Forest Plot Generator for Meta-Analysis
Create a forest plot from your study results
Events & totals exampleUse this forest plot generator to display the results of a meta-analysis online. Enter summary data from at least two independent studies in the table above. The calculator produces a forest plot with study estimates, 95% confidence intervals, and a pooled effect for your selected model. Each row represents one study estimate.
- Choose the format of your study data and the effect measure.
- Choose what the plot shows, such as the weights, and add direction labels.
- Read the forest plot: each study with its 95% confidence interval, the pooled effect as a diamond, and the prediction interval below it.
How to make your forest plot
- Choose your data format and effect measure. Under Effect sizes, select Standard error or 95% CI to match your reported results.
- Enter, paste or import study data in the table above, then assign its columns to the required inputs.
- Choose fixed-effect or random-effects pooling based on your research question and study designs.
- The Forest plot appears below the pooled results. Changes to the data or model update it automatically.
Which data can I use?
| Data available per study | Select | Possible effect measures |
|---|---|---|
| Events and total participants in each of two groups | Events & totals | Risk ratio, odds ratio, risk difference |
| Sample size, mean, and SD in each of two independent groups | Means & SDs | Mean difference or Hedges' g |
| Pearson correlation and sample size | Correlations | Correlation r |
| Effect estimate and its standard error or 95% confidence interval | Effect sizes | A common effect scale, or log OR, log RR, log HR |
For precomputed odds, risk, or hazard ratios, enter the natural logarithm of the ratio and the standard error of that log estimate. The forest plot displays the results as ratios on a logarithmic axis. With 95% CI selected, enter each estimate with its confidence limits instead (log limits for ratios). Use normal-approximation intervals that are symmetric on the analysis scale. A standard deviation is not a substitute for a standard error.
What does the generated plot show?
- Study markers: each marker locates a study's estimated effect. Marker size reflects its weight in the selected model.
- Horizontal lines: the lines show 95% confidence intervals. The Study results table lists the same estimates and limits.
- Pooled diamond marker: the diamond locates the pooled effect; its horizontal error bar shows the pooled 95% confidence interval.
- Prediction interval: for random effects, a dotted line in the row below the diamond shows the 95% prediction interval for the true effect of a new, comparable study.
- Null line: the dashed line marks 1 for ratios and 0 for differences or correlations.
- Not estimable: for odds and risk ratios, a study with no events, or only events, in both groups is listed but not pooled.
In the plot options, you can show the estimates with their 95% CI beside the plot, a weight column, and a summary line with heterogeneity (τ², I², Q) and the test for the overall effect. Studies can be ordered as entered, by effect, or by weight. Under Direction labels, add text such as "Favours treatment" on either side of the null line.
An interval crossing the null line includes the no-effect value. Consider the effect's size, direction, and uncertainty together. Which direction favors a group depends on the outcome and group order; the plot cannot determine that research context for you.
Models and practical limits
Fixed-effect pooling uses inverse sampling-variance weights. Random-effects pooling adds an estimate of between-study variance, REML by default or DerSimonian-Laird, to each sampling variance before calculating weights. Fixed-effect confidence intervals use a normal approximation on the analysis scale; random-effects intervals use a t distribution with k - 1 degrees of freedom when the Hartung-Knapp adjustment is on (the default), where k is the number of included studies. Turn it off for normal-approximation pooled intervals. The Hartung–Knapp adjustment control appears when Random effects is selected. A pooled confidence interval describes uncertainty in the average effect; the prediction interval describes variation in true effects and assumes normality on the analysis scale. Prediction intervals can be unreliable with few studies. The calculator does not produce subgroup or network meta-analysis plots.
Use comparable outcomes, consistent effect directions, and independent study estimates. Use the table filter to exclude a study without deleting its values. Use export / import above the table to download or reopen your input data.
To examine differences between studies, use the heterogeneity calculator. To plot effects against standard errors, use the funnel plot generator. The meta-analysis calculator guide explains all input formats and calculation methods. For background on the plotted intervals, see the confidence interval tutorial.
Method reference: Cochrane Handbook, Chapter 10 describes inverse-variance pooling and forest plots.