line through it marks that study's confidence interval - wider lines mean less certainty, usually from fewer participants. The Cochrane Handbook for Systematic Reviews of Interventions (chapter 10), the standard methods reference, adds a detail most readers miss: "the area of the block indicates the weight assigned to that study in the meta-analysis." At the bottom, a diamond marks the pooled estimate across all studies, with the diamond's width representing the pooled confidence interval, normally narrower than any single study's line because pooling data narrows the estimate of the mean.

A schematic, simplified to three studies:

Study A   |------o------|
Study B      |--o--|
Study C   |----o------------|
Pooled         <>
          -----|----|----|----|-----
              12.5  13.0  13.5  14.0  cm

Study A and Study C give roughly the same estimate but with different precision; Study B sits slightly apart. The pooled diamond lands near the center of mass of all three, weighted more heavily toward the studies with narrower lines.

The weighting is not arbitrary. A study contributes to the pooled diamond in rough proportion to how precisely it estimated its own mean, which usually tracks its sample size - a study of a few hundred men earns a wider line and less influence on the diamond than a study of several thousand. That is a deliberate feature of the pooling method, not something to correct for as a reader; it is exactly why the diamond can sit closer to the large, tight studies than a simple unweighted average of all three lines would.

What to actually look for

The useful thing a forest plot does that a single pooled number cannot is let you see whether the individual studies cluster around the pooled estimate or scatter widely around it. If every study's line overlaps comfortably with the others, the pooled diamond is a fair summary of a genuinely consistent body of evidence. If the lines barely overlap, or some sit clearly apart from the rest, the pooled diamond is still the correct calculation, but it is averaging over disagreement rather than confirming consensus. Meta-analyses usually print a number for this beside the plot, I², which Cochrane describes as the percentage of variability "due to heterogeneity rather than sampling error," with 75-100% counted as considerable. What causes that disagreement between pooled studies is its own subject and is not covered here; this post is only about how to read the picture once it is drawn.

A second thing worth checking, once the overall pattern is clear, is whether any single study's line sits so far from the rest that it looks like it belongs to a different population entirely. A forest plot makes an outlier study visible in a way a table of numbers does not, because the eye catches a line that fails to overlap the others faster than it catches one number sitting a row apart from its neighbours in a column. That visibility is the main reason meta-analyses publish the plot rather than only the pooled figure and its interval.

It is also worth noticing the order the studies are listed in, since that is a choice the authors made rather than an inherent property of the data. Some plots order studies by publication date, which lets a reader see whether estimates have drifted over time as methods changed; others order by sample size or by the estimate itself, which makes clustering easier to spot but can obscure a chronological trend. Neither ordering is wrong, but knowing which one you are looking at changes what pattern you should expect to see at a glance.

Where you will encounter one

Veale et al. (2015), BJU International, pools measured studies the way any meta-analysis does, and a forest plot is the standard tool for showing how its contributing studies relate to each other and to the combined figure - how the paper's pooling worked in more detail sits in its own post. If you ever see a headline figure quoted from a meta-analysis without its forest plot, that is not automatically wrong, but it is withholding the one piece of the paper that shows whether the studies underneath the number actually agreed.

None of this applies to a photograph-derived score. There is no pooling of independent studies behind a rating from Rate Cock, a numeric score from a similar tool, a human judge's verdict, or an AI model's read of an image - a forest plot is not a tool any of those numbers would ever need or produce.