High I-squared values are one of the most common problems researchers face when synthesizing evidence. A standard forest plot tells you that heterogeneity is present but not which specific studies are driving it or whether those studies are also distorting your pooled estimate. That is where the Galbraith plot and the Baujat plot become essential.

These two diagnostic plots are available as dedicated tabs inside our free Forest Plot Generator. This guide explains exactly what each plot shows, how to read the output, and how to use the findings to make defensible decisions about your meta-analysis model.

Why Standard Forest Plots Miss the Source of Heterogeneity

A forest plot orders studies vertically and shows each one's confidence interval relative to the null. You can visually detect heterogeneity when the squares are spread widely, but the plot does not distinguish between two fundamentally different situations.

In the first situation, a small number of outlier studies sit far from the rest while the majority of studies cluster tightly. Removing or down-weighting these outliers would substantially reduce I-squared without changing the direction of your overall conclusion.

In the second situation, heterogeneity is diffuse and no single study is responsible. Here, subgroup analysis or meta-regression is the appropriate next step because there is no outlier to remove.

Acting on the wrong diagnosis wastes time and introduces bias. The Galbraith and Baujat plots give you the evidence to distinguish between these two situations before you decide how to proceed.

The Galbraith Plot (Radial Plot): Detecting Outliers by Precision

The Galbraith plot, also called a radial plot after its creator Robert Galbraith, was developed specifically to visualize heterogeneity in a way that separates imprecise studies from true outliers.

How it is constructed. Each study is plotted as a single point. The x-axis is the reciprocal of the study's standard error (1/SE), which represents precision: studies with larger samples and smaller standard errors appear further to the right. The y-axis is the study's z-score divided by its standard error (the standardized effect). A regression line is drawn through the origin with a slope equal to the pooled effect estimate. Two outer lines parallel to this regression line mark the boundaries at plus and minus two standard deviations.

What the position means. Studies that fall within the two outer lines are consistent with the pooled estimate. Studies that fall outside these lines are statistical outliers. Critically, because the x-axis encodes precision, you can immediately see whether the outliers are small imprecise studies (clustered near the origin) or large precise studies (far to the right). A large, precise study that is an outlier is far more concerning than a small imprecise one, because it carries more weight in the pooled estimate.

A worked interpretation example. Suppose your meta-analysis has I-squared of 72% and 14 studies. You open the Galbraith plot and see that 12 studies cluster tightly near the regression line while 2 studies sit outside the upper boundary, both positioned at mid-range precision. This pattern suggests localized heterogeneity. You examine those 2 studies in detail, find they used a different comparator intervention, and run a subgroup analysis separating them from the rest. I-squared in the main subgroup drops to 18%.

Common misreading to avoid. Do not treat every point outside the boundary lines as a study to exclude. The two-SD boundary is a statistical flag, not a clinical judgment. Always examine why a study is an outlier before deciding how to handle it.

The Baujat Plot: Separating Heterogeneity Contribution from Influence

The Baujat plot, developed by Baujat and colleagues in 2002, adds a second dimension that the Galbraith plot does not capture. While the Galbraith plot identifies outliers in terms of effect size, the Baujat plot simultaneously measures two distinct concepts: how much each study contributes to overall heterogeneity, and how much each study influences the pooled estimate.

How it is constructed. The x-axis shows each study's contribution to the overall Q statistic, which is the raw measure of heterogeneity. The y-axis shows the influence of each study on the pooled effect, measured as the squared difference between the overall pooled estimate and the pooled estimate calculated when that study is left out.

Quadrant interpretation. The Baujat plot naturally divides into four quadrants around the median values of each axis.

Studies in the upper-right quadrant contribute heavily to both heterogeneity and influence. These are your priority studies for investigation. They are making your I-squared high and they are moving your pooled estimate significantly. If any single study belongs in this quadrant, examine its methods, population, risk of bias, and publication context carefully.

Studies in the lower-right quadrant contribute to heterogeneity but do not strongly influence the pooled estimate. These studies have unusual effect sizes compared to the rest but carry less weight. Their presence inflates I-squared without necessarily biasing your summary estimate.

Studies in the upper-left quadrant are highly influential but not strong contributors to heterogeneity. These are often high-quality, precisely estimated studies that align with the pooled estimate while pulling it in a consistent direction. They deserve attention in a sensitivity analysis but are rarely problematic.

Studies in the lower-left quadrant are neither heterogeneous nor influential. These are well-behaved studies that contribute to precision without complicating interpretation.

Using Both Plots Together: A Diagnostic Workflow

Using the Galbraith and Baujat plots sequentially gives you a robust, two-step heterogeneity investigation that is easy to report in your methods section.

Step 1: Open both diagnostic tabs in the Forest Plot Generator after running your main analysis.

Step 2: In the Galbraith plot, identify any studies outside the two-SD boundary lines. Note their precision level (x-axis position) and their direction of deviation (above or below the regression line).

Step 3: In the Baujat plot, locate those same studies. Check which quadrant they fall in. If the outlier studies from the Galbraith plot cluster in the upper-right Baujat quadrant, you have a strong signal that a small subset of studies is responsible for both your elevated I-squared and any instability in your pooled estimate.

Step 4: Run a sensitivity analysis removing the identified studies one at a time. The Sensitivity Analysis Tool automates this process and flags which removals produce the largest changes in the pooled estimate and heterogeneity statistics.

Step 5: Document your decision. Never silently exclude a study identified through these plots. In your manuscript, state that Galbraith and Baujat diagnostics were used, identify which studies were flagged, explain the methodological reason for any exclusion or subgroup separation, and present both the full-set and the sensitivity results.

Reporting These Diagnostics in Your Manuscript

Many journals now expect authors to go beyond I-squared and tau-squared when reporting heterogeneity. Mentioning Galbraith and Baujat plots in your methods section, even briefly, signals methodological rigor.

A sample methods sentence: "Heterogeneity was assessed using I-squared and tau-squared statistics. Galbraith radial plots and Baujat plots were used to identify individual studies contributing disproportionately to heterogeneity and to assess their influence on the pooled estimate."

Include both plots as supplementary figures if the journal allows it. Reviewers who specialize in systematic reviews and meta-analysis will recognize and appreciate the thoroughness.

For publication bias assessment alongside these diagnostic plots, the Funnel Plot Generator provides Egger's test and trim-and-fill adjustment, completing the standard diagnostic battery.

Key Takeaways

FAQ

What is a Galbraith plot used for in meta-analysis?

A Galbraith plot, also called a radial plot, visualizes heterogeneity by plotting each study's standardized effect against its precision. Studies that fall outside the two outer boundary lines are statistical outliers contributing disproportionately to overall heterogeneity, making them candidates for further clinical and methodological investigation.

How is the Baujat plot different from the Galbraith plot?

The Galbraith plot identifies outliers in terms of effect size relative to precision. The Baujat plot measures two separate properties: each study's contribution to the Q heterogeneity statistic (x-axis) and each study's influence on the pooled estimate (y-axis). The Baujat plot's quadrant structure lets you distinguish studies that inflate I-squared from studies that actually shift your summary effect.

Should I exclude outlier studies identified by these plots?

Not automatically. A statistical outlier is a flag for investigation, not a deletion instruction. Examine the flagged study's population, intervention, risk of bias, and follow-up period. If there is a pre-registered or clinically justified reason for the difference, a sensitivity analysis or subgroup analysis is the appropriate response rather than silent exclusion.

What does the Q statistic contribution on the Baujat plot x-axis measure?

The x-axis on a Baujat plot shows how much each study adds to the overall Q statistic, which is the chi-squared test for heterogeneity. A high x-axis value means the study's effect estimate is far from the weighted mean, inflating the total Q and therefore the I-squared percentage.

Can I use these plots with only a few studies?

Yes, but interpret with caution. With 5 or fewer studies, the two-SD boundaries on the Galbraith plot have low reliability and the Baujat quadrant analysis has limited statistical power. Report the plots as exploratory tools rather than definitive diagnostics when your study count is small.

Do these plots replace funnel plots for publication bias?

No. Galbraith and Baujat plots diagnose heterogeneity sources. Funnel plots, along with Egger's test and trim-and-fill analysis, address small-study effects and potential publication bias. Both sets of diagnostics serve different purposes and should be used together for a complete meta-analysis quality check.

Need help with your systematic review or meta-analysis? Get a free quote from our team of PhD researchers.