Meta-analysis software: how to choose between R, RevMan, Stata and the rest
The software does the arithmetic of a meta-analysis: it converts each study's result to a common effect size, weights the studies, pools them, measures heterogeneity and draws the forest and funnel plots. Several programs do this, and the right one depends on your skills, your budget and how unusual your analysis is.
This guide compares the common options and shows how to choose. Whatever you pick, the numbers you put in matter more than the program, and the program and its version must be stated in your methods.
What the software must do
- Effect measures: the ones that fit your outcome type, such as odds ratio, risk ratio, mean difference and standardised mean difference.
- Models: fixed-effect and random-effects pooling, with a choice of estimator for the variance between studies.
- Heterogeneity: tau-squared, I-squared and, for a random-effects model, a prediction interval.
- Exploration: subgroup analysis, meta-regression and leave-one-out sensitivity analysis.
- Plots: forest plots and funnel plots, with tests for small-study effects when there are enough studies.
- Repeatability: export of the results and the data, so the analysis can be run again.
The main options
Prices and licence terms change, so check each program's own site before you decide.
| Software | Type and cost | Good for | Watch out for |
|---|---|---|---|
| RevMan (Cochrane) | Point-and-click; built for Cochrane reviews. Check Cochrane's current licence terms. | Standard pooling, forest and funnel plots and risk of bias figures, in the format Cochrane reviews use. | Fewer advanced methods than R or Stata. |
| R (metafor, meta) | Free and open source; you write code. | The widest range of models and plots, and a script that makes the analysis repeatable. | A learning curve if you have not used R. |
| Stata (meta commands) | Paid; you type commands. | Teams that already use Stata; its built-in meta commands cover pooling, meta-regression and plots. | The licence cost. |
| JASP and jamovi | Free; point-and-click, built on R. | Learning, teaching and simple analyses. | Fewer options than R itself. |
| Comprehensive Meta-Analysis | Commercial; point-and-click. | Users who want a spreadsheet-style interface and the common models. | The licence cost, and a less transparent record than a script. |
| Excel-based tools | Free workbooks and add-ins, such as Meta-Essentials. | Small analyses, and checking a result from another program. | Formula errors are easy to make, and advanced methods are missing. |
Let SynthGap run the first pass
Enter an area. SynthGap proposes candidate topics, checks each against existing systematic reviews, and lists the primary studies behind it. Start on the Free plan: no card needed, and every run shows every topic it finds.
Sign up freeOther options
Python's statsmodels library has basic meta-analysis functions. Specialist programs and R packages exist for network meta-analysis and for diagnostic test accuracy. For an unusual design, such as a multilevel model, ask a statistician which tool fits.
How to choose
- A Cochrane review: use RevMan, because it is the format the review is published in.
- You want control and repeatability: use R, with a script you save.
- Your group already uses Stata: use its meta commands.
- You are learning or teaching: start with JASP or jamovi.
- An unusual design, such as a network meta-analysis or diagnostic accuracy: ask a statistician. R packages cover most of them.
What to report, and where AI fits
PRISMA 2020, item 13d, asks for the model, how you assessed heterogeneity and the software package used. Give the program, its version, the packages and the estimator.
An AI assistant can explain output, suggest a model and draft code, but the pooled result should come from the program, not from the assistant. See AI tools for meta-analysis.
Sources
The figures and rules on this page come from these sources.
- Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71.
Questions
Which software is easiest for a beginner?
Point-and-click programs such as JASP, jamovi or RevMan are the gentlest start. Move to R when you need more methods or want a script you can run again.
Is R necessary for a meta-analysis?
No. Many published meta-analyses used RevMan, Stata or Comprehensive Meta-Analysis. R is the most flexible and it is free, which is why it is common in methods papers.
Can Excel do a meta-analysis?
Yes for simple cases, with a purpose-built workbook, but formulas are easy to get wrong and advanced methods are missing. Use it to check a result rather than as your main tool.
Does the software choice change the result?
It can. Programs differ in their default estimator for the variance between studies and in how they calculate confidence intervals, so the same data can give slightly different answers. State the settings you used, and keep them the same across analyses.
Keep reading
General methodological guidance, not medical advice. SynthGap's suggestions are AI-generated from the published literature, so verify them before registering a protocol.